Writing and Cognition, Volume 4: Research and Applications (Studies in Writing) (Studies in Writing) (Studies in Writing)

Else_SW-Torrance_PRELIMS.qxd 9/29/2006 9:25 AM Page i WRITING AND COGNITION: RESEARCH AND APPLICATIONS Else_SW-To...

Author: Mark Torrance | Luuk van Waes | David W. Galbraith

162 downloads 3312 Views 4MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Else_SW-Torrance_PRELIMS.qxd

9/29/2006

9:25 AM

Page i

WRITING AND COGNITION: RESEARCH AND APPLICATIONS


9/29/2006

9:25 AM

Page ii

STUDIES IN WRITING Series Editor: Gert Rijlaarsdam

Recent Titles in this Series: Teaching Writing in Chinese Speaking Areas Shum and Zhang Writing in Context(s) – Textual Practices and Learning Processes in Sociocultural Settings Kostouli Effective Learning and Teaching of Writing Rijlaarsdam, Van Den Bergh and Couzijn

Related Titles: Writing Hypertext and Learning: Conceptual and Empirical Approaches Bromme and Stahl Powerful Learning Environments: Unravelling Basic Components and Dimensions De Corte, Vershaffel, Entwistle and Merriënboer

Related Journals: Learning and Instruction Educational Research Review Assessing Writing Computers and Composition Journal of Second Language Writing


9/29/2006

9:25 AM

Page iii

WRITING AND COGNITION: RESEARCH AND APPLICATIONS EDITED BY

MARK TORRANCE Staffordshire University, UK

LUUK VAN WAES University of Antwerp, Belgium

DAVID GALBRAITH Staffordshire University, UK

Amsterdam ● Boston ● Heidelberg ● London ● New York ● Oxford Paris ● San Diego ● San Francisco ● Singapore ● Sydney ● Tokyo


9/29/2006

9:25 AM

Page iv

Elsevier The Boulevard, Langford Lane, Kidlington, Oxford 0X5 1GB, UK Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands First edition 2007 Copyright © 2007 Elsevier Ltd. All rights reserved No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333; email: [email protected]. Alternatively you can submit your request online by visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting Obtaining permission to use Elsevier material Notice No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress ISBN-13: 978-0-08-045094-0 ISBN-10: 0-08-045094-6 ISSN: 1572-6304

For information on all Elsevier publications visit our website at books.elsevier.com

Printed and bound in The Netherlands 07 08 09 10 11 10 9 8 7 6 5 4 3 2 1

Else_SW-Torrance_CONTENTS.qxd

10/24/2006

4:54 PM

Page v

Contents

Contributors

ix

Chapter 1. Introduction David Galbraith, Luuk van Waes and Mark Torrance

1

Section 1. Interactions among Writing Processes

11

Chapter 2. Parallel Processing Before and After Pauses: A Combined Analysis of Graphomotor and Eye Movements During Procedural Text Production 13 Denis Alamargot, Christophe Dansac, David Chesnet and Michel Fayol Chapter 3.

From Written Word to Written Sentence Production Guido Nottbusch, Rüdiger Weingarten and Said Sahel

Chapter 4.

Influence of Typing Skill on Pause-Execution Cycles in Written Composition Rui Alexandre Alves, São Luís Castro, Liliana de Sousa and Sven Strömqvist

31

55

Chapter 5. The Word-Level Focus in Text Production by Adults with Reading and Writing Difficulties Åsa Wengelin

67

Chapter 6. GIS for Writing: Applying Geographical Information Systems Techniques to Data Mine Writings’ Cognitive Processes Eva Lindgren, Kirk P. H. Sullivan, Urban Lindgren and Kristyan Spelman Miller

83

Chapter 7. Verbal and Visual Working Memory in Written Sentence Production Ronald T. Kellogg, Thierry Olive and Annie Piolat

97

Chapter 8. Effects of Note-Taking and Working-Memory Span on Cognitive Effort and Recall Performance Annie Piolat

109


vi

10/24/2006

4:54 PM

Page vi

Contents

Chapter 9. The Dynamics of Idea Generation during Writing: An Online Study Huub van den Bergh and Gert Rijlaarsdam

125

Chapter 10. Skilled Writers’ Generating Strategies in L1 and L2: An Exploratory Study Sophie Beare and Johanne S. Bourdages

151

Section 2. Effects of Writing on Cognition

163

Chapter 11. The Writing Superiority Effect in the Verbal Recall of Knowledge: Sources and Determinants Joachim Grabowski

165

Chapter 12. The Effect of Writing on Phonological Awareness in Spanish Sofia A. Vernon

181

Chapter 13. Developmental Trends in a Writing to Learn Task Perry D. Klein, Jennifer S. Boman and Melanie P. Prince

201

Chapter 14. Approaches to Writing Ellen Lavelle

219

Chapter 15. Cognitive Processes in Discourse Synthesis: The Case of Intertextual Processing Strategies Rachel Segev-Miller

231

Chapter 16. Preformulation in Press Releases: What the Writing Process Tells us about Product Characteristics Kim Sleurs

251

Section 3. Writing Media

265

Chapter 17. Talking to Write: Investigating the Practical Impact and Theoretical Implications of Speech Recognition (SR) Software on Real Writing Tasks Noel Williams, Peter Hartley and Vanessa Pittard Chapter 18. How do Writers Adapt to Speech Recognition Software? The Influence of Learning Styles on Writing Processes in Speech Technology Environments Mariëlle Leijten

267

279


10/24/2006

4:54 PM

Page vii

Contents Chapter 19. Longitudinal Studies of the Effects of New Technologies on Writing: Two Case Studies James Hartley Chapter 20. Learning by Hypertext Writing: Effects of Considering a Single Audience Versus Multiple Audiences on Knowledge Acquisition Elmar Stahl, Rainer Bromme, Marc Stadtler and Rafael Jaron

vii

293

307

Chapter 21. Supporting Individual Views and Mutual Awareness in a Collaborative Writing Task: The Case of Col•laboració Henrry Rodriguez and Kerstin Severinson Eklundh

323

References

335

Author Index

365

Subject Index

375

List of Volumes

379

This page intentionally left blank

Else_SW-Torrance_LOC.qxd

10/6/2006

1:47 PM

Page ix

Contributors

Denis Alamargot Laboratory LMDC-CNRS, University of Poitiers, Poitiers, France, [email protected] Rui Alexandre Alves Faculdade de Psicologia e de Ciências da Educação, Universidade do Porto, Porto, Portugal, [email protected] Sophie Beare Teachers of English as a Second/Foreign Language Programme, Algonquin College of Applied Arts and Technology, Ottawa, Canada, [email protected] Jennifer S. Boman Department of Psychology, University of Western Ontario, Ontario, Canada, [email protected] Johanne S. Bourdages Faculty of Education, University of Ottawa, Ottawa, Canada, [email protected] Rainer Bromme Psychological Institute III, University of Muenster, Muenster, Germany, [email protected] São Luís Castro Faculdade de Psicologia e de Ciências da Educação, Universidade do Porto, Porto, Portugal, [email protected] David Chesnet MSHS-CNRS, University of Poitiers, Poitiers, France, [email protected] Christophe Dansac Laboratory LTC-CNRS, University of Toulouse, Toulouse, France, [email protected] Michel Fayol LAPSCO-CNRS UMR 6024, Laboratoire de Psychologie, Sociale et Cognitive, Université Blaise Pascal, Clermont-Ferrand, France, [email protected]


x

10/6/2006

1:47 PM

Page x

Contributors

David Galbraith Centre for Educational Psychology Research, Staffordshire University, Stoke-on-Trent, UK, [email protected] Joachim Grabowski University of Education, Heidelberg, Germany, [email protected] James Hartley Teaching Quality Enhancement Group , University of Bradford, Bradford, UK. [email protected] Peter Hartley School of Psychology, Keele University, Staffordshire, UK, [email protected] Rafael Jaron Psychonomics, Cologne, Germany. Rafael, [email protected] Ronald T. Kellogg Department of Psychology, Saint Louis University, St. Louis, USA, [email protected] Perry D. Klein Faculty of Education, University of Western Ontario, Ontario, Canada, [email protected] Ellen Lavelle Excellence in Teaching and Learning Initiative, Southern Illinois University Edwardsville, USA, [email protected] Mariëlle Leijten Faculty of Applied Economics, Department of Management, University of Antwerp, Antwerp, Belgium, [email protected] Eva Lindgren Faculty of Teacher Education, Umeå University, Umeå, Sweden, [email protected] Urban Lindgren Department of Social and Economic Geography, Umeå University, Umeå, Sweden, [email protected] Kristyan Spelman Miller Department of Applied Linguistics, The University of Reading, Reading, UK, [email protected] Guido Nottbusch Faculty for Linguistics and Literature, University of Bielefeld, Germany, [email protected]


10/6/2006

1:47 PM

Page xi

Contributors

xi

Thierry Olive Centre National de la Recherche Scientifique & Université de Poitiers, Poitiers, France, [email protected] Annie Piolat Department of Psychology, Centre for Research in Psychology of Cognition, Language and Emotion, Université de Provence, Aix-en-Provence, France, [email protected] Vanessa Pittard British Educational Communications and Technology Agency, Coventry, UK, [email protected] Melanie P. Prince Department of Psychology, University of Western Ontario, Ontario, Canada Gert Rijlaarsdam Graduate School of Teaching and Learning, University of Amsterdam, The Netherlands, [email protected] Henrry Rodriguez Human-Computer Interaction group, School of Computer Science and Communication, Royal Institute of Technology, Stockholm, Sweden, [email protected] Said Sahel Faculty for Linguistics and Literature, University of Bielefeld, Germany, [email protected] Rachel Segev-Miller The Center of Academic Literacy, Kibbutzim College of Education, Tel-Aviv, Israel, [email protected] Kerstin Severinson Eklundh Human-Computer Interaction group, School of Computer Science and Communication, Royal Institute of Technology, Stockholm, [email protected] Kim Sleurs Department of International Business Communication, University of Antwerp, Antwerp, Belgium, [email protected] Liliana de Sousa Instituto de Ciências Biomédicas Abel Salazar, Universidade do Porto, Porto, Portugal, [email protected]


xii

10/6/2006

1:47 PM

Page xii

Contributors

Marc Stadtler Psychological Institute III, University of Muenster, Muenster, Germany, [email protected] Elmar Stahl Psychological Institute III, University of Muenster, Muenster, Germany, [email protected] Sven Strömqvist Department of Linguistics and Phonetics, University of Lund, Lund, Sweden, [email protected] Kirk P.H. Sullivan Department of Philosophy and Linguistics, Umeå University, Umeå, Sweden, [email protected] Mark Torrance Centre for Educational Psychology Research, Staffordshire University, Stoke-on-Trent, UK, [email protected] Huub van den Bergh Utrecht Institute of Linguistics, Utrecht University, The Netherlands, [email protected] Luuk Van Waes Department of Management - Faculty of Applied Economics, University of Antwerp, Belgium, [email protected] Sofia A. Vernon Faculty of Psychology, Universidad Autonoma de Queretaro, Queretaro, Mexico, [email protected] Rüdiger Weingarten Faculty for Linguistics and Literature, University of Bielefeld, Germany, [email protected] Åsa Wengelin Centre for Languages and Literature / Linguistics, Lund University, Lund, Sweden, [email protected] Noel Williams School of Cultural Studies, Sheffield Hallam University, Sheffield, UK, [email protected]

Else_SW-Torrance_ch001.qxd

9/27/2006

1:19 PM

Page 1

Chapter 1

Introduction David Galbraith, Luuk van Waes and Mark Torrance

1.1. Writing Research on writing tends to treat writing as if it referred to a single, homogenous activity (as we have just done in this sentence). Looking at the books on the shelves in the office where we are writing this, we can see: “The Science of Writing”, “The Psychology of Writing”, “Cognitive Processes in Writing”, “Dynamics of Writing”, “Writing without Teachers”, “How We Write”, “When a Writer Can’t Write”, “The Psychology of Written Composition”, and “Knowing What to Write”. All of these titles imply a single object of study, about which it is possible to make a set of general claims. But does it make sense to talk about “writing” as a single, homogenous activity, supported by a particular set of cognitive processes? Or is writing, instead, a heterogeneous set of practices linked only by the fact that the final product is in a written form? If there are some general features common to all forms of writing, how important are these relative to the specific features associated with writing in particular contexts? The most obvious candidate for a universal feature is the fact that writing is a visual representation of spoken language. Whatever specific form of writing a writer is engaged in, they will have to be able to transcribe language according to the conventions of a particular writing system, and be able to decode visual symbols into their corresponding meanings. Learning how to do this is not a trivial achievement: It is a main focus of early years education and, in modern literate cultures, is the subject of huge investments of economic and social resources. It is also something which, unlike speech, significant number of people across the world fail to master, either for socio-cultural reasons or because of cognitive difficulties like dyslexia. But this huge social and personal investment does not necessarily mean that the cognitive skills involved are intrinsically valuable. In the brave new world predicted by William Crossman (2004), in which speech recognition/synthesis technology has advanced sufficiently for machines to take over the process of transcription, writing in this sense will ultimately disappear. We will all instead interact with thought through speech, just as nature intended. Once we have, as a society, overcome the primitive Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Galbraith, D., van Waes, L., & Torrance, M. (2007). Introduction. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 1–10). Amsterdam: Elsevier.


2

9/27/2006

1:19 PM

Page 2

David Galbraith et al.

requirement for individuals to do this individually, and delegated it instead to machines, people across the world will finally have equal access to a vast storehouse of cultural knowledge, and societies will be able to switch the social and economic resources invested in learning to read and write to other more pressing needs. According to this prediction, then, writing is a relatively peripheral activity, and the skills involved have little value in themselves. Another view, expressed by Dan Sperber in a target article contributed to a virtual symposium “investigating the impact of the Web on reading, writing and the diffusion of knowledge” (Sperber, 2003) is that writing, or more precisely reading, has sufficient cognitive benefits for it to persist despite such advances in technology. He argues that although the manual process of producing text may well wither away (much in the way that word-processing has replaced typing), reading will remain, not the least because the opportunity to re-read what one has written enables the writer to produce more sophisticated text. It is, in his opinion, the opportunity for “simultaneous reading of what you write” that makes “the process of writing uniquely valuable”. Similarly, although textto-speech software makes it feasible for people to listen to text rather than to read it, he argues that the opportunity text provides for re-scanning, and the control it provides for the reader over pacing, make it unlikely that, particularly for complex texts, reading will be abandoned altogether. The accuracy of such predictions depends on precisely what the cognitive processes involved in writing are, and on the relative balance between the costs involved in the learning and teaching of writing and the benefits of writing as a distinctive medium for thought and communication. It is also worth noting that the benefits that Sperber attributes to being able to re-read what one has written, and to the visual scanning of text during reading, are only potential benefits. They depend on whether the writer does actually exploit the opportunities afforded by the visual form of writing. Studies of children, at least, suggest that this is not something that they do spontaneously (Bereiter & Scardamalia, 1987). Furthermore, there is relatively little consistent evidence that removing visual feedback has effects on higher level features of writing, even among older writers (see, e.g. Olive & Piolat, 2002), though this may reflect the relative lack of research on the issue rather than the final word on the topic. A second candidate for a universal feature of writing is the fact that it involves the creation of an autonomous cognitive object, which can be stored and accessed independently of its creator. In this view, the defining feature of writing is the fact that it is not face-toface conversation. It includes not just the basic processes involved in transcribing speech into a visual form but also the higher mental processes involved in creating a permanent and extended text, which is adapted to an absent reader’s needs and which satisfies the writer’s communicative goals. Such broader features of writing as a process would remain even if the process of transcription were automated. Bereiter and Scardamalia (1987) argue that the shift from face-to-face conversation (in which the speaker can rely on an interlocutor to request elaboration and clarification) to writing (in which the writer has to provide their own prompts) is fundamental to learning to write. Furthermore, the processes involved in doing this often require the writer to re-evaluate and modify their existing knowledge. To use Bereiter and Scardamalia’s terms, writing becomes a knowledge-transforming activity rather than a knowledge-telling activity. Others (e.g. Olson, 1994) focus


9/27/2006

1:19 PM

Page 3

Introduction

3

less on the processes involved in creating these objects and more on the effects of the objects themselves and the cultural practices with which they are associated: Our literature, our science, our philosophy, our law, our religion, are, in an important way, literate artefacts. We see ourselves, our ideas and our world in terms of these artefacts…. The topic of literacy is all about the special, indeed peculiar properties of these artefacts … and about the kinds of competence, the forms of thought and the modes of perception that are involved in coping with, indeed exploiting, this world on paper (Olson, 1994, pp. xii–xiv). Writing, then, is not just a speech written down. It is an object designed to be understood when its creator is no longer present, and in terms of other objects produced in the same circumstances. Learning how to write involves learning how to create such objects, and the effects that writing has on cognition depend on the particular processes involved in creating them. Of course, such effects are, like the opportunity afforded by writing for re-reading, only a potential effect of writing. It depends, as we have just indicated, not only on the writer’s capabilities and their conception of writing as a task, but also on the state of their existing knowledge in the context of the task at hand. Often, the writer may be able to rely on previously produced text having the requisite form, and use their knowledge of these texts to summarise or otherwise paraphrase a pre-existing knowledge-object, without themselves having to engage in the processes required to produce it from first principles. For example, a student might produce a scientific report by mimicking the forms they have encountered in their reading, rather than by deducing the features required from their knowledge of the scientific method and of the knowledge and opinions of their readers. Also, the nature of this received knowledge will vary greatly depending on the type of writing involved: fiction or fact, art or science, poetry or prose, as well as the range of forms associated with specific academic disciplines. Nevertheless, we would argue that a fundamental common factor underlying different forms of discourse is the fact that the products of writing exist as independent objects separate from the immediate context of creation, and that this would remain so even if they were stored in electronic form, and produced and comprehended via speech.

1.2. Cognition One of the notable features of writing research to date is that it has been primarily concerned with the higher level thinking processes, and has paid relatively little attention to the basic processes involved in translating thought into visual form. In part this appears to be because of an implicit assumption that language processing is essentially a matter of translating content between an internal medium of thought and an external medium of communication, and that this is unaffected by whether language is spoken or written. The natural implication being that understanding writing involves understanding the processes involved in producing and evaluating thoughts rather than the processes involved in translating these thoughts into language.


4

9/27/2006

1:19 PM

Page 4


This assumption is most obvious in Hayes and Flower’s original model of the cognitive processes involved in writing (Hayes & Flower, 1980), which has had a huge influence on the development of the field. This posited three basic types of processes in writing: (i) planning, which involved setting goals, and generating and organising ideas to satisfy those goals; (ii) translating, which involved translating ideas into words; and (iii) reviewing, which involved reading and editing the already produced text. Bereiter and Scardamalia’s (1987) similarly influential model focussed on differences between expert and novice approaches of writing, and discussed these essentially in terms of differences in the goals towards which writing was directed. Thus, novice writing was characterised in terms of two models. The knowledge-telling model involved associative retrieval of content from memory which was then translated into words. This was contrasted with the knowledge-transforming model employed by experts in which content retrieval was guided by the writer’s communicative goals and by evaluation of the extent to which the text produced so far satisfied those goals. Both models shared a common language processing module which was simply responsible for translating the output of thought into words. The broad conclusion from research influenced by these models was that the essential difference between expert and novice writers was in the complexity of the thinking behind the text. Experts were typically found to develop a more elaborate representation of their goals, which enabled them to create more elaborate plans, and to evaluate and modify these throughout the course of writing. In addition, this more elaborate conceptual representation of goals for the text enabled them to revise more extensively, evaluating their text in terms of its underlying function with respect to their goals, rather than simply considering whether the text was appropriately expressed (see Hayes and Flower, 1986; Bereiter and Scardamalia, 1987, for reviews). Note, incidentally, that in identifying more elaborate revision as a characteristic of writing expertise, these findings support the idea that the availability of the text in visual form plays an important role in writing, and suggest that, even if appropriate speech technology becomes available, it may still be necessary, as Sperber suggests, to provide a written transcript of the text. Although these models do seem to capture the more deliberate components of the writing process, and interventions inspired by them (as well as Kellogg’s (1994) research on drafting strategies) do appear to lead to substantial improvements in the quality of children’s writing (e.g. De la Paz & Graham, 2002; Torrance, Fidalgo, & García, in press), there is a growing sense that they only provide a partial picture of the writing process. In particular, they appear to overemphasize the role of the more explicit thinking processes involved at the expense of more implicit processes closely linked to text production itself. They have relatively little to say about how the products of thought are instantiated, moment by moment, in the text, or about how thinking interacts with the externally represented text as it is produced. Perhaps in recognition of this, Hayes’s more recent revision of his original model (Hayes, 1996) appears to relax the distinction between “thinking” and “translation” processes. In the new model, the planning component of the original model has been replaced by reflection, which, though it still includes planning, also encompasses other kinds of inferential processing and less goal-directed forms of thinking. The translation component is now characterised as text production, implying a less purely linguistic process, and perhaps encompassing content generation closely tied to the formulation of thought in language.


9/27/2006

1:19 PM

Page 5

Introduction

5

The reviewing component has been abandoned altogether. Instead, the equivalent of the reading sub-component of the reviewing process is characterised as text interpretation, implying that this is not simply a matter of decoding text into meaning, but also includes automatic inferences. Revision is now seen not as distinct type of process on its own but as a complex process made up of text interpretation, reflection and text production. In a previous volume that two of us edited (Galbraith & Torrance, 1999) we argued that the research it reported reflected a shift within the field away from models which focussed exclusively on a view of writing as problem solving towards a view of writing as text production. This involves viewing text production not just as a passive translation of content determined by higher level thinking processes but as playing an active role in the generation of content itself. Galbraith (1999) has characterised this as a knowledge-constituting process, operating according to different principles from those involved in knowledge transforming, and optimised under different conditions. In some respects this is analogous to Kintsch’s (1998) claim that many of the processes involved in text comprehension, which had previously been attributed to explicit problem solving, can be accounted for in terms of a more implicit constraint satisfaction process. For present purposes, the main implication is that writing is not a one-way process of planning followed by translation but a two-way interaction between reflection and text production processes taking place both within and across drafts. In order to capture this interaction as it unfolds, we need to develop more dynamic models of writing of the kind described by van den Bergh and Rijlaarsdam (1999 and this volume), and to be able to examine the interaction between processes as they occur on-line during writing (Dansac & Alamargot, 1999). We also need detailed models of how different processes are coordinated in working memory (see Kellogg, 1996, for an early model; Torrance & Galbraith, 2006, for a recent review of progress in this area).

1.3. This Volume We see this volume as contributing to the further development of a more dynamic model of writing. This builds on the early research that focussed on reflective processes, and the initial research on text production described in the previous book that we edited, and provides a more detailed picture of the processes involved as they take place on-line. 1.3.1. Interactions among Writing Processes The first part of the book focusses on interactions between processes during writing, moving from an initial emphasis on the basic processes involved in text production through to more reflective processes and their interaction with these basic processes. In the first paper, Alamargot, Dansac and Chesnet describe a recent study using the Eye and Pen system and examining how text production interacts with eye movements at different points in writing. They find important differences in the way these processes are coordinated depending on the working memory capacity of the writers. In demonstrating the intricate interactions between eye and pen movements that take place during writing, they cast doubt on the easy assumptions about the equivalence of speech and writing made


6

9/27/2006

1:19 PM

Page 6


by proponents of speech-input word processing. This is followed by a paper by Nottbusch, Weingarten and Sahel investigating the extent to which cognitive processes intervene during the motor execution of written words and sentences. Their finding that in writing, unlike speech, cognitive processes do recur during motor execution, suggests again that assumptions about the equivalence of speech and writing may be premature. The next two papers focus on the effects of lower level processes on the writer’s ability to carry out higher level processes. Alves, de Sousa and Strömqvist use keystroke analysis to examine the effect of motor execution skill on higher level processing. They find that slow typing interferes with text generation, making it less fluent, with more and longer pausing between episodes of text production. They suggest that this reflects a move away from the kinds of parallel processing described by Alamargot, Dansac and Chesnet towards more serial processing, and that this may inhibit the writer’s ability to retrieve content (cf. Grabowski, this volume). In the following paper, Wengelin describes research investigating whether dyslexics’ difficulties with spelling have effects on other aspects of their writing. She finds that their processing difficulties are reflected in less diversity in vocabulary when they write compared to when they speak, and suggests that this is because the cognitive resources dyslexics have to devote to spelling are not available for word searching processes, hence reducing the extent of search in memory (cf. Grabowski, this volume). In contrast to the two previous papers, these papers point to some of the potential benefits of speech technology for writers with low-level production difficulties. All the first four chapters report research involving the collection of detailed online records of writers’ behaviour. These methods show considerable promise in developing understanding of the cognitive processes that underlie text production. Moment-bymoment measures of writing activity, particularly when the writer is performing a realworld writing task, can however result in the generation of an overwhelming large amount of data. In order to avoid this, such technological developments need to be accompanied by equivalent developments in analytic techniques. Lindgren’s paper describes how Geographical Information Systems (GIS) can be used for data mining and visualising information about the cognitive activities involved in writing. She uses data from a single case study to show how the technique can be used to analyse the interaction between different processes during writing, and then demonstrates how the technique can be used to carry out a range of different kinds of data analysis. A recurrent theme in cognitive research on the writing process is the fact that writers are constrained by the capacity of their working memory. This limits both the amount of time the content that is to be expressed can be held in consciousness, and the number of different writing-related cognitive operations that can be performed at one time. Much of the difficulty that is experienced by both developing writers and experienced writers producing complex texts, can, arguably, be traced back to working memory-related issues. Papers by Kellogg, Olive and Piolat, and by Piolat and Gérouit take up this theme. Kellogg et al. describe a study investigating the role of verbal and visual components of working memory in written sentence production. They conclude that, while verbal working memory is involved in sentence generation in general (in contradiction to claims that sentence generation is a modular process in which linguistic encoding is automatic), visual working memory is only involved when conceptual planning is concerned with visual content of some kind. Piolat and Gérouit investigate the effect that individual differences in working


9/27/2006

1:19 PM

Page 7

Introduction

7

memory capacity and different forms of note-taking have on the effort involved in comprehending spoken and written information. In line with the other research in this volume, they find that the effects of writing (in the form of note-taking in this case) are not monolithic but are the result of a complex interaction between the individual capacities of the student, the form of note-taking used and the context in which it takes place. They also find that, as suggested by Sperber in support of his prediction that speech synthesis is unlikely to replace reading, listening involves more effort than reading, particularly in the case of readers with low memory capacity. The final papers in this section examine more reflective processes in writing and how they interact with text production in the generation of content. Van den Bergh and Rijlaarsdam describe two innovations in the use of verbal protocols to study writing. First, they categorise the processes involved in a more fine-grained way than has been done in the past, distinguishing in particular between content generated during reflection and content generated in the context of text production. Second, they employ multi-level modelling to capture how these processes fluctuate during writing, uncovering individual differences in the way that reflective processes and text production are combined, and demonstrating that each is related to the quality of the final product, depending on the point in writing at which they are carried out. In the final paper in this section, Beare describes research using verbal protocols to explore the writing strategies used by bilingual writers in their first and second languages. She finds that writers use similar strategies in the two languages. 1.3.2. Effects of Writing on Cognition The first section of this volume explores the ways in which writers coordinate the various mental processes involved in writing. In the second section, the direction of effect is reversed and chapters explore the ways in which writing might affect cognition. The first three papers in this section focus on lower level processes involved in text production, investigating whether the expression of content in written form per se has effects on cognition. The remaining four chapters then move on to consider the effects of more reflective processes, of specific strategies and of specific genres. In a paper exploring what he terms the “writing superiority effect” Grabowski focusses on the effect of writing on the retrieval from memory of already-learned knowledge. Over a series of studies, Grabowski has consistently found that writing leads to better retrieval than speaking. In his chapter in this volume, he describes research designed to identify the limits of the phenomenon and to understand the relevant causal factors. He concludes that the effect occurs because writing — in the sense of transcribing conceptual material into a linguistic form — requires less cognitive load than speaking (at least in skilled adults) and hence leaves more resources available for memory search. More specifically, he rules out the text’s function as an external store as an explanation for writing superiority and claims instead that better recall is a consequence of the slower output of writing compared to speaking. An important implication of this research is that effects of writing on cognition may not just be a consequence of higher levels of metacognitive and text production processes, but also a consequence of very basic processes involved in translating conceptual material into linguistic form.


8

9/27/2006

1:19 PM

Page 8


The next chapter in this section focusses on the role of writing in learning the structural properties of language itself. Vernon describes a series of studies examining the relationship between writing ability and the development of phonological awareness in Spanish children. Her assumption is that the presence of language in an external visual form facilitates the development of awareness about various linguistic properties of speech. The studies she describes show that children’s ability to carry out various phonological tasks is related to their level of writing ability, and that the children in general can carry out these tasks more easily when stimuli are externally represented in writing rather than spoken. The next two papers are concerned with the effects of higher level writing processes on cognition. Klein, Boman and Prince investigate the relationship between writing processes, as revealed by thinking aloud protocols collected from different age groups, and improvements in scientific understanding, as revealed by the quality of explanations of physical phenomena produced before and after writing. Their results suggest that four different activities — text production, metacognitive (or reflective) processes, overall writing strategy and reviewing of experimental data — make independent contributions to the development of writers’ understanding. Although these results support the emphasis that classic cognitive models give to reflective processes in knowledge-transforming effects, they are also consistent with claims that text production processes play an independent role, and that the coordination of these processes in the service of higher level goals is a crucial ingredient. In the next paper, Lavelle describes her work investigating individual differences in approaches to writing. This differs from previous work on individual differences in writing in that it goes beyond a description of the way writers combine components of the writing process to incorporate differences in conceptions of writing, and hence in the goals that writers have when they write. She suggests that one of the main dimensions differentiating between writers is the extent to which they view writing as the development of personal understanding rather than as simply a matter of complying with external constraints, and that this high level difference in orientation is associated with lower level differences in the way components of the writing process are combined during writing. This opens up the possibility of further research linking such individual differences in approach to writing to differences in the way writers coordinate processes during writing (as described by van den Bergh and Rijlaarsdam in this volume) and to differential effects of writing on the development of understanding (as reported by Klein et al. in the preceding paper). The final two papers in this section examine how the practices associated with particular writing genres shape the cognitive processes involved in writing, and hence, by implication, the way that the cognitive effects of writing are shaped by specific disciplinary practices. Segev-Miller’s paper uses a range of verbal-report methods (including process logs, think aloud protocols and interviews) to investigate the strategies writers use in academic, “writing-from-sources” tasks. Such complex tasks involve precisely the kind of interaction with “autonomous cognitive objects” that we characterised as one of the generic features of writing. Segev-Miller finds that a major source of differences between the writers she studied is not so much in the way they define the task but rather in the strategies they use to carry out the task and in how these are coordinated. She suggests that writing instruction in this area needs to go beyond task definition to focus on how to carry out the processes involved. The following paper by Sleurs uses a similar range of methods, but with a view to examine how the features of a specific genre (press releases) are shaped by the competing demands of the


9/27/2006

1:19 PM

Page 9

Introduction

9

interested parties, rather than just to identify the processes involved in completing a welldefined task. She contrasts the conception of press releases derived from text analysis — in which their features are seen as functioning to facilitate copying from release to press outlet, with a conception of press releases as also serving to satisfy client demands and attracting journalists’ attention to this particular release rather than another, and shows how the writer involved gradually develops the text as they try to reconcile this set of conflicting demands. This close study of one particular individual in a particular context shows the dynamic shifts in the process and their relation to the demands of the genre. 1.3.3. Writing Media The final section of the book moves away from a direct concern with the cognitive processes involved in writing to consider the impact of new writing media on the writing process. The first three papers focus on changes to text production processes and look directly at the impact of current speech recognition technology on writing, assessing how readily different writers adapt to it and the effect it has on their writing. The final two papers focus on the different kinds of knowledge-objects that can be created using electronic media, assessing the effects these have on cognition and considering how to design systems to best support writers’ needs. Williams, Hartley and Pittard point out the lack of systematic empirical research and theory assessing the impact of speech recognition software on writing. Having outlined a range of theoretical approaches to writing they infer a set of predictions derived from these theories about the likely impact of S-R technology. They then describe a small-scale study using questionnaire responses to investigate its impact on more and less experienced writers. They conclude that S-R technology has very mixed effects. Although these in part can be attributed to limitations in current technology, they highlight the extent to which the effects depend on differences in individual writing strategies and on expertise, with novice writers in particular often finding it frustrating. They also suggest that existing theoretical perspectives do not provide an adequate account of the likely impacts of S-R technology. Leijten’s chapter then describes detailed observations of two writers, differing in their characteristic learning style, as they learned to use S-R technology. In line with Williams et al.’s findings, she reports that the two writers differed very strongly in how readily they adapted to the technology and in the extent to which they still relied on the traditional keyboard for important elements of the writing process. In particular, while one writer used the technology almost exclusively as a means of formulating text, but very little to interact with text as it was produced or to revise text once it had been produced, the other writer showed much more evidence of interacting with the text, using the S-R mode to revise as well as produce initial text. Finally, Hartley describes two longitudinal case studies investigating whether changing technology affects the writing style of experienced writers. The first study examines the impact of changing from composing longhand to using a word processor on the stylistic features of the texts produced by three experienced writers. The second study examines the impact of moving from typing on a keyboard to using S-R technology. In both cases, although there were marked differences between writers, he finds very little evidence of major changes in style as writers adopt different technologies, or shift from writing to speech as an input medium.


10

9/27/2006

1:19 PM

Page 10


The final two chapters investigate aspects of reflective processing in the use of new media. The paper by Stahl, Bromme, Stadtler and Jaron parallels the earlier paper by Klein et al. in focussing on the effect of writing on cognition, but investigates this in the context of the specific constraints imposed by the new form of knowledge-object constituted by hypertext. They argue that the multi-linear structure of hypertext requires a particular sensitivity to the demands of multiple audiences, and show that when writers are explicitly asked to create hypertext for multiple audiences they develop more complex hypertext structures, show more evidence of reflective processing and gain more knowledge about the domain than writers who are only asked to create a hypertext for a single audience. This illustrates how the specific features of literate artefacts influence the modes of thought involved in creating them. Rodriguez and Eklundh then conclude the book by describing their work developing a web-based system designed to support collaborative writing activities. This involves designing the knowledge-object so that it can be adapted to the needs of different members of the collaborative team. In particular it is designed such that individual contributors can access aspects of the object relevant to them, while at the same time maintaining a global overview of the context in which their contribution is taking place.

Acknowledgements We thank series editors and invited reviewers for helpful comments, and chapter authors for their time, effort, and patience. The preparation of this book was partly funded by the University of Antwerp, Belgium.


10/24/2006

9:18 PM

Page 11

SECTION 1: INTERACTIONS AMONG WRITING PROCESSES



10/24/2006

9:18 PM

Page 13

Chapter 2

Parallel Processing Before and After Pauses: A Combined Analysis of Graphomotor and Eye Movements During Procedural Text Production Denis Alamargot, Christophe Dansac, David Chesnet and Michel Fayol

Although writing pauses can be considered as main location of high-level processes, the latter can also occur in parallel with graphomotor execution. When a writer composes a text from source documents, the combined analysis of eye and pen movements makes it possible to identify some of these parallel processes and infer their nature. The present study demonstrates (1) that parallel processes differ in nature according to whether they precede or follow a writing pause, and (2) that the frequency and duration of parallel processing phases depend partly on the writer’s cognitive capacities.

2.1.

Theoretical Framework

A survey of writing studies raises two key questions (Alamargot & Chanquoy, 2001, 2002). The first one concerns the engagement of writing processes. Is it possible to identify and describe moments within the course of written production when processes take place in parallel with graphomotor execution? The second question is to what extent these moments of parallel processing depend on the writer’s cognitive and writing abilities. The study presented in this chapter sets out to investigate these two issues by providing a methodological paradigm based on the combined analysis of graphomotor and eye movements of writers while they compose a procedural text from source documents. 2.1.1.

Processing and Writing Pauses

Far more so than oral production, written composition is characterised by a considerable amount of pausing, which can occupy up to 60 or 70% of total composition time. For this Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6 Alamargot, D., Dansac, C., Chesnet, D., & Fayol, M. (2007). Parallel processing before and after pauses: A combined analysis of graphomotor and eye movements during procedural text production. In Rijlaarsdam, G. (Series Ed.); M. Torrance, L. van Waes, & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 13–29). Amsterdam: Elsevier.


14

10/24/2006

9:18 PM

Page 14

Denis Alamargot et al.

reason, various researchers have tried, over the last 20 years, to explain the temporal course of writing pauses according to their linguistic and semantic context (Caccamise, 1987; Chanquoy, Foulin, & Fayol, 1990; Dansac & Alamargot, 1999; Foulin, 1995, 1998; Foulin & Fayol, 1988; Matsuhashi, 1981, 1982; Matsuhashi & Cooper, 1978; Passerault, 1991; Schilperoord, 1996a; Schilperoord & Sanders, 1999). Largely inspired by studies of oral production (Ford & Holmes, 1978; Goldman-Eisler, 1972; Holmes, 1984, 1988; Levin, Silverman, & Ford, 1967; Maclay & Osgood, 1959; Tannenbaum, Williams, & Wood, 1967), these researchers usually interpret writing pauses on the basis of four assumptions (Foulin, 1995): (1) Pause duration varies as a function of the complexity of the processes engaged in; (2) Pause position within the hierarchical structure of the text indicates the nature of this processing; (3) The processes that occur during a pause concern the part of the text that will be written immediately afterwards (processing adjacency principle); (4) As the more demanding controlled processes cannot be engaged in parallel with graphomotor execution, they impose a writing pause (processing sequentiality principle). According to Schilperoord and Sanders (1999), these assumptions are indeed relevant to the production of short and simple texts which resemble written transcriptions of oral discourses rather than written compositions per se (Knowledge Telling strategy; Bereiter & Scardamalia, 1987). However, in the case of more complex text composition (e.g. arguments and summaries), involving more sophisticated composition strategies, the writer has to plan the content more and (re)read the text produced so far in order to revise it or resume it (Knowledge Transforming strategy; Scardamalia & Bereiter, 1991). So, owing to these additional processes, the temporal course of text composition differs markedly from that of the course of oral production. According to Chesnet and Alamargot (2005), when the differences between oral and written production are taken into account, an interpretative framework for writing pauses emerges, which is different from that of oral pauses. The slower pace of written production, compared with that of oral production, makes it easier for the writer to engage in parallel processing during graphomotor execution (Foulin, 1995). This notion calls into question the processing sequentiality principle, according to which pauses are the exclusive locations of high-level processing. Thus, transcription phases are not simply phases of graphomotor execution of a product that has already been generated. 2.1.2.

Parallel Processing During Writing

The engagement of processes in parallel with the execution of the message has been dealt with several times in the literature, for both oral and written production. Ford and Holmes (1978) studied the oral production of isolated two-clause sentences with a double-task methodology. They found a lengthening of reaction times during the execution of the end of the first clause. This result was attributed to a greater cognitive load, due to the parallel engagement of the planning process for the second clause (see also Holmes, 1988). In writing, analyses of think-aloud protocols, pauses and/or double-task performances also indicate the presence of high-level processing (such as planning and revising) during transcription (Brown, McDonald, Brown, & Carr, 1988; Chanquoy et al., 1990; Olive & Kellogg, 2002). The existence of high-level processing paralleling transcription would not appear to be in any doubt (all writers occasionally find their thoughts straying from what their pen is writing). The double issue at hand is more about determining the nature of these parallel processes and


10/24/2006

9:18 PM

Page 15

Eye Movement and Parallel Processes

15

the reasons why, although parallel processing is obviously possible, graphomotor execution sometimes has to be interrupted, and pauses occur. Some answers are suggested by the capacity theory developed by Just and Carpenter (1992) and adapted to written production by McCutchen (1996; see Chanquoy & Alamargot, 2002 for a synthesis). 2.1.3.

Capacity Theory and Parallel Processing

According to Just and Carpenter (1992), all the processes involved in linguistic activities draw on a single, limited pool of key resources. These processes can operate in parallel as long as their cumulative demands do not exceed the total amount of resources. When processing demands exceed available resources, the theory predicts: (i) a slowdown in the execution of the processes involved, that can entail (ii) the sequencing of these processes (some processes are postponed). The demands of a given process depend on (i) its degree of automation (the more automatic it is, the smaller its demand) and/or on (ii) the availability of knowledge involved in this process (highly activated knowledge is faster and easier to retrieve). In the case of writing, lexicon access (McCutchen, Covill, Hoyne, & Mildes, 1994) and graphomotor execution (Alvès, Castro, de Sousa, & Strömqvist (In this book). Bourdin & Fayol, 2000, 2002) can both be automated with practice. However, it is difficult to render the processes involved in content generation, such as planning and reviewing (at pragmatic and semantic levels), automatic, because they constantly require a high level of attentional control. This is the reason why conceptual processes — also called high-level processing — are regarded as the most demanding (Kellogg, 1996). According to capacity theory, it can be argued that variations in writing processing demands are responsible for variations in transcription fluency and for any pauses that occur. Within this theoretical framework, pauses can no longer be regarded as events of outstanding importance, but simply as the lower limit of writing speed. The presence or absence of graphomotor activity thus becomes less relevant as an indication of the processes involved in production than variations in writing speed. Accordingly, the boundary of a writing pause should not be regarded as the boundary of a processing activity, and it can be argued: (i) that a process started in parallel with graphomotor execution can be resumed within the pause, and (ii) that a process started within a pause can be pursued in parallel with a new phase of graphomotor execution. 2.1.4.

Theoretical Assumptions

One assumption is that processes operating in parallel with the graphomotor activity can be classified according to their temporal relationships with pauses. When parallel processes immediately precede a pause, it may be that they are too demanding to be maintained in parallel. They slow the writing down, which results in a pause. According to McCutchen (1996), the conceptual processes involved in text content generation could bring about just such a temporal event because of their high demands. Conversely, the fact that parallel processes occur immediately after a pause could reveal the presence of less demanding processes. In this case, the processes would be more likely to concern formulation (lexical and grammatical choices) and graphomotor programming than more demanding content-related processes.


16

10/24/2006

9:18 PM

Page 16


On the basis of these assumptions, it can be expected that pausing and parallel processing are determined by two parameters: the writer’s cognitive capacity and the demands of all the processes involved at the time. Within this framework, parallel processes may occur at any time during transcription, provided that: (1) sufficient cognitive resources are dedicated to the writing process; (2) the translation is sufficiently automated (concerning syntactic, lexical and graphomotor processes); and (3) domain knowledge is sufficiently activated for it to be retrieved at a lesser cost. In these circumstances, even very demanding processes can be executed during graphomotor activity without a pause (see Olive & Kellogg, 2002).

2.2.

Research

This research had two aims: (1) to identify and describe moments in the course of composition when processes are engaged in parallel with graphomotor execution, before, after or far from a pause; (2) to assess to what extent these moments of parallel processing depend on the writer’s cognitive capacity. As regards the first question, pinpointing parallel processes during writing is far from easy, from a methodological point of view. This difficulty can be solved, at least in part, by studying text composition from source documents, using the Eye and Pen device (1) Alamargot, D., Chesnet, D., Dansac, C. & Ros, C. (2006). Eye on Pen: a new device to study the reading during writing. Behaviour Research Methods, Instruments and Computers, 38(2), 287–299. (2) Chesnet & Alamargot, 2005). 2.2.1.

Methodological Considerations

A great deal of research has used eye movements to determine how information is processed while reading and comprehending sentences or texts (see, for a review: Kennedy, Radach, Heller, & Pynte, 2000; Rayner, 1998) or while typewriting words during copying tasks (Inhoff & Gordon, 1998). Within the field of text composition, researchers have only recently begun to use eye movements as an indicator (see Alamargot & Chesnet, 2001; Caporossi, Alamargot, & Chesnet, 2004; Chesnet & Alamargot, 2005; Strömqvist & Holmqvist, 2001). The task environment plays a central role in writing because of the presence of the text produced so far and, potentially, of documentary sources. Recording eye movements during writing should provide valuable cues about the engagement and course of the writing processes based on this visual information. Asking writers to compose a text from composite sources, e.g. mixing conceptual and linguistic information, should make it possible to detect the presence of planning process (based on conceptual information) and formulating processes (based on linguistic information). An analysis of eye fixations on the text produced so far should provide information about the reading process underlying text revision and formulation. Given the aim of this research, the assumption was that analysing the visual information encoded during the transcription period would reveal some of the parallel processes and give an indication of their nature. The Eye and Pen device makes a synchronous recording of writing activity (movements of the pen, recorded by a digitising tablet) and eye activity (eye movements in the task environment, recorded by an eye-tracking system). Associated with a writing environment comprising the text produced so far and sources


10/24/2006

9:18 PM

Page 17


17

containing referential information and lexical information, the Eye and Pen device enables the researcher to identify the engagement of conceptual processes (based on the visual encoding of referential information) and linguistic processes (based on the visual encoding of lexical information and of the text produced so far). 2.2.2.

Method

In this exploratory study, the graphomotor and eye activity of 16 adult writers (4 males and 12 females, all graduate students) were recorded while they were composing a procedural text. Participants were first familiarised with the assembly of a model turbine (see Figure 1). They were then asked to write a text enabling a reader to assemble the model correctly. At any time during the production phase, the writers could refer to five categories of information, always available within the task environment (1) the photographs of each of the nine objects making up the turbine, (2) the labels of each of these objects, (3) the photographs of the three assembly steps, (4) the labels of each of the assembly steps and (5) the text produced so far. Prior to the composition task, in order to assess their writing abilities and knowledge activation, participants were submitted to four measures: writing memory span, lexical fluency, graphomotor fluency and referential domain expertise. Writing memory span (hereafter Writing Span) is a working memory test adapted to writing from the sentence production span task procedure used by Daneman and Green (1986). The words used to prompt production were taken from the French version of the reading span

Figure 1: The referential domain available within the task environment.


18

10/24/2006

9:18 PM

Page 18


test (Daneman & Carpenter, 1980) by Desmette, Hupet, Schelstraete, and van der Linden (1995). The participants’ mean score was 3.65 (sd .99), ranging from 2.5 to 5.5. As recommended by Desmette et al., the percentage of correct sentences (mean: .50, sd .19, range: .27–.87) was used in the analyses, as it was more telling. Lexical Fluency was measured by writing down as many words as possible in the space of 60 s, alternating two noun categories: fruits and first names (Isaacs & Kenni, 1973). The observed values ranged from 13 to 24 (mean 17.75, sd 3.29). Graphomotor Fluency was calculated according to mean pause duration in a simple writing task (writing the alphabet, the participant’s first name and surname) that did not involve any demanding conceptual or linguistic process (see Chuy, Alamargot, & Passerault, 2004). The observed values ranged from 106 to 187 ms (mean 144 ms, sd 22 ms). Finally, domain expertise was assessed by measuring the time each participant took to assemble the model turbine, the longer the assembly time, the lower the expertise (observed mean = 494 s, sd 265, range: 184–1143 s). 2.2.3.

Analyses

The synchronous recording of eye and pen movements using the Eye and Pen device made it possible to ascertain the nature and duration of the eye fixations as a function of the writer’s graphomotor activity (Alamargot et al., 2006; Chesnet & Alamargot, 2005). Data analysis reveals that, from time to time during production, writers are able to continue transcribing whilst encoding visual information that is physically distant from where the pen is being moved. In order to pinpoint these so-called Parallel Events (PEs), we noted the distance between eye position and pen position at all times. We considered that the writer’s gaze was directed towards another source of information each time this distance between the point of fixation and point of transcription was greater than 4 cm. Given the distance between the writer’s eyes and the sheet of paper, this distance corresponded to an angle of 6.7° from the axis of the gaze. According to Rayner (1998), the mean size of a saccade (angular value) is 2° when reading, 3° when searching for visual information, and 4° when perceiving visual scenes. Moreover, the foveal field of vision is located within 2° of visual angle around the point of fixation, while the parafoveal field of vision is between 2° and 5° around the point of fixation. Choosing an angular limit of 6.7° means adopting a very conservative threshold. It certainly means that a great many parallel processes pass unnoticed, particularly when the gaze moves along the vertical axis, where small distances represent greater changes. Nevertheless, with an angular distance of 6.7°, it is reasonable to suppose that the point of inscription (i) is not in the foveal or the parafoveal field of vision, and (ii) that a saccade is needed in order to return to this point. The Eye and Pen device associated with these analysis criteria made it possible to determine accurately: (1) which characters were written during a PE; (2) the speed at which these characters were written (defined as the distance covered by the pen divided by the duration of the movement); and (3) the nature of the visual information gazed at during the PE. By analysing the context in which PEs occurred, it was possible to distinguish between four different types of PE, on the basis of their temporal relationships with pauses: 1. Independent parallel events: the parallel event occurs without any pause in its context, i.e. during the flow of transcription. Without pausing, the subject turns away to gaze at


10/24/2006

9:18 PM

Page 19


19

something other than her own line of writing, then returns to the ongoing line without any interruption in the writing. 2. Pre-Pause parallel events: the writer continues to write (a) letter(s), while her eyes look for information elsewhere. A writing pause then occurs (i.e. the pen stops moving for at least 130 ms) before the gaze returns to the pen. 3. Post-Pause parallel events: immediately after a pause, when the writer’s gaze was directed away from the point of transcription, writing resumes before the gaze returns to the pen. 4. Complex parallel events: some of the PEs were classified as complex ones, as they featured alternating writing and pausing phases (without the subject looking at the ongoing line of writing). The task environment comprised two kinds of information: the referential domain (photographs of objects and assembly steps, labels for each of them) and the text produced so far. In order to determine the nature of the processes involved during parallel events, we examined the breakdown of fixated information (in frequency as well as in duration), (1) comparing referential domain vs. text produced so far, and (2) dissociating the four kinds of information source comprising the referential domain. These analyses were performed comparing the three kinds of PE (i.e. independent, pre-pause and post-pause parallel events).

2.3.

Results

The presentation of the results will be structured as follows. In Section 1, a single text will be presented to illustrate what can be observed using the paradigm. Section 2 will feature a description of the parallel event parameters (frequency, duration, position as regards word boundaries). Section 3 will address the question of the nature of the processes engaged in during PEs, by analysing the type of visual information that was processed. Finally, Section 4 will examine the predictions of the capacity theory. 2.3.1.

Qualitative Analysis of a Text Produced by a Subject

The following example of a text produced by one of the subjects (see Figure 2) shows the PEs that were identified within it. The underlined characters were transcribed during PEs and the index figures indicate the nature of visual information that was fixated during the PE. The qualitative analysis of this text is quite a good illustration of how PEs occur and the type of processing that takes place within these events. Consider for instance the word “moteur” (6th word in the first line). This word was the location of two PEs. During the first letter, ‘m’, the writer encoded information from the photographs of the objects. During the second PE, while the writer was writing the ‘r’, he encoded information from the photographs of the assembly steps. The word “commande” (end of the 4th line, second column) was written entirely blindly, the eyes gazing first the photographs of the steps (while writing ‘comm’), then at the labels of these steps (while writing ‘ande’). However, this text also shows that processing during PEs does not solely consist in encoding referential information. It may also concern the text produced so far, particularly the spelling.


20

10/24/2006

9:18 PM

Page 20

Denis Alamargot et al. Text by subject 06 Prendre tout d'abord la capsule m2oteur1 et Les3 d1eux co5nstr6uctions en faisant attention y joindre First of all take the motor que les the two constructions making sure that capsule and join it to the la capsule d'engrenag1e en faisant bien différents eng1renages permettant la mise en attention à ce1 the gearing capsule making fonctionnement different gearings needed to run sure that de l'appareil soit bien reliés.3 the device are well que les deux branches de cette dernière connected. soit vers la the two branches of the latter Pre3nd5re2 la bat3t1erie et y fixer le m3o1dule d3e are facing the comm1ande3 Take the battery and attach the première.5 Les différents cables de la control module to it capsule doivent être former. The different Les d1eux branchements batteries/1module de1 cables of the capsule must be commande The two connections between situées vers l'extérieur pour perme5ttre le batteries / control module branchement placed towards the outside to doivent être réalisé après assemblage de la1 make a proper batterie ,1 sur la c3onstruction ca2psule moteur2 définitif.4 connection. et c4ouple engrenage must be made after assembling the battery, on the motor capsule 4 2 Prendre ensuite la cap sule vide et p lacer construction and gearing couple c4inq raccords4 et la buse sur les s5ix branches de cette dernière. Next take the au moyen des deux raccords restants. Les placer empty capsule and place five links and the in between the two remaining links. Put them duct on the six branches of the latter. sur le dessus et y en5clencher1 la batterie1 cable Placer ensuite4 les 4 ob4turate4ur1s en bleu à gauche et cable rouge5 à1 droite. on the étoile autour de la Next place the 4 radial above and clamp the blue battery cable to the shutters around the left and the red cable to the right. capsule et jo5indre5 en face de la buse4 le ventilateur.2 capsule and join opposite the ventilation duct Une fois ces opérations terminées, il est temps d'assembler Once these operations are complete, it is time to assemble Notes: Highlighted text written while gaze was directed elsewhere. 2

1

photographs of the

3

assembly steps; photographs of the objects; labels of the assembly steps; 4 labels of the objects; 5 text produced so far; 6 blank part of the sheet. English translation in italics.

Figure 2: An example of text produced by a subject, showing the PEs.

The word “situées” (beginning of the 5th line) is quite a good example. The writer initially misspelt this word, writing ‘situé’ (and thus forgot to mark the number of the related noun, “cables”), and continued writing the rest of the sentence. When the word “permettre” was written, a PE occurred, during which the writer looked back to the beginning of the line, particularly the word ‘situé,’ which was incorrectly spelled. Following this PE, after completing the current word, the writer immediately (only 826 ms after the end of the last character) added ‘es’ (our underlining) to the end of the misspelled word (unfortunately making a gender error this time). The detection of the grammatical error and the computation of its correction may have been carried on in parallel with the graphomotor execution,


10/24/2006

9:18 PM

Page 21


21

although the actual editing was delayed, probably to avoid interrupting word transcription process already underway. From a quantitative point of view, this text comprised 55 PES for a text composed of 165 words (850 characters), i.e. one PE every three words on an average. During these PEs, 106 characters were written, i.e. 12.4% of the whole text. Note that only 33 PEs (about 60% of the total number) were located at the beginning or end of a word. This rough description is enough to demonstrate that PEs are quite frequent phenomena, wherein various processes may take place. 2.3.2.

Description of General PE Characteristics

On an average, we observed 43 PEs per text produced, for a mean PE duration of 593 ms. The total duration of PEs per text averaged out at 25.52 s, representing 10.2% of total transcription time (excluding prewriting pause and any pause exceeding 130 ms). These results confirm that the phenomenon cannot be neglected, and that the parallel mode must be considered as a ‘natural’ mode for executing at least some of the processes. Pre-Pause PEs were the most common, in frequency as well as in total duration, representing 47.3% of the total number of PEs and 36.6% of total PE time (i.e. time spent on parallel processing). Independent PEs represented 22.2% of the total number of occurrences and 25.6% of total PE time. As regards frequency, Post-Pause PEs and Complex PEs were equivalent (15.3% and 15.1% respectively), although Complex PEs represented a greater proportion of total PE time (21.4%) than Post-Pause PEs (16.2%). In order to be able to examine the distribution of PEs according to word boundaries (some Complex PEs covered two words) and because Complex PEs also included pauses (sometimes longer than the chosen threshold of 130 ms), this kind of PEs was eliminated from the analyses presented below. On an average, PEs appeared in 33.51% of the words. PEs were considered to be at the beginning of a word when the first character written in parallel was in first position in the word, at the end when the last character of the PE was the last in the word, and in-between the word in every other case. This analysis is shown in Table 1. It would appear that the different types of PE were highly correlated with the position in the word (χ2 362.64, df 4, p .00001). Post-Pause PEs were more frequent at the beginning of words (83.5%), Independent PEs were mostly in-between (61.2%), and Pre-Pause PEs came at the end (75.4%). When this analysis is made for each individual participant, all but two participants presented this pattern, with a significant association between type of PE and position (mean χ2 22.58, df 4, p .01). The two subjects who did not present a significant association produced a low number of PEs, especially Post-Pause PEs. As a partial summary, it should be pointed out that when PEs were classified according to their proximity to pauses, the Pre-Pause PEs were the most frequent. The three major kinds of PE differed as regards their position in the words, in a manner that was consistent with the fact that words are generally preceded and followed by pauses. Note, however, that although word boundaries also seem to have been relevant to parallel processing phases, about one-third of the PEs occurred in places that were less predictable.


22

10/24/2006

9:18 PM

Page 22

Denis Alamargot et al. Table 1: Number (italics) and frequencies of the different PEs according to their position within the word.

Independent PE Pre-Pause PE Post-Pause PE Total

2.3.3.

Beginning

In-between

End

Total

33 23.7% 15 6.9% 81 83.5% 129

85 61.2% 53 19.1% 14 14.4% 152

21 15.1% 209 75.4% 2 2.1% 232

139 277 97 513

Nature of the Visual Information Encoded During Parallel Events

2.3.3.1. Encoding referential information vs. text As regards frequency, visual information encoded during a PE concerned the referential domain as often as the text, as far as the Independent PEs were concerned (51% vs. 49%), whereas the referential domain was more often explored than the text during Pre-Pause PEs (62% vs. 28%) and PostPause PEs (58% vs. 42%). However, the differences between the three kinds of PEs did not reach significance (F(2–30) 1.47, NS), while the greater difference between Independent PEs and Pre-Pause PEs was only marginally significant (LSD test, p .10). Contrary to the results for frequency, the total duration of visual exploration of the two sources of information during PEs clearly shows that the referential domain was explored for longer than the unfinished text. During the three types of PE (Independent-, Pre- and Post-Pause PEs), the domain was fixated for longer (85, 87 and 79% of total gazing time respectively, with a mean duration of gazing time per PE of 453, 326 and 300 ms respectively). The mean duration of text exploration was considerably shorter (116, 38 and 94 ms respectively). Again, the differences between the types of PE were not significant (F(2–30) 1.42, NS). 2.3.3.2. Relation between the type of PE and the type of information encoded when exploring the referent Four different sources of information were identified within the domain, and Table 2 shows the frequencies and the percentages of the total time spent gazing at each of these sources for all three types of PE. The most interesting differences concerned the encoding of the object labels and step photographs. Object labels were more frequently explored during Post-Pause PEs than during Independent PEs and Pre-Pause PEs (35.82% of the encoding phase vs. 30.93% and 26.85%), although the greater difference between Post- and Pre-Pause PEs was only marginally significant (LSD test, p .10). Visual exploration was dedicated to step photographs more often during Pre-Pause and Independent PEs than during Post-Pause PEs (25.28% and 21.85% vs. 14.85%, F(2;30) 4.57, p .025). Post-hoc comparisons showed that the difference between Post- and Pre-Pause PEs was significant (LSD test,


10/24/2006

9:18 PM

Page 23


23

Table 2: Breakdown of the number of encoding phases (italics) and the time spent gazing at each kind of source during PEs (%). Independent PE

Pre-Pause PE

Post-Pause PE

Object photos

23.48 27.65

23.86 22.85

28.87 29.27

Step photos

21.85 35.16

25.28 44.71

14.85 19.28

Object labels

30.93 29.96

26.85 23.42

35.82 39.00

Step label

23.73 7.26

24.01 9.02

20.46 12.45

p .01), whereas the difference between Post-Pause and Independent PEs was only marginally significant (LSD test, p .055). As regards the amount of time spent exploring each source of referential information, there was a marginally significant effect of the type of PE on the time spent exploring object labels (F(2–30) 2.66, p .086). During Post-Pause PEs, the proportion of time dedicated to encoding these labels was greater than it was during Pre-Pause PEs and Independent PEs. However, only the difference between Post-Pause PEs and Pre-Pause PEs reached significance in post-hoc comparisons (39% vs. 23.42%, LSD test, p .05). The amount of time spent encoding step photographs followed the same pattern as the frequency, with a significant effect of the type of PE (F(2–30) 6.16, p .05). Less time was spent encoding step photographs during Post-Pause PEs (19.28%), than during either Independent PEs (35.16%, LSD test, p .05) or Pre-Pause PEs (44.71%, LSD test, p .005). A comparison between Independent PEs and Pre-Pause PEs revealed no significant difference. A partial summary of the results indicates that Independent PEs and Pre-Pause PEs were quite similar as regards the nature of the visual information processed during the parallel event. For both types of PE, writers gazed more frequently and for a greater proportion of their gazing time at referential information about the objects, particularly the assembly steps. This suggests that during these PEs, the processes that operate in parallel are more concerned with conceptual matters, and may be dedicated to planning the content of the text, in close relation with the steps of the assembly task. On the other hand, the information fixated during Post-Pause PEs was more often related to lexical problems, particularly object labels. This could attest to the presence of a parallel formulating process and the encoding of lexical information enabling formulation to proceed. One interesting point, which will be discussed further on, is that whereas formulation processing took place during the graphomotor execution following a pause (in Post-Pause PEs), it would appear that the demands of conceptual processes (during Pre-Pause PEs) may lead to a pause in graphomotor execution. This point has two possible consequences.


24

10/24/2006

9:18 PM

Page 24


(1) One can imagine a model of pauses filled by two kinds of processes operating consecutively. The first part of the pause would be occupied by conceptual processes generating forthcoming text content (possibly triggered during a Pre-Pause PE), whereas the second part of the pause would be dedicated to the linguistic formulation of already planned content (processes which can be pursued in parallel with writing during a Post-Pause PE). (2) The fact that conceptual processes carried out in parallel with graphomotor execution frequently lead to a pause could be the consequence of demands being too great for graphomotor execution to be maintained. Conversely, processes involved simply in linguistic formulation would only place a low demand on available resources and could therefore be pursued in parallel with graphomotor execution. These assumptions are consistent with the predictions of capacity theory, according to which (1) writing speed should be lower during PEs, due the demand on resources made by parallel processes; (2) there should be a statistical relationship between the characteristics of PEs and some of the writers’ cognitive abilities, particularly writing production span, lexical fluency, graphomotor fluency and referential expertise. 2.3.4.

Parallel Events and Writer’s Capacity

2.3.4.1. Diminution of writing speed during PEs According to Just and Carpenter (1992), the division of cognitive resources between two processes operating in parallel should reduce the speed of execution of at least one of them if the mental load is too high, eventually causing it to stop altogether when overload is reached. In order to confirm this phenomenon and the presence of costly parallel processes involved during PEs, the writing speed during PEs was compared with the mean writing speed observed outside PEs for the entire text production (excluding the time spent pausing). Table 3 summarises the mean speeds and the comparisons that were performed. The results indicate that, on an average, writing speed was significantly lower during PEs. Other comparisons revealed no significant difference between the different types of PE. Thus, the presence during PEs of processes involved in encoding visual information from the task environment causes graphomotor execution to slow down. According to the capacity theory, this is because allocating resources to these processes diminishes the resources available for graphomotor execution (Bourdin & Fayol, 2002). This in turn raises the question of whether the operation of these processes in parallel with graphomotor execution depends on the writer’s cognitive capacity.

Table 3: Mean speed of graphomotor execution (cm/s) and comparisons between the speeds during PEs and outside PEs.

Student t (df 15) P

Outside Pes

Independent PE

Pre-Pause PE

Post-Pause PE

2.76 / /

2.25 2.12 .05

2.38 3.12 .007

1.99 4.01 .001


10/24/2006

9:18 PM

Page 25


25

2.3.4.2. Relationship between PE duration and frequency and writer’s capacities Multiple linear regression analyses were conducted in order to explore the relationships between either the frequency of PEs in the text (weighted by the number of words in the text) or the total amount of time the PEs lasted for (weighted by the total writing time) as dependent variables, and the four capacity measures as independent variables. Regression analyses were conducted separately for each type of PE, using the forward stepwise method. The results are summarised in Tables 4a (frequency) and 4b (time). The variance of Independent PE frequency is explained both by lexical fluency and by graphomotor fluency. Fluent lexical access and fast graphomotor execution processes were associated with a higher number of Independent PEs. The occurrence of Independent PEs depends partly on how easy and fast the low-level processes involved in formulation and graphomotor execution are. The automation of lexical and graphomotor processes, by means of reducing their cognitive demands, allows writers to operate parallel processes during writing more frequently, and without any pauses. Regarding Pre-Pause PEs, these are more frequent when writers have poor workingmemory abilities (negative correlation with Writing Span) and less referential expertise in the domain (indicated by longer assembly times in the domain expertise variable, positively correlated with Pre-Pause PE frequency in the regression model). Whereas the presence of Independent PE could be associated with greater composition expertise, the Table 4a: Variance explanation in the stepwise regression analysis for the dependent variable no. of PEs/text length. Variable

Beta Correlation

Partial correlation

Semit (13) partial correlation

p level Entered into equation

Independent PE frequency: R2 .65; F(2.13) 12.4; p .001 Writing span Lexical fluency Graphomotor fluency Domain expertise

0.05 0.41 0.52 0.12

0.07 0.52 0.61 0.20

0.04 0.36 0.46 0.11

0.25 NS 2.21 0.05 2.81 0.025 0.71 NS

No Yes Yes No

Pre-Pause PE frequency: R2 .43; F(3.12) 3.11; p .10 Writing span Lexical fluency Graphomotor fluency Domain expertise

0.52 0.22 0.26 0.57

0.52 0.2 0.32 0.56

0.46 0.17 0.25 0.51

2.13 0.10 0.76 NS 1.18 NS 2.37 0.05

Yes No Yes Yes

Post-Pause PE frequency: No variance explained Writing span Lexical fluency Graphomotor fluency Domain expertise

0.20 0.11 0.07 0.16

0.20 0.11 0.07 0.16

0.20 0.11 0.07 0.16

0.77 0.42 0.26 0.62

NS NS NS NS

No No No No


26

10/24/2006

9:18 PM

Page 26


Table 4b: Variance explanation in the stepwise regression analysis for the dependent variable total duration of PEs/total writing time. Variable

Beta Correlation

Partial correlation

Semit (13) partial correlation

p level Entered into equation

Independent PE time: R2 .76; F(3.12) 13; p .0005 Writing span Lexical fluency Graphomotor fluency Domainexpertise

0.12 0.56 0.44 0.18

0.19 0.70 0.62 0.33

0.09 0.47 0.38 0.17

.66 NS 3.40 0.005 2.74 0.025 1.23 NS

No Yes Yes Yes

Pre-Pause PE time: R2 .41; F(3.12)2.77; p .10 Writing span Lexical fluency Graphomotor fluency Domain expertise

−0.44 0.18 0.42 −0.36

−0.45 0.19 0.47 −0.38

0.38 0.14 0.41 0.32

1.74 NS 0.63 NS 1.85 0.10 1.45 NS

Yes No Yes Yes

Post-Pause PE time: R2 .13; F(1.14)2.15; NS Writing span Lexical fluency Graphomotor fluency Domain expertise

0.36 0.02 0.21 0.27

0.36 0.02 0.22 0.26

0.36 0.02 0.20 0.24

1.46 0.08 0.82 0.98

NS NS NS NS

Yes No No No

presence of Pre-Pause PEs would reflect difficulties in retaining content in working memory before putting them down in the text. Due to this lower capacity, the processes triggered in parallel with writing would not leave sufficient resources for the formulating process, thereby forcing the graphomotor execution to stop. According to this interpretation, the increased presence of Pre-Pause PEs would indicate a failure to pursue processes engaged in parallel with graphomotor execution. This failure would be due to insufficient domain expertise and working memory capacity for text composition. With regard to the total time occupied by PEs, we find the same result for Independent PEs: an increase in the total duration of Independent PEs was associated with a greater automation of lexical and graphomotor processes. Moreover, graphomotor fluency seems to play an equivalent role in Pre-Pause PE duration. Pre-Pause PEs last longer when graphomotor execution processes are automated. Whereas the number of Pre-Pause PEs depends on working memory and domain expertise, their duration is related to the automation of graphomotor processes. In other words, the more automatic they are, the sooner parallel processes are triggered, and the longer the PEs last. Lastly, it should be noted that no variable can explain the variance of Post-Pause PEs, regarding either their frequency or their duration. This absence of any result confirms that Post-Pause PEs probably have a particular cognitive status, different from that of the other parallel events.


10/24/2006

9:18 PM

Page 27


2.4.

27

Discussion

The results show that the Eye and Pen design and the method used made it possible to isolate phases during which parallel processes occurred and to describe both qualitatively and quantitatively their relationships with writing pauses. Parallel events last longer than half a second (593 ms), and the fact that they represent no less than 10% of total executing time, suggests that they constitute an important phenomenon. Moreover, PEs can be classified according to their proximity to pauses and the nature of the visual information that was fixated. Processes that occur in parallel far from any pause or just before a pause (during Independent PEs and Pre-Pause PEs) are more concerned with content generation (i.e. conceptual processes). Processes that occur during PEs immediately after a pause (during Post-Pause PEs) are more concerned with the formulation of already prepared content (i.e. linguistic processes). The execution of these processes conducted in parallel with graphomotor execution is probably demanding because it causes a decrease in the writing speed. This competition phenomenon, interpreted within the framework of the capacity theory, confirms the need to share cognitive resources between low- and high-level writing processes. In this context, we show that the frequency and duration of Independent PEs and Pre-Pause PEs are partly dependent on the writer’s abilities (lexical fluency, graphomotor fluency, writing production span and domain expertise). Nevertheless, whereas Independent PEs are related to higher abilities (lexical fluency, graphomotor fluency), Pre-Pause PEs are more frequent in the case of lower abilities (production span and expertise). This suggests that Pre-Pause PEs can be regarded as Independent PEs which, due to the writer’s limited capacities, have been interrupted by a pause. This pause reflects the difficulty of pursuing high-level (conceptual) processes and graphomotor execution in parallel. This interpretation is supported by the decrease in writing speed during PEs, whatever their pausing context is. On the whole, all these results confirm the idea that writing pauses may not always depend on strategic choices on the part of the writer, but may also rely on cognitive capacity: in some cases, the writer may be unable to pursue some of the processes engaged in parallel with graphomotor execution. This interpretation highlights two points which merit further discussion: (1) the status of pauses as the main location of high-level processing and (2) the fact that although cognitive overload may sometimes cause pausing, not all pauses result from cognitive overload. First, with regard to the status of pauses, the present findings on parallel processing suggest that pauses should be regarded not so much as discrete phenomena but as the lower limit of a variable writing speed. According to Kellogg (1996), writing processes may operate in parallel as long as the cognitive resources are not exceeded. Thus, pauses should be considered as the ultimate consequence of cognitive overload, namely a decrease in graphomotor execution speed until it reaches zero. Pause interpretation should then be included within a larger framework, i.e. the analysis of increasing and decreasing writing flows, reflecting the presence or absence of processes with various cognitive loads, operating in parallel with writing. As it happens, the results of the present study show that pause boundaries are not a good indication of controlled and demanding processes newly starting or stopping. They indicate that these processes may begin about 593 ms before the pause occurs, just as they may continue beyond the end of the pause, making pauses very


28

10/24/2006

9:18 PM

Page 28


rough indicators which fail to pick up on all the processes (such as those operating during independent parallel events). Second, as to whether the writer’s cognitive resources play an important role in determining the frequency and duration of parallel events, they may not be the only set of parameters governing the occurrence of pauses. Linguistic units, particularly word boundaries, seem to impose a rhythm on writing. The present study found that, whatever their nature, a high proportion of PEs occurred at word boundaries, in a manner that was consistent with their type: Pre-pause PEs often at the beginning of a word, Independent PEs above all in the middle, and Post-pause PEs at the end. However, this was not an absolute rule, and a considerable proportion of PEs (about a quarter) did not respect this relationship. For instance, that 25% of Pre-Pause PEs appears at the beginning or within a word tends to demonstrate that the capacity constraint also apply without giving importance to linguistic units. The complex relationship between the different types of PE and their position within the word suggests that two different systems of constraints may coexist: on the one hand, capacity constraints, which would cause pauses to occur whatever the unit being written, and, on the other hand, graphomotor and lexical constraints, that would sometimes postpone some processes until unbreakable units, such as the word, had been completed. 2.4.1.

Conclusion

The two aims of this study were (i) to identify and describe moments when processes are executed in parallel with graphomotor execution and (ii) to find out how these moments are determined by the writer’s cognitive capacity. By using the Eye and Pen device (Alamargot et al., 2006; Chesnet & Alamargot, 2005), it was found (i) that parallel processing differs in nature according to whether it precedes or follows a writing pause, and (ii) that the frequency and duration of parallel processing moments depend partly on the writer’s cognitive capacity. These findings represent a significant advance at the theoretical level, as they provide an initial answer to the question of the objectivation of parallel processing during writing. The parallel processes studied here were based on the encoding of visual information. At the theoretical level, the present study raises at least three issues. 1. It is now obvious that text production processes (with the obvious exception of reading and the exproprioceptive monitoring of graphomotor execution) can take place without visual feedback, via the processing of knowledge stored in long-term memory. This kind of processing, based on internal sources of information (e.g. domain knowledge retrieval) was not directly investigated in this study. This means that instances where processing occurs in parallel with graphomotor execution are probably even more numerous than those described here. One future avenue of research will thus be to analyze how processing based on internal sources is combined with and even interacts with the processing of external sources, during pauses or periods of transcription (Chuy et al., 2004). 2. In this study, the nature of the processes engaged in was interpreted according to the visual information that was fixated. However, this interpretation was based on at least two premises. For a start, fixated information is not necessarily encoded in working memory. During text production, the visual focus does not necessarily correspond to the attentional


10/24/2006

9:18 PM

Page 29


29

focus (staring into space, averting the gaze; Glenberg, Schroeder, & Robertson, 1998). Accordingly, these phenomena now need to be identified, in particular by analysing the characteristics of fixations and saccades during writing, in relation to the associated writing performance. Secondly, the relationship between the fixated information and the writing process may not be univocal. Reading a lexical item may trigger conceptual processing or, conversely, non-linguistic information may trigger lexical retrieval. In the case of the present experiment, a more fine-grained analysis needs to be carried out of the temporal relationship between the nature of the visual information encoded and the nature of the processing which immediately follows this information gathering. This analysis alone will reveal if, and when, a lexical item read in the task environment is indeed produced in the phrase. 3. The parallel processes identified in this study were strongly associated with the handwriting situation. The question now is how far the same types of parallel processing can be observed in a typewriting situation. In the case of typewriting, there is a spatial dissociation between the feedback displayed on the screen and the writing process on the keyboard. When a typewritten text is composed from source documents, moments of parallel processing are probably distributed differently, as the gathering of visual information is no longer divided between two spaces (source and writing on the page) but between three (source, keyboard and screen). Only a comparative study of handwriting and typewriting (Nottbusch, Weingarten & Sahel, this book), based on a careful analysis of the associated eye movements, will make it possible to determine the effect of medium on the dynamics of text production. Lastly, at a methodological level, the ease with which the parallel processes were pinpointed in this study and the accuracy of their descriptions confirm the usefulness and relevance of the Eye and Pen device. The latter opens up new perspectives as regards the scientific study of writing processes and the dynamics of their relationships during the writing activity. The role of reading during writing and the dynamics of content generation (Breetvelt, van den Bergh, & Rijlaarsdam, 1994; van den Bergh & Rijlaarsdam, 1999) can now be further investigated.


Else_SW-Torrance_CH003.qxd

9/27/2006

10:15 AM

Page 31

Chapter 3

From Written Word to Written Sentence Production Guido Nottbusch, Rüdiger Weingarten and Said Sahel

In this article, we present an experimental paradigm of time course measurements for the study of written language production at the levels of words and sentences. It is shown that time measurements can give insights into cognitive processes involved in writing. First, we outline the methodology and discuss it in the context of written language production. Second, we summarize a series of experimental studies on written word production. The main result is that, at the beginning of word writing, the segmental information — the grapheme or letter sequence — is not completely specified. Instead, depending on linguistic word structure, the motor buffer has to be reloaded. This indicates that cognitive processes are operating during writing. Third, we present an original study on written sentence production. It shows that the time course of writing sentences is influenced by sentence planning and syntactic structures. Depending on the task type, pauses occur at the boundaries of major syntactic constituents and words as well as within these boundaries. At this point, we also discuss the relationship of written word production and sentence production with respect to time course measurements. Finally, some perspectives for future research are outlined.

3.1. Introduction Real-time research on written language production has up to now focused on two major aspects: written word production on the one hand and written composition (see e.g., Alves, Castro, de Sousa, & Strömqvist, this volume; Schilperoord, 1996b, for a review) on the other. Concerning written word production, a number of studies mainly focused on writing as skilled motor behavior (on typing: Cooper, 1983; Gentner, 1983; Gentner, Larochelle, & Grudin, 1988; Terzuolo & Viviani, 1980; Viviani & Laissard, 1996; on handwriting: van Galen, 1991; van Galen, Meulenbroek, & Hylkema, 1986). These studies promoted the general view that the time course of writing is more or less independent of higher cognitive Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Nottbusch, G., Weingarten, R., & Sahel, S. (2007). From written word to written sentence production. In G. Rijlaarsdam (Series Ed.), and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in Writing, Vol. 20, pp. 31–53). Amsterdam: Elsevier.


32

9/27/2006

10:15 AM

Page 32

Guido Nottbusch et al.

processes. Observable time patterns were, to a large extent, attributed to biomechanical constraints. A main point of this assumption was the probabilistic strength of a sequence of letters for a given interval and, accordingly, the size of the basic unit. Linguistic constraints were mainly limited to varying frequency measures (digraph, trigraph, word). A common idea found in the aforementioned studies is that of a general motor program, which is assumed to receive completely specified sets of lexical-orthographic information (especially van Galen, 1991). Although there were some hints at syllabic (e.g., Shaffer, 1978; Zesiger, Orliaguet, Boë, & Mounoud, 1994) and morphological influences (Orliaguet & Boë, 1993; Pynte, Courrieu, & Frenck, 1991), written word production has not been sufficiently analyzed in terms of linguistic units like syllables or morphemes. However, compelling evidence has been provided that the time course of motor execution in writing is dependent on linguistic processes involved in written language production (Nottbusch, Weingarten, & Will, 1998; Weingarten, 1997; Weingarten, Nottbusch, & Will, 2004; Will, Weingarten, Nottbusch, & Albes, 2005). In Section 3.2, the time-based approach, developed by our research group, will be described and discussed in the context of written language production. In Section 3.2.1, the results of a series of experimental studies on written word production obtained on the basis of our experimental paradigm are summarized. In Section 3.2.2, we present an original study in which this paradigm was expanded to written sentence production.

3.2. Methodological Considerations To make time measurements productive, some methodological considerations are necessary (for a more detailed description, see Weingarten et al., 2004). The experimental paradigm presented here is based on the analysis of discontinuous typing in which participants are asked to type words following the presentation of various stimuli. This provided two basic types of time information: initial latencies (the time intervals between the presentation of the stimulus and the first keystroke) and interkey intervals (the time intervals between successive keystrokes). Studies in typing (e.g., Gentner, 1983; Larochelle, 1983; Ostry, 1983) have listed a number of factors influencing keystroke timing: typing speed, layout of keyboard (peripheral keys being typed slower than more central keys), typing skills (letter repetition being faster than alternating hands/fingers only in non-fluent typists), frequencies of letters and words, and, most importantly, letter context. Shaffer (1978) and Gentner (1983) found effects on keystroke timing in up to three preceding characters and one succeeding character with the immediately preceding character being by far the most important (reduction of variability of the interkey intervals (interquartile range) by about 43%). As it is almost impossible to construct a sufficient quantity of word material with an identical format (three preceding letters and one succeeding — or even more) that, at the same time, include different kinds of linguistic boundaries (see below), we decided to compare only letter transitions between identical pairs of letters — we call them digraphs. For example, we compare interkey intervals of the digraph in rider and ridge. The reduction to the immediately preceding character causes no serious problems, because the effects of other letters are rather small.


9/27/2006

10:15 AM

Page 33

Written Words and Written Sentences

33

To investigate the influence of linguistic word structure on the time course of writing, we defined the variable “type of boundary” to denote the initial locations of various types of linguistic units contained in the word. Interkey intervals are divided into four groups: • syllable+morpheme denotes interkey intervals at combined syllable and morpheme boundaries • interkey intervals for characters at the onset of syllables alone • interkey intervals at ‘pure’ morpheme onsets • interkey intervals for characters at all other positions within a word, i.e., at all positions within a syllable or within a morpheme For example, in the English word ‘c-o-n-f-o-u-n-d-e-d’ the following interkey interval types would be identified: a syllable+morpheme interkey interval between ‘n-f’, a syllable boundary interkey interval between ‘n-d’, a morpheme boundary interkey interval between ‘d-e’, all other interkey intervals are of within-syllable type. If we test differences between these types of interkey intervals, we compare only sets of identical digraphs. In an explorative study (Section 3.2.2), we extended this paradigm from the word level to the sentence level. This approach combines the benefits of a large number of participants typing exactly the same words with the advantage of measurements of typed production, providing data of unit initial as well as unit internal latencies. The analytical unit therefore is the single letter (cf. Schilperoord, 1996b: 77 for spoken (dictated) production). Analogous to Chanquoy, Foulin, and Fayol (1995), we used additional syntactic boundary types: in our case the largest type of boundary is the sentence initial latency which denotes the time span between the presentation of the stimulus and the first keystroke; this implies that sentence initial latencies are opposed to sentence internal latencies but, at the same time, constitute word initial latencies. Furthermore, we define various onsets of syntactic constituents, which are also a special type of word initial latency. A special problem in time measurements in sentences is the space bar. In Section 3.4.3, we illustrate how it can be accounted for. 3.2.1. Written Word Production From a chronometrical perspective, writing words can be characterized as producing a hierarchically ordered set of linguistically motivated units (see Weingarten et al., 2004). At the top level of this hierarchy is the graphemic word. At the word level, the longest delays are found word initially, i.e., between the presentation of the stimulus and the first keystroke. These delays are correlated to various variables such as word frequency, word length, and task type: high-frequency words are initiated faster than low-frequency words and short words faster than long ones; written picture naming causes longer delays than copying a written word. This indicates that before starting to write a word the mental lexicon is addressed and basic word information is retrieved (cf. Bonin, Fayol, & Gombert, 1998). From a linguistic point of view, polysyllabic and polymorphematic words can be analyzed in two ways: (1) as a succession of morphemes, or (2) as a succession of syllables. In some cases, syllable and morpheme boundaries coincide and in others they do not (see Section 3.4.2). Looking at pure morpheme boundaries, surprisingly the delay is no longer than at “simple” letter boundaries (Will et al., 2003; Weingarten et al., 2004). On the other


34

9/27/2006

10:15 AM

Page 34


hand, combined syllable+morpheme boundaries lead to the longest word-internal delays. Delays at these types of boundaries correlate negatively with the word frequency indicating word-internal lexical access (see below). For all other types of interkey intervals, no significant differences were found (Will et al., 2003). Our finding, that the time course of single word typing was not affected by morpheme boundaries, cannot be seen as a counterevidence to the hypothesis of Orliaguet and Boë (1993) that morphemes constitute processing units of typing. But, obviously, not all types of morphemes lead to increased latencies at their onsets. If we look at syllable+morpheme boundaries and simple morpheme boundaries in German, an important morphological difference between the different types of morphemes has to be noted: stems and prefixes always constitute syllable+morpheme boundaries, but never morpheme boundaries (which are not syllable boundaries at the same time), whereas a large number of suffixes comprise morpheme boundaries. To clarify the impact of morphological structure on the time course of word writing, we conducted a series of experiments in which different types of syllable+morpheme boundaries were compared (Nottbusch, Sahel, Grimm, & Weingarten, in preparation). In these experiments, Nottbusch et al. (2005) found a significant difference in timing between syllable+morpheme boundaries that precede a word constituent including a stem (e.g., [hotel cook]) compared to those that do not (e.g., <Eitel-keit> [vanity]). Units of the former type are stems or of the form prefix+stem, whereas the latter type are exclusively suffixes. By definition, suffixes never precede stems as parts of the same word constituent. These results lead us to postulate lexical constituents as dynamic units of word writing. They always include a stem and their lexical dependency is indicated by their correlation with word frequency. A further important result from Nottbusch et al. (2005) is that, at these syllable+morpheme boundaries, the whole word is addressed and not just the base, i.e., the specific word constituent is initiated at this location: syllable+morpheme delays are significantly affected by whole word frequency but not by base frequency. This finding indicates that, word initially, polymorphematic words are not accessed and processed in a compositional manner. Rather, it seems that a word-internal re-access of the representation of the whole word, which — in the case of infrequent items — may have been composed earlier in production, takes place. In future studies, morphemes not coinciding with syllable boundaries should be examined. The effect might be too small to be detected with the chronometric methodology developed so far: it could be a distributed process, or possibly the morphemes are planned at the onsets of larger lexical constituents, or processing costs might be too low to influence the time course of typing. In the hierarchy of interkey intervals, syllable onsets lead to the second largest delays. Delays at syllable boundaries are significantly smaller than those at syllable+morpheme boundaries but still significantly larger than delays at simple letter boundaries (Weingarten et al., 2004). Accordingly, in the hierarchy of processing units, we postulate a syllable tier below the level of lexical constituents. Delays at syllable boundaries are not influenced by word frequency; we thus assume that these units are generated post lexically (see below). One might argue that the observed syllabic effect could be a side-effect of subvocal articulation during writing (e.g., Wilding & Mohindra, 1980). To rule out this possibility, participants were asked to write words while singing a tone. As the syllable effect did not diminish even with this additional task, other explanations must be taken into account. Two


9/27/2006

10:15 AM

Page 35


35

possibilities can be considered: the syllable effect arises as a consequence of central phonological processes accompanying writing and the second possibility is that it is an autonomous temporal pattern of written language generation. In a study with hearing impaired adults, Nottbusch, Grimm, Weingarten, and Will (2005) showed that this group of people wrote with almost the same syllabic patterns as unimpaired individuals, although it is obvious that hearing impaired individuals have different spoken language skills. This result suggests that the syllabic structure in writing is not guided by the phonological system, at least not by a phonological system set up by spoken language. Furthermore, results from a neurolinguistic case study (Sahel, Nottbusch, Blanken, & Weingarten, 2005) demonstrate that syllabic structures are based on postlexical phonological processes. Taken together, these results show that prelingually deaf people do have access to suprasegmental phonology; it also points to the abstract nature of phonological processes being accountable for syllabic structures. Another question that arises with the syllable effect concerns the role of the mental lexicon. As mentioned above, delays at syllable boundaries do not correlate with word frequency. This is seen as the first hint at an effect that is independent of the lexicon. This assumption is supported by results from studies using pseudowords as stimuli (Weingarten et al., 2004, Nottbusch et al., 2005); it is found that syllable boundaries were roughly of the same size as in German words. As pseudowords cannot have a lexical entry, their syllabic structures cannot be derived from the mental lexicon. Therefore, it can be assumed that there is a separate mechanism responsible for the generation of the syllabic structure. Moving down the processing hierarchy, from the graphemic word to lexical constituents and then to the syllable tier, we can now ask whether there are even smaller processing units such as syllable constituents. In Weingarten et al. (2004), it was shown that syllable onsets with two consonants lead to longer interkey intervals than syllables with only one onset consonant. This indicates a split of syllabic units into onsets and rhymes. As the study did not directly address this question, this finding needs further corroboration. A further split of syllables needs to be reported, although it cannot be associated with linguistic word structures. If a syllable contains more than five letters — in German there are many syllables of this length (for example in the CELEX database containing 8854 different syllables, 2720 have more than five letters (Baayen, Piepenbrock, & van Rijn, 1993)) — they tend to be split after about the fourth letter. Presumably, this indicates a limitation of the motor buffer. In the paradigm of time course measurements of word writing we want to present here, no further processing units can be reported yet. Nevertheless, we want to hint at results from other methodological approaches such as error analysis in brain damaged (e.g., Ward & Romani, 2000; McCloskey, Badecker, Goodman-Schulman, & Aliminosa, 1994; Caramazza & Miceli, 1990) and healthy individuals (e.g., Logan, 1999), and word completion tasks (Weingarten, 2005) that provide evidence for graphemes as processing units below the syllable tier and above the terminal letter tier. The results from time measurements can be summarized as follows: When a writer commences to write a word, graphemic information for the onset of the word is generated as well as word structure information, whereas the segmental information is certainly not completely specified. Instead, in a hierarchy of processing steps, segmental information is retrieved and, in the case of complex words, a re-access of the whole word form is


36

9/27/2006

10:15 AM

Page 36


necessary. The processing system requires the splitting of the graphemic words into relatively small units to make them operable for the output system. Preferably, the design of these units is in accordance with the linguistic word structure. 3.2.2. Written Sentence Production Very few studies dealing with temporal aspects of the written production of linguistic units larger than words (phrases, clauses, sentences, paragraphs, and texts) have been reported (Chanquoy, Foulin, & Fayol, 1990, 1995; Matsuhashi, 1981, 1982). The main purpose of these studies was to determine whether planning and production processes during writing proceed according to syntactic constituents like phrases, clauses, or sentences. This was accomplished by exploring the relation between pause durations and these linguistic units. In a single study, Matsuhashi (1982) analyzed, on the basis of videotapes, pauses made by an adult during text writing. From two texts produced under two different writing conditions (generalizing and reporting), she singled out the 10 longest pauses in the first 100 words. The author divided the pauses into those occurring at a sentence boundary and those occurring within the sentence. In the generalizing condition, 5 out of 10 pauses occurred at sentence boundaries, while the remaining five pauses took place within sentence boundaries. In the reporting condition, only three pauses occurred at sentence boundaries, while the remaining seven pauses took place within sentence boundaries. This finding led the author to the conclusion that planning processes during writing were not influenced by linguistic units. However, this conclusion must be considered a little premature. The difference made by the author concerning the classification of the pauses is fairly rough: the internal pauses were considered without any further differentiation. Chanquoy et al. (1990, 1995), using a much more detailed typology of linguistic units in their analyses of pauses, reached a different conclusion regarding the influence of syntactic constituents on the pause length. Chanquoy et al. (1990) video recorded 16 adults writing a text. They found a relation between the pre-writing time (the time elapsing between the end of the instruction and the beginning of graphic activity) and syntactic complexity. The pre-writing time was higher when participants were asked to write three propositions as a single sentence than as three separate sentences. They further found an interaction between syntactic complexity and pause length after the second proposition: pauses were longer when the participants wrote three sentences instead of one, indicating that the third proposition was planned ahead of time only when participants were asked to write one sentence. In a further study with 10 adults (Chanquoy et al., 1995), the authors found a correlation between the pause length and the extent of the following units (paragraph, sentence, clause, phrase, word): the larger the unit, the longer the preparation time. The longest pauses were measured at the initial locations of paragraphs, sentences, and clauses. The authors took this finding as an indication that the initial locations of these syntactic constituents are the main points for the planning and production processes. All these studies investigated the temporal aspects of written production by analyzing videotapes of handwriting. In the study presented here, we expanded our word writing paradigm (Section 3.2.1) on sentence production and explored the temporal aspects of written production by analyzing the time course of the typing of single sentences in German adults.


9/27/2006

10:15 AM

Page 37


37

Our main aim was to ascertain the effect of different syntactic structures on written sentence production. The main hypothesis was that the syntactic structure of a phrase or a sentence influences the planning process. To verify this hypothesis, we compared the production of coordinated and subordinated noun phrases. We expected the planning processes for these two structures to be allocated to differing locations in different ways, becoming evident in different time courses. Furthermore, we hypothesize that planning a syntactic structure benefits from syntactic structures that have been used shortly before. New structures were assumed to require longer planning times than structures being activated already. Therefore, the syntactic number of the noun phrases was varied. To explore the impact of task variation on the planning and production processes in written sentence production, participants were asked to type the same sentences after two presentation modes: pictorial presentation (picture describing) and written sentence presentation (copying). In the case of picture describing, participants had to compose a sentence describing the spatial order of several geometric shapes. In the second mode, participants copied the sentence presented. The hypothesis was that producing a sentence using the pictorial stimulus should lead to longer delays in typing than simply copying the sentence because of the additional cognitive load.

3.3. Method 3.3.1. Participants Nineteen students from the University of Osnabrück took part in the experiment (15 female, 4 male). All participants were native speakers of German, aged between 21 and 32 (M: 24.1, SD: 3.2), and all were naïve with respect to the purpose of the experiment. 3.3.2. Stimuli The participants were asked to type 24 German sentences twice: after a pictorial presentation (picture describing) and after a written presentation (copying). The sentences were made up of two different syntactic structures. The two structures differed in respect of their subject noun phrase. The subject noun phrase of the first type of sentences (N = 12) consisted of two coordinate noun phrases which were linked by the conjunction [and] (corresponding to Figures 1A and 1B), while the subject noun phrase of the second sentence type (N = 12) consisted of a noun phrase which was linked by the preposition <mit> [with] to a second, subordinate, noun phrase (corresponding to Figure 1C and 1D). In both types of sentences, the grammatical category ‘number’ of the predicative (3rd) noun phrase was varied, so that the predicative noun phrase was singular in half of them (Figures 1A and 1C) and in the other half it was plural (Figures 1B and 1D). 3.3.2.1. Pictorial stimuli In the picture-describing task, six colored shapes served as stimuli: a blue star, a green cross, a brown rectangle, a black arrow, a red triangle, and a yellow circle. To compel the participants to produce the to-be-elicited sentences, six


38

9/27/2006

10:15 AM

Page 38


Figure 1: Sentence structures to be elicited by the corresponding elicitation pictures (A–D) shown in Figure 2. The shaded table elements indicate the linguistic units where the sentence structures differ from each other. combinations of these colored shapes were used. For each of these six combinations, the shapes occurred in four different constellations. The shapes that should elicit the subject noun phrase are separated by a line from the shape(s) that should elicit the verb phrase including the predicative noun phrase. The former shapes are located to the left of the parting line, while the latter shape(s) is/are located to the right. The constellations A and B in Figure 2 were designed to elicit sentences with two coordinate noun phrases as subject ( [the black arrows and the red triangles], corresponding to Figures 1A and 1B). The two constellations differed in the number of circles to the right of the parting line. A, the constellation with one circle, should elicit a predicative singular noun phrase, while B, the constellation with two circles, should elicit a predicative plural noun phrase.


9/27/2006

10:15 AM

Page 39


39

Figure 2: Examples of elicitation pictures with four different constellations (A–D) corresponding to the sentence structures shown in Figure 1. (The colour names on the shapes are just given for the grayscale illustration and were not shown in the original colored task.) The constellations in C and D in Figure 2 were designed to elicit sentences with subject noun phrases with a second, subordinate noun phrase ( [the black arrows with the red triangles]). The number of shapes in the second noun phrase (triangles in this example) had to be four, because the grammatical category ‘number’ should be plural. As in A and B, the constellations in C and D differ in the number of circles to the right of the separating line. 3.3.2.2. Written stimuli In the copying mode, 24 sentences were used as stimuli. They were the same 24 sentences which should be elicited in the picture-describing task. 3.3.3. Procedure The stimuli were presented on a 17⬙ computer screen with a resolution of 1024 ⫻ 768. In the pictorial mode, the stimulus picture, having a size of 475 ⫻ 210 pixels, appeared on the upper half of the screen one second after the return key was pressed. The pictorial stimulus remained displayed during the typing of the sentence. The text typed by the participant appeared in the lower half of the screen. The participants started the next trial by pressing the return key. The times elapsing between the appearance of the stimulus and the beginning of the typing activity (initial latency) as well as between successive keystrokes on the keyboard (interkey intervals) were measured. In the pictorial mode, each participant was presented with 24 pictures and was asked to type 24 sentences using one of the four syntactic structures they learned during the twophase practice preceding the main test. In the first phase, the participants were shown the six different-colored shapes used in the main test, one after the other, and asked to name each of them by producing a noun phrase containing a definite article, a colour adjective, and a noun (e.g., <der schwarze Pfeil> [the black arrow]). In the second phase, the participants were trained in the experimental conditions. They were presented with eight pictures, one after the other, similar to those used in the main test. These eight pre-test


40

9/27/2006

10:15 AM

Page 40


pictures were not part of the main test. Each of the four different constellations depicted in Figure 2 appeared twice. On the basis of these pictures, the task was accomplished. The participants were asked to name the shapes to the left and right of the separating line by producing noun phrases similar to those specified in the first phase of the training. Subsequently, they were instructed to connect the two shape noun phrases to the left of the separating line by the conjunction [and], if the shapes were next to each other and by the preposition <mit> [with], if one shape was placed on top of the other shapes (see Figure 1: A and B vs. C and D). Finally, participants were asked to produce the complete sentence by connecting the noun phrases for the shapes located to the left of the separating line with the noun phrase for the shape located to the right by the copula <sind> [are]. They were informed that the whole sentence is the target sentence, i.e., the sentence they should type when they are presented with a picture in the main test. If necessary, the practice phase was repeated. The experimental procedure for the copying task was identical to that of the picturedescribing task described above with the exception that written sentences (without line break) instead of pictures appeared as stimuli in the upper half of the computer screen. Again, participants were asked to type 24 sentences, which were presented in a random order. The typed text appeared in the lower half of the screen (same font, same size). The stimulus sentence disappeared from the screen as soon as the participant pressed the return-key after completing the sentence, the same as with the picture in the picturedescribing task. After one second, the next stimulus sentence was displayed. The order of the two presentation modes was counterbalanced over subjects, i.e., one half started with the picture-describing task, the other half with the copying task.

3.4. Results Because typing speed varied over participants (average typing rate (in characters per second): 2.44, ranging from 1.51 to 4.32), we went beyond traditional descriptive pause analysis and abstained from the approach of defining pauses by a certain temporal length. Accordingly, we did not count pause frequencies but analyzed all interkey interval data. The experimental design (all participants writing exactly the same sentences) allows for direct comparisons of performance differing only by the factors in question (presentation mode, word position in sentence, syntactic structure/features). 3.4.1. Errors In total, 912 sentences were obtained, 456 in each presentation condition. The number of sentences written entirely correctly was 511, 220 in the picture-describing task and 291 in the copying task. The occurrence of an error, if noticed by the typist, clearly influences further production (slowing down, correction, new orientation, etc.). As we cannot be totally sure, if a slip of the key has been noticed by the typist or not, our analyses are based on errorless writing. Sentences with errors were considered in the analyses only up to the word where the first error occurred, i.e., if a participant slips in the third word of a sentence, all data from the error to the end of the sentence were discarded. This results in


9/27/2006

10:15 AM

Page 41


41

50.092 single keystrokes being used in the analyses out of a possible 66.958. This meant a data loss of 25.2% due to errors (23.321 of 33.516 keystrokes in picture describing (data loss = 30.42%); 26.821 of 33.516 in copying (data loss = 19.98%)). Unfortunately, this procedure inevitably causes measurements to ‘thin out’ toward the end of the sentences, i.e., at the beginning of the sentence we usually obtained data from all 19 participants, but measurements for the full stop at the end of the sentence are represented by an average of 10.7 participants per sentence (picture describing: 9.2; copying: 12.2). 3.4.2. Patterns of Word Typing in Sentence Production In a post hoc approach, we tested whether within-word interkey intervals were affected by syllable+morpheme boundaries, as is found in single word typing tasks (Section 3.2.1). The words occurring in the sentences contained six digraphs (<ec>, , , , , ) that occurred at a syllable boundary as well as within-syllable simple letter transitions (most of the cases arose from singular/plural pairs, e.g., vs. ). Although the number of cases was quite low, ANOVAs, based on measurements averaged over participants with the main factor being type of boundary, showed a significant difference in both modes (picture describing: F(1, 104) = 14.647, p < .0005, copy: F(1, 104) = 18.244, p < .0001), within-syllable interkey intervals being typed faster than their syllable boundary counterparts (Figure 3A). Two digraphs occurred as within-syllable interkey intervals, as well as at combined syllable+morpheme boundaries ( and ). In both cases, time is measured before the character <e>, known as being the most frequent character in German (Meier, 1967) and one of the keys typed fastest (Weingarten et al., 2004). Still, the syllable+morpheme interkey intervals in the words and were prolonged compared to the within-syllable interkey intervals in or and or <Stern(e/en)> (Figure 3B). ANOVAs, with type of boundary as the main factor, indicate

Figure 3: Mean interkey intervals for digraphs occurring at syllable boundaries as well as within syllables (A) and for digraphs occurring at both syllable+morpheme boundaries and within syllables (B).


42

9/27/2006

10:15 AM

Page 42


significant differences in both modes (picture describing: F(1, 82) = 316.1, p < .0001, copy: F(1, 82) = 198.8, p < .0001). Another possible influence on within-word typing patterns could be the linear position of the word within a sentence. The colour adjectives were (apart from the sentence initial determiner) the only words that appeared in the same grammatical form (with the same inflectional ending ) in all sentences, i.e., the typed form was the same whether they were part of the first, the second, or the third noun phrase, irrespective of the number/position of the latter noun phrase. Figure 4 shows the time course for all 12 occurrences of the colour adjective [yellow] averaged over participants in the picture describing (A) and copying modes (B). The diagrams in Figure 4 indicate that the linear within-sentence position of the word does not affect the overall patterns. ANOVAs were performed for each within-word character separately with ‘within-sentence position of the word’ as the main factor. The differences failed to reach significant levels (all p > .1). Similar results were obtained for the other colour adjectives. Influences of within-sentence position on the word initial latencies will be discussed in the following paragraph. 3.4.3. Effects of Word Class and Position A closer look at word initial latencies shows that the latencies for the three nouns occurring in the stimulus sentences are constantly high — irrespective of within-sentence position — compared to the other word classes used (1st noun, picture describing: 889 ms, copying: 729 ms; 2nd noun, picture describing: 899 ms, copy: 746 ms; 3rd noun, picture describing: 897 ms, copy: 756 ms). In contrast, before the adjectives the word initial latencies increase with the within-sentence position (1st adjective, picture describing: 591 ms, copying: 443 ms; 2nd adjective, picture describing: 638 ms, copy: 546 ms; 3rd adjective, picture describing: 728 ms, copy: 586 ms), i.e., the further back the adjective is in the sentence, the longer the latency. In the case of the determiners, the sentence initial latency is at the same time the word initial latency for the first determiner and cannot be compared

Figure 4: Mean interkey intervals for the word occurring as 2nd, 6th, or 11th word in the sentence in the picture-describing task (A) and the copying task (B).


9/27/2006

10:15 AM

Page 43


43

directly with the others. The latency located before the 2nd determiner, though, is shorter than that before the 3rd in both modes (2nd determiner, picture describing: 452 ms, copy: 372 ms; 3rd determiner, picture describing: 550 ms, copy: 584 ms). Apart from the nouns, the word initial latencies for the verb are the longest in both modes (picture describing: 738 ms; copy: 829 ms). The most interesting time pattern, however, is that concerning the duration of latencies located before striking the space bar. At two positions within the sentence, the space latencies are nearly doubled compared to the remaining space latencies: the space before the 3rd word (conjunction or preposition) and the space before the 7th word (the verb) (Figure 5). 3.4.4. Effects of Presentation Mode, Ordination Type, and Number on the Time Course of Sentence Production Of more interest than total production times, however, were the influences on the time course of typing. As all 19 participants wrote the same 24 sentences and all sentences were constructed in the same way (using one noun phrase followed by a conjunction or a preposition and a second noun phrase, followed by a verb, a preposition and a third noun phrase; see Figures 1 and 2), it was possible to explore the effects of the three factors on the time course at every position in the sentences: from the first key up to the full stop. Accordingly, separate analyses were conducted for the sentence initial latency as well as the initial

Figure 5: Mean word initial latencies (IL) for all words, latencies before spaces, and mean word internal interkey intervals (IKI) split by ordination type (coordinated vs. subordinated) and presentation mode (picture describing vs. copying). For the sake of clarity, the sentence initial latency (SIL) and the factor number are omitted in this figure.


44

9/27/2006

10:15 AM

Page 44


latencies of the subsequent words, the mean word-internal interkey intervals of all words, as well as for the space prior to every word. Word-internal interkey intervals were grouped by their means, as we can assume that word internal patterns are not affected in a critical way (see Section 3.4.2). This results in 36 mixed-model ANOVAs, with presentation mode, ordination type, and number as fixed factors and participants as a random factor. The mean values and standard deviation split by the fixed factors as well as the significant effects are summarized in Table 1. 3.4.4.1. Modal effects The most striking effect is that sentence production in the picture-describing task is delayed compared to the copying task. This is not only the case at phrase or word initial positions but almost throughout the production of the whole sentence. There are only three positions within the sentences where the differences between both productions were non-significant or where the production in the copying task is even slower than the production in the picture-describing task. The first position is found in the region of the conjunction/preposition initiating the second noun phrase (Table 1: IL_W4). Here, the copying mode is slower than the picture-describing mode: F(1, 733.4) = 57.8, p < .0001; random factor participants, SD: 273.3, SE: 48.2, all other factors and interactions p > .05. The same is true for the initial latency of the 8th word, the verb (IL_W8): F(1, 618.1) = 8.7, p < .005; random factor participants, SD: 357.8, SE: 63.0, all other factors and interactions p > .1. Finally, no significant mode effect can be found at the beginning of the third noun phrase (space_9, IL_W10, IKI_W10). Interestingly, sentence production generally seems to slow down in these positions, as can be seen from the comparison of delays at spaces (space_3 and space_7, see above). But the slowdown is stronger and includes the initial latencies of the related words in the copying mode only, leading to a more distinct separation of the phrases. The additional delays present in the picturedescribing task, in contrast, are distributed over the complete production and less clustered. Interactions with the other main factors are only marginal, e.g., at the 3rd space, the interaction with ordination and the interaction with number arise from an extraordinary high mean value for the subordinated, plural case in the copying mode. 3.4.4.2. Effects of ordination The differences between the production time for sentences with a coordinated 2nd noun phrase (see shaded left part of Figure 1A and 1B) versus a subordinated 2nd noun phrase (Figure 1C and 1D) were significant at the following locations in the sentence: sentence initial, around the 4th and 5th word (conjunction/preposition and beginning of the 2nd noun phrase), and before word 11 and word 12. The three locations, however, can be differentiated by the direction of the difference. The production of the coordinated structure is initiated faster than the subordinated one (sentence initial latency (SIL): F(1, 873) = 7.1, p < .01; random factor participants, SD: 932, SE: 158, all other factors and interactions p > .05; mean interkey intervals for word 1 (IKI_W1): F(1, 855) = 5.5, p < .05; random factor participants, SD: 145.6 SE: 24.5, all other factors and interactions except main effect mode p > .05). The coordinated structure is also produced faster during the conjunction/preposition (IKI_W4: F(1, 731.2) = 34.2, p < .0001; random factor participants, SD: 101.5, SE: 17.2, all other factors and interactions except main effect mode p > .05), at the 4th space (F(1, 729.5) = 22.5, p < .0001; random


Table 1: Means (and standard deviations) for sentence initial latencies (SIL), word initial latencies (IL) for all words, latencies before spaces, and mean word internal interkey intervals (IKI) for presentation mode, ordination type, and number. Coordinated

Subordinated

Plural

Singular

Plural

Singular

Sign. effects

2092 (1040) 2110 (1147)

2251 (1316) 2281 (1468)

2142 (1022) 2259 (1683)

Ordination*

IKI_W1

Picture Copy

309 (145) 311 (150)

329 (166) 301 (151)

337 (182) 317 (203)

356 (228) 309 (159)

Mode** Ordination*

space_1

Picture Copy

269 (186) 225 (156)

291 (243) 213 (148)

299 (218) 220 (119)

330 (427) 220 (134)

Mode***

Picture Copy

551 (368) 419 (256)

624 (507) 444 (349)

605 (406) 448 (325)

583 (359) 456 (336)

Picture Copy

328 (165) 307 (144)

327 (150) 289 (123)

318 (144) 309 (149)

340 (151) 313 (167)

Picture Copy

372 (276) 262 (191)

397 (345) 302 (275)

423 (332) 280 (224)

383 (327) 263 (161)

IL_W3 (noun)

Picture Copy

886 (647) 699 (463)

939 (611) 702 (361)

822 (485) 767 (669)

903 (646) 745 (473)

Mode***

IKI_W3

Picture Copy

327 (151) 292 (117)

327 (142) 291 (126)

336 (162) 303 (137)

338 (139) 294 (116)

Mode***

Picture Copy

676 (457) 558 (468)

752 (678) 555 (431)

603 (435) 724 (666)

712 (481) 616 (521)

IL_W2 (adj)

space_2

space_3

Mode*** Mode *** Ordination× number*

Mode×ord Mode× number*


IKI_W2

Mode***

Page 45

1934 (1013) 2194 (1319)

10:15 AM

Picture Copy

9/27/2006

SIL (det)

45

Subordinated

Plural

Singular

Plural

Singular

Sign. effects

539 (308) 790 (512)

516 (334) 694 (487)

591 (549) 804 (648)

Mode***

IKI_W4

Picture Copy

289 (123) 251 (102)

296 (131) 258 (108)

307 (134) 298 (122)

329 (150) 314 (166)

Mode** Ord***

space_4

Picture Copy

253 (169) 242 (273)

267 (193) 215 (140)

328 (221) 256 (135)

337 (275) 297 (208)

Mode** Ord***

IL_W5 (det)

Picture Copy

461 (378) 339 (327)

477 (342) 414 (529)

415 (272) 342 (301)

451 (397) 392 (294)

Mode**

IKI_W5

Picture Copy

291 (171) 227 (120)

277 (177) 248 (132)

339 (432) 255 (147)

286 (104) 275 (168)

Mode*** Ord*

space_5

Picture Copy

297 (206) 218 (142)

268 (179) 243 (180)

326 (238) 249 (146)

330 (215) 286 (222)

Mode*** Ord***

IL_W6 (adj)

Picture Copy

610 (526) 466 (316)

608 (477) 594 (493)

619 (500) 514 (348)

714 (479) 608 (440)

Mode** Number*

IKI_W6

Picture Copy

332 (138) 312 (146)

321 (147) 292 (115)

331 (163) 297 (138)

338 (143) 328 (151)

mode* ordination× number*

Picture Copy

397 (319) 318 (247)

399 (282) 318 (286)

409 (456) 331 (287)

430 (353) 326 (294)

mode***

Picture Copy

960 (654) 745 (417)

855 (692) 689 (343)

899 (564) 768 (453)

879 (478) 778 (474)

space_6

IL_W7 (noun)

mode***

Page 46

589 (447) 831 (538)

10:15 AM

Picture Copy

9/27/2006

IL_W4 (con/pre)


Coordinated


46

Table 1: (continued)


Table 1: (continued) Coordinated

IKI_W7

IL_W8 (verb)

Space_8

Space_9 Picture

Singular

Sign. effects

Picture Copy

314 (140) 295 (132)

320 (133) 297 (119)

324 (148) 309 (141)

330 (140) 311 (125)

mode*

Picture Copy Picture Copy

667 (594) 667 (516) 772 (597) 863 (821)

698 (617) 682 (592) 770 (561) 824 (585)

584 (477) 620 (579) 688 (388) 788 (629)

609 (490) 724 (543) 716 (491) 840 (548)

Picture Copy

307 (130) 273 (139)

307 (133) 273 (110)

286 (127) 273 (111)

302 (129) 268 (105)

Picture Copy

256 (233) 216 (185)

303 (341) 200 (81)

271 (172) 229 (263)

262 (182) 234 (189)

mode**

Picture

546 (549)

585 (495)

615 (486)

537 (389)

mode***

Copy

499 (389)

376 (187)

449 (279)

486 (410)

mode×ord× number*

Picture Copy

291 (131) 238 (121)

289 (138) 247 (124)

274 (149) 251 (116)

277 (144) 263 (138)

Picture 406 (260) Copy 332 (262) 574 (397) 529 (389)

335 (241) 326 (231) 527 (400)

398 (313) 389 (411) 570 (554)

399 (396) 413 (657)

Copy

666 (631)

517 (477)

504 (389)

number* 643 (525)

mode** mode***

mode***

IL_W10 (det)

Page 47

IKI_W9

Plural


IL_W9 (prep)

Singular

10:15 AM

IKI_W8

Plural

9/27/2006

space_7

Subordinated

47

Subordinated

Plural

Singular

Plural

Singular

Sign. effects

329 (253) 341 (239)

291 (153) 280 (196)

298 (126) 326 (220)

number*

space_10

Picture Copy

365 (334) 243 (180)

325 (215) 292 (270)

289 (144) 243 (113)

293 (148) 263 (148)

mode*** ord*

IL_W11 (adj)

Picture Copy

786 (577) 573 (484)

680 (456) 598 (375)

609 (484) 574 (475)

832 (654) 595 (358)

mode***

IKI_W11

Picture Copy

368 (153) 320 (147)

329 (141) 309 (148)

327 (149) 326 (143)

353 (155) 322 (143)

mode*

space_11

Picture Copy

496 (419) 280 (191)

423 (427) 268 (177)

370 (311) 284 (191)

378 (276) 308 (251)

mode***

IL_W12 (noun)

Picture Copy

1065 (707) 742 (443)

883 (718) 727 (447)

865 (616) 785 (458)

761 (447) 766 (408)

mode** ord*

Picture Copy Picture Copy

345 (142) 273 (124) 795 (545) 677 (466)

332 (124) 314 (127) 799 (485) 653 (528)

328 (140) 312 (139) 773 (502) 749 (477)

345 (144) 322 (128) 763 (622) 679 (456)

mode** number**

IKI_W12 full stop

Note: Significant effects are noted in the rightmost column, *p < .05; **p < 0.005; ***p < .0005.

Page 48

323 (152) 277 (206)

10:15 AM

Picture Copy

9/27/2006

IKI_W10


Coordinated


48

Table 1: (continued)


9/27/2006

10:15 AM

Page 49


49

factor participants, SD: 117.1, SE: 20.5, all other factors and interactions except main effect mode p > .1), during the determiner of the 2nd noun phrase (IKI_W5: F(1, 719.6) = 4.02, p < .05; random factor participants, SD: 108.2, SE: 19.2, all other factors and interactions except main effect mode p > .05), and at the 5th space (F(1, 716.4) = 12.9, p < .0001; random factor participants, SD: 119.7, SE: 20.8, all other factors and interactions except main effect mode p > .05). During the 3rd noun phrase, however, the effect is reversed. At the 10th space, and before the final noun (word 12), the coordinated structure is produced slower than its subordinated counterpart (space_10: F(1, 557.1) = 6.7, p = .01; random factor participants, SD: 109.6, SE: 19.8, all other factors and interactions except main effect mode p > .1; IL_W12: F(1, 513.6) = 4.38, p > .05; random factor participants, SD: 288.2, SE: 52, all other factors and interactions except main effect mode p > .05). In short, at the beginning of the sentence and at the beginning of the second noun phrase coordinated structures can be produced faster than subordinated ones. During the third noun phrase subordinated structures seem to have a timing advantage. Interactions with the other main factors were, again, only marginal. 3.4.4.3. Effects of number Number was varied in the 3rd noun phrase only. Although a significant effect of number was found for the initial latency of the 6th word, the colour adjective of the second noun phrase, F(1, 702.7) = 6.2, p > .05; random factor participants, SD: 161.4, SE: 31.7, all other factors and interactions except main effect mode p > .1. If the number of the following noun phrase was plural (as in the shaded right part of Figure 1B and 1D), the initiation of the word was faster than if it was singular (as in Figure 1A and 1C). Effects of the same kind, the plural form being produced faster than singular form, were found during the 3rd noun phrase for the initial latency of the 10th word (the determiner), the mean interkey intervals for the same word, and for the 12th word (the noun). IL_W10: F(1, 576.2) = 4.5, p > .05; random factor participants, SD: 238.2, SE: 44, all other factors and interactions p > .05; IKI_W10: F(1, 572.8) = 6.2, p > .05; random factor participants, SD: 114.6, SE: 20.5, all other factors and interactions p > .1; IKI_W12: F(1, 508.2) = 8.1, p = .005; random factor participants, SD: 105.8, SE: 18.1, all other factors and interactions except main effect mode p > .5. Recapitulating this means that, at some locations during the production of the 3rd noun phrase, plural forms could be generated faster than singular forms.

3.5. Discussion In the present study, we investigated the temporal aspects of written sentence production by using an experimental paradigm that has previously been found to be successful in assessing the time course of single word typing. The expansion of this experimental paradigm, from the word level to the sentence level, allowed us to gain insights into the time course of word internal typing patterns in a sentence context compared to isolated word typing; furthermore, we can inspect influences of syntactic structures on word and phrase initial latencies.


50

9/27/2006

10:15 AM

Page 50


The analyses at word level revealed a similar pattern of performance to that found in single word typing tasks. The time course of word typing was affected by ‘pure’ syllable boundaries as well as combined syllable+morpheme boundaries. The interkey intervals for digraphs located at syllable boundaries were longer than for identical digraphs occurring within syllables. The interkey intervals for digraphs occurring at syllable+morpheme boundaries were also longer than for corresponding digraphs located within a syllable. For both comparisons, the analysis revealed a significant difference. Furthermore, it has been found that within-word typing patterns were not influenced by the linear position of the word in the sentence as indicated by the similar patterns of performance found for the colour adjectives occurring in the first, second, and third noun phrase positions of the elicited sentences. The finding that the general temporal patterns of word production are very similar in tasks requiring the typing of isolated words and in those requiring the typing of words occurring in a sentence context indicates that the lexical, as well as the postlexical, mechanisms involved in the written word production are extraordinarily stable. Comparing word-level and sentence-level written production, one can state that the sentence initial latency for written sentence production can be compared to the word initial latency for word writing: both initial latencies are strongly prolonged compared to subsequent interkey intervals. At the sentence and phrase level, the analyses revealed that the time course of written production is influenced by two main factors: the nature of the syntactic structure of the sentence and the presentation mode. The influence of the syntactic structure on the time course of sentence typing is revealed by the important finding that the production time for sentences with coordinate subject noun phrases (as in Figures 1A and 1B) was shorter than for those with a subordinate subject noun phrase (as in Figures 1C and 1D) at several locations in the sentences before the verb. The production time for sentences with a subordinate subject noun phrase was longer initially, at the beginning of the prepositional phrase (containing the 2nd noun phrase) as well as within the phrase. This finding indicates that sentence planning does not only take place at the initial locations of sentences and phrases but also within these major syntactic constituents. This is in line with the result of Chanquoy et al. (1995) who found that pauses during handwriting occur at sentence and phrase boundaries as well as at word boundaries. However, our finding that the syntactic structure has an impact on the time course of sentence typing did not confirm the finding of Chanquoy et al. (1995) that the nature of the syntactic structure of a sentence, or phrase, did not influence the pause duration. This discrepancy could be attributed to differences concerning the experimental paradigms and in the accuracy of the measurements used in the present study and in those of Chanquoy et al. (1995). In addition, the data support our hypothesis that a syntactic structure, activated and used shortly before, can be accessed easier and faster than a structure that has to be constructed. This claim is supported by two findings. First, in the sentences with a subordinate subject noun phrase (see Figures 1C and 1D), the noun phrase and the preposition <mit> constitute a prepositional phrase. This prepositional phrase has the same internal syntactic structure as the prepositional phrase to be produced after the verb. This seems to ease the production of the prepositional phrase after the verb, as the production of the subordinated structure is


9/27/2006

10:15 AM

Page 51


51

faster than that of the coordinated one at the 10th space (before the colour adjective) and initiating the final noun. Therefore, it can be assumed that the production of the prepositional phrase, following the verb, causes less cognitive load for the sentences with subordinate noun phrases compared to those with coordinate noun phrases (Figures 1A and 1B). Second, this pattern corresponds to traces of syntactic pre-planning becoming evident through the significant effect of the number of the predicative (3rd) noun phrase at the end of the sentence on the initial and subsequent keystrokes of the determiner and the mean interkey intervals for the noun. Here, the production was faster when the number of the 3rd noun phrase was plural (equivalent to that of the 2nd noun phrase, as in Figures 1B and 1D) and slower when it was singular (different to that of the 2nd noun phrase, as in Figures 1A and 1C). Again, this effect could be attributed to the facilitation of a previously constructed syntactic structure. Concerning the influence of the presentation mode it was found that the typing time for sentences in the picture-description task was longer than for the copying task, the effect being spread over almost the whole sentence. This finding indicates a larger cognitive load for the picture-description task. This assumption is supported by the finding that the picture-description task was more error-prone than the copying task. On the other hand, a comparison of the time course of the picture-description task and the copying task indicates that the planning and production processes were not identical in the two modes. In the copying task there seems to be a clear division into three production entities, corresponding to major syntactic constituents: the first noun phrase, the second noun phrase or prepositional phrase initiated by the conjunction/determiner, and the verb phrase including the third noun phrase. In contrast, the division of entities is less clear in the picture-describing task. Here, the additional load on the production system, obviously related to object identification and subsequent lexical activation, seems to be distributed over the whole production process, even at the letter level, leading to smaller steps in sentence production. These findings pose severe problems for models that assume that the linguistic processes involved in producing sentences are strictly serial. Instead, processes of different hierarchical statuses have to be assumed to work in parallel with motor execution. The results clearly indicate that, in both modes — similar to single word writing — not all information needed to write a sentence or phrase is available at the onset. Instead, the planning processes seem to continue within these syntactic constituents. Moreover, these planning processes were reflected in a distributed fashion, i.e., over a series of successive keystrokes, sometimes spanning more than one word (e.g., the effect of ordination on the keystrokes around the conjunction). This becomes even clearer in the picture-describing task, where the general production time was longer than in the copying task.

3.6. Summary and Further Perspectives The measurement of the time course of written language production has been proven to be a very useful methodology for the study of cognitive processes involved in this type of action. This is true, especially for typing, because in this mode of writing we obtain timing data for each letter transition. It has been shown in a series of experiments that typing


52

9/27/2006

10:15 AM

Page 52


does not proceed as a steady production of segmental units, e.g., letters or words, in equal time steps. Instead, words and sentences have to be split into smaller linguistic units, presumably due to the limitations of processing modules. The size of these units, delays at their onsets, and correlations of these delays with, for example, frequency distributions, allows for the temporal specification of cognitive processes in writing. In this respect, temporal studies of word and sentence writing are far more advantageous than of speaking, due to the fact that, in the latter case, no empirical evidence has yet been found to point to the fact that the time course of production is determined by cognitive processes going on during motor execution. In most studies time measurements in spoken word production are confined to initial latencies (reaction times), prior to the beginning of the motor execution. On the other hand, in spoken sentence production pausing has been investigated extensively but here too, time structures do not result from cognitive processes exclusively but also depend on the conditions of communicationtoo. A major question in time measurements addresses the types of units that are observable in motor execution. In our studies, we could not support results presented by Ostry (1983) who found that, at the word level, there is always a delay after about four letters with the number of letters being the major determinant. Thus, we deny that the observed chunking in writing proceeds as a direct function of counting segments. Certainly, there are limitations of motor buffers determined by the sheer size of a unit, but they only come into play when no linguistically motivated segmentation is at hand. At the word level, this is the case when a syllable is made up of more than about five letters. Here we find a split of a syllable without observing linguistic structures (e.g., syllable constituents). The most important determinant of dynamic units in word writing is in accordance with linguistic word structure. Two major processing steps shall be mentioned here: lexical processes and post-lexical processes. The mental lexicon is addressed word initially and also, in the case of complex words, word internally. We call those sub-word units lexical constituents, indicating that they depend on stored lexical information (Weingarten et al., 2004). Post-lexical processes lead to the formation of still smaller processing units which correspond with syllables. We reported neurolinguistic evidence that, on the one hand, the syllabic structures are based on postlexical phonological processes (Sahel et al., 2005). On the other hand, evidence from deaf writers suggested that these phonological processes do not depend on spoken language experience (Nottbusch, et al., 2005). Hence, we conclude an abstract level of postlexical phonological syllabification. Time structures of written sentence production are determined by the properties of the words involved and also by superordinate conceptual and syntactic tasks. For experimental research, this poses the methodological problem of having to control far more factors. In the study presented here, it has been shown that general temporal patterns of word production are the same in various conditions of isolated word writing and word writing in a sentence context. This indicates an extraordinarily stable lexical, as well as post-lexical, mechanism involved in written word production. Temporal patterns of written sentence production can be characterized by three aspects: (1) an important part of sentence planning is accomplished sentence initially; (2) major syntactic constituents lead to delays around their onsets; and (3) additional and/or parallel processes can cause a general slowdown of the production process.


9/27/2006

10:15 AM

Page 53


53

In future studies, five main goals should be pursued: 1. Time measurements provide clear evidence for the cognitive load posed by different syntactic constructions and conceptual tasks and the exact place in time, where this load is operant. Future research should investigate various syntactic patterns with the aim of developing a dynamic model of written sentence generation. 2. The combination of time measurements with methods of cognitive neuroscience appears to be very promising. Brain imaging as well as electrophysiology can clarify the neural substratum of behavioral data gained by time measurements of the type proposed here. 3. A conjunct analysis of typing or handwriting movements and eye movements during sentence production has shown great potential (cf. Alamargot et al., this volume). By combining both techniques in combination, we would be able to decide if pausing is triggered by eye movements toward the object in question (e.g., in case of loss of information in working memory) or not. 4. Performance in picture describing and sentence copying will probably be influenced in different ways by concurrent visual and verbal memory tasks. Presumably, error rate and production speed in picture describing will be further augmented by an additional task backed by visual working memory than by an additional task using verbal working memory alone. Opposite, or equivalent, effects of both concurrent tasks can be expected in the copy task (cf. Kellog et al., this volume). 5. As we do not write isolated sentences in real life but texts, models of written sentence production have to be integrated into models of text production. From a methodological point of view, this poses the problem of how the multiple factors at work can be controlled. Mixtures of problem solving and preconditioned writing tasks could be a compromise solution.

Acknowledgments We thank three anonymous reviewers for their constructive suggestions and Philip Cummins for checking the final version. This research was supported by a grant from the German Science Foundation (DFG, Schwerpunktprogramm ‘Sprachproduktion’) to Professor Rüdiger Weingarten.



9/27/2006

11:05 AM

Page 55

Chapter 4

Influence of Typing Skill on Pause–Execution Cycles in Written Composition Rui Alexandre Alves, São Luís Castro, Liliana de Sousa and Sven Strömqvist

It is well known that the cognitive cost of programming motor movements in writing can be considerably high if execution is not automatized. However, it is not clear how this cost might affect the on-line production of a written text, namely the distribution of pauses vs. execution periods. Narratives were collected using ScriptLog. Keystroke interval within a word was measured and used to distinguish between fast typists — for whom execution was presumably automatic, and slow typists — for whom execution required attention. The relative distribution of pauses vs. execution periods between two consecutive pauses was examined. Results showed that the time ratio between pauses and execution differs between groups. Relative to fast typists, slow typists make more pauses, and have shorter execution periods. These results are discussed in light of two phenomena: the trade-off between execution and formulation processes, and the adoption of serial vs. parallel ways of composing.

4.1. Introduction There is agreement in writing research that motor execution can have a cognitive cost (Bereiter & Scardamalia, 1987; Cooper & Matsuhashi, 1983; Fayol, 1999; Graham & Harris, 2000; Kellogg, 1996; Martlew, 1983; McCutchen, 1996). Nevertheless, there are few attempts to assess this cost. Exceptions are studies using written serial recall tasks (Bourdin & Fayol, 1994, 2000; Penney & Blackwood, 1989), and, more recently, written composition tasks (Kellogg, 2001; Olive & Kellogg, 2002). These studies, reviewed below, have shown that the cognitive burden from motor execution can be detrimental to both children and adults. However, it remains unclear how the on-line production of a written text is affected by motor execution skills. Here, we review this

Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Alves, R.A., Castro, S.L., & de Sousa, L. (2007). Influence of typing skill on pause–execution cycles in written composition. In Rijlaarsdam, G. (Series Ed.); M. Torrance, L. van Waes, & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 55–65). Amsterdam: Elsevier.


56

9/27/2006

11:05 AM

Page 56

Rui Alexandre Alves et al.

question, and report a study in which we explore how different levels of typing skill affect the on-line production of a narrative text. Penney and Blackwood (1989) have asked college students to recall lists of digits either by typing, or by handwriting. They found a decrease in the recall of the last serial positions of the lists when the responses were typed, but not when they were handwritten. The suppression of the recency effect was attributed to the participants’ low typing skill. Similarly, Bourdin and Fayol (1994) have used serial recall and compared spoken and handwritten responses from adults and children. They found that children performed worse than adults if the responses were handwritten, but not if they were spoken. They explained this finding as the lack of automaticity in low-level writing processes, such as handwriting and spelling, in beginning writers. Convincingly, they have supported this interpretation by showing that the performance of adults could be brought to levels similar to those of the children if they were required to write with untrained cursive capital letters. Thus, it seems that if attention has to be divided between the execution of untrained motor programs and the maintenance of memory traces, trade-offs are likely to occur and performance deteriorates. Situations of divided attention are paramount in text production. In recent years, this feature has been captured by an increasing focus of writing research on working memory (WM). A good illustration is Kellogg’s (1996) model which incorporates the demands of writing processes like formulation, execution, and monitoring on the multi-component WM model proposed by Baddeley and Hitch (1974). Regarding the execution process, Kellogg asserts that resources from the central executive are needed to program the motor movements in writing, but he adds that “execution can, when well-practised, proceed virtually automatically” (p. 59), thus allowing a more efficient management of demands from the formulation and monitoring components. However, if execution is not automatized, the simultaneous operation of the two other components might be impaired or impossible. Two recent studies tested this prediction of the model. Kellogg (2001) addressed the question whether writing components compete for the same WM resources, using a reaction time (RT) interference paradigm (RTs to auditory probes are collected in single task — baseline, and while writing; interference RTs are taken as estimates of the spare capacity, higher scores indicating less available resources; for a complete description of this procedure see Olive, Kellogg, & Piolat, 2001; and for recent implementations, see chapters in this volume by Piolat, and by Kellogg, Olive, & Piolat). Kellogg (2001) manipulated the demands of planning by varying the type of text to be composed (narrative, descriptive, or persuasive), and the demands of execution by varying output mode (handwriting, or typing on a keyboard). He found that when planning demands are relatively low, as in writing a narrative as opposed to a persuasive text, interference scores are smaller not only during planning, but also during execution and monitoring. Similarly, when the execution cost was lifted, as in writing by hand as opposed to typing, interference scores were lower. These findings suggest that different writing components share a common pool of resources (i.e. the central executive in Baddeley’s terms), so that if a given component requires less capacity, others can make use of it. However, because writing is typically a demanding and effortful task, competition among writing processes is most often the case. How do adult writers manage to produce a text when required to use an untrained response mode, thus having to deal with the cognitive cost of execution? Olive and


9/27/2006

11:05 AM

Page 57

Typing Skill and Pausing

57

Kellogg (2002) asked third graders and undergraduates to write a persuasive text and then to copy it. Half of the adults composed and copied the texts using their usual cursive script, and the other half used an unpracticed uppercase script. RT interference was measured in three conditions: (1) while copying (transcription), (2) in writing while pausing for longer than 250 ms (composition), and (3) in writing and not pausing (transcription ⫹ composition). Olive and Kellogg found less RT interference in the transcription condition of adults using cursive script. This finding indicates that, for adults, the execution of handwriting is automatized and allows other processes to be activated simultaneously. Indeed, the highest RT interference (in adults, cursive script) was found in the transcription ⫹ composition condition. A different picture emerged from the children’s results, where transcription yielded the highest interference. Thus, children devoted more resources to motor transcription than to composition. The interference score of the transcription ⫹ composition condition was intermediate, and not reliably different from the other two. This indicates that transcription per se overloads the attention capacity of children, who may not be able to activate other writing processes during motor execution. Interference scores in the adult group writing in uppercase (the unpracticed script) were similar in the three measurement conditions. This indicates that the presumably more effortful execution exerts a toll on all other components of the writing process. Olive and Kellogg suggest that when execution is less practiced writers might strategically alternate between planning, monitoring, and execution, that is, they would adopt a serial mode of composing. Further evidence in favor of a serial mode of composing under highly demanding execution comes from Olive and Piolat (2002), who found that suppressing visual feedback during a composition task leads to similar interference RTs whether the writers were pausing or handwriting, thus showing a similar pattern to the writers using the uppercase script in the other study. The studies reviewed above demonstrate that motor execution affects writing processes, and, overall, point to the role of typing proficiency. For example, if heightened demands from motor execution lead to the adoption of a serial mode of composing, then this is the mode that adults who are not proficient typists should use. In order to investigate this issue, however, we must be able to measure typing proficiency. Strömqvist (1999) proposed that, in a composition task, the median keystroke interval within a word is the most reliable indicator of typing proficiency. The reasons for this are that within-word strokes are very common and fast, and their timing is marginally influenced by planning or monitoring. Here, we called this measure typing speed, and used it to distinguish between slow and fast typists. We assumed that for slow typists execution is resource demanding, and for fast typists execution is virtually automatic. We distinguished these two levels of typing skill on the basis of a median split procedure, and explored the composition process in both groups. The rationale reviewed above instigated us to query for differences in the distribution of pauses and execution periods. One of the most striking observations of a writer producing a text is that huge amounts of time are spent not “writing”, this is, not executing typing or handwriting. For instance, Wengelin (1999) reported that college students spend 41% of their writing time in pauses longer than 2 s. Why do writers spend so much time pausing? What are they doing while pausing? As noted by Schilperoord (2001), writers can pause for several reasons: physical causes (e.g. fatigue, motor execution of typing or handwriting), socio-psychological causes (e.g. writer’s block, daydreaming), or cognitive causes (e.g. writing processes,


58

9/27/2006

11:05 AM

Page 58


cognitive overload). As pauses due to physical reasons are usually very brief, Schilperoord suggested a cut-off value of 1 s in order to exclude them from analysis. Specifically addressing writing on a keyboard, Wengelin (1999, see also this volume) argued that a good pause criterion should take into account the typing skill of the writer, and should be set well above the typing speed of the slowest writers. In our study, a pause criterion of 1 s would be too low to account for the pauses in the slow group. Thus, to make a fair comparison, we raised the cut-off value to 2 s. This is also a very common pause threshold (e.g. Levy & Ransdell, 1995; Severinson-Eklundh & Kollberg, 1996a; Strömqvist & Ahlsén, 1999), which is particularly suitable to examine high-level processes in writing. Although what happens during pauses is important, it is at best only half of the picture. One should also look at periods between consecutive pauses — what we call execution periods. It is probably in execution periods that storage and processing demands are higher, and where Flower and Hayes’s portrayal of a writer as “a thinker on a full-time cognitive overload” (1980b, p. 33) is most accurate. While typing, writers must literally keep in mind the representation of what they intend to write, pay attention to the output being produced, maybe plan further segments or revise the already written ones, or even pay attention to finding the keys on the keyboard. As discussed above, all these functions are likely to involve the central executive, whose capacity is well known to be limited (Baddeley, 1996, 2000). How does a slow typist manage this situation where a limited amount of resources has to be distributed among so many processes? One possibility is that slow typists have more difficulty in sustaining execution periods for as long as fast typists, since more resources are directed towards motor execution proper, and thus fewer resources are available to the other processes involved in on-line writing. This would lead to shorter execution periods in slow typists as compared to fast typists. Another possibility, suggested by Olive and Kellogg (2002), is that slow typists leave high-level processes unattended while typing, and pause to activate them. This would imply a serial mode of composing, a sign of which would be a greater number of pauses. Together with the study of the on-line processes in slow and fast typists, we also explored the linguistic characteristics of the texts produced. We looked at holistic ratings of text quality, and explored lexical measures, namely the amount, type, and diversity of the words used. Does the quality of the narratives differ between slow and fast typists? This question is of concern because it has been found that the quality of a text can be affected by difficulties with the mechanics of writing, both in children (Bereiter & Scardamalia, 1987; Graham, 1990) and in adults (Bourdin & Fayol, 2002; Olive & Kellogg, 2002).

4.2. Method 4.2.1. Participants Twenty-one first-year college students (mean age: 19.3 years; 11 female) from the University of Porto participated in this experiment. All participants had previous experience with writing on a computer keyboard, although frequency of using the computer and of writing on the keyboard varied among participants.


9/27/2006

11:05 AM

Page 59


59

4.2.2. Materials The picture story “Frog, where are you?” (Mayer, 1969) was used to elicit written and spoken narratives. The booklet is composed of 24 pictures that portray the adventures of a boy and his dog in search of their missing frog. The written narratives were collected using the computer program ScriptLog 1.04 (Strömqvist & Malmsten, 1998) running on a Macintosh computer, which was also used for the presentation of the pictures. 4.2.3. Procedure Participants were instructed to tell a story from the pictures, either by writing on a computer, or by speaking into a microphone. Both tasks were performed without time limit. The order of these tasks was counterbalanced, so that 11 participants started by writing, and 10 by speaking. Data were collected in individual sessions that lasted for one hour on average. Before starting on the narratives, participants gave written answers to demographic questions, and reported their frequency of computer and keyboard usage in a Likert five-point scale. Also, they were allowed to leaf through the picture booklet. They were told that the pictures would be presented once at a time on the computer screen, and that they were required to produce text for each one of them. The presentation of the successive pictures was self-paced. When advancing to the next picture, the text written for the previous one was removed from the screen. 4.2.4. Treatments and Analysis Although spoken narratives were collected, here we examined only the written ones. The narratives were transcribed and coded in CHAT (Codes for the Human Analysis of Transcripts) format to allow analysis by the Computerized Language Analysis software, CLAN (MacWhinney, 2000). CLAN was used to measure word length and frequency, to calculate lexical density, and to assess vocabulary diversity. Lexical density indicates the proportion of content words relative to total number of words. Nouns, verbs, adjectives, and modal adverbs ending in “-mente” (Portuguese equivalent to “-ly”) were classified as content words. Vocabulary diversity was assessed with the D measure (McKee, Malvern, & Richards, 2000). This measure was chosen instead of the more common Type-Token Ratio (ratio of different words to total words, TTR) because it is not influenced by sample size, a problem that affects TTR. D was computed through a mathematical modeling procedure; it ranged from 5, a value typical for a 5-year-old child, to 120, for a sample of academic writing (Malvern & Richards, 2002). Two experienced teachers of Portuguese, blind to the study, assessed independently the quality of the written narratives using Likert scales, ranging from 1 (very low quality) to 5 (very high quality). They rated each narrative on five scales: Overall Quality, Formal Use of Language, Creative Use of Language, Volume of Information, and Narrative Structure. Disagreements between judges higher than one point occurred only once in each scale. They were resolved through discussion between the judges so that only one-point disagreement remained. In order to establish a more conservative estimate of inter-rater agreement, the Weighted Kappa (Cohen, 1968) was computed. Moderate scores of


60

9/27/2006

11:05 AM

Page 60


agreement were found between judges in the five dimensions mentioned above, respectively, Kw⫽ .50, .50, .48, .63, .69. The narratives were analysed with ScriptLog on-line analysis module (Strömqvist & Malmsten, 1998; see also Strömqvist et al., 2006). One of the analyses was typing speed. To prevent typing speed from being inflated by extreme values (writers might just start pauses in the middle of a word), we computed first the median of the within-word keystroke interval for each participant, and then a mean for the whole group (Wengelin & Strömqvist, 2000). Further analyses required the establishment of a pause criterion, which, as discussed in the Introduction, was set at 2 s of keyboard inactivity. Using this criterion, the overall writing time can be divided between time spent in pauses, and time spent in execution periods. An execution period was defined as an instance of keyboard activity between two consecutive pauses in which at least one word is typed. We measured the duration and the number of words of each execution period. Transition times in selected discourse contexts (e.g. word, clause, and sentence) were also examined.

4.3. Results We will start with a brief survey of the results for the whole group, and then concentrate on the comparison between the slow and the fast subgroups. On average, participants spent 48 min on the writing task (SD ⫽ 23 min), 54% of which in execution periods, and 46% in pauses. Average fluency was 12.2 wpm, and average typing speed was .32 s. The narratives were written with about 500 tokens, 48% of which were content words (see Table 1 for more information). Generally, the variables analyzed here were not influenced by the fact that some participants wrote their stories after having Table 1: Average writing time and lexical measures of the narratives for the whole group, and split by slow vs. fast typists. Groups

Total writing time (min) Total pause time (min) Total execution time (min) Typing speed (s) Fluency (wpm) Number of words Word length (in characters) Content words Lexical density Different words Vocabulary diversity Note: Standard deviations in parentheses.

All (N ⫽ 21)

Slow (n ⫽ 10)

48.4 (22.8) 22.5 (14.4) 25.9 (10.5) .32 (.14) 12.2 (5.3) 514 (191) 4.4 (.2) 245 (86) .48 (.03) 227 (72) 72.4 (13.7)

59.1 (24.5) 31.1 (15.0) 28.0 (11.7) .44 (.1) 7.9 (2.8) 431 (164) 4.4 (.2) 206 (74) .48 (.03) 195 (56) 70.6 (14.2)

Fast (n ⫽ 11) 38.6 (16.7) 14.6 (8.3) 23.9 (9.3) .21 (.03) 16.2 (3.6) 589 (189) 4.4 (.2) 280 (84) .48 (.02) 257 (75) 74.0 (13.8)


9/27/2006

11:05 AM

Page 61


61

produced them orally. However, there were two exceptions: pause length and vocabulary diversity. Writing a story that had been previously produced in the spoken modality was associated with shorter pause length [M ⫽ 4.6. vs. 5.3; F(1, 19) ⫽ 4.4, p : .05], and less vocabulary diversity [M ⫽ 66.2 vs. 78.1; F(1, 19) ⫽ 4.6, p : .05]. A median split-half procedure was applied to typing speed in order to categorize subjects as slow or fast typists. Ten participants had a typing speed higher than .27 s, and were considered slow typists; half of them had started with the written narrative. Eleven participants had a typing speed equal to or lower than .27 s, and were considered fast typists; 6 of them had also started by writing (then produced the spoken narrative). The comparison between slow and fast typists is, thus, not biased by the order in which the spoken and written narratives were produced, since in both groups about half of the participants started in one condition, the other half in the other one. The slow group consistently reported less use of computer (M ⫽ 2.6 vs. M ⫽ 4.4), and less writing on keyboard (M ⫽ 2.1 vs. M ⫽ 3.8) than the fast group, respectively, F(1, 19) ⫽ 10.7, p : .01, and, F(1, 19) ⫽ 9.6, p : .01). This relates faster typing speed to greater amount of practice, and gives credit to self-report measures as a reliable means to screen typing automaticity. Not surprisingly, slow typists took longer to compose their texts. On average they spent 59 min in the writing task, whereas fast typists spent 38 min [F(1, 19) ⫽ 5.1, p : .05]. Consequently, in 1 min slow typists produced only half the words produced by fast typists [F(1, 19) ⫽ 34.0, p : .001] (see Table 1). However, the difference in total writing time does not extend to both components of composition time, i.e. pauses and execution periods. Compared to fast typists, slow typists had more overall pause time [F(1, 19) ⫽ 9.9, p : .01], but similar overall execution time (F : 1). If, at this rough description, slow typists spend more time pausing, what happens at the level of pause–execution cycles? Table 2 clarifies this question. Slow typists spend more overall time pausing not because their individual pauses are longer [F(1, 19) ⫽ 1.8, p ⫽ .19], but because they make a higher number of pauses [F(1, 19) ⫽ 9.0, p : .01; d ⫽ 1.3]. Regarding execution periods, while slow typists can sustain execution for 7.7 s, fast typists do it for a longer time, 11.6 s [F(1, 19) ⫽ 10.2, p : .01; d ⫽ 1.4]. During their execution periods, slow typists produce half the words produced by the fast typists. The higher cost of execution for slow typists is well demonstrated

Table 2: Characteristics of pause–execution cycles (whole group, and slow vs. fast typists). Groups

Number of pauses Pause length, P (s) Execution period length, E (s) Number of words typed Execution cognitive cost (P/E)

All (N ⫽ 21)

Slow (n ⫽ 10)

Fast (n ⫽ 11)

266 (157) 5.0 (.8) 9.7 (3.4) 4.0 (2.0) 5.8 (2.6)

358 (161) 5.2 (.9) 7.7 (2.3) 2.5 (.9) 7.6 (2.4)

184 (102) 4.7 (.7) 11.6 (3.2) 5.4 (1.7) 4.2 (1.5)

Note: Standard deviations in parentheses. Execution cognitive cost is the ratio of pause time over execution time (multiplied by 10, for ease of presentation), computed individually, and then averaged.


62

9/27/2006

11:05 AM

Page 62


in the ratio between pauses and execution, which we will call execution cognitive cost. For clarity, the ratio was multiplied by 10. Thus, for each 10 s of execution, slow typists paused for 7.6 s, while fast typists paused for only 4.2 s. In Table 3, instances of keyboard inactivity (absolute pauses) in selected discourse contexts — sentence, clause, and word — are presented for comparison between the slow and fast subgroups. There are only two contexts in which there are no statistical differences, Opening Sentence and After Comma. In all other contexts, slow typists have longer absolute pauses, although these differences seem to be of greater magnitude at the word level. Now, let us move from processes to products. Do the observed differences between slow and fast typists have an impact on the characteristics of their final stories? Differences were found in Text Length, Content Words, and Different Words (see Table 1): slow typists produce smaller texts [F(1, 19) ⫽ 4.1, p ⫽ .05] with less content words [F(1, 19) ⫽ 4.5, p : .05], and with less different words [F(1, 19) ⫽ 4.4, p : .05]. Since the last two differences are dependent on text size, the basic finding here is a tendency for slow typists to produce smaller texts. Narratives composed by both groups are similar in terms of Word Length, Lexical Density, and Vocabulary Diversity. As these measures are typically sensitive to text quality, the fact that there are no differences is an indication that the major difference between the stories composed by slow and fast typists is at the lexical level, and concerns number of words. Subjective ratings of the quality of the narratives seem to be concordant with the characterization of the written products as described above. Experienced judges did not rate differently the stories written by slow and fast typists according to Overall Quality, Formal Use of Language, Creative Use of Language, and Narrative Structure (see Table 4). The only difference occurred on Volume of Information: the stories written by slow typists were judged as having less information [F(1, 19) ⫽ 4.6, p : .05]. This result is consistent with the results, described before, regarding the total number of words, and particularly content words.

Table 3: Average duration of keyboard inactivity (in seconds) in selected discourse contexts (whole group, and slow vs. fast typists). Groups All (N ⫽ 21) Opening a sentence ._â Closing a sentence a^. After closing a sentence.^_a Before comma a^, After comma ,^_a New word after comma,_â Opening a word a_â Within a wordaâ Closing a word a^_a

1.6 (.9) 2.1 (.9) .91 (.67) 2.5 (1.4) .45 (.76) 1.1 (.9) .70 (.33) .32 (.14) .30 (.20)

Slow (n ⫽ 10) 2.0 (.7) 2.6 (.8) 1.3 (.8) 3.2 (1.6) .77 (1.0) 1.7 (1.0) .95 (.24) .44 (.10) .41 (.23)

Fast (n ⫽ 11) 1.3 (1.0) 1.7 (.7) .6 (.4) 1.8 (1.0) .17 (.04) .6 (.36) .47 (.21) .21 (.03) .20 (.08)

Note: Standard deviations in parentheses. “a” stands for any letter, “_” for spacebar, and “^” for absolute pause. Periods and commas are indicated as such.


9/27/2006

11:05 AM

Page 63


63

Table 4: Average ratings of text quality (whole group, and slow vs. fast typists). Groups

Overall quality Formal use of language Creative use of language Volume of information Narrative structure

All N ⫽ 21

Slow (n ⫽ 10)

Fast (n ⫽ 11)

2.6 (1.0) 2.7 (.9) 2.5 (.9) 3.2 (1.0) 2.9 (1.1)

2.3 (1.1) 2.6 (1.0) 2.3 (1.0) 2.8 (1.1) 2.5 (1.2)

2.9 (.7) 2.9 (.9) 2.8 (.9) 3.6 (.8) 3.1 (1.0)

Note: Standard deviations in parentheses.

4.4. Discussion The results from this study have shown that the on-line writing of slow typists is characterized by shorter execution periods, and higher number of pauses than that of faster typists. These differences have considerably large effect sizes (thus indicating robust differences between the groups), and they can be a sign of different strategies concerning the on-line management of the writing processes. Slow typists seem to be comparable to the participants in the studies by Bourdin and Fayol (1994), and by Olive and Kellogg (2002), who were instructed to write using unpracticed capital letters. Like the writers in the first study, slow typists may be suffering from a trade-off between the execution and the formulation systems. Not having mastered typing skill, slow typists may tend to forget part of what they had initially planned. They might be pausing to reread the text, and recover a lost idea. In order to cope with the limited cognitive resources and the high demands of execution, like the writers in the second study, slow typists might be using a serial way of composing. They may be devoting pauses to high-level writing processes, and execution periods to typing. Being unable to think and type at the same time, they might be alternating between execution, formulation, and monitoring, as suggested by Olive and Kellogg (2002). While lack of typing automaticity is a prime factor to explain these findings, it might not be the only one, and other factors, too, might have played a role. Although their putative contribution cannot be ascertained with the present experimental design, it should be noted that the distinction between slow and fast typists would capture differences attributable to other explanatory factors only coincidently. Furthermore, when considering other possible factors, it is important to keep in mind that the subgroups studied here were divided looking at within-word keystroke, which is possibly the most sensitive context to typing proficiency (Strömqvist, 1999), and that self-report measures of keyboard and computer usage reliably distinguished both groups. An alternative explanatory factor might be WM: the present findings might be due to lower WM capacity in slow typists. With less available cognitive capacity, overload should occur more often. However, this seems not to be the case. In a recently completed study, where WM capacity was measured independently, we replicated the finding reported here, and did not find differences in WM capacity between slow and fast typists (Alves &


64

9/27/2006

11:05 AM

Page 64


Castro, 2004). Furthermore, WM capacity was not related with the length of execution periods, instead, it was related to pause length — writers with larger WM spans made longer pauses. As noted by Torrance (personal communication), slow within-word typing rate could also be due to higher-level features of the writing task (e.g. lexical choice, syntactic planning, content-determination). There is evidence showing that difficulties with higher-level processes can be detrimental to lower-level processes (for a review, see Fayol, 1999). So, rather than a single bottom-up trade-off, possible top-down influences also need to be considered when analyzing the present findings. A comparison between typing and handwriting production tasks by slow typists should help clarify this issue. Future research also needs to clarify the functional role of the pauses — if slow typists pause more often, and what is happening during these pauses? Two directions are promising to shed light into these questions. One is to use the triple task technique (Piolat & Olive, 2000) specifically on pauses. The other is to compare eye movements between slow and fast typists. Alamargot, Dansac and Chesnet (see this volume) report evidence on Parallel Events (PE) occurring during graphomotor execution. Concerning the distinction made here, one straightforward prediction is that PEs would be less common in slow typists. Regarding absolute pauses, our results replicate the well-established finding that pause length tends to decrease as one moves from larger to smaller discourse units (Chanquoy, Foulin, & Fayol, 1996; Foulin, 1998; Schilperoord, 1996a, 2001). It also seems that slow and fast typists are different as soon as they start typing. However, since no reliable differences were found at the start of sentences, both groups might be devoting similar time and resources to planning at starting points of the written discourse. The differences between slow and fast typists were generally of greater magnitude at the lexical level, but with the present design it is not possible to ascertain specific effects, at the lexical level, from motor execution and translating skills (such as lexical choice and access). The finding that slow typists tend to produce shorter texts is not trivial, if one takes into account that there was no time limit for the composition task. For slow typists, being concise can be a strategic way of dealing with the high cost of motor execution; conciseness was not associated with lesser text quality, but it is probably the reason why their texts were judged to have less information. Although our distinction between slow and fast typists is concordant with the distinction between serial and parallel ways of composing, contrary to Olive and Kellogg (2002), in this study the final products of composition were similar in several lexical measures, and in text quality. Even though execution seems to be a burden for slow typists, it was one that they carried without prejudicing the quality of their narratives. This is not surprising, because they wrote in a well-known discourse genre, and no time limit was imposed. But, were we to alter one or more of those variables, we would predict differences in some of the lexical measures, maybe also on text quality. Overall, our findings show that typing speed is a reliable way to assess the degree of typing automaticity, and that the distinction between slow and fast typists is a proper way to study the cognitive cost of execution. Furthermore, they reveal how effortful execution can be for slow typists, and that the lack of typing automaticity can substantially alter the composition task.


9/27/2006

11:05 AM

Page 65


65

Acknowledgments The study reported in this paper was supported by a grant from the Portuguese Fundação para a Ciência e a Tecnologia (FCT) to the first author (SFRH/BM/374/2000), and by the Center for Psychology at the University of Porto, Language Project, also funded by FCT. The authors thank R. Malatesha Joshi, Thierry Olive, Annie Piolat, Mark Torrance, and two anonymous reviewers for very helpful comments on earlier versions of the manuscript.



9/27/2006

11:34 AM

Page 67

Chapter 5 The Word-Level Focus in Text Production by Adults with Reading and Writing Difficulties Åsa Wengelin

This chapter investigates how the spelling problems of writers with dyslexia affect their written language production process and other aspects of their texts. A group of writers with dyslexia is compared with a group of normal writers, investigating differences in the production process, as well as vocabulary differences in the finally edited texts. Their written vocabulary is also compared with their spoken. The results show that poor writers pause more frequently during writing, especially within words, and make a higher proportion of spelling-related editings. These differences in the process are associated with the differences in vocabulary within the writing conditions that do not occur in the spoken conditions. It is suggested that these differences are related to specific differences in writing process, which are due to extra cognitive demands that encoding of words imposes on the poor writers, and poor writers’ use of conscious strategies to avoid difficult words.

5.1.

Introduction

This paper is an attempt to synthesize studies earlier presented in Wengelin (2000, 2001a, 2001b, 2002). Typically, we classify someone as a poor speller when he or she has produced a text with many spelling errors and think of their spelling difficulties in terms of the number and type of spelling errors they have made in a written text or in a spelling test. However, what we usually do not know is how much difficulty the writer has had on top of the errors we find in the text. There are two broad possibilities. One is that the writer is not very aware of her difficulties or does not care about spelling and hence the errors we find in the finally edited text are errors she did not know and/or did not notice that she had misspelled. Another is that the writer is very aware of her difficulties, wants to write as correctly as possible, and has hesitated on many more words than the misspellings we have


Wengelin, A. (2007). The word-level focus in text production by adults with reading and writing difficulties. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 67–82). Amsterdam: Elsevier.


68

9/27/2006

11:34 AM

Page 68

Åsa Wengelin

found in her text. Perhaps she has even chosen some words or expressions in order to avoid words she finds difficult to spell. Among writers with dyslexia, the second type is the most common. As Taube (1988) points out, persons with dyslexia have a very low self-esteem when it comes to reading, writing, and learning. This makes it probable that they spend a lot of time and effort on the lower levels of the writing processes, such as lexical choice and encoding of words. This probably makes them less fluent or even disfluent writers. This is interesting for cognitive as well as educational reasons. From the cognitive point of view, several studies (e.g., Just & Carpenter, 1992; Kellogg, 1999; Lea & Levy, 1999) have suggested that the attention capacity that is available for different cognitive processes is limited. And as, for example, Bereiter and Scardamalia (1987) and Fayol (1999) have pointed out, writing is a complex task that involves several costly components that affect or even steal capacity from each other. For example, McCutchen, Covill, Hoyne, and Mildes (1994) showed a clear relation between translation fluency and writing ability. Differences in sentence generation and lexical retrieval were related to individual differences in writing skill. Both younger writers and skilled adult writers (see also McCutchen, 1988) seemed to benefit from fluent translation processes. If these processes operate fluently, they draw little on the limited working memory processes, something that provides the writer with more resources for planning and revision. Further, Snellings (2003) suggested that the speed of lexical retrieval influences the fluency of writing. Moreover, Fayol (1999) suggests that three different components of the writing process — the transcription process per se, lexical access (including graphemic representations), and spelling irregularities — can have an impact on the cost of writing. For poor spellers, all three of these components may need a lot of cognitive capacity, something that will steal resources from the actual “composition” processes. Hence, a writer who is very aware of her writing problems is likely to be a disfluent writer. From an educational point of view, teachers can get a more accurate picture of a student’s difficulties by studying the whole process of writing rather than just the finally edited product. An important question is how difficulties with low-level processes like lexical retrieval and encoding of words influence the finally edited product and its perceived quality. McCutchen et al. (1994) argue that if translation processes are not fluent then the writing process as well as the written product will be affected. In extreme cases where translation processes demand a lot of working memory resources, they may interfere with planning and revision as well as lead to problems with the production of grammatical sentences. Similar results have been obtained by Alves et al. (this volume), who showed that fast typists made fewer and shorter pauses than slow typists and had longer and more productive execution-periods. Moreover, Ransdell and Levy (1996) showed that writing fluency (as measured by words per minute) is the single most important predictor of writing quality in university students. Hence a conscientious poor speller — and therefore probably a disfluent writer — may also produce texts which are perceived as low quality, with insufficient coherence, and low lexical diversity, etc. The aim of this study is two-fold. The first is to investigate whether (and, if so, how) the spelling difficulties of writers with dyslexia are manifested in the written language production process. The focus will be on word-level “disfluencies,” i.e., pauses and editings related to single words. If it is the case that the poor spellers spend a lot of time and


9/27/2006

11:34 AM

Page 69

Word-Level Focus in Writing Difficulties

69

effort on lexical choice and word encoding, we would expect them to have a higher proportion of word-level disfluencies than other writers do. The second is to investigate one aspect of the finally edited product — and of writing quality — namely vocabulary diversity. A high-quality written text is typically associated with a higher lexical diversity and a higher lexical density than a spoken text. If these subjects tend to show a high proportion of word-level disfluencies, it appears reasonable to believe that their lexical diversity will be affected. The question is whether there is a relation between a high degree of word-level disfluencies and the lexical properties of a text. The properties investigated are lexical diversity and lexical density. Both will be described in the section on “Vocabulary” below. 5.1.1.

Disfluencies in Written Language

In the study of spoken language, the so-called “disfluencies” (e.g., pauses, speech repairs, repetitions, false starts, hesitations, etc.) have long been viewed as indicative of the mental processes underlying speech production. Spoken language is typically produced under “on-line” processing constraints. This means that there is typically a receiver present, and that the sender risks making her speech unintelligible if she cuts it up with too many pauses or changes. Moreover, the sender cannot save, review, and revise her message until it is perfected. The signal “disappears” into the air once the words are spoken. According to Grabowski (this volume), differences like these make writing less resource demanding than speaking. However, for beginning writers (e.g., Bereiter & Scardamalia, 1987) and for people with writing difficulties (e.g., Wengelin, 2002), the absence of a receiver may make writing more resource demanding than speaking. Pauses in language production have been argued to enable the speaker to gain time for processes having to do with the continuing choice of content and types of expression, i.e., planning (see for example Allwood, Nivre, & Ahlsén, 1990), but also for the execution and monitoring of utterances. A common view of disfluencies is that they also indicate problems in the planning or execution of utterances, for example, difficulties in word finding and conceptualization. In spoken language, pauses have been shown to occur frequently at clause boundaries, sometimes between words within sentences, but almost never within words (Goldman-Eisler, 1968). Swerts (1998) further showed that discourse boundaries between larger units, such as sentences, were more predictive of pauses than smaller units, such as words. In the typical writing situation, there is no receiver present and therefore no need for the sender to gain “interaction” time. The writer can, in other words, make as many pauses as she likes, anywhere she likes. Despite this, Wengelin (2001b) obtained similar results for writing as for speaking concerning the distribution of pauses. Although pauses are more frequent within words in writing than they are in speaking, intra-word pauses are still rare in the writing processes of skilled writers. Moreover, just as in speaking, pauses are more frequent at boundaries between smaller units (relative frequencies) than at boundaries between larger units. However, if a person is a poor speller — for example due to dyslexia, as in this paper, this distribution may look different. It appears reasonable to believe that poor spellers will interrupt their flow of writing when they run into words the spelling of which they do not know. We could expect poor spellers to have a higher proportion of pauses within words as well as a higher proportion of word-level editings.


70

9/27/2006

11:34 AM

Page 70

Åsa Wengelin

5.1.2.

Vocabulary

Typically, the vocabulary used in written text is both more dense, i.e., has a higher proportion of open-class word tokens, and more diverse than that used in spoken text. Strömqvist et al. (2002) argue that this is mainly due to the difference in processing constraints between speaking and writing. They propose (in line with Chafe, 1982) that the reuse of frequent words may be a natural strategy to reduce the cognitive load imposed by the on-line constraints (especially time) of spoken language and that a wider range of lexical choices can be afforded in writing where the on-line constraints are relaxed. It could also be the case that more “small words” — closed-class words — are used in speech because (just like disfluencies) they allow time for planning on different levels — for example, lexical choice. They showed not only that this modality effect exists across age groups from grade 4 to university students but also that it increases with age and writing experience. Vocabulary differences between the texts produced by writers with dyslexia and those by other writers could have several sources. One possibility is that poor spellers who are also poor readers have a smaller total language input, due to very little reading (and hence less written language input) and therefore a smaller vocabulary altogether. In that case, we would expect their vocabulary diversity to be lower than that of the other writers in both spoken and written conditions. A smaller written language input could also lead to a limited “written-language awareness,” in which case we would expect the writers with dyslexia to have similar values for lexical density and lexical diversity in both spoken and written language, and that these would be lower than those of the written language of the control group. By comparing the spoken and written language of the two groups we can rule out the possibility that the differences we may find in the written language of the subjects with dyslexia are a general language or learning disorder, rather than specific to their written language. If the problem of the subjects with dyslexia were specific to written language, we would not expect any differences between the groups concerning their spoken language. If this is the case and there is no difference between spoken and written language of these subjects with dyslexia, their written texts will be very “spoken-language” like. One possible explanation for such a result could be a limited awareness of the differences between spoken and written language — i.e., a limited written-language awareness. Another possibility is that writers with dyslexia — who are well aware that they are poor spellers — use strategies to avoid words they find difficult. This could lead to a smaller lexical diversity but should not influence the density. Target words that the writer is unable to spell are often substituted for synonyms or near-synonyms (e.g., Wengelin, 2002). Obviously, these synonyms are of the same part of speech as the target words — and thereby of the same category (open- or closed-class). Hence, lexical density is not affected. However, the diversity could be affected if a poor speller restricts her writing to words she feels secure about. Word substitutions of this category are often preceded by “extra” pauses or series of editings of the original word. Finally, it could be the case that the lexical retrieval and/or the encoding of words in writing have such a high cognitive cost for the subject’s dyslexia that they need to reuse words just like in speaking. This would probably influence mainly lexical diversity — and


9/27/2006

11:34 AM

Page 71


71

density to a certain extent — as well as the pause patterns, but not the editing patterns to the same extent. Such results would agree with earlier-mentioned studies (e.g., Just & Carpenter, 1992; Kellogg, 1999; Lea & Levy, 1999; McCutchen et al., 1994) on fluency and cognitive capacity.

5.2.

Method

5.2.1.

Subjects

The current paper reports analyses from 11 Swedish adults with dyslexia and 10 adults with no history of reading and writing difficulties.1 All subjects had experience with typing. They all had Swedish as their first language and had finished secondary school. The subjects with dyslexia had all been diagnosed as dyslexic, in adulthood, and before entering the study. Their reading and spelling abilities were tested again at the beginning of the present study. The reading tests included word decoding, reading speed, and reading comprehension. All subjects were generally poor decoders, slow readers, and poor spellers. For the results of the tests see Wengelin, 2002. 5.2.2.

Data Collection

This study deals with the production of “spontaneous texts,” elicited by means of keystroke logging, which is an increasingly popular method of studying writing in real time. See Nottbusch, Grimm, Weingarten, and Will (this volume) for a further discussion about realtime research on written production. All subjects wrote five texts on five preset topics: a personal narrative (I have never been so afraid), a picture-elicited narrative (The frog story, Mayer, 1969), a route direction, an argumentative text (a letter to the editor), and a job application. The two narratives and the route direction were also collected from the same subjects in an oral mode. For the current paper the frog story is left out since the pictures elicit many more pauses than the other tasks. The written texts were collected by means of the Apple Macintosh version of the key stroke logging program ScriptLog (Strömqvist & Malmsten, 1998). See also Alves et al. (this volume). The oral stories were told as monologues and video recorded. They were then transcribed in CHAT format. The advantages of the CHAT format are that it is widely used in spoken-language research, is well documented, and works as the input format to the CLAN package. Also, the finally edited versions of the written texts were transformed to CHAT format, so that they could be compared with their spoken correspondents by means of the CLAN package. The CLAN package is a battery of programs for corpus analysis. For example, it calculates not only simple measures like word frequencies and collocation frequencies, but also performs more advanced calculations like vocabulary diversity, etc. See MacWhinney (2002) for a description of CHAT and CLAN.

1

The data in the present study were collected as part of the research program “Reading and Writing Difficulties of Disabled Groups,” sponsored by the Swedish Research Council for Social Research (SFR).


72

9/27/2006

11:34 AM

Page 72

Åsa Wengelin

5.2.3.

Keystroke Logging by Means of ScriptLog

As mentioned above, the writing activities were recorded by means of the keystroke logging program “ScriptLog.” ScriptLog records writing activities that take place on a computer (word processor). It keeps a record of all events on the keyboard, mouse clicks, the screen position of these events, and their temporal distribution. In this study, ScriptLog’s own editor was used. This is a simple editor that is similar to NotePad in Windows. It allows all basic editing functions — moving around in the text, cutting, copying, and pasting — but does not allow the writer to change font or size or include clipart or other more advanced editing functions. The supported functions could be conducted by means of the mouse, the menus, or keyboard shortcuts. From a ScriptLog record, you can then derive not only the finally edited text, but also the “linear” text with its temporal patterning, pauses, and editing operations. A short example from one of the subjects with no reading and writing difficulties is given in example 1.a and 1.b: Example 1.a. Aldrig har jag varit så rädd som när jag senaste gången rökte i min fars bil. (“I was never so afraid as the last time I smoked in my father’s car.”) Example 1.b.⬍154.22⬎ Aldrig ha>r jag varit så rädd⬍15.07⬎ ⬍DELETE20⬎ r jag varit så rädd som när jag rökte i min fars bil sistr jag varit så rädd” (“I have never been so afraid”). She pauses for 15.07 s, possibly rereading the first words, noticing the “⬎” that she has mistakenly inserted in the word “har” (“have”). She deletes 20 characters including the “⬎” and then repeats the “r jag varit så rädd,” which she continues with “som när jag rökte i min fars bil sist” (“as when I smoked in my father’s car last time”). She pauses for 5.20 s and then deletes the word “sist” (“last”). She pauses again and deletes another 21 characters back to the word “jag” (‘I’), changing the structure and lexical choice of the sentence from “rökte i min fars bil sist” (“smoked in my father’s car last time”) to “senaste gången jag rökte i min fars bil” (“the latest time I smoked in my father’s car”). She pauses again for 2.47 s before finishing the sentence with a full stop and a carriage return. 5.2.4.

Analyses

Four types of analyses will be presented. First, quantitative analyses of productivity, in terms of number of words produced, and spelling errors in the written texts are presented. The aim of this is mainly to provide a background to the difficulties of the subjects with the reading and writing difficulties. Second, some analyses of “disfluencies” in the written production are presented, with both pauses and editings being counted as disfluencies.


9/27/2006

11:34 AM

Page 73


73

Further, the two groups are compared for number of pauses, distribution of pauses, and number of editings. Editings categorized as related to spelling problems are analyzed in more detail. The use of disfluencies to cope with spelling problems is then discussed from a qualitative point of view. Third, two measures of vocabulary — lexical density and lexical variation — are analyzed. Finally, the relations between disfluencies and vocabulary are analyzed by means of regression analyses. 5.2.4.1. Disfluencies Pauses were analyzed for frequencies in different locations, with a specific focus on word-level pauses versus other pauses. The distribution of pauses was then used to construct “pause profiles,” which were compared across the two groups. The first task of a pause analysis is to define a pause criterion. Despite the fact that defining a pause in writing is a complex task, which really presupposes that individual typing speed is taken into account, most studies have used a cut-off point, which is common for all subjects in the study. Since — to my knowledge — no good method for defining individual pause criteria has been developed, the current study follows the same tradition. The cut-off points used in different studies have varied from 250 ms (e.g., Olive & Kellogg, 2002) to 5 s or more (e.g., Jansen, van Waes, & van den Bergh, 1996), depending on the purpose of the study, the input mode (handwriting or typing), and the characteristics of the writers. In this study, it is assumed that to make a pause a writer has to “interrupt” her typing considerably longer than the “normal’” transition time between two keystrokes. Hence, a pause is a transition time between two keystrokes, which is longer than what would be expected to be necessary merely for finding the next key. This definition was operationalized as a transition between two keystrokes that was longer than 2 s. There were two reasons for this. First, this pause criterion is at least twice as long as a “normal transition” even for the slowest writer. A normal transition time was defined as the median transition time between lower-case letters within words. The mean (standard deviations within brackets) median intra-word transition time was 0.247 s (0.05) for the control group and 0.491 s (0.17) for the writers with dyslexia. The maximum median intra-word transition time was 0.796 s and was produced by a writer with dyslexia. Second, several earlier studies of typing have used a two-second criterion (e.g., Chanquoy, Foulin, & Fayol, 1996; Spelman Miller, 2000d) and this makes the current studies comparable to earlier work. See Wengelin (2002) for a more thorough discussion about pause definitions. Editings were analyzed for frequency and for the content of the editings. This paper will focus on editings related to spelling. An editing related to spelling is defined as an editing caused by a spelling problem. A spelling problem is here defined as a situation in which a writer is uncertain about the correct spelling of a word. If the writer makes the wrong choice when she runs into a spelling problem, a spelling error occurs (e.g., Nauclér, 1980). Spelling errors should be distinguished from typos. Typos are “slips of the keyboard,” i.e., errors that occur despite the writer’s knowledge of how they are spelled. This distinction is operationalized as follows: A typo is a substitution of a letter, within a word, which is adjacent to or holds the corresponding position on the keyboard for the other hand. Another common case is the omission of a letter. Typos are usually corrected almost immediately and are rarely left in the final text. Further, words with typos are usually only changed once, and then always to the correct version. A spelling error on the other hand is


74

9/27/2006

11:34 AM

Page 74

Åsa Wengelin

an error that breaks a spelling rule. A spelling editing is an editing that changes the spelling of a word, either from a spelling error to the correct spelling, from a spelling error to another spelling error, or even from a correct spelling to a spelling error. Spelling editings that involve only one or two letters are here categorized as “basic spelling editings.” In order to investigate the strategies used to cope with a spelling problem all editing sequences that include a basic spelling editing are analyzed. 5.2.4.2. Vocabulary As was mentioned above, two vocabulary measures were used. Lexical density was calculated as the percentage of open-class (nouns, verbs, adjectives ,and adverbs derived from adjectives) words (tokens) in the texts. Lexical density is also an indicator of sentence structure since it involves the use of closed-class words, such as prepositions, conjunctions, articles, etc. A text with lower lexical density has more of those, which gives a different sentence structure. The measure chosen for the analysis of lexical diversity is theoretical vocabulary2 (e.g., Menard, 1983; Broeder, Extra, & van Hout, 1986; Voionmaa, 1993). This measure was chosen since the different sizes of the text in the corpus makes Type/Token ratio impossible to use. The formula of theoretical vocabulary calculates the expected amount of types in a text of a certain length L, by calculating the mean number of types for all possible ways to select these L-word tokens from the original longer text. In principle, it picks a random number of words, e.g., 100, from a corpus and then counts the number of types within these 100 words. Since the subset of words could be picked in many different ways, the theoretical vocabulary is calculated as the mean of the number of types for all possible ways of choosing the 100 words from the corpus. The advantage with this measure is that all texts investigated can be “reduced” to the same length and therefore theoretical vocabulary can be used with more certainty than types per token when comparing corpora of different sizes. The disadvantage is that the size of the reduction is determined by the shorter texts, if texts are compared (Voionmaa, 1993).

5.3. 5.3.1.

Results Productivity and Spelling

The four-text written language subcorpus of the control group consists of 9319 words, the corresponding subcorpus of the subjects with dyslexia 6746. There were considerable individual differences — especially in the control group — and no consistent effect was found for either group or text type concerning text length. The two-text spoken subcorpus of the control group consists of 9207 words and the corresponding subcorpus of the subjects with dyslexia 5949. Just as for the written language, no main effect was found for either group or text type concerning text length. Finally, no effect was found for modality (spoken or

2

Thanks to Leif Grönqvist who computerised this measure.


9/27/2006

11:34 AM

Page 75


75

written language). The group means of the individual mean text lengths and the mean percentage of misspelled words are shown in Table 1. In the full corpus of subjects with dyslexia, 854 (≈7.1%) of the word tokens were misspelled and they were evenly spread across the different genres. In the subcorpus used for the current study, 518 words (≈7.2%) were misspelled. This gives a mean of 7.7% misspelled words per person (SD 2.8%) and text. A detailed description of the errors is given in Wengelin (1998) and Wengelin (2002). In all the texts (including the frog stories) of the control group only 31 errors were left in the finally edited texts. Since there was no time limit for the writing tasks, these subjects probably had no difficulties in finding and correcting the few errors they had made. Spelling will not be further discussed for this group. Based on the spelling analysis, we could expect the writers with reading and writing difficulties to be quite concerned about their spelling and spend a lot of effort on spelling — if they are aware of their difficulties. If so, this would be expected to show up in their pausing and editing patterns. 5.3.2.

Disfluencies

As would be expected, a t-test showed that the groups differed concerning overall disfluency (df/key) rate (t(19) ⫽ –6.55, p ⬍ 0.001). Table 2 shows the characteristics of the disfluency rates for the two groups. A closer look reveals that if we split up the disfluencies in pauses and editings we find that while both groups have similar rates (no significant differences found) of editings (ed/key) the group with dyslexia appears to have a much higher rate of pauses per key-

Table 1: Mean text lengths, both groups; misspelled words, the dyslexic group (standard deviations in brackets). Text length

Con N ⫽ 10 Dys N ⫽ 11

Misspelled words

Speaking

Writing

347.45 (264.50) 221.78 (104.11)

232.98 (127.04) 153.32 (27.53)

0.0 7.7% (2.8)

Table 2: Mean disfluency rates, both groups (standard deviations in brackets). Groups

df/key* pause/key* ed/key *p ⬍ 0.001.

t

Dys N ⫽ 11

Con NE ⫽ 10

0.14 (0.028) 0.41 (0.016) 0.03 (0.008)

0.06 (0.024) 0.11 (0.034) 0.02 (0.013)

⫺6.55 ⫺5.75 ⫺2.39


76

9/27/2006

11:34 AM

Page 76

Åsa Wengelin

stroke (pause/key, t(19) ⫽ ⫺5.75, p ⬍ 0.001). And this would be true if we knew that all subjects typed with similar speeds. This is not the case. As was mentioned in the discussion about the pause criterion, the subjects with dyslexia were significantly slower than the control subjects (see also Wengelin, 2002). Considering that the correlation between typing speed and pause frequency is very high for this group (r ⫽ ⫺.712, p ⫽ .11), the general pause frequency may not be a very informative. Therefore, finding out how the pauses are distributed across different contexts is more interesting than the mere frequencies. 5.3.3

Pause Distributions

A χ2(p < 0.001) shows that the two groups have different pause distributions concerning word-related pauses. The subjects with dyslexia have both more word-related pauses in relation to other pauses and more intra-word pauses in relation to pauses between words than the control subjects. These differences are illustrated in Table 3 as relative pause frequencies.3 The first three columns show the mean frequency of pausing in contexts before, within, and after words for the two groups. Notice, that for each word there is only one possibility of pausing before (between the space and the first letter) and after the word (between the last letter and the space) whereas there are several possibilities between letters within words. Therefore, in order to increase readability, the numbers in brackets show the relative frequencies based on number of words instead of number of possible contexts. The fourth column shows the mean of these means, i.e., the mean frequency of the wordrelated pauses. The final column shows the corresponding mean for pausing in other contexts. For example, pauses could occur in relation to punctuation marks, editings, etc. These are contexts in which planning and monitoring beyond the micro level are more likely and therefore they have relatively high pause frequencies. All pauses within and between words within sentences were counted as word-related pauses. An obvious problem with this is that some of these pauses co-occur with phrase and/or clause boundaries, contexts in which planning or monitoring is more probable than

Table 3: Relative pause frequencies in different contexts, both groups. Other contexts

Word-related pauses

Con N ⫽ 10 Dys N ⫽ 11

Before words

Within words

After words

Mean

7.1% 18.9%

0.3% (1.31%) 3.4% (12.5%)

2.8% 9.2%

3.4% 10.5%

Mean 17.3% 39.1%

Note: For context within words the numbers outside the brackets are frequencies related to all possible contexts within words (i.e., in the word “p_a_u_s_e” there are four possible pause contexts), whereas the numbers within the brackets are related to the number of words, in order to facilitate comparisons with the contexts before and after words.

3

The relative frequencies are more readable than the absolute numbers used for the χ2 test.


9/27/2006

11:34 AM

Page 77


77

micro-level problem solving. However, considering how rare such boundaries are compared to other word boundaries, the high frequency of word-level pausing is still a strong indication that the subjects with dyslexia use a lot of effort on the word level. 5.3.4.

Editings

Despite the fact that there were no differences between the groups concerning editing frequency, the different usages of editings will be explored a little bit further. Concerning the editings, less than 2% of the editings in the control corpus (out of 1201) were editings related to spelling and all of these were basic spelling editings. By contrast, of the 1255 editings made by the subjects with dyslexia, 335 (26.7%) editings were coded as related to spelling problems. The mean percentage of editings related to spelling per person and text was 28% (SD 15.1) in the group with dyslexia. Analysis of these editings suggested that two main types of strategies were used. The first type was a strategy to find the correct solution, and all subjects used this strategy at least once, namely to rewrite the word in different versions, until it “looked” correct. The second type was avoidance strategies, and they appeared on several linguistic levels as follows: — Change the problematic word to a synonym or near-synonym (usually after having tried strategy 1 first) — Evade the problem by exchanging the problematic word for a whole phrase. — Rewrite (or even delete) the whole sentence or more. Individual subjects used the different strategies to different extents, but all subjects used the “rewrite word” strategy. Six of the eleven subjects changed words to synonyms, phrases ,and other sentences or deleted the problematic chunk. The five others seemingly only rewrote the difficult words. It could be the case that these subjects use their pauses to make “covert editings,” i.e., to change their lexical choice before they have started to write it. This could perhaps be revealed in a think-aloud protocol analysis, but the present data do not allow for further speculations on that topic. Of the basic spelling editings found, 50% corrected a misspelling while 50% changed a correct spelling into an incorrect spelling, or changed an incorrect spelling into another incorrect version. 5.3.4.1. Summary of the disfluency results The results showing that the subjects with dyslexia have a high proportion of word-related pauses — specifically intra-word pauses, and make many spelling-related editings, could indicate that they have very low confidence in their own spelling abilities. The editing strategies found show that — at least for some of the subjects — the lexical choice is influenced by their spelling abilities. However, a more general question is how the final written product is affected by the word-level focus during the production process. Below, the vocabulary of the finally edited texts will be explored further, and compared to the spoken language of the same subjects. 5.3.5.

Vocabulary

As was mentioned above, written language is typically expected to have a higher lexical density than spoken language, and this would be expected to be the same for both groups.


78

9/27/2006

11:34 AM

Page 78

Åsa Wengelin

A two-way mixed ANOVA revealed an effect for modality (F(1, 40) ⫽ 37.365, p ⬍ 0.001). There was also an interaction effect between modality and group (p = 0.015). The difference is smaller (but consistent, p ⫽ 0.018)) for the group with dyslexia than the group without difficulties (p ⬍ 0.001). The subjects with dyslexia showed a mean lexical density of 33.6 in writing compared with 30.7 in speaking while the subjects without difficulties showed a mean lexical density of 38.2 in writing compared with 30.7 in speaking. In other words, the groups differ in writing but not in speaking. The writing of the subjects with dyslexia is closer to their speaking than that of the control subjects. The interaction is plotted in Figure 1. Written language would also normally be expected to have a higher lexical diversity than spoken language. Lexical diversity was calculated on a “reduced” text size of 100 words using the theoretical vocabulary measure described earlier. A two-way mixed ANOVA showed a main effect for modality (F(1, 40) ⫽ 18.44, p ⬍ 0.001) and an interaction effect between modality and group (p ⫽ 0.035). Only the control subjects made a difference between speaking and writing (p ⫽ 0.003). Their mean lexical diversity in speaking was 65.36 while their mean lexical diversity in writing was 72.11. The subjects with dyslexia on the other hand had a mean of 64.19 in speaking and 66.22 in writing. They wrote as they spoke. Once again, they did not differ from the subjects without difficulties in speaking. This is shown in Figure 2. To sum up, there is no difference between the two groups for either lexical density or lexical diversity in speaking. In writing, on the other hand, the group with no difficulties obtained higher values on both measures. In other words, the problem of the subjects with dyslexia appears to be writing specific, and it seems unlikely that the difference between the groups is mainly due to a significantly smaller vocabulary in general. 5.3.6.

Correlations Between Disfluencies and Vocabulary

Let us now turn to the relation between the disfluencies and the vocabulary measures. If it were the case that the subjects with dyslexia had “a lower written-language awareness”

40 39

R&W Con

38 37 36 35 34 33 32 31 30 Speaking

Writing

Figure 1. Lexical density in spoken and written language, both groups.


9/27/2006

11:34 AM

Page 79


79

73 72

R&W Con

71 70 69 68 67 66 65 64 63 Speaking

Writing

Figure 2. Lexical diversity in spoken and written language, both groups. than the control subjects, something that was partly indicated by their lower lexical diversity and lexical density, we would not expect a high correlation between these variables and any kind of disfluencies. Intuitively, disfluencies should be related to problem awareness, problem solving, and hesitation. On the other hand, a high degree of “spelling problem awareness” does not necessarily exclude a low “written-language awareness.” If, on the other hand, spelling strategies were important, we would expect a correlation between the vocabulary measures and the most spelling-related disfluencies, namely editing frequency and intra-word pause frequency, Finally, if lexical retrieval were the main problem, we would not expect a correlation between these “spelling-related” disfluencies and vocabulary variation, but rather between pause frequency in general and the lexical measures. The significant correlations (Pearson two-tailed) between the variables for the whole sample are shown in Table 4. The Table 4 gives the correlations calculated for the whole sample together. However, if the groups are analyzed separately, no such correlations are found for the control group. For the subjects with dyslexia, the only significant correlation occurs between lexical diversity and editing frequency (Pearson r ⫽ ⫺0.680, p ⬍ 0.05). In order to find out how well the disfluency variables could predict the lexical measures, a multiple regression was conducted. The analyses were conducted first for all writers together. The sample is really too small for group-wise analyses of correlation and regression. However, in order to get an impression of the two groups, group-wise analyses were also conducted. For all writers together, a model of intra-word pausing (t(20) ⫽ ⫺2.44, p ⬍ 0.05) and editing frequency (t(20) ⫽ ⫺4.03, p ⫽ 0.001) predicted about 62% of the variation in lexical diversity (r 2 ⫽ 0.62, p ⬍ 0.001). Both were unique predictors. When conducting the analysis group-wise, no effect was found for the control group. For the writers with dyslexia, on the other hand, editing frequency and intra-word pausing appear to predict about 55% of the variation in lexical diversity (r 2 ⫽ 0.55, p ⬍ 0.05). Of these two variables only editing frequency was a unique predictor (t(10 ⫽ ⫺2.95, p ⬍ 0.05).


80

9/27/2006

11:34 AM

Page 80

Åsa Wengelin

Table 4: Correlations between lexical measures and disfluency rates, both groups together.

Pause frequency Intra-word pausing Editing frequency

Lexical diversity

Lexical density

⫺0.550 (p ⫽ 0.009) ⫺0,523 (p ⫽ 0.014) ⫺0.701 (p ⬍ 0.001)

⫺0.528 (p ⫽ 0.013) ⫺0.455 (p ⫽ 0.037) ⫺0.468 (p ⫽ 0.031)

Concerning lexical density, 28% (r 2 ⫽ 0.28, p ⬍ 0.05) of the variation could be explained by general pause frequency (t(20) ⫽ ⫺2.71, p ⬍ 0.05). No other variable appeared to contribute to the model. However, this effect completely disappeared when the groups were analyzed separately.

5.4.

Discussion

To sum up, spelling problems (that the writers are aware of) do indeed appear to influence the production process of writers with dyslexia. They appear to be manifested in a high proportion of intra-word pauses and in many spelling-related editings. The vocabulary measures of the writers with dyslexia also distinguish them from writers without writing problems. However, there was no difference for either lexical density or lexical diversity between the two groups in speaking. These results suggest that the problem is specific to writing. Moreover, it appears unlikely that the difference between the groups is due to a more restricted vocabulary in general. Let us therefore consider other possible explanations. Since the subjects with dyslexia have lower results than the controls for both measures, a possible explanation is that they have a lower written-language awareness than the controls, possibly owing to less written language input explained by less reading. Another possibility is that the effort of encoding written words takes cognitive capacity from other processes such as vocabulary choice and sentence structuring. A third possibility is that the vocabulary of these writers is affected by their strategies to avoid words they find difficult to spell. Finally, we could not rule out the possibility that the subjects with dyslexia have a more limited lexical access in writing than the controls. Considering the small sample and the fact that the results of the correlation/regression analyses are not as clear as the results concerning differences between written and spoken language, the following should be taken as an open discussion about possible explanations and hypotheses for further investigation. Let us first consider written-language awareness. If it were the case that the subjects with dyslexia had lower written-language awareness, we would have expected both the lexical density and the lexical diversity to be similar in spoken and written language. However, the fact that these writers actually did show a significant difference between the two modalities in lexical density indicates that they can manage to make a certain difference, i.e., they are not completely unaware of the different structure of written language. The conclusion that written-language awareness is not the main explanation is supported by the result that the vocabulary measures did correlate with the disfluency measures.


9/27/2006

11:34 AM

Page 81


81

Lexical diversity could be predicted by intra-word pausing and editing frequency. Lexical density could be predicted by general pause frequency. If low written-language awareness had been the main explanation, we would have had no reason to expect such correlations. However, these results were for the whole sample. When analyzing the groups separately, no significant correlation was found for the control group and the only unique predictor of any of the vocabulary measures was editing frequency, which together with intra-word pausing predicted 55% of the vocabulary diversity. The strong correlation between editing frequency and lexical diversity makes a strong case for the editing strategies as the main explanation of the low lexical diversity. This is also supported by the results of the vocabulary analyses. Considering that open-class words are often exchanged for other open-class words (exchanges of words for synonyms), spelling strategies should not influence the lexical density to the same extent as the diversity. If the target word that the writer was unable to spell were an open-class word, it would most likely not be exchanged with a closed-class word, and therefore the lexical density would not be affected to the same extent as lexical diversity. However, with the use of the limited vocabulary of words the writer is able to spell, the diversity would most likely be lower than it normally would be in written language. Even though the effect was weak, the subjects with dyslexia did show a difference between spoken and written language for lexical density, but not for lexical diversity. Spelling-related editing strategies are probably caused by encoding problems. Finding substitutes for the target word is effortful. So is trying to find the correct spelling of the target word. Both these processes could possibly steal cognitive capacity from other processes, in which case the vocabulary effect could be “double” — both editing strategies and lack of cognitive capacity due to encoding problems would in themselves lead to less vocabulary diversity. Since it is the case that both lexical diversity and lexical density are lower in written language for the subjects with dyslexia than for the control subjects, a possible interpretation is that the cost of encoding words in writing for these subjects “substitutes” for the cognitive constraints imposed by the on-line conditions in speaking. Therefore, the subjects with dyslexia do not manage to vary the vocabulary, and structure their texts, in writing in the same way as subjects who don’t have such cognitive constraints, and their writing becomes more like their speaking. Considering Grabowski’s (this volume) claims about the writing superiority effect, the current results indicate that the writing superiority effect does not have the same validity for poor writers as for skilled writers. That encoding problems are involved is further supported by the high degree of intraword pauses, and the result that intra-word pausing contributes to the prediction model of lexical diversity. However, the results do not allow us to distinguish between encoding problems and editing strategies. The role of lexical retrieval is unclear. The fact that intra-word pausing correlates with lexical diversity could indicate that encoding problems, as opposed to lexical retrieval, are the major obstacle, since such pauses happen after the word has been started. However, the small sample, and the fact that this correlation was significant only when the whole sample was included, make it impossible to speculate further about this issue. However, considering McCutchen et al’s (1994) strong results about the relation between lexical retrieval and fluency, further investigation of this issue during text production is needed.


82

9/27/2006

11:34 AM

Page 82

Åsa Wengelin

To conclude, this study suggests that writers with dyslexia are disfluent writers, that there is a relation between the disfluencies and the final edited text, and that the characteristics of the texts produced by the subjects with dyslexia are writing specific. There is little doubt that the cost of execution is high for subjects in the current study. According to Alves et al. (this volume), there may be a trade-off between cost of execution and formulation processes. The sample of the current study is too small for such conclusions. Nevertheless, it appears as if the most likely explanation of the lexical characteristics of the texts of the subjects with dyslexia is a combination of encoding problems and editing strategies, something that would agree with Alves et al’s results.

Else_SW-Torrance_Ch006.qxd

9/27/2006

11:52 AM

Page 83

Chapter 6

GIS for Writing: Applying Geographical Information Systems Techniques to Data Mine Writings’ Cognitive Processes Eva Lindgren, Kirk P H Sullivan, Urban Lindgren and Kristyan Spelman Miller

This chapter presents the use of the Geographical Information Systems (GIS) for data mining and visualising information about cognitive activities involved in writing. The information can be collected from various sources, such as keystroke logs, manual analysis of stimulated recall sessions and think-aloud protocols. After an introduction to the GIS, an English as a foreign language (EFL) writing session is used to explain how to create the various GIS layers from the different information/analysis sources, and show how they can be easily data mined using the GIS techniques to improve our understanding of the cognitive processes in writing. The illustrative graphs used to provide an insight into the methodology are based on keystroke-logged data, manual researcher-based analyses and coded stimulated recall data that were collected after the writing session. Also a tool for visualisation and data mining, the GIS technique can support analysis of the interaction of cognitive processes during writing focusing on the individual writer, differences between writers or the writing processes in general. Depending on the research question, GIS affords the possibility to aggregate data to the level of writers, de-aggregate data in any way chosen or display data as attributes of individuals.

6.1. Introduction The increasing range and effectiveness of research methodologies used experimentally to obtain information about the cognitive processes in writing has created a problem for the researcher. The researcher is presented with an enormous amount of detailed information to synthesise into a cohesive whole. Further, the range of approaches used to analyse the Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Lindgren, E., Sullivan, K.P.H., Lindgren, U., & Spelman Miller, K. (2007). GIS for writing: applying geographical information systems techniques to data mine writings’ cognitive processes. In G. Rijlaarsdam (Series Ed.); M. Torrance, L. van Waes, & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in writing, Vol. 20, pp. 83–96). Amsterdam: Elsevier.


84

9/27/2006

11:52 AM

Page 84

Eva Lindgren et al.

data collected about writing from keystroke logging, think-aloud protocols and videorecorded writing sessions, among others, often highlight different aspects of the writing process and are difficult to synthesise well to describe the cognitive processes behind the writing session. There is a need to be able to efficiently data mine the mass of collected data. Berry and Linoff (1997) defined data mining as “the exploration and analysis, by automatic and semiautomatic means, of large quantities of data in order to discover meaningful patterns and rules” (p. 5). Lindgren and Sullivan (2002) proposed the LS-graph, as shown in Figure 1, as a method of using the information collected by keystroke logging techniques and combining them with researcher-coded layers of the information. The graph includes four layers. The light vertical lines indicate a revision, the top thin light line indicates the total character production including deleted characters at each point in time, the middle bold line indicates the characters/text retained after deletions at each point in time and the point at which the writer is working on the text at any point of time is indicated by the revision route indicated by the lower thin dark line. By using the concept of layering the graph enables visual analysis of occurrences in time and space of, for example, text-based revisions, the large unfilled dots, surface revisions, small filled dots and balance revisions, large filled dots (Kollberg, Lindgren & Sullivan, 2000). As long as a piece of data can be assigned a coordinate, defined by time of occurrence and place of occurrence, it can be plotted on the graph and represented as a separate layer. It is thus possible to add information to the graph from manual analysis of, for example, additional data collection methods, such as the occurrences of think-aloud data or which revisions were discussed in a post-writing reflection session. The LS-graph is a static representation of the writing process. It permits layering of data, yet does not facilitate automatic data mining of the underlying data. Thus, the researcher is unable to access the information that is the basis for a particular LS-graph feature through interaction with the graph. The graph is not ‘point-and-clickable’. The LS-graph allows the researcher to discover that a particular set of events occurred at a certain point during the writing process and from this it is possible to go back to the data to examine the event in more detail. The need to manually return to the data to examine an event in detail is timeconsuming and is complicated if a set of different data sources is being used. An approach to overcome the aforementioned problem with the LS-graph was proposed, yet not developed, by Lindgren and Sullivan (2002). They proposed the use of the GIS techniques, such as Arc View or Map Info (e.g. Robinson, 1998; Wegener, 2000), for interactive data mining of information obtained about a writing session, or sessions, from a diverse range of collection and analysis methodologies. This paper demonstrates how the GIS techniques can provide an effective method of data mining data collected about writing sessions and provide the researcher with new or deepened insights into the cognitive systems involved in writing. GIS techniques are superior to those of the LS-graph and can be used in writing research to assist the researcher to detect and delimit the impact of medium, genre and environment upon any one or more, implicit or explicit, aspects of the cognitive processes in writing. Before demonstrating how GIS can be used to help the researcher data mine collections of information about writing sessions, a detailed introduction to GIS and the sorts of geographical questions it was designed to examine is given. How the functions available in


9/27/2006

11:52 AM

Page 85

GIS for Writing

85

1800 1600 1400 1200 1000 800 600 400 200 0 0

500

1000

1500

2000 Time (s)

2500

3000

3500

Figure 1. The LS graph with four layers generated from log-data and one layer generated from hand-coded data. (Source: From “The LS graph: A methodology for visualising writing revision” by E. Lindgren & K. P. H. Sullivan, 2002, Language learning, 52, p. 586. Copyright 2002 by Language Learning Research Club, University of Michigan. Reprinted with permission.) GIS can be used to data mine in the field of writing research is then illustrated with data from a high school EFL learner. The illustration demonstrates how GIS can be applied to writing research and provides new insight into writing’s cognitive systems. The analysis methods used in the illustrative example are introduced prior to the data mining examples.

6.2. Geographical Information Systems Two fundamental questions in geography are: ‘why is it different here as compared to there?’ and ‘how do these differences change?’ These simple questions connect to patterns and processes because whatever occurs, occurs in space and time. Tangible artefacts as well as mental constructs mediated through human actions have a spatial extension in the stream of time. Acknowledging that most data collected for analysis have spatial references and that their spatial distribution can reveal additional information makes a strong case for the GIS. The first applications of GIS date back to the beginning of the 1960s and from then on computer scientists, computer cartographers and geographers have jointly developed the techniques and methods available in today’s software packages (e.g. ARC/INFO, IDRISI and MapInfo). The applications of GIS are by no means restricted to geography and its sub-disciplines. Within the environmental and social sciences, there are a number of spatial models drawing on GIS-functionalities. Atmospheric models for weather forecasts, hydrological models for rainfall simulations, models for subsurface contamination at hazardous disposal sites and forest growth models are a few examples from the environmental sciences. Within the social sciences, there are, for instance, economic modelling with a spatial


86

9/27/2006

11:52 AM

Page 86

Eva Lindgren et al.

dimension, geographical models designed for diffusion of innovations, sociological models of segregation of socio-economic or ethnic groups in urban areas, transport engineering models of travel and goods transports (Wegener, 2000). The plethora of different applications supports the notion of GIS as an analytical tool appropriate for studies on various scales and in different disciplines. Before presenting the motives for adopting GIS in the field of writing research some of the basic functionalities of GIS are introduced in the following paragraphs. In geography, the GIS is often described as being located in the point of intersection between cartography, database management and spatial analysis. To some users GIS is about digital maps and it is used as a visualisation device for the production of various thematic maps. Indeed, digital maps are much more flexible than traditional paper maps, but GIS is more than desktop mapping, not least because of its built-in database functionality. Behind a digital map there is at least one set of spatially referenced attribute data, which is stored in a database often called table. In a vector GIS-model the table is a matrix of data where the rows represent a geographical object and the columns represent the attributes of these objects. Geographical objects can take the form of points, lines or polygons. For example, on a map of Sweden, cities would be represented as a point object, the road network as a line object, and the counties as a polygon object. For each of these geographical objects specific information can be added. In the table showing cities, attributes like number of inhabitants, unemployment rate and average housing prices can be included. The usefulness of the system is further improved by the integration of relational database functionality. In case we like to study income differentials across the urban system, we would need an indicator of income, e.g. income from work, but this attribute is not available in the current data. However, by using the city code as a primary key the wanted information could be joined from another table in the system or from an external data source. In the GIS, geographical objects are organised in three types of layers (point, line and polygon), which is a central component in spatial analysis. The combination of two or more layers in the same map, i.e. overlay analysis, is a powerful tool for detecting new relationships in the dataset. In addition, the overlay analysis may create further questions, which give rise to more systematic examination of the available dataset. Since the GIS has a database structure, specified queries can be made in order to highlight certain issues. In our example, we would like to know in which counties the 10 cities with the highest average income are located. The cities that meet this criterion can be exported to a new layer, which may be used in the next step of the analysis. Hypothetically, cities with high average income serve as an economic engine for their regions, and offer a diversified and dynamic labour market with low unemployment rates. Is it the case that the chosen cities coincide with the 10 cities having the lowest unemployment rates? This and much more complicated questions can be answered by means of a GIS. The example above is a method called selection by attribute. Another way of making selections is by location. In case we are interested in a specific county and would like to have a map only showing the road network and cities within its borders, this ‘cookie-cutting’ procedure is pertinent. A commonly used functionality in GIS is the buffer, which is a circular area around a point, line or polygon. For example, in a study on accessibility to service and jobs in the county one would like to investigate the ratio of persons who live close to the major roads and public transports. Occasionally the problem at hand is related


9/27/2006

11:52 AM

Page 87

GIS for Writing

87

to proximity. Suppose we have coordinates for every comprehensive school in a city and we would like to create catchment areas that minimise the distance pupils have to cover. These polygons (Thiessen) have the property of creating boundaries in which the distance from any location within the polygon to the polygon’s centroid (school) is shorter than to any other polygon centroid. Moreover, in some GIS software, it is possible to calculate density surfaces based on irregular point data. Every year a number of real estates are sold in the city, but this dataset creates a point pattern of property values that does not say much of values at other sites. However, by putting a raster (grid) over the map and for each cell of the raster calculate average property values within a pre-specified radius a smoothed property surface is obtained. This surface makes a thematic map over estimated property values in the city (Fotheringham, Brunsdon, & Charlton, 2000). The described properties of the GIS makes it well suited for analysis of data on different aggregation levels. A simple definition of aggregation is the grouping of objects together into a new object. For example, in terms of socio-economic variables, such as income from work, income in a city can be represented on a continuum from one figure denoting average level to as many figures as there are individuals with income from work. In this way, data can be totally aggregated, disaggregated in anyway chosen or displayed as attributes of individuals. All these levels of aggregation, from macro to micro, can be handled by the GIS. Income of individuals living in a city can be mapped in a point layer that clearly shows local variations, whereas average income levels across districts can be mapped in a polygon layer presenting divergences between different parts of the city. In this way, the analysis of city income can fully benefit from data stemming from different aggregation levels.

6.3. Applying GIS to Writing To illustrate how the GIS can be applied to writing research in order to data mine information about the writing process that has been collected by researchers, the data collected about the writing of one text by a 13-year-old native Swedish writer, Hanna, in her non-native English will be used. Although the GIS writing map landscapes presented here, following on from the LS-graph (Lindgren & Sullivan, 2002), use time in seconds as the x-coordinate and the number of keystrokes (or written characters) as the y-coordinate, it is possible to use any coordinate system provided that each layer of data is placed within the same coordinate system. This is necessary as layers can only be investigated simultaneously if they have identical coordinate systems. Hence, different layers can contain information collected from one writing process as well as several different writing processes from the same or different writers. It is the ability to reveal patterns between, as well as within, layers that come from a range of sources that creates the power of applying GIS techniques to writing research. The GIS software has been designed so that the user can define how particular occurrences of features between layers should be represented visually. In the following sections a step-by-step illustration of how GIS layers can be explored to provide meaningful insight into writing processes is presented. The composition task, data-collection and analysis techniques used to create the GIS layers are explained before applied to the data mining analysis.


88

9/27/2006

11:52 AM

Page 88

Eva Lindgren et al.

6.3.1. Setting Up the Layers for Data- Mining Using GIS When data-mining, the researcher is looking to discover meaningful patterns. Hanna wrote the piece of EFL writing that is used to illustrate the types of patterns, which can be found using GIS-based techniques. Hanna’s task was to write to a pen-friend in Australia and describe her home town. She was asked to write about half of one A4 page and was given up to half an hour to complete the task. The writing session was keystroke logged using the computer keystroke-logging program JEdit (Cederlund & Severinson Eklundh, 1989; Severinson-Eklundh & Kollberg, 1996b). JEdit records every key pressed, every mouse movement made, every deletion undertaken and every paste action taken during a writing session along with the time of each of these events. The program also registers where these actions occur in the text, for example, at the point of inscription or in previously written text. Other examples of keystroke-logging programs are presented in Wengelin (see this volume) and Leijten and van Waes (2004). Figure 2 shows a short extract of the JEdit log file of Hanna’s composition. From the JEdit log-file it is possible to create GIS layers representing the total character production including deleted characters at each point in time, the number of characters retained after deletions at each point in time and the point at which the writer was working on the text at any point in time. 6.3.2. The GIS Database Files An example GIS database file is shown in Table 1; this table presents a few seconds of the data automatically generated from the JEdit log-file shown in Figure 1. For each second of the writing session, automatically generated information about revisions and pauses is shown. It is important to remember that the coordinate system for all layers must be the same. As stated earlier, in the examples in this chapter, the x-axis is the time in seconds and the y-axis the number of typed characters. This coordinate system is used, therefore, in this chapter for both the layers automatically generated from keystroke logs and those generated from manual coding. Other coordinate systems can be used, but must be applied consistently to all data layers; another system may, for example, be more appropriate if none of the data layers is collected using keystroke-logging techniques. From database files of the type illustrated in Table 1, it is possible to build GIS representations of the writing process. Figure 3 illustrates one possible GIS representation of the writing in English by Hanna. This representation is built upon automatically generated data, an extract of which is presented in Table 1. In order to create the dark grey ‘number of typed character’ layer in the graph, the column ‘Time’ was used as x-coordinate and the column ‘Number of typed characters’ as y-coordinate. This created a point layer. A GIS functionality in the program was used to transform the point layer into a polygon representing ‘number of typed characters’. The layers, ‘text length’, light grey polygon, and ‘revision route’, dark line, were created in a similar manner by using the ‘Time’ column together with a second column to create the coordinates for each occurrence. The graph further shows the position of all revisions (dots) and pauses (triangles). It also gives an indication of pause length to the research through the size of the triangles that is plotted. This information is clearer and more accessible after zooming in to examine the detail.


9/27/2006

11:52 AM

Page 89

GIS for Writing

89

Figure 2. An extract from the JEdit log-file of a 13-year-old Swedish student’s composition in English. Pausological information is presented as bracketed numbers (e.g. representing a pause of 2.4 s); cursor movements are indicated by the left/right/up/down arrows; backspace deletions are indicated by the crossed arrow symbol; space bar presses are indicated by the _ symbol, and all other characters (letters and punctuation marks) are also shown. The running time of the writing event (in seconds) is given down the left-hand margin.

Figure 3. One possible GIS representation of the writing in English by a 13-year-old Swedish writer based upon a database automatically generated from a log-file of the writing session. 6.3.3. Using Manual Analysis Layers for Data Mining One of the strengths of using GIS to examine the writing process is that an infinite number of layers can be created provided that each layer of data is placed within the same coordinate system. These layers can, for example, contain information collected from one writing process and/or information collected from several different writers and/or collected using different research methods. It is, therefore, possible, for example, to link layers automatically generated from log files with manual researcher analyses of a writing session. The possibilities this opens will be illustrated here using the automatically generated layers

Revision length

Revision location

3

I

1

53

D

1

Pause length (s)

Pause location

Text length

No. of typed characters

56 59

59

57

65 68 65 64

68

64 64

5

D

1

64

Note: Revision type “I” indicates an insertion, and “D” indicates a deletion.

5

2.4

67

69 69

Page 90

4

Pause no.

11:52 AM

Revision type

9/27/2006

40 41 42 43 44 45 46 47

Revision no.

Eva Lindgren et al.

Time (s)


90 Table 1: An extract from a GIS database file that has been automatically generated from a few seconds of the JEdit log-file (as shown in Figure 2).


9/27/2006

11:52 AM

Page 91

GIS for Writing

91

illustrated above, two manual analysis layers of the same writing session and information as to whether a revision was discussed in a post-writing stimulated recall session or not. The two manual analyses made were one using Spelman Miller’s (2000c) framing devices, and another using the Lindgren and Sullivan (2006) revision analysis taxonomy. The theoretical basis of these analysis methodologies is outlined before how GIS can be used to data mine the combination of layers based on different research and analysis methodologies is demonstrated. The researcher need to be clear about their research motivation for combining layers of data; exploration of bulks of information with no clear research question can lead to misinterpretations due to the possibility of arbitrary interpretations of some patterns. The manual-analysis data layers used to illustrate GIS’s relevance for writing research were selected to investigate our writer’s use of discourse features in conjunction with pauses and revisions. 6.3.3.1. Framing device analysis The framing device analysis is a means of categorising pause location in terms of the discoursal function of the unit of language produced prior to the pause. The framework is described extensively elsewhere by Spelman Miller (2000a, 2002b, 2002c), but in short, it consists of a scheme for describing units of production based on the notion of topic. Pause location in cognitive studies of writing is generally defined in strictly formal ways, based on the grammatical status of the units, that is, defining the pause at word-, phrase- or clause-internal levels. As an alternative, the unit, or stretch of language bounded by pauses, can be considered in terms of the discourse role it might fulfil in introducing, maintaining and developing topic in the discourse produced. A set of so-called ‘framing devices’ is proposed, consisting of elements at phrase or clause level, which serve to frame or set up the rest of the message. The notion of framing refers to the processes by which the writer makes choices about topicalisation in clauses, and about the topical relations across clauses. The notion of topic from which it derives is one of sentence rather than discourse-level, and draws on Hallidayan concepts of theme by identifying such elements or structures as subject theme, adjunct theme, textual (non-experiential theme), empty theme (it-, existential there structures), and thematised structures, including finite and non-finite clauses. Some examples of the category of framing device identified are given below. The underlined element consists of the unit defined as a framing device. • Subject theme: This hypothesis might be attributed to the claim of lateralisation. This will be further discussed below. • Adjunct theme: Around puberty, human beings will face lateralisation of the brain. Among individual factors, those which are widely recognised are age, aptitude, motivation and personality. • Textual theme: To start with, in an attempt to present a theoretical view of motivation, Skehan put forward four hypotheses. In addition, Long provides evidence to suggest that native-like acquisition is not possible after the sixth year of life. • Empty theme: There are debates on the methodology of experiments carried out. It may be the case that in the beginning students might learn faster. • Thematised structure: Since I was a child, my big dream was to become an English teacher. • After reviewing the literature, I will move on to present the research questions.

Revision Revision Revision Revision Stimulated Pause Pause Pause Text length location category effect recall no. length (s) location length

I

1

53

132

1

1

56 57

59 59

4

D

1

64

131

1

0

5

D

1

64

0

1

0 5

65 64 65 64

68 68 69 69

67

Subject theme

Page 92

3

2.4

No. of typed characters

11:52 AM

Revision type

9/27/2006

40 41 42 43 44 45 46 47

Revision no.

Eva Lindgren et al.

Time (s)


92 Table 2: An extract from a GIS database file that contains both data automatically generated from a few seconds of a JEdit log-file and additional manually coded data about the location of framing devices.


9/27/2006

11:52 AM

Page 93

GIS for Writing

93

Figure 4. Graph showing the search process (left pop-up window) for pauses more than or equal to 2 s at framing device locations.

The significance of these framing device categories for the study of writing processes is that it offers a layer of interpretation beyond the purely grammatical in describing the language, which is produced on-line. This allows the researcher to identify patterns of production and to relate pausing and formulating behaviour to discoursal features of the language being produced. In comparative research on L1 and L2 writers, for example, different practices were observed between writers in terms of their preferred location for pausing (Spelman Miller, 2000c). Some L2 writers, for example, paused frequently and for long durations after framing devices, suggesting that they tended to focus on the concerns of establishing or maintaining the topic first within the sentence before proceeding. Others showed tendencies to pause at clause and sentence final position, and not to interrupt the flow of production after framing devices. The EFL text process used as an example here was analysed according to which pauses occurred at framing device locations, and a column with the results was added to the database table, an extract of which is shown in Table 2. The expanded GIS database file, as illustrated in Table 2, can be used to investigate the writing process in a number of ways. For example, Figure 4 illustrates a search within the pause layer for pauses more than, or equal to, 2 s that occur at framing device locations. The search window shows the search function, and the triangles in the graph represent the result of the search: locations in the writing process where the writer has made a pause of at least 2 s that has coincided with a framing device.


94

9/27/2006

11:52 AM

Page 94

Eva Lindgren et al.

6.3.3.2. Revision analysis taxonomy The second manual analysis undertaken on Hanna’s writing of a text in English and added to the GIS analysis was the online revision taxonomy developed by Lindgren and Sullivan (2006). In her study of L2 writers, Spelman Miller (2002c) found interesting patterns of revision behaviour at framing device locations, and the addition of the online revision taxonomy data provides an opportunity to demonstrate how data mining searches that link appropriate GIS layers can assist the researcher to, for example, overview and locate combination of events for closer investigation. The online revision taxonomy (Lindgren & Sullivan, 2006) applied to Hanna’s English writing defines revisions first, according to their position in the text and second, according to their effect on the text. The revisions are divided into two main categories: pre-contextual revisions and contextual revisions. Contextual revisions are conducted within an externalised full context and are divided into subcategories depending on whether the revisions preserve or affect the meaning of the text (cf. Faigley & Witte, 1981). Contextual revisions are divided into Form and Conceptual revisions. The former include revisions that do not affect the content of the text, such as spelling, grammar and punctuation. The latter category includes revisions of a conceptual character, such as content or style revisions. Pre-contextual revisions are conducted at the point of inscription. They are carried out before a full context has been externalised as written text and can represent form as well as conceptual revisions. Pre-contextual revisions are results of the writer elaborating with form, content or style issues in the course of writing and can thus hint at important decision-making junctures in the unfolding discourse. Example (1) shows a revision in which pre-contextual revision is used to change the first person singular pronoun ‘I’ into the plural ‘we’ and thereby include the audience in the activities described. (1) Since I live by the sea I [DELETE 1] we can spend a lot of time there... When revisions occur in connection with pauses, they form ‘revision units’ (Lindgren & Sullivan, 2006). Some of these units occur at framing device locations (Spelman Miller, Lindgren & Sullivan, 2004). Example (2) shows an on-line sequence including a revision unit. The writer pauses after the framing device ‘There is’, deletes ‘is’ and replaces it by ‘are’. This revision unit indicates that the writer is elaborating with the continuation of the topic. (2) There is [DELETE 2] are dep [DELETE 3] debae [DELETE debates on… 6.3.3.3. Stimulated recall data layer After having conducted a search between layers in GIS, a closer investigation of the outcomes can be undertaken through the use of an important and useful functionality in GIS, that is, the possibility to point and click on the circles, squares and so on that are placed in the graphic representation of the writing process as the result of a search and access the information behind them. It can also be through a search for links between even more layers of data. To illustrate these points, a stimulated recall data layer was added that made it possible to ascertain which of the located pre-contextual revisions the writer defined as being of a conceptual character. This layer was collected after the student had completed the writing task. The student took part in a stimulated recall session with a fellow student immediately after having completed writing. During the stimulated recall session the students replayed their key-


9/27/2006

11:52 AM

Page 95

GIS for Writing

95

Figure 5. A graph showing the result of two search processes. The first, pre-contextual revisions defined by the writer as conceptual in a stimulated recall session, represented as dots. The second, co-occurrences of this revision type with pauses at framing device locations; the graph shows the resultant data-point and its underlying data (left pop-up window). stroke logged texts and discussed the revisions they had made, what they had been thinking about when they were writing and not writing, and what they were trying to achieve with their texts. These sessions were tape-recorded. The first author then listened to the recordings and coded them (see Lindgren & Sullivan (2003) or Lindgren (2004) for more details about this procedure). The data were then entered into a GIS layer data-sheet. Each code and its details were placed within the same coordinate system as all other layers. In Figure 5 one new layer showing revisions has been added to the graph presented in Figure 4; a complex data mining operation involving a combination of revision category, stimulated recall behaviour and pause category formed the search criteria. The creation of layers representing the outcome of a data mining operation is interactive; the researcher can easily adjust the data mining criteria during layer creation. For the creation of the new layer showing revision in Figure 5, a search was conducted for pre-contextual revisions that the writer had defined as conceptual in the stimulated recall session. The selected revisions were used to create a layer consisting of only the selected revisions; these were then plotted on the graph as dots in Figure 5. Another search was then conducted for pre-contextual revisions of a conceptual character that co-occur with pauses at framing device locations; Hanna produced one such revision unit. This revision is indicated by the arrow in Figure 5. The combination of layers created from analyses of Hanna’s writing process based on theories of revision and the location of pauses in the writing process, which are


96

9/27/2006

11:52 AM

Page 96

Eva Lindgren et al.

presented as dots and triangles in Figure 5, helps the researcher locate discourse boundaries where the EFL writer is elaborating the topic. Furthermore, Figure 5 shows an example of another Arc View function ‘identify results’. Once an occurrence has been found, it is possible to point and click on it and ‘identify results’. That is, see the data that are coded at this point in the writing process. The data-mined data point is shown as a square on the graph. In the ‘identify results’ window it is possible to see that the revision was of a conceptual character and that it co-occurred with a 4-s pause of framing device character. The ‘identify results’ window also shows the time point at which the revision occurred 400 s into the writing session; this facilitates easy access to the primary data in the log-file and replay of the relevant part of the logged writing session.

6.4. Conclusion This paper has demonstrated a novel approach to data mining writing data using the GIS; how the GIS can be used to search through data to locate patterns, plot them and allow the researcher to point and click to view the underlying data has been illustrated. The types of data used in the examples presented in this chapter are illustrative and have concentrated on a single text by one writer. However, as the GIS methodology permits a relational database connected to the system to include data relating to the analysis of many writers’ texts, the analysis is not restricted to one writer at the time. It is this feature of the GIS that makes it possible to aggregate data in any way justified by the research question. The researcher, however, needs to be aware of the limitations of the method and not aggregate or de-aggregate without methodological justification, nor draw conclusions without considering the issue of aggregation. In spite of this proviso, the methodology is flexible and permits the researcher to define the aspects of writing that are of interest for investigation. It facilitates the rapid over-viewing of an individual’s writing process and helps the researcher access detailed characteristic processes for each writer. That “writers and not revisions should be the unit of analysis” is something that Rijlaarsdam, Couzijn and Van den Bergh (2004, p. 207) have recently stressed; this contrasts with the focus of much of the writing research to date on one or two aspects of the writing process. The use of the GIS-based data mining places the writers, rather than particular aspects of their writing such as pauses or revisions, in focus and can facilitate the observation of the “strings of cognitive activities” (Rijlaarsdam et al., 2004, p. 207) that interact to create the whole due to its power to data mine large numbers of data and analysis types.


9/27/2006

12:01 PM

Page 97

Chapter 7

Verbal and Visual Working Memory in Written Sentence Production Ronald T. Kellogg, Thierry Olive and Annie Piolat

During written sentence production, semantic content is planned and then encoded into a linguistic expression. Verbal working memory may enable these required computations by temporarily storing word representations. By contrast, visual working memory may only be needed when the semantic content activates imaginal as well as prepositional codes. College students wrote in longhand definitions of either concrete or abstract nouns, while concurrently performing either a visual or verbal working memory task. Both concrete and abstract nouns would disrupt the verbal task, but only concrete nouns would disrupt the visual task. The definitions were richer in detail for the concrete words, suggesting imagery was involved, compared to the abstract words. As predicted, only these image-evoking words slowed reaction times on the visual working memory task compared to baseline, control measurements. Both high and low imagery words interfered equally with the verbal working memory task. The results are discussed in terms of planning and translating ideas and processes involved in sentence generation.

7.1.

Introduction

Working memory refers to a system for temporarily maintaining mental representations that are relevant to the performance of a cognitive task in an activated state. There have been a wide range of theoretical approaches to working memory proposed in the literature (Miyake & Shah, 1999), but in the present chapter we adopt the view that working memory comprises multiple components. The original Baddeley (1986) model postulated a phonological loop for storing and rehearsing verbal representations, a visuo-spatial sketchpad for visual object representations and their locations, and a central executive for attentional and supervisory functions. The evidence to date suggests that the visual and spatial components are distinct from one another and that the phonologically based verbal store is separate from Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Kellogg, R.T., Olive, T, & Piolat, A. (2007). Verbal and visual working memory in written sentence production. In G. Rijlaarsdam (Series Ed.), and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in Writing, Vol. 20, pp. 97–108). Amsterdam: Elsevier.


98

9/27/2006

12:01 PM

Page 98

Ronald T. Kellogg et al.

a fourth semantic store (e.g., Jonides & Smith, 1997; Haarmann, Cameron, & Ruchkin, 2002; Martin, Shelton, & Yaffee, 1994). In writing, the central executive may be the most important component because it appears to be involved in planning ideas, translating ideas into text, and reviewing the ideas and text produced thus far (Kellogg, 1996). Only motor transcription operates effectively with minimal or no executive attention and this is true only when handwriting or typing is well practiced and automatized in adults. For young children, the attentional demands of even handwriting are a major impediment to fluent and effective composition (McCutchen, 1996). When motor transcription is laborious, the central executive and possibly other components of working memory are diverted from planning, text generation, and reviewing (Bourdin & Fayol, 1994; Olive & Kellogg, 2002). A psychometric study of individual differences in children’s working memory indicated that only measures related to executive functions in the verbal domain accounted for a large source of variance in compositional quality and fluency (Swanson & Berninger, 1996). What, then, might the roles be for the other components of working memory in text production? The role of verbal working memory in sentence comprehension has been extensively investigated (e.g., Caplan & Waters, 1999; Just & Carpenter, 1992). Because of the difficulties involved in studying production (Bock, 1996), it is not surprising that less is known about the working memory requirements of sentence generation and writing extended texts. For spoken sentence generation, it is necessary to translate the conceptual contents of the message to be communicated into a grammatically correct string of words and encode these words phonologically (Bock & Levelt, 1994). Is grammatical encoding modular or is it dependent on the general cognitive resources of working memory? For written sentences, there must also be a stage of orthographic encoding to spell each word. Phonological encoding may also be involved in written sentence production, because one route to word spelling is a conversion of phonemes to graphemes (Caramazza, 1991). Are the grammatical, phonological and orthographic encoding stages required in written sentence generation limited by the availability of working memory? What are the working memory demands of planning the conceptual content before it is translated into a sentence? In the present chapter, we seek to take a preliminary step in addressing these fundamental, unanswered questions. Specifically, we tested the hypothesis that one or more aspects of translating the conceptual content of a sentence into a well-formed linguistic structure requires verbal working memory. It is not possible with our procedures to isolate the individual demands of grammatical, phonological, and orthographic encoding. Rather, we sought to measure whether all three of these taken together required verbal working memory. We further tested the hypothesis that the visual component of working memory is involved in planning the conceptual content of a sentence when the writer manipulates images of objects and events. Thus, our focus is on distinguishing between the working memory demands of planning conceptual content, on the one hand, and translating this content into a linguistic expression, on the other. 7.1.1.

Background

A model of sentence generation: Sentence generation entails planning conceptual content and then linguistically encoding it into a grammatical string of words. Imaginal and propositional representations are translated into the ordered words of a sentence through grammatical and


9/27/2006

12:01 PM

Page 99

Verbal and Visual Working Memory

99

phonological encoding. Grammatical encoding includes functional processing, in which lexical entries are selected and their semantic-syntactic functions assigned, and positional processing, in which the lexical forms are retrieved and the sentence constituents assembled (Bock & Levelt, 1994). Orthographic encoding is further needed in written production prior to handwritten or typed motor output (Carramaza, 1991). Unlike spoken production, spelling is required for written production. Phonological encoding appears to provide one route for determining the needed graphemes, whereas another route is based on lexical-orthographic representations (Badecker, Hillis, & Caramazza, 1990). Verbal working memory: Power’s (1985) results first suggested that a heavy load on verbal working memory reduces slightly the length of generated sentences. He studied spoken production while participants concurrently retained either three or six digits in verbal working memory. The six digit, but not the three digit, load tended to shorten sentence length, but the effect was weak and proved reliable only in a materials’ analysis of variance and not in the subjects’ analysis. Sentence length was reliably reduced by a six digit concurrent load in a text production task (Ransdell, Levy, & Kellogg, 2002), however. Participants were interrupted as they wrote short essays to encode and later to recall sets of six digits. Repeated sets were studied and then tested throughout the writing session. The average sentence length was reduced from 12 words in the control to 8 words in the six-digit condition. By contrast, a light load on verbal working memory in the form of irrelevant speech has little, if any, effect on sentence length. The sentence length effect might imply that verbal working memory is essential for unimpeded sentence generation, at least in written if not spoken output. When verbal working memory is distracted by the requirement to encode and retain six digits, then fewer words per sentence are possible. It is unclear, however, if the sentence length effect is due to the loading of verbal working memory or other components of the system. For example, it might be argued that attentional and other executive functions were solely responsible because coordinating two tasks requires the central executive (D’Esposito et al., 1995). To resolve this issue, Kellogg (2004) modified the task designed by Power (1985). At the start of each trial, participants received two noun prompts to include in each sentence. As in Power’s experiment, either a moderate or heavy load was placed on verbal working memory by requiring the retention of either three or six digits during sentence production (Baddeley & Hitch, 1974). As an additional condition, the visual and spatial components of working memory were loaded by requiring the retention of a visual image in order to decide if a probe coincided with a part of this image (Podgorny & Shepard, 1978). All dual task conditions demanded executive functions of working memory but they differed in the demands placed on the verbal versus visual and spatial components. If sentence length is reduced only in the digit conditions, then it can be concluded that verbal working memory is necessary for unimpeded sentence generation. In two experiments, the sentences generated while maintaining six digits in working memory contained reliably fewer words than those generated under no load, a visual-spatial load, and a three-digit load. A half word reduction was observed. This matched the findings of Power’s results with spoken production, where six digits allowed fewer words per sentence (M ⫽ 7.25) than did a no load control (M ⫽ 7.75). A load of three digits produced a small, unreliable reduction in both the spoken and written experiments.


100

9/27/2006

12:01 PM

Page 100


The sentence length effect appears to be a result of the heavy demands placed on verbal working memory by the concurrent retention of six digits. A less extreme load on verbal working memory and a load on the visual and spatial components of working memory did not reliably shorten sentence length, despite their demands on executive functions (D’Esposito et al., 1995). If all three load tasks placed equal demands on attention and other functions of the central executive, it could be concluded that verbal working memory is the critical resource needed for normal sentence length. On the other hand, it could also be argued that the six-digit condition recruited more executive attention to rehearse the digits because they exceeded the four-item capacity of the verbal store (Cowan, 2000). Thus, the safest conclusion from these results is that the sentence length effect was caused by an increase in demands on both the executive and verbal components of working memory in the six-digit condition (see also Ransdell et al., 2002). A six-digit load seems to target sentence length precisely without affecting other linguistic features. Just as many clauses were generated while retaining six digits as with none when complex sentences were requested by the experimenter. Neither grammatical nor spelling errors were impacted. Reading level, vocabulary complexity, sentence complexity, and use of passive constructions were all the same for a no load as for a six-digit load. It could well be argued from these null effects that sentence generation is largely automatic and modular. The cognitive resources of working memory were, for the most part, unnecessary to plan, encode, and execute a single written sentence. But for the persistent reduction in sentence length, when the load on verbal working memory was great, the modularity hypothesis accounts for the data. An important boundary condition of this finding is that each sentence was generated in isolation from the next. In text production, there must be coherence among sentences and the working memory demands would be greater than in isolated sentence generation. Accordingly, Ransdell et al. (2002) found that six digits held in working memory disrupted the subjectively rated quality of the essays written in addition to sentence length. Thus, the production of a coherent extended text is highly unlikely to be modular, as argued by McCutchen (1984, 1988). Planning versus translating: The locus of the sentence length effect is unclear. One might argue that a concurrent load of six digits disrupts the planning of conceptual content or the translation of content into sentences. Planning produces abstract propositions and images that must be encoded linguistically. This translation process entails grammatical, phonological, and orthographic encoding. If one assumed that the six-digit load disrupted only the planning of content, then it could be argued that the linguistic processes involved in generating an isolated sentence are indeed modular. If fewer propositions were retrieved or generated before initiating the sentence, then grammatical and perhaps subsequent stages of encoding might be automatic but a reduction in sentence length and typing time would be observed nonetheless. It should be noted that abstract propositions are presumably maintained in semantic working memory (Martin et al., 1994) and concrete images in visual working memory (Sadoski & Paivio, 2001). To trace the sentence length effect to planning, one must assume that the six-digit load disrupted semantic working memory instead of or in addition to verbal working memory. This hypothesis could be tested by manipulating the meaningfulness of the concurrent task. Six meaningful nouns should reduce sentence length, whereas six nonsense syllables should not.


9/27/2006

12:01 PM

Page 101


101

One aspect of Kellogg’s (2004) findings suggest that planning is not the culprit in the sentence length effect, however. Electrophysiological recordings have revealed that the comprehension of sentences containing three unrelated nouns activates brain regions subserving semantic working memory to a greater degree than those containing three related nouns (Haarmann et al., 2002). It takes more time to initiate sentence production for two unrelated noun prompts that are weakly activated in semantic working memory compared to two related nouns (Power, 1985; Rosenberg, 1977). This relatedness effect ought to increase under a six-digit load condition on the assumption that the concurrent retention of six digits disrupts planning. But Kellogg’s (2004) results showed that the difference in initiation time between related and unrelated prompts was no greater under a six-digit load than in a no-load control. Despite reasonably large sample sizes (N ⫽ 48 in Experiment 1 and N ⫽ 72 in Experiment 2), the relatedness effect remained stable across the load conditions. Thus, it appears that some aspect of linguistic encoding rather than planning is disrupted by a six-digit load. This interpretation corroborates the finding that grammatical encoding can be disrupted by a heavy concurrent load on working memory. Fayol, Largy, and Lemaire (1994) reported that maintaining five words in working memory can cause subject-verb agreement errors in the written transcription of orally presented French sentences. Errors occur under memory load when verb agreement depends entirely on orthography, but are markedly reduced when phonological cues are available (Largy & Fayol, 2001). Individual differences in the capacity of verbal working memory as measured by reading span and skill in lexical selection and retrieval are also reliably correlated (Daneman & Green, 1986; McCutchen, Covill, Hoyne, & Mildes, 1994). Such relationships should not be observed if linguistic encoding is entirely modular and not dependent upon verbal working memory. Visual working memory: It has been argued thus far that verbal working memory probably supports one or more aspects of linguistic encoding rather than the planning of conceptual content. What about the visual and spatial components of working memory? Kellogg (1996) proposed that these components might play a limited role in text production. Specifically, generating concrete ideas would invoke visual imagery and organizing ideas would invoke arranging them spatially (Sadoski & Paivio, 2001). Some evidence supports the view that visual and spatial working memory plays a more limited role in text production than does verbal working memory. In a text production task Lea and Levy (1999) found that a concurrent visual-spatial tracking task in fact disrupted the fluency of written composition by 13% relative to a writing only control condition. A concurrent phonological task disrupted fluency still more (21%) and showed more task errors compared to performance on the visual-spatial task. Swanson and Berninger’s (1996) concluded that individual differences in the capacity of visual-spatial working memory fail to correlate with children’s writing performance, whereas they do so with their reading performance. Also, Kellogg’s (2004) visual-spatial load on working memory failed to shorten length reliably whereas a heavy load on verbal working memory did so. Still, none of these previous studies directly tests the hypothesis that visual working memory is only needed for planning ideas in the form of concrete images. Further, it remains uncertain whether the disruptions in sentence production obtained when a heavy load is placed on verbal working memory are not at least partly attributable to the central executive. Retaining six digits (Kellogg, 2004) or five words (Fayol et al., 1994) while writing a sentence could place a heavy load on both the central executive and the verbal component of working memory. An experiment designed to address these issues is needed.


102 7.1.2.

9/27/2006

12:01 PM

Page 102

Ronald T. Kellogg et al. Rationale

The roles of verbal and visual working memory in the written production of a single sentence were examined in the experiment reported here. Participants wrote in longhand definitions of either concrete or abstract nouns, while concurrently performing a task demanding either visual or verbal working memory. The visual and verbal versions of the task were designed so to equate the demands made on the central executive. We predicted that defining both concrete and abstract nouns would disrupt the verbal task, but only concrete nouns would disrupt the visual task. An interaction of task and materials would rule out the view that the verbal interference is caused by diverting the central executive from writing. If that were so, then the concrete and abstract nouns ought to disrupt equally the visual task, too. The predicted interaction, therefore, would support the hypothesis that verbal working memory per se is necessary for linguistically encoding the content of any sentence, whereas visual working memory is needed only in planning sentences with concrete nouns. Sadoski, Kealy, Goetz, & Paivio (1997) argued that writing definitions of concrete words draws on imaginal as well as propositional representations. In support of this claim, they reported that participants initiate production faster and compose more detailed, higher quality definitions of concrete compared with abstract nouns. Further, they reported using imagery more often in defining concrete relative to abstract words. Here we sought to replicate their findings and extend them by including verbal and visual concurrent tasks. Participants composed definitions while concurrently performing a working memory task designed to require either the verbal or the visual component, plus the executive functions demanded in juggling the task concurrently with writing (D’Esposito et al., 1995). It was predicted that in composing definitions of only high imagery, concrete words would interfere with a visual working memory task. Because the maintenance of word representations was hypothesized to be necessary in linguistically encoding any written sentence, we expected that both low and high imagery words would disrupt the verbal working memory task. The verbal working memory task required detecting visually presented phonological segments (ba or da) on a 30 s variable interval schedule and deciding rapidly if the stimulus matched the last one presented. In reading these stimuli, phonological representations are known to be activated as well as orthographic representations (Massaro & Cohen, 1994). It was desirable to equate the verbal and visual tasks with respect to presentation modality, varying only the kind of materials used, so that the phonological segments were read rather than heard. The visual working memory task was identical to the verbal task, except that objects (triangle or circle) instead of phonological segments were presented. The secondary task of tone detection provides a measure of the executive attention required by writing (Kellogg, 2004). Here, the secondary task required more than detecting a stimulus, focusing attention on it, and scheduling a response. It was also necessary to store the stimuli and update the contents of either verbal or visual working memory. The two tasks were expected to be equally difficult to perform under baseline conditions, assuming that each makes the same demands on the central executive. Thus, any differences observed under dual task conditions should be attributed to the effects that writing has on the verbal versus visual stores themselves.


9/27/2006

12:01 PM

Page 103


7.2.

103

Method

College students (N = 60) were assigned in equal numbers to one of four groups defined by the factorial combination of materials (concrete versus abstract nouns) and tasks (verbal versus visual). In each condition, participants wrote definitions of 10 nouns while concurrently performing a working memory task that required the detection of a visually presented target and a speeded decision regarding whether to respond. They were instructed to respond by clicking a mouse button whenever the target was different from the last one presented. Thus, the task required maintaining the most recent target in working memory, detecting a new target, matching the new target to the one in memory, deciding to respond or to inhibit responding, and updating the most recent target. Reaction time (RT) was measured in ms along with the percentage of correctly detected targets. The instructions and stimuli were presented with a modified version of SCRIPTKELL (Piolat, Olive, Roussey, Thunin, & Ziegler, 1999). For the verbal working memory task, each target consisted of a pair of phonological segments (ba and da). On a 30 s variable interval schedule, one of the two targets appeared in large letters on a computer screen. The screen was below a glass desktop on which the participants wrote the definitions on paper in longhand using their dominant hand. The mouse was positioned near their non-dominant hand for responding to each non-repetition target. Thus, in the sequence ba ba da ba, the participant was instructed to respond to da and the final ba. For the visual working memory task the same procedure was followed, but the materials were large visual shapes (triangle or circle). Baseline measurements were collected for the working memory task in isolation, so that the degree of interference in RT could be determined. Also, the definition task was performed as a control condition without the concurrent task for 10 min. The data were collected in three blocks. The procedure began with the working memory control block for 12 min., followed by the definition control block for 10 min. Extra time was given for the working memory task to insure it was adequately learned. Finally, the dual task condition was tested for 10 min. The instructions for each condition were read on the computer screen before beginning each block. A total of 40 nouns were selected from the Colorado word norms for inclusion in the study (Toglia & Battig, 1978). Half were concrete nouns (e.g., house, wheat, pencil) that had been rated as easy to image visually. The other half were abstract nouns and were difficult to image (e.g., freedom, moment, duty). Other properties of the nouns, such as familiarity, pleasantness, and length were approximately equal in the two kinds of materials. The concrete nouns were randomly divided into two sets of 10 and one was assigned to the baseline block and the other to the dual task block. The same procedure was followed for the abstract nouns. The 10 nouns were listed on a page with space provided for each definition. The instructions for the definition task followed those given by Sadoski et al. (1997). Participants were asked to write a “dictionary-style definition for one usage of each word.” They were encouraged to “write a clear definition of each word you try” and to not worry “if you fail to define all ten words.” Finally, they were told to “write clearly at a normal rate and do not be excessively concerned with the vocabulary, grammatical correctness, spelling, or editing of your definitions.” The latter instruction was designed to place


104

9/27/2006

12:01 PM

Page 104


emphasis on sentence generation itself as opposed to the monitoring of production. The definitions were scored on a three-point scale. A score of 0 was assigned if no definition was given for a term in the 10 min. allowed. A score of 1 was given if a sketchy, poorly detailed definition was provided. This rating was used when only a single assertion was made in defining the noun. A score of 2 was given for a complete, well-detailed definition that included two or more assertions. A total score was calculated across all 10 nouns with a maximum score of 20. Two judges rated each definition and the inter-judge reliabilities were r ⫽ .89 for baseline sentences and r ⫽ .90 for experimental sentences. The ratings averaged across the two judges were used in the analyses described next.

7.3. 7.3.1.

Results Definitions

Replicating Sadoski et al. (1997), an analysis of variance (ANOVA) conducted on the definition scores produced a reliable main effect of materials, F (1, 56) ⫽ 41.25, MSE ⫽ 5.64, p ⬍ .001. The concrete nouns (M ⫽ 18.9) received higher scores overall than the abstract nouns (M ⫽ 16.1). No other effects were reliable. The kind of working memory task performed did not affect performance on the primary composition task. This indicates that participants gave priority to the composition task. In fact, the scores were just as high in the dual task condition as in the baseline or control condition (see Table 1). The secondary task of detecting either a new phonological segment or a new shape did not disrupt sentence production processes. This result further strengthens the case that sentence production proceeded normally as would be expected if writing took priority.

Table 1: Means and standard errors of definition scores. Materials Task Baseline control Verbal Visual Dual task Verbal Visual

Concrete

Abstract

18.8 (.35) 18.7 (.41)

16.2 (.59) 16.0 (.62)

19.1 (.87) 19.0 (.26)

16.6 (.60) 15.8 (.67)

Note: Standard errors are given in parentheses (n ⫽ 15).


9/27/2006

12:01 PM

Page 105

Verbal and Visual Working Memory 7.3.2.

105

Sentence Length

The definitions of concrete nouns were also reliably longer than those of abstract nouns, F (1, 56) ⫽ 11.55, MSE ⫽ 20.71, p ⬍ .01, as shown in Table 2. This outcome is consistent with the finding that concrete nouns elicited richer, more detailed definitions than did abstract nouns. There were no other reliable sources of variance. A reduction in sentence length was not observed with the modest loads placed on visual and verbal working memory here. Consistent with previous findings (Kellogg, 2004), sentence length appears to be truncated only when a heavy load is placed on verbal working memory (e.g., retaining six digits). The visual and verbal WM tasks were non-reactive in that production processes did not appear to be altered judging from sentence length. 7.3.3.

Secondary Task Accuracy

The next question concerns the difficulty of the two secondary tasks. There were no reliable differences in the accuracy of detecting targets between the verbal and visual tasks when they were performed in isolation in the control condition (Table 3). This is an important result because it demonstrates that the two tasks were of comparable difficulty. It supports our assumption that the visual and verbal tasks made equal demands on executive attention when performed without concurrent writing. Adding the secondary task to writing reliably interfered with the accuracy of responding to the secondary task. Overall, the percentage of correct responses dropped from 87.9% during baseline to 83.1% during dual task conditions. The reduction in accuracy was about the same for either the verbal or visual task and for both types of nouns, indicating that the writing and secondary tasks competed for one or more resources of working memory. It is likely that the executive functions required to carry out the two tasks at once were overloaded.

Table 2: Means and standard errors of words per sentence. Materials Task Baseline control Verbal Visual Dual task Verbal Visual

Concrete

Abstract

11.1 (.64) 12.2 (.89)

9.6 (1.27) 16.0 (.65)

11.5 (.78) 13.0 (.26)

9.3 (1.17) 8.7 (.64)



106

9/27/2006

12:01 PM

Page 106

Ronald T. Kellogg et al. Table 3: Mean percentages of correct responses to targets and standard errors. Materials Task Baseline control Verbal Visual Dual task Verbal Visual

Concrete

Abstract

88.6 (3.0) 89.8 (2.2)

83.4 (4.5) 89.9 (2.4)

84.4 (2.6) 87.1 (2.1)

77.9 (5.1) 82.9 (2.5)


Clearly, the effect was not limited to verbal or visual working memory but affected both equally as would be expected if the shared resource were the central executive. The main effect of measurement condition was the only reliable effect, F (1, 56) ⫽ 15.23, MSE ⫽ 48.4, p ⬍ .001, in the ANOVA. Thus, performance suffered a little when the working memory tasks were combined with writing definitions, but accuracy was still high with fewer than 20% missed targets. Moreover, accuracy was unaffected by the kind of working memory task performed or the kind of words defined. Participants were able to maintain accuracy but it is possible that they did so only by slowing their response times, which are presented next. 7.3.4.

Secondary Task RT

The mean RT to hits in the target detection task are presented in Table 4 for the various conditions. The mean RT did not differ reliably among the four conditions during the baseline control measurements. Thus, the participants were able to perform either the control visual or control verbal task just as accurately and as rapidly. Again, this outcome is important in establishing that the two tasks are of equal difficulty when performed in isolation. The mean RT increased reliably when tested in the dual task situation in all cases except the abstract-visual condition. The largest increase observed was in the concrete-visual condition. This pattern was supported by a reliable measurement X materials X task interaction, F (1, 56) ⫽ 9.48, p ⬍ .01. There was also a reliable main effect of measurement condition, F (1, 56) ⫽ 53.51, p ⬍ .001 and materials X measurement interaction, F (1, 56) ⫽ 4.57, p ⬍ .05, MSE ⫽ 17.0 for all effects. Thus, writing definitions slowed the time needed to respond to the targets in the verbal task regardless of whether low or high imagery nouns were used. The visual task, on the other hand, showed RT interference only with concrete


9/27/2006

12:01 PM

Page 107


107

Table 4: Means and tandard errors for secondary task reaction time (ms). Materials Task Baseline control Verbal Visual Dual task Verbal Visual

Concrete

Abstract

891(61) 790 (44)

841 (57) 909 (65)

1033 (66) 1094 (75)

1028 (75) 967 (62)

Note: Standard errors are given in parentheses.

nouns that presumably evoked images in planning the content of the definition. The selective increase for concrete nouns only in the visual task suggests RT interference effects are not entirely accountable in terms of competition for executive resources of working memory. The lack of interference with the visual task for abstract nouns is problematic for an account based on sharing attention to cope with the dual task demands.

7.4.

Discussion

It was hypothesized that verbal working memory supports necessary processes in written sentence production, whereas visual working memory supports optional processes associated with the planning of image-based semantic content. Here, writing a definition to either a concrete or an abstract noun slowed the responses made to a concurrent verbal working memory task. This outcome is consistent with the notion that linguistic encoding is not modular and required the use of verbal working memory. As expected, the visual working memory task was slowed only by concrete nouns. The planning of conceptual content prior to grammatical and phonological encoding was assumed to be sensitive to whether images as well as propositions were retrieved and maintained in working memory. Sadoski et al. (1997) concluded that concrete words activate both imaginal and propositional representations in planning the content of definitions. Dual coding supports superior definitions. The present study replicated their results and further showed that image-based conceptual content interferes with a concurrent visual working memory task. Past studies have shown that heavy loads on verbal working memory disrupt sentence production (Fayol et al., 1994; Kellogg, 2004; Power, 1985; Ransdell et al., 2002). It could be that part of these effects were because the central executive was also heavily loaded. Here we observed an interaction of task and materials, despite that both tasks made the same demands on the central executive and were equally difficult during baseline conditions. The present findings strengthen the case that verbal working memory per se is involved in translating an idea into a sentence.


108

9/27/2006

12:01 PM

Page 108


One might object that the visual task allowed verbal encoding of the shapes because triangle and circle were easily named. If that were the case, then interference should have been found for both concrete and abstract nouns in the visual as well as the verbal task. Even so, in future work, it would be useful to replicate the findings using unfamiliar objects that cannot be readily named and discriminated on the basis of verbal labels. The selective disruption caused by concrete nouns on the visual task should be replicated and contrasted with a spatial task in which participants respond to location rather than shape. We anticipate that neither concrete nor abstract nouns will slow responding to spatial location, although other kinds of writing tasks may well show such interference (e.g., defining the location of a landmark within a city). The demands of sentence production on working memory are possibly highly task specific. In line with this speculation, Passerault and Dinet (2000) reported that writing fluency is disrupted by a visual concurrent task when composing a descriptive text but not when composing an argumentative text. 7.4.1.

Conclusion

Both, the sentence length effect caused by a heavy, six-digit load on verbal working memory and the present findings, indicate that verbal working memory is necessary for unimpeded sentence generation. By contrast, visual working memory appears to play a more selective role. Further, the visual working memory task investigated here revealed RT interference with writing the definitions only for concrete nouns that are readily imaged. Visual working memory, then, appears to be involved in sentence generation at the planning stage, when conceptual representations activate images as well as abstract prepositions. The design of the experiment presented here offers one way to separate planning and translating ideas into sentences. However, the linguistic encoding of ideas involves multiple stages. It is unclear whether the interference observed on the verbal working memory task arose from grammatical encoding, phonological encoding, orthographic encoding, or some combination of the three. Moreover, these different kinds of linguistic encoding may be interactive rather than cascaded stages of processing (Dell, Chang, & Griffin, 2001). From a connectionist perspective, it makes little sense to ask whether the interference observed in the present experiments is isolated to one or more of these levels of encoding. Rather, it is likely that either all three levels of representation depend on the computational resources of verbal working memory or none do. The present findings speak against the modularity hypothesis of automatic linguistic encoding, but further research is needed to address specifically how written sentence generation depends on verbal working memory.

Acknowledgements The authors thank Adele Handley, Julie Lee, Kevin Kelley, Bridgett Gregg, and Tyler Mork for their assistance in conducting the study reported here. This research was supported in part by NATO Collaborative Research Grant No. LST.CLG 974939.


10/24/2006

9:56 PM

Page 109

Chapter 8

Effects of Note-Taking and Working-Memory Span on Cognitive Effort and Recall Performance Annie Piolat

This experiment was designed to assess the cognitive effort (measured using fast reaction times) made by student participants with different working-memory spans when they took notes as they listened to or read a lecture. The students had to use their conventional note-taking technique or an outline-based technique (pre-printed note sheets showing the title and subtitles). Note quantity and formatting (lists of ideas) were measured, as well as the revisions made on the notes taken. The results pointed to the importance of working memory in note-taking. Note-taking while listening demanded more attentional resources than note-taking while reading. Compared to conventional note-taking, the pre-outlined technique triggered a change in note format, but without an increase in cognitive effort. The note takers with a greater working-memory span adapted more easily to the situation that required the most attentional resources, i.e., note-taking on pre-outlined sheets while listening to the lecture.

8.1. Introduction This study had three aims. The first was to look at how much cognitive effort is allocated during note-taking under different conditions: listening to a lecture or reading the same lecture in written format, while taking notes in the conventional way or using a preoutlined technique. The second aim was to determine whether a note taker’s particular characteristics (such as memory span) affect the level of involvement in the task. The third was to validate the hypothesis that note formatting — which reflects the way information is processed by the note taker — varies with the note taker’s working-memory span, the


Piolat, A. (2007). Effects of note-taking and working memory span on cognitive effort and recall performance. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 109–124). Amsterdam: Elsevier.


110

10/24/2006

9:56 PM

Page 110

Annie Piolat

information-intake mode (reading or listening), and the note-taking technique (conventional or pre-outlined). 8.1.1. Note-Taking: A Learning Effect Note-taking is traditionally considered to serve as an external storage mechanism (Kiewra & Frank, 1988; Lindberg-Risch & Kiewra, 1990). Note takers listening to a lecture or reading a text take notes in order to compile a written record of information they will utilize later. But analyses of the conditions and functional features of note-taking have shown that other cognitive operations are carried out by note takers in addition to the simple transcription of information (for a review, see Piolat, 2006; Piolat, Roussey, & Barbier, 2003). For example, studies on the knowledge acquired by students who did or did not take notes during class have shown that the information-encoding process that occurs during note-taking triggers extensive memory storage (Kiewra, 1987). In other words, the sheer fact of taking notes is thought to provoke the “internal” memorization of the information written down (Castello & Monereo, 1999; Foos, Mora, & Tkacz, 1994; Laidlaw, Skok, & McLaughlin, 1993; Norton & Hartley, 1986; Roussey & Piolat, 2003; Williams & Eggert, 2002). This may seem paradoxical, since the very reason for taking notes is “to be sure not to forget” any information. There are two explanations of why memorizing might occur during note-taking. One possibility is that it results from the selecting and organizing processes note takers carry out as they attempt to confine their notes to the most useful information (Faber, Morris, & Lieberman, 2000; Lonka, Lindblom-Ylänne, & Maaury, 1994; Morgan, Lilley, & Boreham, 1988; Nist & Hogrebe, 1987; Oakhill & Davies, 1991; Slotte & Lonka, 1999; Spires, 1993). Another possibility is that memorization is triggered by the decisions note takers make in order to put information into notes, which are never an exact copy or verbatim transcription of what was read or heard (Einstein, Morris, & Smith, 1985; Hadwin, Kirby, & Woodhouse, 1999; Kiewra, Benton, & Levis, 1987; Kiewra, DuBois, Christensen, Kim, & Lindberg, 1989). Note takers not only devise ways of abbreviating words (Branca-Rosoff, 1998; Faraco, Barbier, Falaise, & Branca-Rosoff, 2003), but also utilize various other note-formatting devices as they attempt to create a written rendition of a succession of ideas, for example, or of a hierarchical relation between a main idea and the sub-ideas used to develop it (e.g., indentation with bullets; see Piolat, 2006). Idea-selection strategies (rough or even fine-grained sorting of critical ideas, etc.) and note-formatting strategies (lists of sentence fragments, diagrams, etc.; Gruneberg & Mathieson, 1997) are deliberately employed by note takers as they use a given note-taking method (e.g., linear approach, outlining, key-word tree structures; Piolat, 2001; Slotte & Lonka, 1999, 2001). Many studies have focused on the question of the effectiveness of different learning methods (Boyle & Weishaar, 2001; Dye, 2000; Foos et al., 1994; Kiewra, 1991; Kiewra & Benton, 1988; Kiewra, DuBois, Christian, McShane, Meyerhoffer, & Roskelley, 1991; Kiewra, Benton, Kim, Risch, & Christensen, 1995; Robinson & Kiewra, 1995; Robinson, Katayama, DuBois, & DeVaney, 1998; Roussey & Piolat, 2003; Ruhl & Suritsky, 1995). Kiewra et al. (1991), for example, assessed the performance of students employing different note-taking methods during a lecture (conventional format, outline, or matrix framework). Compared to linear note-taking, which is the most common method


10/24/2006

9:56 PM

Page 111

Note-Taking and Working-Memory

111

among students, pre-outlined methods were found to enhance retention. In particular, the use of a matrix framework appears to make it easier to both memorize and connect factual information, whereas note-taking in a pre-printed outline seems to mainly promote the retention of facts. 8.1.2. Cognitive Effort and Note-Taking Contexts As noted by Piolat, Olive, and Kellogg (2005), all the works mentioned above focus on the integration of information in long-term memory. These studies however did not explore the role of working memory in note-taking. For Hartley and Davies (1978), attentional capacities of note takers decrease as a function of several factors such as the importance given to the course and to the information that is delivered. Kiewra (1988b, 1989) mentioned the role of working memory in note-taking, indicating that quantity and quality of notes might be quite different according to the “working-memory skill” of note takers. Other researchers have analyzed the nature of the information being noted and concluded that the attentional capacity of note takers decreases throughout a course or lecture (Scerbo, Warm, Dember, & Grasha, 1992). Some other authors have considered that the role of working memory when note-taking is a means to decrease cognitive load during reading (Yeung, Jin, & Sweller, 1997), or in problem solving (Cary & Carlson, 2001; Cohn, Cohn, & Bradley, 1995). The notes, as an external memory, thus support retention in working memory of intermediate information or solutions that will be used for comprehension or for elaborating a final solution. Researchers have however neglected to study the relation between working memory and note-taking, which is critical given that taking notes involves juggling comprehension and production processes under, at times, severe time pressure. In order to put note takers in situations requiring different amounts of cognitive effort, two information-intake modes were set up: listening and reading. Note-taking while reading clearly generates a lighter cognitive load than note-taking while listening to a lecture (Gérouit, Piolat, Roussey, & Barbier, 2001; Roussey & Piolat, 2003). During a spoken lecture, note takers must simultaneously understand and write, and they must keep up with the speaking pace of the lecturer. In a reading situation, they can carry out these operations one at a time, allocating to each one the amount of time they deem necessary. The present study was conducted to determine whether note takers manage to carry out the necessary processes involved in various note-taking situations. More specifically, the idea was to detect potential trade-offs that might lead note takers — who cannot allocate more resources to the different processes — to accord more time to note-taking when the situation permits. Trade-offs of this type should contribute to efficient processing and thereby promote information intake in the course of note-taking. Stated from a different angle, the idea is to find out whether note takers take advantage of the self-paced feature of reading to write more things down as they access text information. Two note-taking techniques were studied. Depending on the experimental condition, participants had to take notes either in the conventional way without any particular instructions, or in an imposed, pre-organized way on pre-printed sheets containing an outline. Sanchez, Lorch, and Lorch’s (2001) work has suggested that the use of outlines is a good way of selecting text information. As Kiewra et al. (1991, 1995) suggested, note takers can take advantage of pre-printed note sheets to sort information more effectively and write


112

10/24/2006

9:56 PM

Page 112

Annie Piolat

down more information than they normally would. However, unless they are trained, note takers have trouble with pre-outlined note-taking sheets (Roussey & Piolat, 2003). In this situation, then, the effort expended by note takers should be greater than in a conventional note-taking situation. Furthermore, the devices they use to arrange information on paper (such as indentation to separate and rank ideas) should be different. Finally, if it is true that the pre-outlined method is more difficult (and thus more costly in terms of cognitive resources), then note takers rereading their pre-outlined notes immediately after taking them may be prompted to revise their notes in order to improve them. 8.1.3. Note-Taking: An Effortful Activity The suggestion that note-taking promotes the internal encoding of information was contested by Haenggi and Perfetti (1992), who showed that it is not the note-taking activity itself, but the type of reprocessing (rewriting notes, rereading notes, or rereading the text) that affects how well information in a long text is learned. The prior knowledge of the note taker and his/her memory capacity are thought to be determining factors in text reprocessing. The fact that merely taking notes while listening to a lecture triggers learning may indeed seem puzzling from the functional standpoint. Apparently, learning in this case is the outcome of how much and how well information is processed during note-taking. The operations that could potentially be carried out are highly diverse, and include listening in order to understand the lecture, noting what needs to be retained, and reading to make sure that what was written down corresponds to what was said and to one’s goals (Piolat, 2001; Piolat et al., 2003). In large part, these operations are deliberate and thus require attentional resources (Hayes, 1996; Torrance & Jeffery, 1999). Consequently, note takers with a limited processing capacity are forced to strategically divide up their attentional resources, placing priority at any given time on understanding, noting information, or reading information already written down. The allocation of resources to the various operations required by this complex activity is achieved by working memory, whose span varies across note takers (Baddeley, 2000). The possible role of a note taker’s working-memory capacity was mentioned in the work by Kiewra (1988a, 1988b, 1989) and Cohn et al. (1995), but this question was not examined in its own right. The present study was aimed at assessing the role of working memory in note-taking (for a review of these issues, see Piolat et al., 2003; Piolat et al., 2005). The working-memory capacity of note takers was evaluated using the memory-span test devised by Desmette, Hupet, Schelstraete, and Van der Linden (1995). By distinguishing note takers according to their memory span, it should be possible to detect the note-taking strategies specific to each working-memory capacity. It should also be possible to determine the adaptiveness of the operating strategies developed by different note takers. To analyze the functional constraints imposed upon the working memory of note takers, the experimental procedure used must be able to detect cognitive involvement in the note-taking activity. One of the functions of working memory is precisely to allocate resources to different processes and to keep knowledge activated in long-term memory so that the note taker can access it (Baddeley, 2000). Resource allocation or cognitive effort can be measured using the dual-task technique that Kellogg (1987) developed to study text writing (Kellogg, 1994, 1999; Olive, Kellogg, & Piolat, 2002; Piolat & Olive, 2000).


10/24/2006

9:56 PM

Page 113


113

Kiewra (1988a, b, 1989) and Cohn et al. (1995) suggested that, as Haenggi and Perfetti (1992) have shown for text reprocessing, note takers with different memory spans should adapt differently to the processing constraints imposed by the experimental design. In this study, four dependent variables will be used: cognitive effort during note-taking, the quantity of notes, the visual marking of ideas, and note revision during rereading. The main expectation is that in the most resource-demanding contexts, participants with high memory spans will have less difficulty compared to participants with low memory spans. In a previous study concerned with the amount and quality of memorized information according to note-taking contexts, we showed that listening in an outline context was the most difficult for note takers (Roussey & Piolat, 2003). It is thus expected that low-span note takers should (a) note down less information, (b) structure their notes less, and (c) correct their notes less afterwards. By contrast, high-span note takers, because they have more resources at their disposal, should better adapt to an unusual note-taking technique (plan) and thus process information more efficiently and rapidly.

8.2. Method 8.2.1. Participants Eighty sophomore students majoring in psychology participated in the experiment. They were randomly assigned to four experimental conditions defined by crossing two two-level factors: note-taking technique (conventional or outlined) and information-intake mode (listening or reading). The students in each group were divided into two categories on the basis of their working-memory span (high span vs. low span) assessed on a reading-span test(developed by Desmette et al., 1995). This test was chosen because it can account for memory span in relation to the comprehension process, which is fully active during note-taking. The average memory spans of the four experimental groups did not differ significantly (F 1). 8.2.2. Materials 8.2.2.1. Oral and written versions of the lecture The information to be learned was provided in an oral or written version (equivalent content). A 12-min excerpt of a university-level correspondence course in literature recorded by Joëlle Gleize of the University of Provence was selected. The course was about Umberto Eco’s postscript to his novel The Name of the Rose. The excerpt (oral version of the lecture) consisted of 1680 words. It was broken down into hierarchical units of content using the judge method (Faraco, Barbier, & Piolat, 2002). The course content was totally unfamiliar to the psychology-student population. To create the written version of the lecture (see Roussey & Piolat, 2003), paragraph boundaries had to be inserted. This was aimed at preventing readers from getting discouraged with an overly compact, continuous block of text. Given that paragraphing has a substantial impact on text interpretation (Heurley, 1997), the paragraphs could not be defined at random and were therefore inserted in accordance with the content units established by


114

10/24/2006

9:56 PM

Page 114

Annie Piolat

three judges. The judges came to a common agreement after an initial phase of individual work. Another measure taken to facilitate reading was to present the text in two columns and on two pages (Verdana font, size 9). A few words were changed here and there to avoid stylistic forms specific to the spoken word. This reduced the written text to 1633 words. It was entitled “Joëlle Gleize’s Commentary on the Postscript to “The Name of the Rose,” a Novel by Umberto Eco”. There were no subtitles. Concerning text organizers, both the oral and written versions contained several discursive indicators (for example, “A first series of remarks can be made…”; “Hence the two driving ideas behind my second point…”; “Second line of reflection…”). These structuring devices, which were present in the original lecture, were retained in the written version to facilitate comprehension (Rickards, Fajen, Sullivan, & Gillespie, 1997; Scerbo et al., 1992; Titsworth, 2001; Titsworth & Kiewra, 2004). Before the experiment, several other sophomore psychology students who were asked to judge the content of the lecture rated it as “difficult”. 8.2.2.2. Note-taking sheets The four note-taking sheets supplied to participants were either blank (free note-taking) or pre-outlined (outlined note-taking). The outlined sheets had headings and subheadings (see Appendix 8.A). The outline was based on the content analysis conducted by the judges. These different sections were placed in empty boxes to make the writing areas stand out. The heading levels were represented by character styles (boldface and italics).The relevance of the outline was evaluated by three other judges, who came to a full agreement. On the outlined sheets, boxes were used to highlight the headings and subheadings and delineate the empty areas where the participants were supposed to write their notes. This presentation brought out the different parts of the outline, without having to rely on numbering or lettering. 8.2.3. Procedure The participants were tested individually in a session that lasted about one hour. Each participant had to perform the following tasks in succession. 8.2.3.1. Fast-reaction training The note takers were trained (30 trials) to react as quickly as possible to an auditory probe (a sound emitted by a computer at random intervals) by clicking on the computer mouse with the hand they were not using to hold the pen when they wrote. Each note taker’s motor reaction time was calculated automatically by averaging over the last 20 trials (for a complete presentation of this method of measuring cognitive effort, see Olive et al., 2002; Piolat et al., 1999). 8.2.3.2. Note-taking of the literature lecture Then, participants in the oral-lecture condition (listening situation) took notes for the 12-min duration of the lecture. Participants in the reading situation were given an unlimited amount of time, but the time actually taken was measured. No constraint was imposed on reading time for the following reasons. Firstly, it is impossible to attribute an identical time to both listening and reading conditions. Observations have shown that readers needed almost twice the time for understanding a text via reading compared to its understanding via listening. Secondly, it was also not very


10/24/2006

9:56 PM

Page 115


115

relevant to leave readers taking notes in an arbitrary way for 24 min. In effect, in the latter case, the most rapid readers could have developed specific memory strategies during the unused time. Finally, it is rather unusual to take notes from a read text under strong temporal pressures. Thus, within the framework of this first study, whereby we compared cognitive effort during reading versus listening, reading was not constrained. Before starting the outlined note-taking in both the reading and listening conditions, the participants were allowed 2 min to examine the content of the four pre-outlined sheets. In the two conventional note-taking conditions, the participants began to take notes immediately. Throughout the note-taking period in all experimental conditions, the participants also had to do as they had done in the training session, that is, react as quickly possible to sounds sent out at random at a mean rate of one every 30 s. Their reaction times were recorded by the computer. 8.2.3.3. Five-minute rereading of notes Next, all participants were asked to review their notes for 5 min to prepare for knowledge questions. They were free to correct them with a pen of a different color from the one used to take the notes. This made it possible to detect any changes in the notes. 8.2.3.4. Reading-span test The participants then took the memory-span test. They were presented with three sets of two to six sentences whose last words they had to memorize and recall (for further information on the testing procedure, see Desmette et al., 1995). Two scores were attributed to each participant, one for the largest number of words remembered in each series (2, 3, 4, 5, or 6), and one for the total number of words correctly recalled on the test as a whole. The participants were classified as high-span and low-span note takers with respect to the medians of the score distributions. 8.2.3.5. Knowledge test Finally, the participants answered knowledge questions at their own pace. The test questionnaire is presented in Faraco et al. (2002) and in Roussey & Piolat (2003). The results of the knowledge test will not be reported here (see Roussey & Piolat, 2003). 8.2.4. Dependent Variables Five dependent variables were chosen. They were designed to account for the following aspects of note-taking: duration (1), cognitive effort (2), characteristics of the notes (3 and 4), and note revisions (5 and 6). 8.2.4.1. Note-taking time (in minutes) This variable only pertained to the reading mode of information intake. Again, in the listening condition, all participants necessarily took notes for the 12-min duration of the lecture, whereas in the reading condition, they could take as much time as they wished. 8.2.4.2. Cognitive effort associated with note-taking (in milliseconds) The note-taking reaction-time mean minus the motor reaction-time mean was the measure of attentional resource allocation to this activity (Piolat et al., 2003; for further information, see Olive et al., 2002; and Piolat & Olive, 2000).


116

10/24/2006

9:56 PM

Page 116

Annie Piolat

8.2.4.3. Note quantity (in words) The term “word” refers here to a group of letters with a blank on either side. A word could be a full word or an abbreviation. The number of words was tallied for each note taker. 8.2.4.4. Visual marking of ideas This variable concerned graphic effects (dashes, asterisks, arrows, etc.) used by the participants to highlight the flow of ideas (see example in Appendix 8.B). For participants to be classified as “note formatters,” their notes had to have at least 10 marks of formatting. 8.2.4.5. Note revision during rereading The participants spontaneously revised their notes after-the-fact by adding visual marks, words, or other markers like underlining. To be classified as “note revisers,” participants had to have at least one addition in two of the three revision categories.

8.3. Results Analyses of variance were used to assess the effects of the three factors: informationintake mode (reading vs. listening), note-taking technique (outlined vs. conventional), and the note taker’s memory capacity (high span vs. low span). 8.3.1. Note-Taking Time During Reading Note takers who used the pre-printed outline worked longer (27.89 min) than conventional note takers did (23.34 min), F(1,36) 3.87, p 0.048 (see Table 1). For both notetaking techniques, low-span note takers have a tendency to take more time (27.98 min) than high-span note takers did (23.25 min), F(1,36) 4.18, p 0.057. 8.3.2. Cognitive Effort Associated with Note-Taking Oral-lecture note-taking was associated with a mean cognitive effort that was significantly greater than when the same lecture was read (362 ms vs. 278 ms), F(1,72)10.86, p .002 (Table 2 and Figure 1). Table 1: Mean note-taking times (in minutes) and standard deviations in the reading condition, by the note takers’ working-memory span (high vs. low) and the note-taking technique (outlined vs. conventional)

High-Span notetaker Low-Span notetaker Total

Outlined technique

Conventional technique

Total

25.79 (7.30) 29.99 (6.13) 27.89 (6.74)

20.71 (8.99) 25.97 (7.89) 23.34 (8.24)

23.25 (8.11) 27.98 (7.57)


10/24/2006

9:56 PM

Page 117


117

Table 2: Mean reaction times (in milliseconds) and standard deviations in the reading and lecture conditions, by the note takers’ working-memory span (high vs. low) and the notetaking technique (outlined vs. conventional). Outlined technique

Conventional technique

Total

314 (0.09) 273 (0.10) 293 (0.09)

231 (0.10) 292 (0.13) 262 (0.11)

273 (0.10) 283 (0.11) 278 (0.10)

358 (0.16) 352 (0.09) 355 (0.13)

351 (0.13) 388 (0.10) 370 (0.12)

355 (0.14) 370 (0.10) 362 (0.12)

Cognitive Effort (wRT in ms)

Outline

Conventional

400 350 300 250 200 Reading

Listening

High-Span Note takers

Outline Cognitive Effort (wRT in ms)

Reading condition High-Span notetaker Low-Span notetaker Total Lecture condition High-Span notetaker Low-Span notetaker Total

Conventional

400 350 300 250 200 Reading

Listening

Low-Span Note takers

Figure 1: Mean cognitive effort (weighted reaction times in milliseconds), by information-intake mode (reading a document vs. listening to a lecture), note-taking technique (outlined vs. conventional), and note takers’ working-memory span (high vs. low). Interactions between the different factors are not statistically significant. A tendency toward significance was obtained for the interaction between note-taking and memory span of note takers, F(1,72) 1.97, p .16. Thus, the effect observed previously was significant only when the students used their conventional note-taking technique (reading 262 ms vs. listening 370 ms), F(1,72) 8.86, p .004. When they took notes in an outline, their average cognitive effort did not differ significantly between the reading and listening conditions (294 ms vs. 355 ms), F(1,72) 2.83, p .097. 8.3.3. Note Quantity The analysis yielded an interaction between the information-intake mode (listening vs. reading) and the working-memory span of the note takers (high vs. low) in the reading


118

10/24/2006

9:56 PM

Page 118

Annie Piolat

condition (Figure 2). High-span note takers took more notes when they listened (259 words) than when they read (214 words) whereas low-span note takers wrote more when they read (265 words in reading vs. 211 words in listening), F(1,72) 7.49, p .01. 8.3.3.1. Visual marking of ideas The significant interaction between the note-taking technique and the information-intake mode indicated that, regardless of their memory span, few note takers in the listening-outline condition used visual formatting marks, X2 (dl1) 4.58, p .035, (see Figure 3). This result is more obvious for high-span note takers (X2 (dl1) 3.38, p .066) than for low-span note takers (see Figure 3).

Conventional

300 280 260 240 220 200 Reading Listening High-Span Note takers

Outline Number of Noted Words

Number of Words Noted

Outline

Conventional

300 280 260 240 220 200 Reading Listening Low-Span Note takers

Figure 2: Number of words noted, by information-intake mode (reading a document vs. listening to a lecture), note-taking technique (outlined vs. conventional), and note takers’ working-memory span (high vs. low).

Outline

Conventional % of Low-Span Note Formatters

% of High-Span Note Formatters

100

Outline

80 60 40 20 0 Reading

Listening

High-Span Note takers

Conventional

100 80 60 40 20 0 Reading

Listening

Low-Span Note takers

Figure 3: Percentage of note takers who used visual marks, by information-intake mode (reading a document vs. listening to a lecture), note-taking technique (outlined vs. conventional), and note takers’ working-memory span (high vs. low).


10/24/2006

9:56 PM

Page 119


119

8.3.4. Note Revising by Addition of Visual Marks, Words, and Underlining For the conventional note-taking technique, the information-intake mode and memoryspan factors interacted significantly, X2 (dl1) 3.83, p .05 (see Figure 4). High-span note takers revised their notes more in the reading than in the listening condition. By contrast, low-span note takers revised their notes more in the listening than in the reading condition. In the listening condition, high-span note takers revised their notes more after having used an outlined framework than a conventional note-taking technique, X2 (dl1) 3.83, p .05 (see Figure 4).

Conventional

100 80 60 40 20 0 Reading Listening High-Span Note takers

Outline % of High-Span Note Revisers

% of Low-Span Note Revisers

Outline

Conventional

100 80 60 40 20 0 Reading Listening Low-Span Note takers

Figure 4: Percentage of note takers who subsequently revised their notes, by informationintake mode (reading a document vs. listening to a lecture), note-taking technique (outlined vs. conventional), and note takers’ working-memory span (high vs. low).

8.4. Discussion and Conclusion A summary of the results and the discussion are presented below considering the following dependent variables: (i) Cognitive Effort and Task-Execution Time; (ii) Note Quantity and Format; (iii) Note Revising by Addition. Finally, research perspectives concerning memory span of note takers are discussed. 8.4.1. Cognitive Effort and Task-Execution Time The importance of cognitive effort is as an indicator of the deliberate engagement of the note taker in an activity. It illustrates the role played by the process used for an appropriate monitoring of knowledge and of the level of involvement of working memory (Kellogg, 1994, 1999; Piolat et al., 2005). The first result in relation to cognitive effort shows that taking notes while listening demanded more cognitive effort than taking notes from a written document. This result is in agreement with the hypotheses and is not surprising. When operations can be


120

10/24/2006

9:56 PM

Page 120

Annie Piolat

sequentially organized (as they are when note-taking during reading) and do not have to be carried out simultaneously (as is the case during listening), the engagement of note takers is less important. However, this conclusion is true only when the students used their habitual note-taking technique. When pace is determined by the teacher’s speaking rate, it is cognitively costly for students to simultaneously perform the operations required to comprehend spoken information and produce written notes. Taking notes from a read text is less demanding. The second result related to cognitive effort shows that the effort expended by the note takers was not significantly dependent upon the note-taking technique. On the other hand, when taking notes while reading, where time was unlimited, note takers worked longer with the outline than when they took notes in the usual way. This increase in task-execution time suggests that the students had trouble using the pre-outlined note-taking sheets. However. the greater amount of time devoted to taking notes enabled writers to reduce the amount of effort involved. The third result — the effect of memory span on the way in which note-taking was monitored — shows that the cognitive effort expended by the students as they took notes did not vary significantly with their working-memory capacity, as measured on the reading-span test. The situations set up in this experiment were apparently difficult ones with a heavy cognitive load, probably due to the lecture content (literature), which was about a topic unfamiliar to the psychology-student population. In the reading situation where task duration was self-defined, note takers with a low memory span worked longer than those with a higher memory span. Note takers thus seem to compensate for a lack of attentional resources by lengthening their execution time. Such a trade-off is strategic and adaptive. 8.4.2. Note Quantity and Format The quantity of notes taken reflects a close monitoring of information when these notes are transcribed in an organized way (Piolat et al., 2003). The main results concerning these variables are the following. The quantity of notes did not differ significantly across note-taking techniques (outlined or conventional). But, the number of words noted, on the other hand, did depend on the memory capacity of the note takers, who reacted differently in the reading and listening situations even though the material to be learned was the same. Note takers with a lower working-memory span wrote down fewer words when listening than when reading, whereas high-span note takers did just the opposite. In the conventional note-taking situation — an experimental setting in which the cognitive effort varied significantly across information-intake modes — the note takers put more into the task while listening than while reading. Note takers with a high memory span wrote more in the former situation than in the latter. The amount of cognitive effort applied was reflected in the note quantity. But low-span note takers, despite their comparable cognitive effort level in the listening situation, wrote down fewer words than the others. These various results point to the difficulty with which low-span note takers select and write information while listening in comparison to high-span note takers. This interpretation needs confirmation, especially since reading note takers with a low working-memory span took advantage of the chance to lengthen the task time and produce many more notes


10/24/2006

9:56 PM

Page 121


121

than note takers with a high memory span. It is important to notice that the number of words written was not a measure of note quality (N.B. An analysis of notes’ quality is currently in progress; see Barbier, Faraco, Piolat, & Branca-Rosoff, 2004). It was indicative, however, of a clearly different strategy in the two groups of note takers, depending on the task constraints. As far as note formatting was concerned, the two groups of students produced comparable notes when working in their usual way (conventional technique). Formatting devices (bullets, dashes, arrows, asterisks) were used to list information by four-fifths of the participants in the reading and listening situations alike. Moreover, in the reading context, note-taking on pre-organized note sheets (outlined technique) did not prevent the use of formatting markers, since the note takers had enough time to use them. By contrast, when they were listening and had to use this technique, the note takers were no longer able to format the information noted. Note-taking while listening already demands substantial attentional resources. In this context, having to take notes on pre-outlined pages appears to be a hindrance for most note takers, since only a fifth of them used formatting markers to organize the noted information. 8.4.3. Note Revising by Addition In the experimental procedure, the note takers were asked to reread their notes before answering knowledge questions (for information on these results, see Roussey & Piolat, 2003). The fact that they corrected their notes during this rereading phase is a sign that the notes were insufficiently detailed (addition of words) or poorly organized (addition of indentation and underlining to highlight statements as subheadings). The extent of the revisions varied with the note takers’ memory span. No matter what note-taking technique was used, a greater number of low-span note takers revised the notes they had taken while listening as compared to notes taken while reading. This fact indicates that these note takers attempted to improve the notes they had been forced to take at a speed imposed by the lecturer. Moreover, regardless of the note-taking technique employed, these participants noted few words. In contrast, more high-span note takers revised their lecture notes but only when they had used the pre-outlined technique, although they wrote down a large number of words. This finding may indicate that these note takers had trouble processing information in the listening situation while taking notes by means of this demanding technique. In addition, the high-span note takers were also the ones who did the most revising when the notes were taken in the conventional manner while reading. Given that the cognitive effort they applied in this note-taking situation was not particularly strong, after-the-fact corrections were probably not a reflection of processing difficulties — these note takers might simply have attempted to compensate for the few notes taken in this situation by adding to them and making them more explicit. 8.4.4. Conclusion Note-taking varies in difficulty with the information-intake mode (reading or listening) and the note-taking technique (conventional or pre-outlined). Note takers devise


122

10/24/2006

9:56 PM

Page 122

Annie Piolat

adaptive strategies in order to adjust their level of involvement in the different note-taking operations. The results of this study suggest that the way note takers adapt is largely dependent upon their working-memory span. The encoding of information in long-term memory has long been a high-priority topic in research on note-taking. The results obtained here point out the important role played by working memory. Note-taking — a common and frequent activity for adults, not only in learning situations (Piolat & Boch, 2004) but also at work (Hartley, 2002) — is a key activity for analyzing the role of working memory in verbal information processing, for the note taker must both understand and produce. In another experiment with the same design and the same population of note takers, Roussey and Piolat (2003) showed that the knowledge acquired by students as they took notes depended upon the conditions in which the activity was carried out, particularly the note-taking technique employed. As already observed by Kiewra et al. (1991), the use of a pre-outlined technique enhances the memorization of essential information and factual knowledge. However, the benefits of this technique are only clear-cut for note takers with a high memory span in a reading situation. The results reported in the present study suggest that, irrespective of their memory span, note takers had trouble using the outlined technique, particularly when they were in the listening situation — their notes were not formatted as much and were revised afterward. In addition, more cognitive effort was applied. Unless they are trained, note takers appear to be disoriented by the pre-outlined note-taking technique in a lecture-listening situation. Thus, depending on their memory capacity and the note-taking conditions, note takers adapt to different extents by varying their level of involvement in the task (cognitive effort) and the features of their notes (number of words, formatting). Whenever possible (in the reading situation here), they also lengthen their note-taking time. As demonstrated for text comprehension (Just & carpenter, 1992; MacNamara, Kintsch, Songer, & Kintsch, 1996) and for written text production (McCutchen, 1996; Ransdell & Levy, 1996), working memory span appears to be an important influence on the processes taking place during note-taking. In future research, it would be worthwhile to find out what means note takers use to adapt to a task they sense as difficult. Depending on the availability of attentional resources — so quickly used up in a note-taking situation — do note takers prefer to concentrate on understanding, even if it means taking fewer notes? Or, on the contrary, do they prefer to put less effort into comprehension in order to write more things down? This strategic trade-off seems to be applicable to note-taking while listening. The strategies devised for note-taking while reading seem to be of a different nature. Some answers to these questions could be found by analyzing the content of notes taken in different contexts. However, the analysis of this private and elliptical writing is very difficult.


10/24/2006

9:56 PM

Page 123


123

8.5. Appendix 8.A. Outline Shown on the Preprinted Note-Taking Sheets in the Two Experimental Conditions with Outlined NoteTaking The following headings and subheadings were placed in empty boxes to make the writing areas stand out. The heading levels are represented by character styles (boldface, italics) (see diagram below). Introduction Information about Umberto Eco Two types of writing: essay writing and novel writing Functions of the Postscript Defining the postscript and differences between the postscript and the novel Functions attributed by Eco to his postscript Justification of thematic and narrative choices Why this novel? How can the reader be involved? The Novel and Reading To write is to build the reader Writing while thinking about what, whom? Writing for what readers? Shaping the reader’s behaviour and questioning To read is to build multiple meanings Role of the title Unexpected connections The text of the novel and other text types Conclusion: Reading, a free and constrained interpretive activity

Introduction

The Novel and Reading

Functions of the Postscript Conclusion

Diagram 1. Diagram of the four sheets handed out to the notetakers in the outline condition.


124

10/24/2006

9:56 PM

Page 124

Annie Piolat

8.6. Appendix 8.B. Note-Taking Example: Visual Formatting Marks (Dashes, Asterisks, Arrows) Produced by a Low-Span Note Taker


9/27/2006

4:23 PM

Page 125

Chapter 9

The Dynamics of Idea Generation During Writing: An Online Study1 Huub van den Bergh and Gert Rijlaarsdam

While the classic models of writing do emphasize, in theory, the dynamic nature of the writing process, the actual analyses, in practice, have tended to present a relatively static view of writing, partly due to the global definition of generating, and partly due to the analyses implemented. This paper aims to demonstrate two ways in which more dynamic models can be investigated, and points to consequences for theoretical insight. First, a more fine-grained categorization of generating activities is used than previous research, assuming different types of generating, functionally determined by the cognitive activity preceding the act of generating, and assuming that the relationship can change during the writing process. Second, to examine the conditional relevance of these activities at different points in writing and the relationship between these activities and text quality at different points in the writing process, sophisticated statistical procedures were implemented. It then shows empirically (1) that different kinds of idea generation do occur with differing overall frequencies, and are more or less prevalent at different points in writing, (2) that these different kinds of generating are related to individual differences between writers, and, most importantly, (3) that the relationship between types of generating and the overall quality of the final text depends on the point in writing at which the activity is carried out. The results demonstrate that the more fine-grained categorization of activities and the identification of when activities occur enable us to improve the prediction of text quality based on observations of the writing process.

9.1.

Introduction

Generating and translating can be considered as the two basic processes in writing. If one has no ideas, or has no access to ideas, one has nothing to write about. If one has ideas but no means to express them, no communication occurs. In this paper we focus on generating ideas. Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Van den Bergh, H., & Rijlaarsdam, G. (2007). The dynamics of idea generation during writing: An online study. In G. Rijlaarsdam (Series Ed.) and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in writing, Vol. 20, pp. 125–150). Amsterdam: Elsevier.


126

9/27/2006

4:23 PM

Page 126

Huub van den Bergh and Gert Rijlaarsdam

If one considers idea generation from a more theoretical point of view, we need to take some distance from the global heuristic models of writing (Flower & Hayes, 1980, 1981; Hayes & Flower, 1980; Hayes, 1996; Bereiter & Scardamalia, 1987) in two respects. In the Hayes and Flower model, idea generation is part of the so-called planning component, in which plans-to-do and plans-to-say lie at the basis of the interaction between three subactivities of planning: idea generation, structuring, and goal setting. In an educational context, writing assignments set more or less fixed parameters for both types of plans (i.e., write a text so and so long for such and such an audience, as a response to such and such a viewpoint or argument). Hence, according to the Flower and Hayes (1980) model, idea generation is mainly constrained by parameters from the task environment (i.e., the assignment as it is presented and the text produced so far). Bereiter and Scardamalia (1987) distinguish two basic configurations: knowledge telling and knowledge transforming. Knowledge telling involves the retrieval of information on the subject matter, and relevant discourse schemas, from long-term memory (see, for instance, Schilperoord, 1996b) and translation of these ideas into language. Successive parts of the text (sentences) reflect more or less directly the spread of activation through associative memory. In knowledge transforming, both sub-processes are involved too, but now mediated by more mature problem-solving strategies by which communicative goals are imposed on the generation process. In the models of writing cited, the researchers stress that in theory they see writing as a dynamic, recursive process, in which generating can follow many other kinds of processes. In their empirical research, however, in practice dynamism is not a central feature. The main emphasis in their empirical work is on the different types of goals towards which generation is directed (content planning vs. rhetorical planning, for example), rather than on the context during writing in which they occur. Alternatively, they emphasize the amount of planning relative to translation. Thus, for Bereiter and Scardamalia (1987) one of the basic distinctions between knowledge telling and knowledge transforming is that knowledge telling involves translating ideas as soon as they are activated, whereas knowledge transforming involves reflecting on ideas, before, while or after writing them down. They support this distinction by differences between children and adults in overall amount of planning and in the amount of reflective planning. For both Flower and Hayes’ (1980) and Bereiter and Scardamalia’s (1987) empirical research, generating is treated as a single, unidimensional activity, measured in terms of total frequency during writing. It is time to elaborate on the multifunctionality of cognitive activities such as generating and on the dynamic character of writing. Therefore, we explore in this paper whether we should distinguish different types of idea-generation processes (multidimensionality), whether these different processes are more likely to occur at particular times or stages in the completion of a writing task (dynamism), and the combined effect of the occurrence of a generating process and the point in the writing process at which it occurs on the quality of the final text (conditional effectiveness). Our approach should result in a better prediction of text quality from planning indices (Hayes & Nash, 1996). 9.1.1.

The Process of Idea Generation

Although different types of generating might be distinguished, the mechanism by which the generation process works might not differ. That is assumed to proceed through spreading


9/27/2006

4:23 PM

Page 127

Dynamics of Idea Generation

127

activation. Ideas occur to the writer as first sets of ‘cognitive nodes’ are activated by (parts of) the assignment, the content of which may (or may not) be put into text, perhaps after a set of structuring activities (i.e., relating ideas, organizing ideas, arranging ideas hierarchically, etc.). As the text grows the internal representation of the text changes accordingly. That is, the nodes and/or their interrelations are constantly changing during text production. Consequently, every change in text representation causes, by means of processes like automatic spreading activation (Anderson, 1983), the activation of nodes which were previously not active. In essence, cognitive processes, such as retrieving information from declarative memory, are bound by the amount of cognitive energy called activation (Rijlaarsdam & van den Bergh, 2005). If the activation level of a cognitive element is low, it is not available in working memory. When the activation is maximal, the concept is in attentional focus, or, to put it differently, it is retrieved into the working memory. “Consequently, only those productions whose conditional specifications match the currently active concepts can be applied” (Schilperoord, 1996b, p. 43). This guarantees that the execution of actions is relevant with respect to the current processing. There are several routes by which a (partial) knowledge structure can become active, although here we only focus on one: spreading activation (see Schilperoord & Sanders, 1999). Spreading activation is what happens when a particular concept is active and starts spreading out its activity to related concepts (Rumelhart & McClelland, 1986; Deane, 1992). Generally speaking, spreading activation accounts for the fact that one concept may facilitate recalling another, associated concept. Which other concepts are activated depends on the structure of the network. The main point to be made here is that the attentional state (working memory) changes due to text production. Application of Andersons’ (1983) theory to writing processes, especially to generation, leads to the hypothesis that every change in internal representation may lead (1) to the activation of different nodes, which were not activated before, or (2) to the activation of the relation between nodes, which were not associated before (compare, Bereiter & Scardamalia, 1987). Hence, generation of ideas does not depend only on task environment factors as the assignment or the text-produced-so-far, or knowledge factors as the discourse scheme at hand (see Schilperoord & Sanders, 1999). Instead, all processes that alter the cognitive representation of the text can set constraints on the generating process. If, for instance, the text is changed, either by adding or revising a clause or sentence, different ideational elements may be activated, and therefore, come into focus. Hence, nodes that were not accessible before are getting attention just because of the growing text, as a result of one cognitive activity or another. So we claim that the activation of nodes might not solely depend on the three mentioned task environment factors. Each cognitive activity in action may affect the cognitive representations of the text, and therefore changes the activation of nodes. Take, for instance, the cognitive activity of generating itself. Generation leads to the activation of nodes, which themselves activate other, strongly related nodes, as part of the associative network of nodes. Another example is the activity of translation of ideas into text. When the activated ideas (propositions, strings, images or in whatever format they are encoded) are put into language, the cognitive representation changes, and hence new nodes will be activated. Thus the translation of ideas into language will trigger new ideas (see, for example, Galbraith, 1996; van den Bergh & Rijlaarsdam, 1996).


128

9/27/2006

4:23 PM

Page 128


We may assume that writers employ specific strategies to guide their generation processes. In these cases, meansend relations are constituted: activity X serves the execution of activity generating. For instance, when a writer is in need of new ideas, he may reread parts of the assignment or the document, as a springboard for generating new ideas or reread parts of the already-written text as a stepping stone for generating. In these instances, the writer implements a goal-directed strategy: If (goal is generating), then first (reread assignment/reread already (part of) written text), then generate (and optionally generate). Even generating can serve as means to generate: If (goal is generating), then generate and generate. In fact, all possible cognitive activities may serve as a means to serve generating, although some will be more often implemented than others, and that writers may vary in the strategies they exploit (inter-writer variability of strategies), and may vary in the way they distribute the strategies over the writing process (inter-writer variability of temporal distribution). In an earlier attempt, we reported various relations between generation and other cognitive activities, showing that various types of generation could be distinguished according to the type of the preceding activity (van den Bergh & Rijlaarsdam, 1999). However, we came to realise that for methodological reasons, these results are hard to interpret. In that study, we took as the unit of observation the combination of generating and some other cognitive activity (e.g., reading the assignment), defined by the type of activity preceding the act of generating. So, we studied the temporal distribution of generating preceded by a specific cognitive activity. Therefore, the distribution of this combination is also determined by the occurrence of the second cognitive activity, which may or may not itself be related to generating; if, for instance, reading the assignment occurred ten times, but only one time before generating, we focused only on the latter; as nine out of ten instances of the former activity are not taken into account, the results may be somewhat distorted. In essence, we assumed that the preceding cognitive activity determined the occurrence of generating; instances in which the preceding activity occurs but generating does not occur, and instances of generating with another preceding activity were not taken into account. Furthermore, it is implicitly assumed that the relation between both cognitive activities changes during the writing process. Hence, the resulting conclusions might (more or less) be built on quicksand. To circumvent this problem in this study, we will estimate the relation between the pairs of cognitive activities, and study the way these pairs influence the occurrence of generating and the way they effect the relation between generating and text quality. In the present study, we aim to establish an empirical basis for the distinction between different types of generating. We will try to show that the distinction between different types of generating enhances our comprehension of written text production. In addition, we intend to show the relevance of the distinction between different types of generating for untangling the relation between writing processes and the quality of the resulting text. That is, we assume writers’ ideas do not pop up at moments they are necessary, but that writer makes active use of available information and available activities to feed the generation process. As said before, the processes by which nodes are activated may be the same for all four types of generation, while the triggering event is different. In this study, we planned to distinguish five cognitive activities that precede generation activities. Next to idea generation as a consequence of reading the assignment (assignment-driven-generation), or idea generation as a result of rereading the text produced so far (rereading-text-driven-generation),


9/27/2006

4:23 PM

Page 129


129

we distinguish between generation as a consequence of translating (translation-drivengeneration), generation based on generation itself (generation-driven-generation), and generation triggered by structuring activities (structuring-driven-generation). Translationdriven-generation occurs when new nodes are activated due to the translation of ideas (propositions, images, strings, etc.) into text. It is the translation process that cues nodes not activated until now. Hence, translation-driven-generation occurs only after a translating (or encoding) phase; generation is preceded by text production. Generation-driven-generation occurs when generating one idea generates the next; generation-driven-generation only occurs if it is preceded by generation. Structuring-drivengeneration shows when structuring of previously generated ideas, or parts of text, lead to the generation of new ideas. After statistical modeling of these five pairs, we had to skip the pair reading-already-written text/generating: none of the models we tried proved to fit. This was not a problem that we were able to solve. 9.1.2.

Distribution of Cognitive Processes Over the Writing Process

A second point to stress concerns the distribution of constituting activities like idea generation over the writing process. When Flower and Hayes (1981) pay attention to the consecutive activities during writing, they conclude that writing is ‘juggling with constraints,’ and at the same time, that in theory, almost every cognitive activity can be preceded by any other cognitive activity or succeeded by any other cognitive activity. We claim, however, that writing is, in fact, more constrained and that in certain stages of writing processes certain cognitive activities (and pairs of activities) are more likely to take place than during other stages of writing. Hence, we assume that cognitive activities are not distributed at random over the writing process. The occurrence of cognitive activities is determined by two factors: (1) the task as perceived by the writer, and (2) the task environment. The first factor determines the global writing strategy, a global configuration of the execution of cognitive activities to be employed (see Levy & Ransdell, 1996). For instance, if the task contains a topic and a genre that is well known to the writer, in general a knowledge telling strategy will do; if the task is relatively new to the writer, relatively more planning is needed in the beginning of the process. When this global strategy is set, the task environment at hand requires for an adaptation from the general pattern. Matching the current situation, the text produced thus far, with the targeted situation, causes the activation of cognitive activities, ‘perceived’ necessary at that moment, while inhibiting other activities. According to our model of writing, the distribution of cognitive activities constituting the writing process is set and guided by a monitor, which gets information from both procedural memory as well as from text produced thus far (see Hayes & Flower, 1980). The monitor is fed by knowledge about a general writing strategy, with some task specific modifications. This knowledge is constructed from former experiences and explicit teaching (Oostdam & Rijlaarsdam, 1995). Some cognitive activities are relatively dominant during one stage, whereas other activities are relatively dominant at another stage. Therefore, the occurrence of cognitive activities has to be studied as a function of changes in the task situation (see Rijlaarsdam & van den Bergh, 1996). The probability of occurrence of a writing activity at a given moment during writing is the relevant feature, rather than the


130

9/27/2006

4:23 PM

Page 130


frequency of occurrence of a cognitive activity. That is, each cognitive activity has, by definition, a probability of occurrence between 0 (does not occur at all) and 1 (does absolutely occur). The probability of occurrence during writing is assumed to be a function of the context made up by other, surrounding cognitive activities. When, for instance, reading of already-written text is going on, the chance that revision will occur is rather large. During the writing process, configurations of chance change, as a result of the changing context, as writing time is progressing and already-written text is restricting the opportunities to occur, like when making a chess move (see Figure 1). Stressing the importance of occurrence at specific moments during the writing process, does raise two types of questions. The first type concerns the temporal distribution of the configuration of probabilities of cognitive activities over the writing process; how does the set of probabilities of occurrence of activities change during writing? In addition, does the interpretation of an activity, or its function in the writing process, vary over the writing process? For instance, in the beginning, an activity like ‘reading the assignment’ serves the aim of ‘tuning to the task’ whereas the same activity during later phases in the writing process serves to check to see whether one is still on the right track, or as an activity to generate new ideas. Hence, the distribution of activities over the writing process seems relevant. The second type of question is about the differences between individuals: do individual differences in cognitive activities emerge not only from differences in frequencies, but also, and perhaps more importantly, from differences between the way individuals distribute the cognitive activities over the writing process? Some writers rely heavily on the planning process at the start of writing, with some revision later in the process, while other writers rely on writing a first draft to create options for thinking, and then revise. These questions set the ground for the third type of question: how are variations in distribution of cognitive activities over the writing process related to the quality of the

Figure 1: A dynamic model of cognitive activities.


9/27/2006

4:23 PM

Page 131


131

resulting text? That is, for studying effective patterns of cognitive activities, it is not only relevant if we know that differences in the number of a certain cognitive activity carried out (i.e., a certain frequency) relate to text quality variance. It is more relevant to find out whether this relation changes over time. In other words, does it matter when a cognitive activity is carried out. For instance, writers starting to write without giving much thought about the assignment, without taking time to consider plans-to-do, or plans-to-say, might not have written a text as well as writers who postponed writing to later stages. Translated to correlation coefficients this comes down to a negative correlation between translation-driven-generation and text quality (TQ) in the initial episodes of the writing process, and (perhaps) a positive relation during later stages of writing. Here the moment in the writing process is considered as an indicator of the changing task situation, or perhaps, to be more precise, as an indicator of rerepresentations of the text produced so far. We hypothesize that generating activities, as they are defined by the preceding cognitive activity (assignment-driven-generation, translation-driven-generation, structuring-drivengeneration, and generation-driven-generation) are not randomly distributed over the writing process (for empirical evidence for other cognitive activities, see e.g. van den Bergh, Rijlaarsdam, & Breetvelt, 1993; Breetvelt, van den Bergh, & Rijlaarsdam, 1994, 1996; van den Bergh & Rijlaarsdam, 1999, 2001; van der Hoeven, 1997). If they prove to be randomly distributed, there is no empirical basis to distinguish these pairs, and our hypothesis of functional pairs will be found to be false. Applying this non-randomness to the hypothesized sub-processes of idea generation, it seems, for instance, likely that assignment-driven generation occurs primarily in the beginning, when writers consulted the assignment. And that in general translation and generation-driven-generation occur later in the writing process. An ideal writer would start with the generation of ideas based on information provided in the assignment. This triggers the process of continuous consecutive text representations, and, therefore, opens the gate to other cues triggering generation. The considerations above lead to two major claims to be empirically verified. First claim is that different types of generation dominate at different moments of the writing process: assignment-driven-generation, translation-driven-generation, generation-driven-generation and structuring-driven-generation are not randomly, but differently, distributed over the writing process. This holds for the general or mean distribution of these cognitive activities for the group as a whole. It also holds for the distribution of each type of generation for individual writers. When we may hold this claim, we have strong indications that different types of generating exist. That is, depending on the moment in the writing process, writers use different strategies to generate new information from long-term memory. Secondly, we claim that different types of generating are differently related to text quality. If we assume that the non-randomness of the distribution of cognitive activities over the writing process is an important characteristic of writing, it follows that the occurrence of cognitive activities during one stage is more important than during other stages. That is, many occurrences at some moment are strongly related to text quality than many occurrences at other moments. Therefore, the correlation between each type of generating and the quality of the resulting text varies during the writing process. Hence, we assume that the correlation between assignment-driven-generation, rereading-text-driven-generation, translation-driven-generation and generation-driven-generation with text quality varies over the writing process. According to the rather simple analysis above, we would expect


132

9/27/2006

4:23 PM

Page 132


assignment-driven-generation to be more relevant or effective during the initial stages of writing, whereas the three other types of generation should be more relevant at later stages of writing.

9.2.

Method

9.2.1.

Design

The study involved real-time measurement of writing behavior, in which writing-aloudprotocols, as well as text quality ratings, were collected. As writing task, we used a documented writing assignment, taken by 36 grade nine students. Each student wrote an argumentative essay under think-aloud conditions on ‘Living alone, yes or no?’ The assignment was accompanied by some documentation –– clippings, quotes –– on the topic and a specification of the communicative situation as a context for the writing task: the underlying reason for writing (contribution to a classroom debate), the focus (personal opinion on the question), the purpose of the text (persuasion), and the audience (peers). Previous to the writing assignment, all students were briefly trained in thinking aloud using math problems. When they got the writing assignment, they were immediately prompted to think aloud. We chose a writing-from-sources task to counteract to some extent the influence of knowledge of the subject matter, and to create an environment in which external sources for idea generation were included. 9.2.2.

Protocol Analysis

Each writing session was audiotaped, and subsequently transcribed. Each protocol was fragmented into units, which consisted of one cognitive process each (e.g., reading the assignment, meta-comments, rereading (part) of the text written, writing, generating, revising, evaluating, etc.); the number of fragments varied between writers from 78 to 1026, and totals to 16,890. Each fragment was coded by two independent raters into 11 categories, derived from Hayes and Flower model, and validated in other studies (Breetvelt, van den Bergh, & Rijlaarsdam, 1994). The agreement between raters was relatively high (92% of the protocol fragments were identically classified); the differences in coding were resolved after a discussion. In the present study, we distinguish between four types of generating ideas, based on the cognitive activity preceding the generating activity itself, being the cue that triggers the generation activity. Hence, assignment-driven-generation occurs (1) if ideas are generated directly after a (re)reading-the-assignment or documentation-activity; translation-driven-generation (2) if ideas are generated directly after producing text; generation-driven-generation (3) if a generating activity follows a generating activity; and (4) structuring-driven-generation, which occurs if an act of generation is preceded by the structuring of previously generated information or structuring of information already written down. In Table 1, a protocol fragment is given for each of these four types of generating activities. From this table it becomes apparent that generating can occur after different cognitive activities.


9/27/2006

4:23 PM

Page 133


133

Table 1: Protocol excerpts illustrating the different types of generating activity. Activity

Protocol fragment (some context; specific fragment in italics)

Assignment-drivengeneration (1)

(Reading assignment) ‘… As adult who starts living alone, you can get a hard time. You must pass the time yourself. Social contacts were left at the parental home’ (Thinking:) ‘No friends anymore’(Writing:) ‘no/less friends’

Translation-driven-generation (2)

(Reading own text:) ‘It is not good that human beings stay alone’ (Writing:) ‘I am’ (Thinking:) ‘I am disagreeing with this statement, I think/ no I don’t agree/ it’s not good … I have to think of something else’

Generation-driven-generation (3)

(Thinking:) ‘ should I continue that line of thinking / eeeehhhm. Something like …/ or did I wrote first something … about … eeh how people live nowadays .../ I think/ eehh/ no, nowadays they don’t find it that strange’ (Writing:) ‘Nowadays they don’t find it that strange’

Structuring-driven- generation (4)

(Structuring:) ‘Oh I have to put the things I don’t agree with first ehh ehh ehh yes first all disadvantages let me think, oh yeah…’ It’s not good for children as well.

Note: The target protocol fragment is shown in italics.

Table 2: Frequency, mean and differences for five cognitive activities (N = 36). Cognitive activity Reading assignment Generating Structuring Translating/writing Total

Frequency

Average

From to

921 1866 541 3491

25.6 51.6 15.0 96.8

2–63 10–246 0–53 22–180

16,980

417.7

78–1016

In Table 2, some descriptive statistics are presented. From Table 2 it appears that of the total number of 16,890 protocol fragments, 1866 are classified as generating. Hence, writers have produced on average (1866/36 ) 51.6 generating fragments. The number of generating remarks, however, varies greatly between writers; one writer produced just 10 generating fragments, whereas another produced 246 fragments in which ideas were generated. These differences between writers do not only pertain to generating, but also to the other cognitive activities.


134 9.2.3.

9/27/2006

4:23 PM

Page 134

Huub van den Bergh and Gert Rijlaarsdam Multilevel Analysis

The data were analyzed by means of multilevel models. The choice for this type of analysis is primarily based on the structure of the data, although there are some practical advantages too. The observations (i.e., protocol fragments) are seen as nested within subjects. As a given activity (at a certain moment) characterizes the writing process of a specific writer, we cannot assume activities as exchangeable, and hence, we have to consider activities as belonging to the writing process of that writer, i.e., nested within writers. A practical advantage results form the large differences between writers in the number of verbalized fragments. To accommodate the necessities of more traditional models, the number of observations is equalized over respondents. Mostly this is achieved by analyzing the frequencies of a certain activity (i.e., the frequency of generating). However, in that case the number of observations is not taken into account, making no difference between a frequency of 10 out of 100 and 100 out of 1000. From a statistical point of view, the expression of scores in terms of percentages is not satisfactory either, as the precision of a score varies with the number of observations (the denominator in frequencies). Therefore, we turn to the (logit of the) probability of occurrence at each point in time during the writing process (see for a demonstration van den Bergh & Rijlaarsdam, 1996). Without going into the details of the statistical modeling here (see the Appendix), we have to estimate the probability of the occurrence of an activity as a function of the passing of time: Y f (time previous activity). Besides the mean change in occurrences of an activity during the writing process, we are also interested in the extent individual patterns differ from this general pattern. Therefore, we have to estimate individual functions, as a separate function for each writer j: Yj fj (time previous activity). The procedure to estimate individual parameters, which relates the occurrence of a cognitive activity to the elapsed time, is quite complex (see for instance, Goldstein, 1991). In essence, however, the procedure boils down to two steps: (1) estimate the mean occurrence, and (2) estimate the individual deviations from the mean. The last step can be done by two procedures: (1) estimating the individual deviations or residuals, or (2) estimating the variance between writers for a given parameter. The second procedure gives the same result, but has the advantage that only one parameter and its variance have to be estimated, and not as many parameters as there are individual writers in a study.

9.3.

Results

The results can be summarized in many ways. We have chosen to summarize them by means of figures. For those who feel more comfortable with numbers, Appendix has been constructed. Alongside the parameter estimates, some information of the multilevel modeling is presented in Appendix. 9.3.1.

Generating and Assignment-Driven Generation

An obvious source for generating new information is, of course, the information provided by the assignment. Figure 2 presents the mean occurrence of generating after reading some


9/27/2006

4:23 PM

Page 135


135

or all of the assignment and generating in general during the writing process. Comparison of both curves shows what the occurrence of generating adds to our knowledge on the occurrence of generating in general. The differences between the two lines are particularly interesting as at those moments the occurrence of reading the assignment influences the occurrence of generating activities. The probability of generating changes over the writing process; after () 20 min the probability of generating is 0.12; i.e., 12% of the observed activities concern generating. In earlier stages of the writing process, it is lower, and in later stages it decreases to (practically) zero. From Figure 2 it becomes apparent that the information in the assignment is important for generating new information; (re)reading (elements of) the assignment increases the probability of occurrence of generating; many generating activities appear to be triggered by information in the assignment. The influence of ‘reading the assignment’ on generating is especially strong during the first part of the writing process. After () 20 min, the probability to generate after having read the assignment is about 0.38; after having read the assignment four out of every ten activities concern generating. Please note that generating is more likely to occur after reading the assignment than in general. Hence, even up to () 60 min the probability to generate after reading elements of the assignment is relatively high as compared to generating in general. So, the assignment greatly enhances generation of information, not only during the initial phases, but also during later phases of the writing process. Two types of patterns emerge from inspection of the individual writers. The first pattern (left-hand panel of Figure 3, writer 2) coincides more or less with the mean pattern, both in the occurrence of generating and generating right after reading the assignment.

Probability of Occurrence

0.5

0.4

0.3

After reading assignment After reading assignment

0.2

Generating

0.1 Generating

After reading assignment

0.0

Generating 0

20

40 60 Time (in minutes)

80

Generating 100

Figure 2: The mean change over the writing process in generating activities after ‘reading the assignment’ and generating in general.


136

9/27/2006

4:23 PM

Page 136


0.7


0.6

2

0.5 0.4 2

0.3 2

2

0.2 0.1

22

0.0 0

20


2 80

100

0.7 19 Probability of Occurrence

0.6 0.5 19

19 19

0.4 0.3

19

19

0.2 0.1 19 0.0 0

20


80

100

Figure 3: The distribution of generating activities over the writing process after ‘reading the assignment’ (dashed lines) and not after ‘reading the assignment’ (solid lines) for two writers.

Generating is clearly stimulated by reading the assignment (6 out of 10 generating activities at maximum), although generating also occurs even if it is not stimulated by reading the assignment. For the second type of pattern, indicated by writer 19, in the beginning generating only occurs after reading the assignment; the probability of generating not after reading the assignment is practically zero. During the writing process, the occurrence of generating


9/27/2006

4:23 PM

Page 137


137

increases continuously. The difference between generating and reading-the-assignmentdriven-generation, however, decreases during the writing process. So, the influence of the assignment on generating activities decreases, until it is negligible (for both writers). In Figure 4, the correlation between generating in general and generating after reading (elements) of the assignment is presented. Both correlations show a steep rise in the beginning, and a fall at the end of the writing process. The correlation between text quality and assignment-driven-generation increases and decreases more markedly than this correlation for generating in general. And, the first correlation reaches its peak earlier in the writing process. Hence, the assignment, and assignment-driven-generation is, as might be expected, important in the beginning of the writing process, whereas at later moments in the writing process the influence of assignment-driven-generation on text quality is hardly noticeable. In other words, generating ideas related to reading the documentation in the assignment is an effective strategy in the first half of the process, while other generating strategies are effective in the second half. 9.3.2.

Generating and Generating-Driven Generation

In Figure 5, the mean change over the writing process for generation and for generationdriven-generation is presented. So, comparison of both curves shows what the occurrence of generating adds to our knowledge on the occurrence of generating in general. Both (mean) curves have more or less a similar course over the writing process. In the beginning, the probability of occurrence of both types of generating is low, but increases steadily during the first 20 min to decrease afterwards. The probability of generating after generating is clearly lower than generating in general. Hence, after an act of generating a writer

0.7

Generation

Correlation

0.5 Generation 0.3

Generation

After reading the assignment After reading the assignment Generation 0.1 After reading the assignment -0.1 0

20


80

100

Figure 4: The correlation (y-axis) between generating (solid) and assignment-driven generating (dashed) over the writing process (x-axis).


138

9/27/2006

4:23 PM

Page 138


0.12


Generating

0.08

After generating

Generating 0.04

After generating Generating After generating Generating

0.00 0

20


80

100

Figure 5: The mean change in generating activities after ‘generating’ and not after ‘generating.’ is less likely to generate. Although the combination generating–generating does occur, generating ideas inhibits the generating of new ideas in the next protocol fragments. We are aware that the mean writer is a construct and may not be exemplary for the individual writer. Therefore, in Figure 6, the probabilities for generating-driven generation and generation not triggered by generation are presented for individual writers. In the left-hand panel of Figure 6 the probability of occurrence is plotted for writer 11. This writer shows a generating behavior that represents more or less the mean writer from Figure 5. In the right-hand panel of this figure the occurrence of generating for two less typical writers is plotted. For writer 11, the general probability of generating increases in the first part of the writing process, but then decreases slowly later on (solid line). However, after having engaged in generating, the probability of engaging in a subsequent act of generating decreases rapidly (the dashed line). The probability of two subsequent acts of generating decreases rather sharply; after () 40 min this probability has decreased to practically zero. This figure shows that writer 11 does engage in generating activities, but only some of them are triggered by generating. Especially during the mid-phase of writing, generating is not likely to trigger another generating activity: for this writer, sequences of generating activities are scarce; the act of generating seems to be embedded in patterns of behavior that do not involve generatinggenerating sequences. Writers 2 and 19 (right-hand panel of Figure 6) show quite a different pattern of generating. Writer 2 shows the general trend of an increase of generating activities during the first part of the writing process followed by a decrease later on (solid line). However, after an act of generating, the probability of generating increases. So, this writer is likely to engage in an act of generating after a previous act of generating; one act of generating is


9/27/2006

4:23 PM

Page 139


139


0.15

0.10 11 11 11 0.05

11 0.00 0

20

11 11 40 60 Time (in minutes)

11 11 80

100

0.7 19 Probability of Occurrence

0.6

19

19

19

0.5 0.4 0.3

19 2

2 2

2

0.2 0.1 0.0

19 19 0

2 2 20


2 80

100

Figure 6: The distribution of generating activities over the writing process after ‘generating’ (dashed lines) and not after ‘generating’ (solid lines) for three writers.

more likely to be followed by another act of generating, than is the case with the average writer (compare Figure 5). Strings of generating activities are produced, without too many occasions where generating is preceded by other cognitive activities. This pattern may indicate that for the writer, generating is a process, not an activity. Writer 19 shows both an atypical generating pattern (solid line) as well an atypical relation between two subsequent acts of generating (dashed line). For this writer, the probability of generating increases during writing; at the end of his writing process almost seven


140

9/27/2006

4:23 PM

Page 140


out of ten activities concern the generation of information (solid line). In the beginning, one act of generating has a (small) inhibiting effect on a subsequent act of generating, which is more or less the general case. After () 20 min, however, during the next 20 min, with the passing of time the probability increases that one act of generating engages in another act of generating. The whole pattern of this writer for generating is typical: it seems that near the end of the process this writer adds a large amount of new information, somewhere within the already-written text, by generating strings of ideas. Which of these deviations from the mean pattern resulted in good texts? Figure 7 presents the correlation between both types of generating and text quality. Suppose we would not have made a distinction between types of generating activities. Then we would have told the story of the solid line: in general generating activities are positively related to text quality in the middle of the writing process. In the beginning and in the end, there is hardly a relationship between generating activities and quality of the written texts (the correlation of .30 being not significant). But the story for generating triggered by generating (dashed line) is different: the correlation is equally absent in the beginning, but then increases more markedly than the mean pattern, to slow down more gradually (see Figure 7). Even near the end of the writing process, generation-driven-generation is positively related to text quality. However, at the end of the writing process generation-driven-generation hardly occurs. So, generation-driven-generation is an act, which affects the resulting text the most in the second half of the writing process. Note that if we had reported the effect of generating-only, we had underestimated the effect of generating on text quality in the second half of the process. In sum, generating as a single activity and generating-driven generating both are related to text quality, in a slightly, though significant different way. Generation in general is

Generation After generation

Correlation

0.6

Generation

0.4

After generation Generation

Generation 0.2

After generation 0.0 0

20


80

100

Figure 7: The correlation (y-axis) between generating (solid) and generating-driven generating (dashed) over the writing process (x-axis).


9/27/2006

4:23 PM

Page 141


141

positively related to text quality, especially between () 20 and () 40 min the correlation between the probability of generating and text quality is strong. For generating-drivengeneration the correlation with text quality is lower in the beginning of the writing process, but decreases less as compared to generating in general. So, in the second half of the writing process, generating-driven-generation seems a sensible strategy to write good texts, as compared to isolated instances of generating. Writers must generate in the beginning, but must not often produce strings of generating, while after the mid of the process, generating is beneficial to text quality, but those writers, who produced strings of generating activities in this phase, deliver better texts. Please note that this does not imply that writer 19, so deviant from the mean pattern, did a good job: it might be that this writer performed other cognitive activities in a less effective pattern. In fact, writer 19 delivered one of the lowest rated texts, but this does not appear to be attributable to his way of generating. 9.3.3.

Generating and Structuring-Driven Generation

The (re)structuring of information might well lead to the activation of nodes that were not activated before. So, (re)structuring of available information might be a sensible strategy for the generation of new information. However, (re)structuring can only be useful if relevant information is available. Therefore, it is expected that especially during later phases of the writing process structuring activities will enhance the occurrence of generating activities. Figure 8 presents the mean probabilities of structuring-driven generation and generation that does not occur after structuring activities. Structuring of generated information does not seem to increase the amount of generating activities. On the contrary, in the beginning of the writing process structuring information


0.25

0.20

0.15 Generating 0.10

0.05

After structuring Generating

Generating After structuring

After structuring 0.00 0

20

Generating Generating 40 60 Time (in minutes)

80

100

Figure 8: The mean change in generating activities after ‘structuring’ and not after ‘structuring.’


142

9/27/2006

4:23 PM

Page 142


inhibits the generation of information somewhat; the probability that generating occurs after a structuring remark is less than after other activities. In the second half of the writing process, however, structuring exerts a positive influence on generating in general (see Figure 8). That is, in the second half of the writing process, generating occurs relatively more frequent after acts of structuring than it does after other types of cognitive activity. Figure 9 presents the individual probabilities for four writers. In the left-hand panel, writer 2 and 11 can be considered as representing average writing processes; they behave more or less the same as the mean writer. Generating occurs less frequently after structuring in


0.3 2

2

0.2

2 11

2

0.1

11 11

0.0

2

11 0

11 20


1.0

80

100

17

17

2

17

17

0.9 Probability of Occurrence

11 2

2

0.8 0.7 0.6

19 17

19

0.5 0.4

19

17 17

17

0.3

19

0.2 0.1 0.0

19 19 0

20


80

100

Figure 9: The distribution of generating activities over the writing process after ‘structuring’ (dashed lines) and not after ‘structuring’ (solid lines) for four writers.


9/27/2006

4:23 PM

Page 143


143

the beginning, whereas in the second halve of the writing process it occurs more frequently after structuring. On the right-hand side of Figure 9, two less typical writers are presented. They are not only atypical in the probability of occurrence of generating activities, but also with respect to the relation between generating and structuring activities. Contrary to the average writer, these two writers increase the frequency of generating activities during the writing process. For writer 17, structuring activities are in the beginning of the writing process a steppingstone for generating activities. Later on, after () 20 min, there is hardly any relation between structuring and generating activities; that is, the probability of occurrence of generating activities does not change due to the occurrence of structuring; generating is as likely to occur after structuring as after any other activity. Although writer 19 shows, as noted before, a deviant distribution of generating activities over the writing process, the general relation between generating and structuring activities remains intact: At the start, generating activities are less likely to occur after structuring, while during later phases of the writing process the occurrence of generating activities are more frequent after structuring. Figure 10 shows the changes in correlations between text quality and generating in general and generating after structuring during the writing process. From this figure it becomes apparent that generating after structuring can have a positive effect on text quality. But this effect is limited to only a very small time-interval. During the first 10 min generating triggered by structuring is even negatively related to text quality. Afterwards the correlation rises quickly until a maximum is reached after 38 min. Later on the correlation with text quality gradually diminishes. If we had restricted the analyses to generating-only activities, we would have interpreted the positive correlation between

0.8

Correlation

Generation After structuring

0.4

Generation Generation Generation

0.0

After structuring

After structuring -0.4 0

20


80

100

Figure 10: The correlation (y-axis) between generating (solid) and structuring-driven generating (dashed) over the writing process (x-axis).


144

9/27/2006

4:23 PM

Page 144


generating and text quality in the beginning of the process as an effect of generating. But now we may nuance this result: it is an effect of generating, but not of the type that is related to structuring. From another perspective, the relation between generating and text quality is underestimated in the period between 35 and 60 min, where the correlation between generating and text quality is lower than the correlation between structuring/generating and text quality. 9.3.4.

Generating and Translating-Driven Generation

Every writer knows that sometimes new ideas pop up while writing. Hence, the act of writing can trigger generating activities. In Figure 11, the effect of translating or writing on generation becomes apparent. From Figure 11 it is apparent that the act of writing itself enhances the probability of occurrence of generating activities. Up until about 60 min, the occurrence of generating after an act of writing is higher than not after an act of writing; after writing the probability that generating will occur is almost doubled. Furthermore, the occurrence of translating-driven generation follows the same general pattern over the writing process as generation in general. It is low in the beginning, reaches a peak at around 20 min to diminish later on to about zero at the end of the writing process. In Figure 12, the individual patterns of occurrence of translating-driven generation for four writers are presented. On the left side of this figure, the patterns for two average writing processes, in this respect, are presented, while the right-hand panel shows two atypical writers. The left-hand side figure shows that the difference between generating in general and translation-driven generation varies between writers. Writer 11 gives evidence of a marked increase in generating activities after writing activities during the first part of


0.20

0.15

After writing

0.10

After writing

0.05

Generating

Generating

After writing Generating 0.00 0

20


80

Generating After writing 100

Figure 11: The mean change in generating activities after ‘writing’ and not after ‘writing.’


9/27/2006

4:23 PM

Page 145


145

his writing process. Afterwards generating activities are less likely to occur after an act of writing, as compared to generating in general. The same holds more or less for writer 2, although the differences are less marked. The right-hand panel of Figure 12 shows two distinct patterns. Both these writers appear to increase the occurrence of generating activities throughout the writing process. Writer 17 almost solely generates information at the end of his writing process. Interestingly, this writer shows a clear decrease in generating activities after an act of


0.3 2

2

0.2

2 11

2

0.1

11 11

0.0

2

11 0

11 20


1.0

80

2

100

17

17

0.9 Probability ofOccurrence

11 2

2

17

17

0.8 0.7 0.6 0.5 0.4

17 17

17

19 19

17

0.3

19

19

0.2 0.1

19 19

0.0 0

20


80

100

Figure 12: The distribution of generating activities over the writing process after ‘writing’ (dashed lines) and not after ‘writing’ (solid lines) for four writers.


146

9/27/2006

4:23 PM

Page 146


writing. Writer 19, who also increases the number of generating activities during his writing process, on the other hand shows a moderate increase in translating-driven generation during the first half of his writing process. In Figure 13, the correlation between translating-driven generation and text quality and between generation (not after writing) and text quality are presented. Both types of correlation show the general pattern, low in the beginning and then rising, to diminish later on. However, the correlation for generation after an act of writing shows a more marked course during the writing process. In the beginning, it is practically absent, but after 30–40 min it is remarkably high (r 0.87), and in the remainder of the writing process, it stays well above the correlation between text quality and generating in general. If we had limited our analyses to generating as decontextualized activity, we would have had underestimated the effect of generating markedly after 20 min of writing. Note that the writing-generating sequence indicates a knowledge telling process (Bereiter & Scardamalia, 1987): it seems that the knowledge telling process is a strong strategy in the middle of the writing process, even when the task involves writing from sources.

9.4.

Discussion

We have shown that preceding cognitive activities influence the occurrence of generating. That is, for a certain moment in the writing process, we can predict whether or not generating is likely to be the next activity if the writer’s current activity is known. Crucial in this respect is the moment in the writing process: previous activities may decrease or increase the likelihood of generation depending on the moment in the writing process. For instance, structuring decreases the probability of generating in the beginning, but

0.9

After writing Correlation

0.7 Generation

0.5 Generation Generation After writing Generation

0.3 After writing

0.1 0

20


80

100

Figure 13: The correlation (y-axis) between generating (solid) and writing-driven generating (dashed) over the writing process (x-axis).


9/27/2006

4:23 PM

Page 147


147

increases the probability of generating later in the writing process. For translating, or writing, the reverse seems to be the case; in the beginning writing appears to stimulate generation, whereas during later phases in the writing process generating is less likely to occur after writing. Individuals appear to differ largely with respect to the distribution of generating over the writing process, and also with respect to the influence of previous activities. Although the general pattern indicates a decrease of generating after structuring in the beginning, and increase in later phases of the writing process, for some individual writers the opposite seems to be the case; structuring enhances the occurrence of generating in the beginning but diminishes the occurrence of generating during later phases of the writing process. Relations between the temporal distribution of generating after a certain activity and text quality appear to be rather self-explaining. For instance, assignment-driven-generation is a rather adequate strategy in the beginning of the writing process, but less effective (in terms of quality of the resulting text) during later phases of the writing process. Translation-driven-generation is related positively to text quality as well, but especially later in the writing process as assignment-driven-generation. And structuring-driven-generation is again positively related to text quality, but at a later moment than translationdriven-generation. Of course, assignment-driven-generation is (positively) related to text quality in the beginning of the writing process; writers need to tune to the assignment; they have to use the available information in the assignment to retrieve information from longterm memory. That translation-driven-generation is related to text quality during later phases is not surprising: it makes sure that given the beginning of the text, the text itself stays coherent. The relationship between structuring-driven-generation and quality is rather complex; from negative in the beginning to high during the middle to low in the end of the writing process. Hence, the effectiveness of structuring-driven-generation seems to rely on the availability of information (in the beginning), but considerations of the order of content elements do enhance the probability to generate new information and is positively related to text quality. Note that this generating strategy almost secures a coherent text, as new information is related to much of the previous information and not only to the information in the previous sentence. In Introduction section, we claimed that cognitive operations can change the mental representation of the text. Hence, by means of automatic spreading activation, new nodes (or new information) would be activated resulting in an increase in generating activity. This line of reasoning has clearly proved too simple. First of all, not all activities that change the cognitive representation of the text do enhance the occurrence of generating activities. For instance, an act of generating clearly changes the text representation, but decreases the probability of generating; after an act of generating the probability of generating decreases. Or, structuring activities seem to diminish the likelihood of generating activities in the beginning of the writing process. More in line with our a priori expectations are the results for ‘assignment-driven-generation’ and ‘translation-driven-generation.’ Both appear to increase the occurrence of generating activities. That is, after reading or rereading part or all of the assignment and after an act of writing, the probability of engaging in an act of generating is relatively high. Second, we would like to point to the large differences between individual writers. The occurrence of an act of generating is largely dependent on the writer. These large differences


148

9/27/2006

4:23 PM

Page 148


limit the value of general patterns. To understand the occurrence of cognitive activities we need to study the patterns of individual writers in detail. All in all, the theory thus far does not explain the occurrence of generating activities. Too many factors that seem important are not taken into account in the existing writing-process theory. In particular, the relations between cognitive activities are not widely studied. We think that we have shown that it is worthwhile to pursue this line of research, and that it adds to our understanding of the writing process. We have also shown that generating ideas is not a one-dimensional activity, but that it can be stimulated by different cognitive activities during different phases of the writing process. Furthermore, the relation between these combinations of activities and text quality appears to dependent on the moment in the writing process; what appears to be effective in the beginning, might not be effective during later phases. Hence, such combinations, of which we only studied four, appear to be conditionally effective. Disregarding relationships between cognitive activities implies that the relationship between activity and text quality is less precise than it could be.

9.5.

Author Note

This chapter presents a reanalysis of data originally presented in van den Bergh, H., & and Rijlaarsdam, G. (1999). The dynamics of idea generation during writing: An online study. In M. Torrance & D. Galbraith (Eds.), Knowing what to write. Conceptual processes in text production (pp. 99–120). Amsterdam: Amsterdam University Press.

9.6.

Appendix

Let Yij be a dichotomous variable, which indicates whether writer j (j = 1, 2, … J) generated information at moment i (i = 1, 2, … Ij; Yij = 1), or not (Yij = 0). In that case, the occurrence of generating over the writing process can be modeled as a function of the moment of occurrence. If we denote the time elapsed since the start of the writing process with tij, then the occurrence of generating can be modeled by means of a polynomial: Logit (Yij) B0 tij0 B1 tij1 + B2 tij2 ⋅⋅⋅ Bk tijk (u0j tij0 u1j tij1 u2j tij2 + ⋅⋅⋅ ukj tijk)

(1)

(i 1, 2, …Ij; j 1, 2, …, J; k 1, 2, …, K) That is the occurrence of generating (Yij 1) is modeled as a polynomial of tij (powers of tij). Such polynomials are known to be extremely flexible, and, depending on the number of coefficients and their numerical value, they can take almost any shape (compare van den Bergh & Rijlaarsdam, 1996). The model consists of two parts: the fixed part and the random part (between brackets). The fixed part describes the mean changes in generating over the writing process, whereas the random part describes individual deviations from this general or mean pattern. The number of coefficients in the fixed as well as in the random part is considered an empirical question.

Else_SW-Torrance_CH009.qxd 9/27/2006

Table A.1: Parameter estimates for generation and generation-driven generation (TQ: text quality; standard errors between parenthesis). Random part of the model (see Equations 2 and 3): variance of B0 tij0 (.17) (.05) (.01) (.18) (.08) (1.3)

.10 .01 –.00 –.07 .66

(.03) (.01) (.08) (.03) (.16)

.01 .02 .01 .06

(.00) (.01) (.01) (.31)

B4 PCA tij0 B5 PCA tij1

TQ

1.05 .35 1.75

(.36) (.15 (.34)

.20 .99

(.08) (.17)

244.9

(57.8)

B0 tij0 –2.11 (.14)

B1 tij1 –.25

(.07)

B2 tij2 –.12

(.02)

B3 tij3 .02

(.00)

B4 PCA tij0 .54

(.21)

B5 PCA tij1 .17

(.06)

TQ 100.9

(2.61)


Fixed part of the model (see Equation 2 and 3)

Page 149

.65 .11 .02 –.38 –.17 4.10

B2 tij2

4:23 PM

B0 tij0 B1 tij1 B2 tij2 B4 AC tij0 B5 AC t1 TQ

B1 tij1

149


150

9/27/2006

4:23 PM

Page 150


In order to estimate the relation with another cognitive activity, we have to take into account that such a cognitive activity can only influence generating if it has occurred before a specific act of generating. So, it is not the occurrence of an activity at moment t, which influences generating at moment t, but an activity at moment t–1. So, the explaining processes always have to be one lag behind the process to be explained. If we denote the previous cognitive activity with PCAij, we can write the model to be analyzed as: Logit (Yij) B0 × tij0 B1 tij1 + B2 tij2 + ⋅⋅⋅ + Bk tijk B(k+1) × tij0 PCAij B(k+2) tij1 × PCAij ⋅⋅⋅ B(k+k) tijk × PCAij

(2)

(u0j tij0 u1j tij1 u2j tij2 ⋅⋅⋅ ukj tijk u(k1) PCAij tij0 u(k2) PCAij tij1 ⋅⋅⋅ u(kk1) PCAij tijk) The first line of the fixed part of the model (B0–Bk) gives the pattern of generating if the previous cognitive activity (PCAij) does not occur. The combination of the first and second line determines the mean pattern of occurrence of generating after the explaining cognitive activity. So, if the mean coefficients (B(k+1)–B(k+k)) reach significance, there is a relation between the occurrence of generating after the explaining cognitive activity. The random parameters determine the individual patterns after the explaining cognitive activity and not after the previous cognitive activity. The relation with text quality (Y2j) is rather easy to model. Just assume a multivariate model, with that equation (2) as the first part and Y2j α 0 vj

(3)

as the second part of the model. Now the residuals are allowed to covary. Please note that these covariances, or when standardized: correlations are a function of (powers of) tij. Hence, the correlations are allowed to change over the writing process. In Table A.1 above, the estimates are presented for generation-driven generation.


9/27/2006

12:14 PM

Page 151

Chapter 10

Skilled Writers’ Generating Strategies in L1 and L2: An Exploratory Study Sophie Beare and Johanne S. Bourdages

The aim of this study is to investigate the use of generating strategies in first (L1) and second language (L2) writing of eight skilled bilingual writers (English/Spanish). The study was guided by the following research questions: What writing strategies are used in facilitating generating by skilled bilingual writers in their first and second languages, and are there differences between the strategies they use in their first and second language? Data were collected through think-aloud protocols. The results of this study confirm the view that skilled bilingual writers use similar writing strategies in L1 as in L2, when using within-subject comparison.

10.1. Introduction Writing processes of second language learners have been at the core of several studies in the last two decades. Among the different research topics related to writing, the differences between first (L1) and second language (L2) writing have been systematically attracting attention with different perspectives. Much of the research on this topic has examined how L2 learners compare to L1 writers (Berman, 1994; Cumming, 1989; Silva, 1993; Jones & Tetroe, 1987) and Berman (1994) found many similarities between L1 and L2 composing. Studies have also been concerned with describing the behaviour of unskilled (novice) L2 writers (e.g. Arndt, 1987; Uzawa, 1996) and skilled L2 writers (e.g. Cumming, 1989; Sasaki & Hirose, 1996), but very few studies (Matsumoto, 1995; Sasaki, 2002) have looked at bilingual writers who may be using their L2 writing for purposes other than academic, for example for professional purposes. This article focuses on generating strategies used by skilled bilingual writers (English/Spanish). These writers are noteworthy as they have achieved a proficiency level similar to L1 writers and cannot be considered as L2 learners in the traditional sense. The


Beare, S., & Bourdages, J.S. (2007). Skilled writers’ generating strategies in L1 and L2: An Exploratory Study. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 151–161). Amsterdam: Elsevier.


152

9/27/2006

12:14 PM

Page 152

Sophie Beare and Johanne S. Bourdages

inquiry into skilled bilingual writers and their L2 writing strategies may prove interesting from the point of view of establishing differences between two “stabilized” systems. 10.1.1. Generating Strategies Among the different components of the writing process, content generating is considered to play a major part (Bereiter & Scardamalia, 1987; Caccamise, 1987; Collins & Gentner, 1980; Gould, 1980; Hayes & Flower, 1980; Levy & Ransdell, 1995). In order to generate, writers must first retrieve information from their long-term memory (Hayes & Flower, 1980). Generating content or generating ideas is explained as a sub-process of the writing process (Hayes & Flower, 1980) this sub-process may be viewed as either automatic or controlled (Alamargot & Chanquoy, 2001). According to Torrance, Thomas and Robinson (1996), choice of the writing task may determine whether the idea generation is automatic or controlled, and in familiar tasks the idea generation could be mainly automatic. The generating sub-process as described by Hayes and Flower (1980) is divided into three steps (Alamargot & Chanquoy, 2001): (1) knowledge retrieval; (2) evaluating appropriateness of the retrieved units of knowledge; (3) evaluating usefulness of the retrieved units of knowledge. In this study, the focus is on the controlled functioning of the generating sub-process that can be observed in the writer’s overt behaviour while writing or reported in the verbal data of the think-aloud protocols. Chenoweth & Hayes (2001) explain that ideas are retrieved and proposed in short “bursts”. At first, these ideas are in pre-linguistic stage, then they are put into propositions or “bursts” in think-aloud protocols. A writing strategy, for the purpose of this study, is defined as a sequence of activities rather than an event (Broekkamp & Van den Bergh, 1996; Fayol & Monteil, 1994; Garner, 1988). For example, a language-switching strategy could involve a sequence of activities in generating content in L2 such as: doing memory search in L1, translating the idea into L2, thinking of another similar idea in L2, switching to L1 and so on. The use of a strategy is said to be controlled consciously by a writer. The writer would know during the writing process when to switch languages and when to translate an idea or a word into L2. A writer may have difficulty retrieving an idea or a word if the particular idea or word was learnt in an L1 context (Friedlander, 1990). The writer may switch to L1 to retrieve the idea and then translate it into L2 (Chenoweth & Hayes, 2001). Moreover, a strategy may be covert, for example when no dictionary is used, but it could be captured by verbal-report data such as thinkaloud protocols. Research has shown that use of L1 in L2 writing is a fairly common strategy (Cumming, 1989; Roca de Larios, Murphy, & Manchon, 1999; Sasaki, 2002). It has also been demonstrated that highly skilled writers in second language who perform equally well in the first language could easily use language switching or translation as a strategy. Sasaki (2002) found that both novice and expert writers used translation strategy from L1 to L2. The results indicated that novice writers stopped more frequently during translation. In another study by Uzawa and Cumming (1989), findings show that English speakers studying a foreign language, intermediate level Japanese, used their L1 to search for ideas, prepare notes, write drafts and organize mentally. A strategy could be used throughout the writing process or for a sub-process, such as generating content (Broekkamp et al., 1996). Among writing strategies, re-reading has also been the focus of several studies (Sasaki, 2002; Levy & Ransdell, 1995). In view of


9/27/2006

12:14 PM

Page 153

Generating Strategies in L1 and L2

153

Bereiter and Scardamalia’s (1987) Knowledge Transforming Model, re-reading can be explained as a strategy used to generate content by reworking the text to create new ideas (Alamargot & Chanquoy, 2001). In their model, Bereiter and Scardamalia (1987) present a knowledge-telling model which assists the knowledge-transforming model in content generating. Chenoweth and Hayes (2001, p. 84) proposed a model for written language production including “text written so far” activating the internal components in their model. These included a proposer that retrieves prelinguistic knowledge or ideas, a translator that then puts these into a verbal expression and a transcriber that transforms the expression into a written sentence. According to this model, re-reading the text written so far would activate generating either through controlled retrieval or reworking what is written so far. Van den Bergh and Rijlaarsdam, in this volume, explain re-reading as an activity in generating, naming it “Rereading-Text-Driven-Generation.” The focus in this study is on the short propositions or “bursts” verbalized by participants in the think-aloud data. Some new propositions may follow the re-reading strategy, some are retrieved by language-switching strategy and other new propositions by simply expressing all kinds of thoughts/ideas until the one the writer wants to keep. We can call this an idea-generation strategy. This may resemble Van den Bergh and Rijlaarsdam’s “Generation-Driven-Generation activity”. 10.1.2. Contributing Factors in L2 Writing Research on bilingual writers indicates that language proficiency is a factor in writing (Berman, 1994; Cumming, 1989; Roca de Larios, Murphy & Martin, 2002; Sasaki, 2002). Although high language proficiency and writing expertise are psychologically different (Cumming, 1989), high language proficiency has a positive effect on the writing product. When their language proficiency improves students produce better texts. Friedlander (1990) studied 28 Chinese-speaking subjects to determine the effects of a first language on writing in English as a second language. The subjects were university students. The author’s results demonstrate that language may constrain writers in a certain way during the writing process; if the writers use the language in which they acquired the topic or the subject, their writing is enhanced. Generating strategies would be affected by this finding, as generating or idea creation in bilinguals may be using both languages to retrieve content, when in difficulty. Previous research also shows variation in other sub-processes of writing. Differences were found (Silva, 1993) among undergraduate students who had advanced levels of proficiency in English as a second language and who displayed a wide range of levels in writing ability. More specifically, the subjects performed less re-reading and reflecting in writing texts in their L2. Silva’s research shows that although general composing process patterns are similar in L1 and L2, there are differences in other aspects of writing: L2 writers did less planning at the global and local levels, and L2 writers did less goal setting and had more difficulty achieving these goals. L2 writers in Silva (1993) had difficulty in organizing generated material in L2 but not in L1. Studies such as Berman (1994), Jones and Tetroe (1987) and Matsumoto (1995) suggest that writers transfer their writing strategies from L1 to L2 provided they possess L2 grammatical proficiency (Berman, 1994). Berman studied 126 secondary school EFL students.


154

9/27/2006

12:14 PM

Page 154


Through analyses of students’ pre- and post-test essay organization and grammatical proficiency, Berman (1994) concludes that many learners transfer writing skills between languages and that their success in doing so is assisted by their grammatical proficiency in the target language. Other researchers (Jones & Tetroe, 1987) analyzed protocols to study the L1 and L2 planning behaviours of six Spanish-speaking L2 writers, all of whom were in a nine-month intensive program preparing for graduate studies. Their proficiency ranged from Band 3–4 (Carroll, 1980) at the beginning of the study, and Band 6 at the end of the study, the highest level being Band 9. Carroll (1980) in his book, Testing Communicative Performance, explains the nine-point language-band system of the English Language Testing Service used in British Institutions, starting with Band 1or 0: non-user; 2: intermittent user; 3: extremely limited; 4: marginal; 5: modest; 6: competent; 7: good; 8: very good and ending with band 9: expert user. Each band gives a description of the writer; for example, band-9 expert writer writes with authority, accuracy and style, has mastery of appropriate and concise English, and band-2 intermittent writer is described as having no working facility with perhaps sporadic uses. The interaction between proficiency L2 and the writing performance ability was the focus of Jones and Tetroe’s (1987) study. The findings show similarity in composing processes between L1 and L2. Matsumoto (1995) interviewed four Japanese university professors on their processes and strategies for writing a research paper in English and found that these professional writers use strategies similar to those used by native English speakers. In addition, she suggests that already existing L1 writing strategies get transferred to L2 writing. Although there has been much research in comparing L1 and L2 writing processes in the last few years, there are still some limitations in the research design from the previous studies. Most of the subjects in the study are learners, frequently at the beginning of undergraduate levels, in college, or university or at the secondary school levels. Many of the subjects may be international students with strong and active L1 capabilities and EFL proficiency in L2. Other students may be permanent residents in English-speaking countries with ESL proficiency, but their L1 is only used at home or in small social circles. They may have stopped writing in L1. In other words, the findings from previous research may be difficult to replicate or to apply to a particular group of writers because of the heterogeneous characteristics of the L2 subjects. The aim of the current study was to fill a research gap by conducting a study with subjects who were fully bilingual and were skilled writers in both L1 and L2. The following are the research questions that were generated by reviewing previous research and from a very small pilot study. Before conducting our research with eight bilingual subjects (Spanish/English), the research instruments and the procedure for the main study were tested. Two subjects, both skilled writers, were used: one monolingual (English) and one bilingual (Polish/English). Polish was chosen as a first language since the first author had knowledge of both languages so that the analyses of the pilot study could be conducted quickly and our preparations for the current study would begin. In the results of the pilot, the bilingual participant had shown differences in generating. In idea-generation without re-reading, her think-aloud recording indicated more activity in her L1 (21% more). But, her activity level in L1 was similar to the monolingual’s L1 (English) results. While using the re-reading strategy to generate, her activity was higher than the monolingual subject. In summary, the between-subjects comparison showed similarities in L1 generation


9/27/2006

12:14 PM

Page 155


155

strategies, but within-subject there were differences between L1 and L2 when generating. We found very minimal language-switching strategy in L2 and none in L1. 10.1.3. Research Questions Research question one: When generating in L2 will the writers reveal more languageswitching strategies than when generating in L1? Studies (Sasaki, 2002; Roca de Larios et al., 1999; Uzawa & Cumming, 1989) reveal more language-switching strategies in L2 than when generating in L1. Research question two: Do bilingual writers use generating strategies at the same level of activity in both L1 and L2 writing? The literature (Berman, 1994; Jones & Tetroe, 1987) provides data that bilingual writers transfer skills from L1 to L2. Thus the level of activity would be similar in both languages. Research question three: Are both groups (native speakers of English and native speakers of Spanish) use the same level of activity in L1 and L2 generating? This last question was based on the small pilot study where both L1 English writer and L1 Polish writer had similar level of activity in generating in their L1.

10.2. Method 10.2.1. Participants The subjects of this study were eight bilingual writers (English/Spanish) skilled in writing in both their L1 and L2. Four participants were native speakers of Spanish born in Mexico, Colombia, Cuba or Chile. The remaining four participants (two born in Canada and two in South America) were native speakers of English who specialized in Spanish at the university level. Both languages, English and Spanish, were used in their work and at home. They studied at the post-graduate level, or worked as professors and/or wrote and published in both languages. They were recommended by professional contacts as bilingual individuals with a high level of proficiency. All of them started learning their L2 either early in life (even before school) or later when they started high school. They frequently came from families that were also bilingual in those two languages and some married similar individuals. Among the participants, there was a fiction writer who published his short stories in English or Spanish, a university professor who published her academic articles in social science journals, in both English and Spanish and an international journalist from Cuba who wrote in both Spanish and English. There were three language teachers who taught both English and Spanish to language learners at language schools; finally, the last two subjects were graduate students who studied literature in their second language, did some teaching at university and wrote essays, poetry and short stories. For subjects’ information, see Table 1. In order to participate in this study, each subject, in addition to showing his/her published work, had to write a paragraph in his/her L2 and score Band 9 on the British Council ELTS test (see Appendix 2). The paragraph was to confirm their good writing skills as they were already identified as bilingual writers. No further assessment was conducted.


156

9/27/2006

12:14 PM

Page 156


Table 1: Participant information. Participanta

Gender

Kevin Eduardo Jorge

Male

20–30 30–40 40–50

Language Teacher Writer, Language Teacher Writer, Language Teacher

Canadian Cuban Chilean

Cathy Maria Paula Mary Kristin

Female

20–30 20–30 20–30 30–40 30–40

University Student (graduate) Language Teacher University Professor, writer University Student (graduate) Language Teacher

Canadian Colombian Mexican Canadian Canadian

Age

Occupation

Origin

a

Not real names.

10.2.2. Procedure Our analysis involved comparison of L1 and L2 writing by the same writers, and also comparison between writers with Spanish and writers with English as their L1. Before the study, the subjects participated in an interview in which they talked about how they wrote in L1 and L2. The participants were trained individually to think aloud in English by the first author before the first session. During the study, the participants were required to write two essays: one in their L1 and one in their L2. Based on prior interviews, two topics were prepared, making sure they were familiar with them in either language. L1 topic: You have lived in two cultures: your own and that of your second language. Compare them. What are some of the noticeable differences? For example, do people in your culture see the collective or group rights as more important than individual rights? L2 topic: You speak two languages and perhaps can teach them both. Compare English and Spanish by discussing some of the difficulties you had learning your second language and some of the difficulties you notice other learners have when learning Spanish or English. The time given for each writing session was two hours. Each session was audio-taped and subsequently transcribed. The interview and writing sessions were held in their work or home offices, or wherever they usually did their writing. The same researcher (SB) was always present during the writing sessions. All communication between the researcher and the subjects was carried out in English and was audio-taped. During the sessions, subjects were allowed to use bilingual dictionaries when necessary. After completing the writing sessions, all Spanish participants were asked to translate their L1 think-aloud transcription into English. It was deemed that an English translation done by the author of the think-aloud session would provide the researchers with more accurate interpretation. These participants were able to do that effectively and they were paid for the translation. Subsequently, the translated protocols were checked by a professional Spanish/English translator.


9/27/2006

12:14 PM

Page 157


157

10.2.3. Protocol Analysis Think-aloud protocols were used to collect data. As mentioned, all protocols in Spanish were translated in English. In the think-aloud transcriptions, the expressions were short and not in a sentence form. The transcriptions were analyzed using the original and translated forms. The coding used was based partially on the coding system that Chenoweth and Hayes (2001) used in their research. The segments or contents that were proposed in new statements were marked and counted. If the other language was introduced the expression was analyzed for new statements or new expressions following it. Re-reading of the text used so far was noted as a strategy if a new proposition or new statements came after it. If a new proposition or new statements came up without re-reading, then the idea-generation strategy without re-reading was used. In the following segment of the written protocol, two types of propositions or statements are shown: (1) a new statement made after re-reading a section of the “written text so far” (2) a new statement made while not re-reading the written text. … I observed…students uhm…who will try to argue (new statement) …while teaching Spanish (new statement)…I observed students who will try to argue (rereading)…who try to transfer (a new statement after rereading) and even translate from Spanish to English (new statement)… Language switching was coded separately. It was noted as a strategy if a new proposition or new statement occurred following language switching. Since the SB had already practised protocol coding during the pilot study, and had a good understanding of the coding done in Chenoweth and Hayes (2001) research, SB coding was conducted. Thus, reliability of the protocol coding would be high.

10.3. Results 10.3.1. Language-Switching Strategies Language-switching strategy occurs during generating when a writer switches from the language of writing, L2, for example, into his/her L1 or from L1 (the language of writing) to L2. Previous research indicates that more language-switching occurs when the writer is less skilled. In this study only three out of the eight participants switched to L1 during generating content in L2 and, among these three, two of them switched to L2 during L1 content generating. Eduardo used only L1 when generating L2 content. Maria and Cathy used the other language when generating in L1 or L2. Moreover, Cathy was the only one of the three who used a bilingual dictionary regularly during her writing process. Eduardo, during his L1 session, generated entirely in L1 (Spanish). He began his L2 session generating in L1, then slowly switched into English as the session progressed. In his think-aloud recording 18% of content generating was done in his L1. The topic may have been less familiar in L2 than L1 (Friedlander, 1989). This writer had equal number of words in both L1 and L2 product, even though his L2 recording for content generating was much shorter than his L1 content generating.


158

9/27/2006

12:14 PM

Page 158


Maria’s think-aloud recording during generating in L1 (Spanish) had the following words “behaviour… I am thinking about the word in English but I want it in Spanish… now I have it” (translation). In two instances, in Maria’s recording L1 (Spanish) languageswitching occurred; they were single words. In L2 the same participant had a similar situation with seven instances. Again, they were words or short expressions. She quickly and discreetly substituted the new word or expression and kept going. Because she switched languages in both L1 and L2, it may be her style that is common among some fully bilingual individuals (as reported by many). However, it occurred more frequently in L2 which may have an impact on the writing fluency. Cathy, on the other hand, would generate whole chunks in her L1 (English) using Spanish such as “I went to Spain in the summer” (translation). Then, she paused and translated into English. She did that two times in her L1, and in L2 (Spanish) about 30% of her generating was done in English. She wrote in her L2 but generated in L1 from time to time, paused and translated before writing. She also used her bilingual dictionary. Before the session, Cathy told the researcher that she liked writing in Spanish, especially short stories and poetry; therefore, we were quite surprised to see so much L1 in content generation of L2 writing. Cathy’s frequent use of L1 in L2 content generation may have many possible explanations: (1) the L2 topic was not as familiar in L2 as in L1. If the writer has not learnt about the topic in the second language, then the writer may switch to L1 while retrieving content from the long-term memory. Cathy’s recording had the largest number of utterances in her think-aloud and, yet, her final product in L2 is of average length; (2) the individual may have had training in translation and has developed a habit that overlaps with the composing process in second language; (3) the writer has experience in L2 writing, but the fluency in writing is weak and the writer’s proficiency of the second language may be lower than that of the others. In summary, only three subjects out of eight used language-switching strategies. For these three participants, the incidents of language switching were higher in L2 than in L1 during content generating. This finding supports other research findings such as Sasaki (2002) who reports that some of her subjects used translation as a strategy, but that it was more common with novice writers. Hypothesis 1 is supported for only three participants; content generating in L2 revealed more language-switching strategies than when generating in L1. Since less than 50% of the participants used this strategy, Hypothesis 1 is not completely supported. The participants in this study had a proficiency level in their L2 that was very close to native speakers. Their high level of proficiency may explain the fact that they did not need to revert to L1 in their composing. Significance testing was not applied because of the sample size being only eight subjects; thus, the findings cannot be generalized. 10.3.2. Two Strategies in Generating: Re-reading the Text So Far and Idea-Generation English L1 writers proposed on average 19 new statements in L1 and 21 when writing in L2. Each statement was generated after a re-reading strategy (reading the text produced so far). On average, Spanish L1 writers proposed 25 new statement/propositions in L1 and 11 new propositions or new statements when writing in L2 after using a re-reading strategy (on the text produced so far) in L2. The English L1 writers seem to come up on average


9/27/2006

12:14 PM

Page 159


159

with a near equal number of new propositions in L1 and L2 (19/21). The Spanish L1 writers did not use the re-reading strategy very frequently. However, they did use idea-generation strategy similarly in their L1 and L2 (23/27 new propositions or new statement) (see Table 2). English L1writers used idea-generation frequently by producing statements or new language in their L1 texts, 116 statements and 50 new statements in L2. The generating in L1 by English writers seems to be very active with some variation between participants. Jorge (Spanish L1), who has published several of his works in both languages, had only 3 new propositions in his L1 generating by re-reading, but 45 new propositions by idea-generation strategy. He did not use the re-reading strategy fully. Perhaps he had the text in his long-term memory and just added more ideas. He could have used a knowledge-telling strategy where retrieving is more automatic. This study deals with the controlled generating but the topics may be more familiar to some writers. In summary, the within-subject comparison has demonstrated similarity in English writers in L1 and L2 generating using the re-reading strategy, and for Spanish writers similarity in L1 and L2 generating using the idea-generation strategy. Since the study was exploratory using a case-studies approach with fully bilingual writers, the findings are interesting but do not have large enough sample to allow inferential statistics. In some ways, this is a pilot for a larger study. Comparison between English and Spanish writers indicated no trend; the levels of activity in generating differed in both groups and the type of strategies they used. Once again, the individual differences were evident.

10.4. Discussion Though the data emerging from this study are exploratory and limited in many ways, they indicate the generating strategies preferred by skilled writers in their L1 and L2. Our hypotheses were not fully confirmed by the study and the small sample prevented us from performing significance testing. Nevertheless, there were new findings that could stimulate replication of the study with a larger sample and the results could contribute to the links between full language proficiency and content generating in writing. The study has been of exploratory nature with the aim to examine the generating strategies in writing of bilingual adult writers. As our results indicate, language switching is not that frequent among highly proficient bilingual writers. There is a need to separate the languages while one writes. Languages are culturally bound and going back and forth from one language to the other may be disruptive. Future studies need to look into the issue of language switching in L2 writers with a proficiency level close to L1 writers. Re-reading was preferred by native speakers of Spanish in their L1 and idea generation was preferred in their L2. The native speakers of English preferred idea-generation in their L1, as well as in their L2. And its use was about the same in both languages. How can we explain this difference? Van den Berg and Rijlaarsdam (this volume) propose a model with four types of generating: Assignment-Driven-Generation (idea generation as a consequence of reading the assignment), Re-reading-Text-Driven-Generation (idea generation as a result of re-reading the text produced so far), Translation-Driven-Generation (idea generation as a result of translating) and Generation-Driven-Generation (generation based

Mean NSS (SD)

Kristin Cathy Mary NES NES NES

Kevin NES

Mean NES (SD)

17 28

3 45

24 8

24.75 (21.98) 22.75 (17.35)

16 40

33 88

7 165

21 171

19.25(10.84) 116 (63.21)

23 37

12 11

2 17

5 46

10.5 (9.33) 27.75 (16.48)

28 7

21 77

16 22

18 91

20.75(5.25) 49.25(40.99)

Note: NSS, Native Spanish Speaker; NES, Native English Speaker.

Page 160

55 10

12:14 PM

Paula NSS

9/27/2006

L1 (Spanish/English) Generating by re-reading Idea-generation(no re-reading) L2 (Spanish/English) Generating by re-reading Idea-generation(no re-reading)

Eduardo Jorge NSS NSS


Maria NSS


160

Table 2: Propositions/statements made during generating.


9/27/2006

12:14 PM

Page 161


161

on generation itself which occurs when generating one idea generates the next). They assume that these four types are differently distributed over the writing process. They also propose that the mean distribution of these cognitive activities would be distributed differently for each type of generation among individual writers. Our results seem to go in that direction. But, a question remained unanswered: How can the difference between L1 Spanish speakers and L1 English speakers be explained in terms of their specific preferences in generating strategies? It is a challenge to find a very homogeneous group, outside the school setting. The individual differences play such a major role, especially in small group samples. Each individual in this study had a particular characteristic in generating writing. For example, one participant did almost no re-reading of the written text when generating in L1, another started generating in L2 solely using L1 and then switched over and continued in L2 only. There were other examples of differences among individuals in generating, stressing their importance.



9/27/2006

1:02 PM

Page 163

SECTION 2: EFFECTS OF WRITING ON COGNITION



9/27/2006

1:02 PM

Page 165

Chapter 11

The Writing Superiority Effect in the Verbal Recall of Knowledge: Sources and Determinants Joachim Grabowski

Language production processes are frequently involved in the diagnosis of knowledge. However, the relation between the knowledge basis and the spoken or written output is seldom addressed. We investigated whether and how language production processes related to speaking and writing systematically influence the results of verbal diagnoses of knowledge. Particularly, the writing superiority effect turned out to be a stable and replicable finding: In adults, writing allows for higher content validity of the indication of knowledge, compared to speaking. A theoretical analysis of the oral- and written-language production processes and the related cognitive load through its particluar subprocesses explains why linguistic output generally does not provide a valid window to cognition. For the diagnosis of knowledge, the advantage of writing as opposed to speaking is experimentally demonstrated. In subsequent experiments, working-memory capacity as well as the correspondence between the verbal modalities of knowledge input and output prove to be determining factors of the writing superiority effect, whereas verbal intelligence as well as stress and arousal seem to exert no influence.

11.1.

Speaking, Writing, and the Diagnosis of Knowledge

This chapter reviews several years of theoretical development and experimental investigation on a rather simple question: Are speaking and writing equivalent with respect to the diagnosis of knowledge? Apart from sign languages as a third full-range means of linguistic expression, spoken- and written-language productions are considered the two predominant mirrors of the knowledge that people have available in their minds. However, do both mirrors reflect the underlying knowledge equally well? — The general answer will


Grabowski, J. (2007). The Writing Superiority Effect in the Verbal Recall of Knowledge: Sources and Determinants. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 165–179). Amsterdam: Elsevier.


166

9/27/2006

1:02 PM

Page 166

Joachim Grabowski

be “no,” and the way in which the two verbal modes diagnostically differ will subsequently be called the writing superiority effect. In the present section, we briefly describe the issue of oral and written knowledge diagnosis, along with some involved methodological problems. Then, we will summarize a series of experiments within a methodical paradigm designed to locate the sources of diagnostic differences between oral and written knowledge recall (Section 2). Next, we will try to gain further understanding of the involved mechanisms by looking for determinants that might increase, alter, or destroy the writing superiority effect; this will be described in Section 3, together with a recent experiment in which the writing superiority effect is scrutinized for the interaction between the verbal modes of input (i.e., reading vs. listening) and output (speaking vs. writing). Finally, we will give a concluding account of what we have learned so far and what we still need to find out about the roles of speaking and writing in the diagnosis of knowledge (Section 4). In oral- as well as in written-language production, an important part of planning does not only relate to retrieval processes from memory, but also to selection processes (Grabowski, 1996; Herrmann & Grabowski, 1995). This means that, for most cases of everyday communication, our rhetorical skills and competence involve to select from among the underlying information available in memory those pieces of information that we will subsequently verbalize in order to successfully pursue our communicative goals. For example, we will omit information that our partner already knows or that does not fit the goals of the present situation. In the case of writing, it is characteristic for the transition beyond a simple knowledge-telling strategy that children learn to not to write down every piece of information that they can retrieve from memory, but carefully decide which information will be needed, or functional, for a good text (Scardamalia & Bereiter, 1987). However, there are also situations in which it is required, or really imperative, to circumvent our competence for selection processes because the task is to verbally convey everything that we know on a certain topic or field. Typically, such situations are related to the diagnosis of somebody’s knowledge. In the educational world, this is regularly the case with all sorts of exams, and also eyewitness reports and other kinds of detailed accounts and descriptions call for completeness rather than selection. In practice, the person (e.g., a teacher) or institution (e.g., the court) interested in what somebody knows about something must definitely decide whether to use speech or writing in order to make somebody’s knowledge verbally appear. From the perspective of laboratory research, however, we can first ask whether speaking or writing would provide the more accurate picture of somebody’s actual knowledge. Of course, it would be a too naive view to assume that verbalized linguistic products of any modality, spoken or written, directly reflect the underlying knowledge; neither spoken nor written utterances can be analogous indicators of the knowledge basis they externalize. Nevertheless, we can look into the content validity of speaking and writing in the diagnosis of knowledge: Is one of the two verbal modalities superior with respect to the diagnostic relation between the thematized information in the observable utterances (⫽ the symptom, as it were) and the potentially, or maximally, available information in memory (⫽ the cause)? Thus, we do not ask the usual diagnostic question: What does somebody know? But: Given a certain knowledge basis, does it appear better, or more complete, in spoken or in written recall?


9/27/2006

1:02 PM

Page 167

Writing Superiority Effect

167

In order to delimit our issue, three things are worth noting. First, although cognitive psychology has extensively dealt with the diagnosis of knowledge up to now, it predominantly addressed the question of the representation of knowledge, i.e., how knowledge is organized in human memory, going on with the assumption that the respective information is cognitively available (Anderson & Schooler, 2000). In contrast, we are interested in how to find out what somebody knows, not considering the cognitive structures, or formats, of the underlying representations. Second, cognitive psychology developed a wealth of laboratory methods for the assessment of knowledge and its structures, from similarity judgements and sorting tasks (Hinsley, Hayes, & Simon, 1977) to modern eye-tracking (Alamargot, Dansac, & Chesnet, this volume). These procedures, however, are little practical in everyday contexts. In contrast, we are interested in the diagnosis of knowledge as it is employed with exams, or eyewitness reports, outside the laboratory, which mainly involves ‘normal’ speaking or writing. Third, it has been shown that utterances resulting from the retrieval of knowledge are strongly influenced by the perspective taken during the perception and cognition of the retrieved information (Anderson & Pichert, 1978). In contrast, we are interested in the verbal externalization of knowledge independent of a particular encoding perspective. When investigating the diagnostic validity of speech and writing, some methodological difficulties arise from the widely known fact that we can’t resolve three variables from one equation, here: the knowledge in memory, the linguistic output, and the diagnostic relation between them. With ‘normal’ diagnostic situations like exams, the linguistic output of the examinee is taken as an indication of his or her knowledge and cognitive abilities, relying on the diagnostic relation between thought and its verbalization, which is taken for granted. Now, this diagnostic relation itself is subject to scrutiny; the observed linguistic output must serve as an indication of the content validity of speaking as compared to writing. Therefore, we need to take the underlying knowledge for granted; i.e., we need to have access to what somebody knows independent of the information given in his or her utterances. However, people substantially differ with respect to the knowledge they have. In order to exert some control on the participants’ knowledge, at best to guarantee that the spoken or written recall of participants in the respective experimental conditions is based on comparable knowledge, we came up with three methodical approaches. First, to reduce the variability of the multiple interconnections within individual knowledge, it is advisable to work with knowledge elements that can be well circumscribed. Here, all knowledge domains appear to be feasible that are countable and deliminated against other domains. This is the case, for example, with geographical knowledge like the recall of the European capitals or of the states of the United States, both adding to some 50 items, which avoids floor effects as well as ceiling effects when asked to recall all of them. Second, a determination, or at least approximation, of the maximally available knowledge in memory can be achieved through a post-test using cued recall. By the use of appropriate cues, we can increase and improve the recall of knowledge, thus getting closer to the maximally available knowledge in memory. For European capitals, for example, the associated states can be an effective cue, and the initial letters can effectively cue the American states. If we then compare the recalled knowledge in speaking or writing with the results of cued recall, we get a measure of underachievement that says how far the participants, in their spoken or written knowledge diagnosis, remained behind their potential


168

9/27/2006

1:02 PM

Page 168

Joachim Grabowski

knowledge. Third, the most consequent way of laboratory experimentation is to induce ‘artificial’ knowledge that is afterwards diagnosed, thus avoiding any effects of previous knowledge. In the following sections, we will meet examples of each method.

11.2. Sources of Diagnostic Differences Between Speaking and Writing Models of the language production process usually assume some hierarchically structured levels, with a stage of pre-linguistic planning at the top, formulation processes in the middle, and articulation, or grapho-motoric execution, at the bottom (Grabowski, 1996; Herrmann & Grabowski, 1995; Levelt, 1989; cf. Alamargot & Chanquoy, 2001). With some simplification, the subsystems of formulation and articulation (or grapho-motoric execution, respectively) can be assumed to be largely automatic in adults, but not in children (Bourdin & Fayol, 1994; Grabowski, 2005), although even adults may experience increased cognitive load in writing due to formulation and execution difficulties (Bourdin & Fayol, 2002; Fayol, Largy, & Lemaire, 1994). However, the decisive factor that puts load on the cognitive system is the high-level processes of planning (Jou & Harris, 1992; Power, 1985). The functional locus of these high-level processes is an instance of central control, which comprises two functionally distinguished parts, a declarative part for storage, and a procedural part for central execution. Our notion of central control follows the concept of working memory with limited capacity (Baddeley, 1986, 1996). Central control is considered responsible, with speaking as well as with writing, for the initiation of the language production process in general, for the retrieval of information from long-term memory, and for the maintenance of this information in working memory as the basis for the ongoing verbalization processes. The appropriate adaptation of verbalized information to contexts and partners through selection processes is a further resource-consuming task of central control. However, the demands of these — and some other — sub-processes on cognitive resources may vary from situation to situation and depend on the speaker’s experience and proficiency. With respect to the diagnostic validity of speaking as opposed to writing, we now concentrate on a theoretical analysis of possible differences between the two verbal modalities’ capacity demands on working memory. The more cognitive load results from the peculiarities of speaking or writing, respectively, the less resources are available for information retrieval and maintenance as input for the further language production process, which in turn should lead to a decrease of the oberservable utterances’ content validity. There are particularly three parameters on which we may assume an important variation between the two verbalization modes (Grabowski, 1996, 1999). 11.2.1.

Discourse Protocol

A prerequisite for linguistic and thematic coherence of utterances or text is to remember what has already been said or written. Discourse protocol refers to the storage in working memory of the previously produced informational contents (Hjelmquist, 1984), as opposed to the verbatim protocol of the linguistic surface of utterances, which is assumed to be stored at an automatic level and which is normally lost after a few seconds (Glanzer, Fischer, & Dorfman,


9/27/2006

1:02 PM

Page 169


169

1984; Sachs, 1967). If a speaker does not know whether he or she has a particular piece of information already produced, the very information may either be repeated (which violates conversational maxims) or omitted (which may impair the success of communication). With speaking, the record of information already produced must be maintained exclusively in memory, because the traces of spoken behavior are volatile. Writing, however, provides an external store, the already written text, that can be re-read and looked up. Therefore, writing involves, when compared to speaking, a potentially relieving factor for working memory through its external store for the discourse protocol. 11.2.2.

Time Per Unit

The motor execution of writing takes longer than the motor execution of speech; articulation proceeds faster than grapho-motoric behavior. Cognitive resources compare, in terms of physics, to power and not to work, which means that cognitive capacity relates to resources per time unit. Since the verbalization of a particular portion of information takes longer in writing than in speaking, the cognitive resources can be used for a longer period of time, from which information retrieval from long-term memory as well as planning processes should benefit. In addition, the knowledge representations just being verbalized may remain longer activated until the associated orthographic representations are read out for writing, which may increase the probability of associated knowledge items to become also activated and retrieved. On this parameter, again, writing involves the potentially less demanding characteristics on working memory. 11.2.3.

Pacing

Speaking calls for continuous progress; we must either go on with our speech production or at least justify any longer pause in order to keep our turn (Sacks, Schegloff, & Jefferson, 1974). With writing, the pacing of language production is self-determined; we can stop the grapho-motoric process, stare at the ceiling and concentrate our attention solely on retrieval or planning processes. This third parameter also implies a potential relief of the cognitive load in writing, compared to speaking. All three parameters say something for the smaller cognitive load of writing compared to speaking, yielding the assumption that the written recall mode should allow for a better and more complete indication of an examinee’s potentially available knowledge: The less load the verbalization process itself puts on the cognitive system, the more capacity should remain for the complete recall of the available knowledge from long-term memory. In order to test this assumption, we designed four experimental conditions that create a variable of increasing cognitive load, according to the theoretical considerations given above (see Table 1). (1) In normal writing, the advantageous values of the three factors are all combined: Visible writing offers an external store for the discourse protocol, proceeds slowly due to the written execution, and is self-paced. (2) In invisible writing, the support from the external discourse protocol is removed, which introduces one additional aspect of cognitive load as compared to the previous condition.


170

9/27/2006

1:02 PM

Page 170

Joachim Grabowski

Table 1: Four experimental conditions that create a variable of increasing cognitive load. Normal writing

Invisible writing

Dictating

Normal speaking

Discourse protocol

⫺

⫹

⫹

⫹

Time per unit

⫺

⫺

⫹

⫹

Pacing

⫺

⫺

⫺

⫹

▲

cognitive load

The ‘invisibility’ of the written traces can be achieved by using ‘magic ink’ or carbon paper, in the case of handwriting, or by programming a text editor, which makes the screen empty after each return key, in the case of typing. (3) In dictating, participants were allowed to use the stop button of a recorder when recalling their knowledge into a microphone (without using the rewind function, which would provide them, again, with an external discourse protocol). Thus, a second additional facet of cognitive load was introduced, namely the faster execution of speech compared to writing. With the use of the stop function, however, the verbalization process remained self-paced and, in a positive sense, interruptable. (4) In normal speaking, eventually, all cognitively rather demanding values of the factors are combined: There is no external store for the discourse protocol, articulation proceeds fast, and the pace of speaking requires some continuous progress in production. With the help of this design, we can investigate whether speaking and writing are equivalent, or not, with respect to content validity. If not, we can additionally localize the effect, discovering the relevant theoretical parameter(s). Next, we briefly report on three experiments in which the above design was used; here, we only refer to the most central results. 11.2.4.

Experiment I

This experiment tested the diagnostic equivalence of speaking and writing with geographical knowledge.1 11.2.4.1. Method Participants were 60 students (20 male, 40 female) from Mannheim University (mean age: 24.6 years), 15 per condition. They were asked, under one of the four conditions, to recall all European states and all European capitals they knew. They were not required to recall only associated pairs of a state and its capital; if necessary, they could solely recall a state without its capital or a capital without its state. The two writing conditions were conducted with handwriting; invisible writing was operationalized with

1 This experiment was conducted by Karin Huber in the framework of her diploma thesis, supervised by the author. We here report on an improved and corrected re-analysis of her data.


9/27/2006

1:02 PM

Page 171


171

magic ink. In the dictating condition, the use of the stop button was shown. Participants were left alone in the room during verbal recall and were given as much time as they needed. Thus, knowledge recall occured as a power test, not as a speed test. Having signalled that they had completed the task as well as they could, participants were presented with a cued-recall post-test, which comprised of an alphabetic list of the European states where the associated capitals were to be filled in. For all conditions, the post-test was given in written modality. One experimental session took about 20 min. 11.2.4.2. Results Results are shown in Table 2. The most interesting result relates to the underachievement in the experimental conditions, compared to cued recall performance. A oneway analysis of variance yields a highly significant effect (F(3, 56) ⫽ 5.43; p ⬍ .01). Post hoc comparisons (Duncan) localize the effect between the two written and the two oral conditions. 11.2.4.3. Discussion Depending on the modality in which the participants’ knowledge was assessed, they ‘knew’ on average three to four items more in the writing conditions than in the oral conditions, which is, compared to the absolute means of capital recall that range between 15 and 18 items correct, some 20% extra just by employing a mode of knowledge diagnosis that appears to save cognitive resources! With respect to the content validity of the verbal modalities, writing appears clearly superior to speaking (⫽ writing superiority effect). The localization of the effect between invisible writing and dictating, on the variable of increasing cognitive load, indicates that the decisive factor appears to be the duration of processing information from memory to physical execution or articulation, respectively. The presence or absence of an external discourse protocol does not seem to play any important role. 11.2.5. Experiment II This experiment tested the diagnostic equivalence of speaking and writing with nonverbally induced knowledge. 11.2.5.1. Method Participants were 52 students (14 male, 38 female) from Mannheim University (mean age: 22.3 years), 13 per condition. First, they were presented with a posterboard on which 40 simple objects were graphically represented (e.g., key, star, tooth, candle). In a pre-test, the objects were controlled for unambiguous naming. The experimenter read Table 2: Means and standard errors for relevant variables of Experiment I. Normal writing Capitals correct 17.7 (1.71) (experimental condition) Capitals correct 20.1 (1.40) (cued recall) Underachievement 2.5 (0.92)

Invisible writing

Dictating

Normal speaking

NOVA

16.7 (1.45)

16.9 (2.47) 14.8 (1.80)

n.s.

17.9 (1.90)

22.6 (2.65) 20.1 (1.95)

n.s.

1.1 (1.19)

5.7 (0.98)

5.3 (0.61)

p ⬍ .01


172

9/27/2006

1:02 PM

Page 172

Joachim Grabowski

aloud each object’s name once; participants were then given 6 min to memorize the objects. Next, the “proverbs” scale from the WILDE intelligence test was administered as a measure of verbal intelligence. This intermediate task was also to suppress immediate memory effects. After that, participants were asked, under one of the four conditions, to recall as many of the previously memorized objects as possible. The two writing conditions were conducted with handwriting. Invisible writing was operationalized by a pile of index cards; participants in this condition were instructed to write on each card only one recalled object’s name and then to throw the card into a kind of mail box before writing on the next one. In the dictating condition, the use of the stop button was shown. Participants were given as much time for recall as they needed. At the end of the experiment, the participants’ listening span was measured (according to Daneman & Carpenter, 1980). 11.2.5.2. Results Results are shown in Table 3. Because in this experiment, the diagnosed knowledge was induced in the laboratory, it is not necessary to control previous knowledge by a post-test. A oneway analysis of variance yields a significant effect (F(3, 48) ⫽ 3.04; p ⬍ .05) for the numer of correctly reproduced items. A planned contrast, following the expectation derived from Experiment I, localizes the effect between the two written and the two oral conditions. The verbal intelligence measure does not differ between the four conditions, nor does it significantly predict, or correlate with, the number of recalled items. The listening-span measure does not differ between the four conditions (which indicates successful randomization). A median split of the listeningspan measures shows that participants in the upper half of listening span capacity perform significantly better on the recall task than the participants of the lower half (28.4 vs. 25.6, p ⬍ .05). However, there is no interaction between the four conditions and listening span with respect to the number of recalled items. 11.2.5.3. Discussion While all participants committed the presented materials to memory ceteris paribus, it depends on the verbal modality of recall how much they will know when their active recollection is tested. Again, a writing superiority effect was observed, localized between the same two conditions as in Experiment I. Verbal intelligence does not play any role in this context. Listening span, as a measure of working-memory capacity, exerts a linear effect on the number of recalled items, but does not interact with the diagnostic modalities. This means that working-memory capacity contributes to the prediction of individual recall performance, but does not, or not sufficiently, explain the

Table 3: Means and standard errors for relevant variables of Experiment II.

Items correct Verbal IQ Listening span

Normal writing

Invisible writing

Dictating

Normal speaking

ANOVA

27.7 (1.47) 107.2 (3.33) 3.8 (0.32)

29.9 (1.21) 112.9 (3.72) 4.1 (0.26)

24.9 (1.43) 107.5 (3.02) 3.5 (0.32)

25.2 (1.27) 101.7 (3.95) 3.8 (0.29)

p < .05 n.s. n.s.


9/27/2006

1:02 PM

Page 173


173

writing superiority effect. (This would only be the case if the listening-span size played a stronger role in speaking than in writing.) 11.2.6.

Experiment III

This experiment tested the diagnostic equivalence of speaking and writing with episodic knowledge. 11.2.6.1. Method Participants were 63 students from Mannheim University, 16 per condition (invisible writing: 15). Episodic knowledge was induced through a film of a theft in an optician’s shop. Participants were instructed to recall the previously observed event such that the listener, or reader, would be enabled to exactly reconstruct the chain of events in every detail. The resulting spoken or written texts were segmented into content units and then classified according to the episodic structure derived from the original film. (The system of event analysis used in this experiment is explained in greater detail in Rummer, Grabowski, & Vorwerg, 1995.) 11.2.6.2. Results and discussion Results are shown in Table 4. Again, normal writing and invisible writing on the one hand, and dictating and normal speaking on the other hand, led on most variables to similar results within the respective pair of conditions, but to significant differences between them. In the two speaking conditions, there were significantly more content units than in the writing conditions that refer to an episode of the event (p ⬍ .05). This is, however, due to the fact that in the speaking conditions, as compared to the writing conditions, much more repetitions of previously already mentioned episodes occured (p ⬍ .001). The number of different episodes recalled in the verbal accounts is the same across all four conditions. Thus, the spoken accounts are longer, but at the cost of redudancy. Further, the grain size of the given information differs across the modalities; the number of generalized propositions (macropropositions according to van Dijk, 1980) is highest with normal speaking and lowest with normal writing (p ⬍ .05). And most important, there is a tendency toward a higher proportion of wrong information (reproduction errors) in the speaking conditions, compared to the writing conditions (p ⬍ .06). With respect to a correct, precise, and repetition-free recall of episodic knowledge, we again find writing superior over speaking.

Table 4: Means and standard errors for relevant variables of Experiment III. Normal writing No. of episodes No. of repetitions No. of different episodes No. of macropropositions No. of reproduction errors

27.4 (1.56) 1.0 (0.30) 26.4 (1.55) 3.7 (0.56) 0.9 (0.22)

Invisible writing

Dictating

Normal speaking

ANOVA

23.1 (2.08) 32.1 (2.53) 29.6 (2.47) p ⬍ .05 1.1 (0.35) 5.8 (1.01) 5.3 (0.76) p ⬍ .001 22.0 (2.01) 26.3 (1.80) 24.3 (2.10) n.s. 5.0 (0.54) 5.9 (0.38) 7.1 (0.91) p ⬍ .01 1.7 (0.72) 2.5 (0.37) 2.3 (0.35) p ⬍ .06


174

9/27/2006

1:02 PM

Page 174

Joachim Grabowski

11.2.7.

General Discussion

The experiments described above provide typical examples of studies that look into the content validity of the various verbal modalities of knowledge diagnosis. The results of Experiment I have also been successfully replicated in a study with students of another university. This study followed the same experimental paradigm, but the writing conditions were operationalized through typing on a computer keyboard; in the invisible writing condition, the European capitals were typed into an editing window on the screen, which was emptied as soon as the enter key was pressed (but, of course, stored in a hidden protocol file). Thus, the superiority of writing over speaking in the diagnosis of knowledge does not depend on the mode of writing (handwriting versus keyboard typing). Particular attention may be given to the fact that in research on the writing superiority effect, list-like encyclopedic knowledge, graphical knowledge, and episodic knowledge can be successfully treated within the same theoretical and methodological context, while general psychology in many cases carefully keeps apart the various types of knowledge representation in memory. Up to now, there was only one attempt to replicate the writing superiority effect that failed; when teacher students tried to recall political knowledge (listing the 16 German states after the reunification, together with their capitals and their political leaders), a floor effect foiled all visible differences. Whenever intelligence measures were taken in experiments on the writing superiority effect, there was no systematic correlation with recall performance. Hitherto, we can summarize the obtained findings in three respects. First, the study of the diagnostic validity of speaking and writing allows for the theoretically and empirically uniform treatment of different kinds of knowledge: encyclopedic, episodic, or experimentally induced. The issue at question appears to be a very general phenomenon. Second, there is strong evidence that writing allows for better content validity than speaking, when the task is to recall the underlying knowledge as completely as possible. Third, the critical difference between speaking and writing appears to relate to the temporal management of recall and execution processes, which seems to be more favorable with writing, in terms of information retrieval from long-term memory, information maintenance in working memory, and finally the verbalization of information through language production processes. In contrast, the presence or absence of an external store for the already verbalized information does not affect the amount of recalled knowledge. Rather, the invisible writing condition produced in all experiments the numerically highest recall rate; a fact that, although not statistically significant, may be worth exploring. The possibility of interrupting the continuous stream of language production in order to direct one’s attention to a particular sub-process, as it is provided in dictating as opposed to normal speaking, seems to have no effect on the diagnostic results as well.

11.3. Determinants of the Writing Superiority Effect So far, we have the writing superiority effect as a stable finding that allows for replication across different tasks and knowledge domains. However, in order to better understand an effect, its scope and its causes, it is helpful to know about the ways in which other factors can influence it. Thus, we are investigating the effects of related possible determinants that


9/27/2006

1:02 PM

Page 175


175

may increase, alter, or destroy the writing superiority effect. (Sometimes, to know what makes an effect disappear may contribute as much to its proper understanding as a series of successful replications.) One possible determinant has already been addressed in the previous section: individual working-memory capacity, which can be measured with the various memory-span tests (Ransdell & Levy, 1999). Working-memory capacity has a linear effect on the amount of recalled knowledge (as with Experiment II), or on the size of underachievement (as with a replication of Experiment I). It improves the recall of knowledge in all verbal modalities, but does not account for the observed differences between speaking and writing. Subsequently, we will report on three experiments in which further factors were studied that could possibly influence the writing superiority effect: stress, self-attention, and the modality of knowledge input. Since we know from the findings described in Section 2 that the relevant difference for the writing superiority effect is between normal speaking and normal writing, these experiments only compare the speaking and writing modality, neglecting the transitional conditions of invisible writing and dictating. 11.3.1.

Experiment IV

It is particularly important to obtain valid results of knowledge diagnosis in the context of examinations. However, examinees tend to experience some increased arousal when their knowledge is assessed by the examiner. In order to find out whether the writing superiority effect also applies in situations like exams, we tested whether (moderate) stress would have an influence on the verbal modalities of knowledge diagnosis. 11.3.1.1. Method In this study, 74 teacher students (8 male, 66 female) from Heidelberg University of Education (mean age: 22.6 years) participated who were randomly assigned to one of the four conditions of a 2 ⫻ 2 design. One independent factor was the verbal modality of recall, with the two values writing and speaking. The other independent factor was low versus moderate stress, induced by an tangram puzzle that the participants had to solve at the beginning of the experiment. The puzzle was either very easy or very difficult, so that almost all, or almost none, of the participants in the respective condition were able to solve it within the given time. Both groups were told, however, that most people would succeed with the puzzle. After that, the geographical task from Experiment I (European states and capitals) was employed. Then, participants underwent a listening-span measure, followed by the cued recall post-test (see Experiment I above). Finally, an anxiety questionnaire was administered that assessed state as well as trait anxiety. 11.3.1.2. Results The anxiety questionnaire showed that the induction of stress was successful. While there was, according to randomization, no difference in trait anxiety between the conditions, participants in the two moderate stress conditions reported significant higher state anxiety than participants in the low stress conditions (T(72) ⫽ 4.1; p ⬍ .001). With respect to underachievement in the recall of European capitals (cued recall minus number of correct capitals recalled in the experimental condition), we received the following means (standard errors given in parentheses): writing/low stress 3.8 (0.66); writing/moderate stress 3.6 (0.60); speaking/low stress 5.8 (0.90); speaking/moderate stress 4.7 (0.77). An analysis of variance yields a significant effect of the verbal modality (F(1, 73) ⫽ 4.3; p ⬍ .05); all


176

9/27/2006

1:02 PM

Page 176

Joachim Grabowski

other effects and interactions, including the influence of listening span, remain below a significant level. 11.3.1.3. Discussion Although moderate stress was successfully introduced through the difficult tangram puzzle task, it does not affect the writing superiority effect, which appears in the expected way as a stable and robust effect. 11.3.2.

Experiment V

An alternative explanation of the writing superiority effect, which is sometimes raised in discussions, relates to the role of self-attention in language production. It is argued that self-attention may be higher with speaking than with writing, because with speaking, the speaker would experience more reflection on his or her own participation in the discourse. Increased self-attention, then, should weaken the cognitive resources available for knowledge recall, thus accounting for the lower content validity of speaking, compared to writing. To test this assumption, we artificially increased self-attention during knowledge recall. If this is a decisive factor, the results in the writing condition should come closer to the speaking results, because speaking, which is anyway characterized by higher selfattention, would not suffer much from a measure that attracts self-attention. 11.3.2.1. Method Participants were 80 students (25 male, 55 female) from Mannheim University (mean age: 23.0 years). The experiment can be characterized as a two-in-one design, combining the European capital task (see Experiment I) and the induced knowledge task (see Experiment II). First, participants saw the posterboard with its 40 graphical objects to be memorized. Then, they were asked to recall all European capitals they knew, under one of four conditions of a 2 ⫻ 2 design. One independent factor was the verbal modality of recall, speaking versus writing. The other independent factor was increased self-attention compared to a neutral condition. Increased self-attention was introduced through a monitor in front of the participants in which they saw themselves while they were performing the knowledge recall tasks. (They were filmed with a camera standing behind a panel.) We will refer to the self-attention values by “with camera” and “without camera.” Next, a listeningspan measure was conducted, followed by the cued recall post-test for the European capitals. Afterwards, participants were asked to recall the previously memorized objects. A self-attention questionnaire was filled out at the end of the experiment. Each participant performed on both recall tasks either with or without increased self-attention. However, the verbal modality changes between the tasks. Participants who orally recalled the European capitals were to recall the memorized objects in the written mode, and vice versa. 11.3.2.2. Results The scores of the self-attention questionnaire did not differ between the conditions. With respect to underachievement in the recall of European capitals, we received the following means (standard errors given in parentheses): writing/without camera 2.5 (0.75); writing/with camera 2.7 (0.54); speaking/without camera 5.2 (0.73); speaking/with camera 4.0 (0.72). An analysis of variance yields a highly significant effect for the verbal modality (F(1, 79) ⫽ 8.4; p ⬍ .01); all other effects and interactions, including the influence of listening span, remain below a significant level. The results of the second


9/27/2006

1:02 PM

Page 177


177

task, recalling the objects previously memorized, show a similar pattern: the verbal modality factor is the only effect that substantially explains for the dependent variable (F(1, 79) ⫽ 3.9; p ⫽ .51). The mean numbers of correctly recalled items are: writing/without camera 26.1 (1.74); writing/with camera 24.4 (1.36); speaking/without camera 23.3 (1.38); speaking/with camera 21.4 (1.33). 11.3.2.3. Discussion The variation of self-attention did not affect the writing superiority effect, which again appears stable and robust. Because the scores of the self-attention questionnaire, which can be taken as a treatment check, did not differ between the conditions, it could be, however, that the measure was not strong enough to really affect the experienced self-attention of the participants. 11.3.3.

Experiment VI

An alternative explanation of the sources of the writing superiority effect points to an influence that resembles the principle of encoding specificity (Tulving & Thompson, 1973). The fact that written recall of knowledge retrieved from memory shows higher content validity than oral recall could go back to the input mode of the learned materials, which might, at least in educational contexts, more often be read than heard. This experiment was designed to test the assumption. 11.3.3.1. Method Teacher students from Heidelberg University of Education (N ⫽ 97; 8 male, 89 female; mean age: 22.8 years) participated in the study and were randomly assigned to one of the four conditions of a 2 ⫻ 2 design. First, they were to memorize a set of 30 nonsense sentences, which were syntactically constructed according to S-P-O or S-P-PP (e.g., “The mouse carries the lawyer,”.“The bicycle crawls into the tree.”). The set of sentences was presented three times in succession. The materials were either read or heard by the participants (⫽ input modality). In the case of reading, they appeared on a screen for six seconds each; in the case of hearing, a previously recorded acoustic file was played. The timing of the sentences’ succession was the same with both input modalities. After the knowledge induction phase, the participants were shown a short entertaining film in order to suppress memory effects in short-term memory. Then, they were asked to recall as many sentences as possible either by writing or by speaking. 11.3.3.2. Results The mean numbers (standard errors) of correctly recalled sentences were as follows: input reading/output writing 12.5 (0.87); input reading/output speaking 8.5 (0.81); input listening/output writing 10.1 (0.95); input listening/output speaking 11.5 (1.20). An analysis of variance yields no main effects of the independent factors, but a highly significant interaction (F(1, 93) ⫽ 7,7; p ⬍ .01). The memorized material is much better recalled in the same modality as it was initially acquired. 11.3.3.3. Discussion In the previous experimental tasks, we often dealt with well-consolidated and integrated information in long-term memory. It could be that information of this kind, particularly in educational contexts, has more often been read than heard throughout the participants’ lives, so that the writing superiority effect would have to be


178

9/27/2006

1:02 PM

Page 178

Joachim Grabowski

re-interpreted as a hidden effect of input modality. However, the effect also appears stable and robust in cases in which knowledge was experimentally induced through graphical objects or in a film, where there was no verbal input modality that could interfere with the mode of recall. In the present experiment, the recall task relates to ‘fresh,’ recently acquired knowledge, the cognitive representation of which seems to remain still bound to the verbal modality in which it was first encountered, as long as there are no further links and connections between the nonsense sentences represented in memory. With time, input repetition, interconnection and consolidation of memory traces, it may be that the fact that certain materials are most often read rather than heard, contributes to the superiority of writing in the recall of knowledge from memory. (A test of this assumption would involve the recall of information that, over years, has certainly been more often heard than read.)

11.4. Concluding Remarks In this paper, a survey of the most central experiments on the writing superiority effect was given. The writing superiority effect appears to be a stable and replicable phenomenon; it reflects different cognitive management strategies in writing and speaking, where the retrieval of knowledge from memory seems to underly different constraints and degrees of freedom depending on the verbal output mode to which the accessed knowledge will be transferred. A closer look into the production times and pausing times, which were recorded in most experiments but not yet analyzed, will possibly provide some further insight into these processes. Differences between spoken- and written-language production appear to apply as early in the process as with the retrieval of information from longterm memory. In order to further clarify the cognitive causes of the writing superiority effect, we have to develop new experimental paradigms that more directly probe into the relevant high-level processes of language production. What do the reported findings contribute to the relation between writing and cognition? Obviously, this relation comes in two directions. When referring to the writing superiority effect, we assume that writing has a very basic impact on cognition, because it activates specific, or additional, aspects of cognitive representations in memory (for example, orthographic versus phonological representations of words for concepts), and it also involves a particular temporal pattern of access to cognitive representations — which may strongly contribute to the writing superiority effect, because the longer activation of an orthographic representation, compared to the execution velocity of speech, may facilitate the association of related concepts and, thus, increase the recall of knowledge within an interconnected domain. At the same time, cognition influences writing, insofar as our cognitive processes are organized such that they, at least in the case of simple verbal recall from memory, ‘feed’ the processes of written-language production better than speaking. It is worth noting that the reported effects relate to recall from long-term memory. They do not, for example, contradict the results from Bourdin and Fayol (1994, 2000, 2002; see also Grabowski, 2005), according to which working-memory performance often decreases due to the higher cognitive load of written compared to oral tasks. The writing superiority effect mainly relates to simply structured, list-like information to be recalled from memory; it has not been applied to cases where available knowledge must be re-organized in


9/27/2006

1:02 PM

Page 179


179

order to plan and write complex texts. After all, the effects of high- and low-level processes that relate either to working-memory functions or to long-term memory functions are not yet sufficiently integrated. Besides the theoretical significance of the obtained results, they bear implications for the appropriate assessment of knowledge in diagnostic situations, particulary in educational contexts and with exams. Written output clearly appears to show better content validity than spoken output, when the task is to recall one’s knowledge from memory as complete as possible. Moreover, the results of Experiment VI indicate that it might be worth directing more attention to a bi-modal consolidation of knowledge in order to achieve some flexibility of its subsequent recall. The diagnostic quality may be impaired when a written test is conducted on contents that have been conveyed by the teacher predominantly in the oral modality.

11.5. Acknowledgments Part of the research described in this paper was supported by grants from the University of Mannheim, and from the Senate of the University of Education at Heidelberg, respectively. I am particularly grateful to Manfred Hofer who most generously offered his institutional facilities, and to Michaela Görlinger who was involved in the implementation of many experiments on the writing superiority effect.



9/27/2006

1:15 PM

Page 181

Chapter 12

The Effect of Writing on Phonological Awareness in Spanish Sofia A. Vernon

The main purposes of the research reported here were: () to examine the relationship between children’s level of writing and phonological awareness and (2) to consider the physical presence of writing as an important variable in phonological awareness studies. We present the results of two different studies concerning phonological awareness in Spanish-speaking kindergarteners. Study 1 comprised a blending task (not previously reported) and two different segmentation tasks, which have previously been reported in Vernon and Ferreiro (1999) and Vernon (2002). The second study deals with a letter-identification task, which has not been discussed in previous papers, and two initial phoneme deletion tasks (Vernon, Calderon, & Castro, 2004; Vernon, 2002). Results for both studies show that in all the tasks, the children’s level of writing explains most variance. The physical presence of writing is also an important variable for segmentation and deletion tasks.

12.1.

Introduction

The question whether writing has an effect upon cognition is a long-standing one. Olson has claimed that writing has historically been “responsible for bringing aspects of spoken languages into consciousness, that is, for turning aspects of language into objects of reflection, analysis and design” (1994, p. 258). Different scripts represent different aspects of language, so what is brought into consciousness depends upon the type of script. Not all aspects of what is said are brought into consciousness, so any writing system can be a “model of some properties of language” (p. 258). Metalinguistic awareness is not unitary, but depends, in part at least, on the proficiency with which each person can use and think about a particular kind of script. The effect of writing on cognition is addressed in different chapters of this book: Klein, Boman and Prince (this volume), for example, finds that elementary school children and Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Vernon, S.A. (2007). The effect of writing on phonological awareness in Spanish. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes, & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in Writing, Vol. 20, pp. 181–199). Amsterdam: Elsevier.


182

9/27/2006

1:15 PM

Page 182

Sofia A. Vernon

university students can both write to learn, and that “metacognitive writing operations make a strong contribution to the level of students’ general writing strategies.” Nottbusch, Weingarten and Sahel (this volume) report a study concerning written language production at the levels of words and sentences. They suggest that segmental information is not completely specified at the beginning of word writing, and hence this is actively processed in the course of writing. In the process of writing sentences, cognitive processes such as sentence planning are also involved. Wengelin (this volume) compares adult dyslexic subjects to normal writers, and finds that spelling difficulties influence the production process (pauses, editing and vocabulary in the finally edited texts). Even if poor spellers do not have a more restricted vocabulary in general, they show lower results concerning lexical diversity and lexical density. Encoding seems to take cognitive capacity from other processes such as vocabulary choice and sentence structuring. This chapter deals with a particular aspect of the relationship between writing and cognition. The main issue that concerns my research programme is whether writing makes a special contribution to phonological awareness. This contribution is examined in two different ways: first, by examining the relationship between the development of children’s knowledge of a particular script (Spanish) and their phonological awareness; second, by considering the physical presence of writing as an important variable in phonological awareness studies. The hypotheses that have guided the research are that: (1) writing development interacts strongly with phonological awareness; (2) phonological awareness develops, and its development is related strongly with writing development, and (3) the physical presence of writing enables a more analytical analysis of words and thus contributes to a better performance in phonological awareness tasks. All these hypotheses lead to the assumption that writing does, in fact, enhance metalinguistic awareness (or at least phonological awareness) as Olson (1994) has pointed out. In order to explore these hypotheses, we adopted Ferreiro and Teberosky’s (1979) writing task in order to distinguish between levels of early writing ability and examined how it was related to various features of phonological awareness. In the studies presented here, several tasks that have been traditionally used to test phonological awareness have been adopted (blending, segmentation and initial phoneme deletion). In one study, we focussed on blending and segmentation. Blending has been amply tested in research because it has been thought to be one of the main skills involved in reading. If children must identify sound units separately through lettersound recognition, they must also be capable of blending these sounds together in order to identify words. This is an easier task than segmentation or deletion (Yopp, 1988). Segmentation of the oral word is particularly important for novice writers, who depend upon their analysis of the sound structure in order to decide how many letters to use for each particular word. It is our belief that as children progress in their writing development their capacity to analyze words into different kinds of segments increases and they become more competent to analyze features that are not readily available in the speech signal. Therefore, children, who are better able to represent language externally, should show more metalinguistic awareness. If the hypothesis that the external representation of language facilitates metalinguistic awareness is correct, then we would expect that letters in the written words would make phonemes and intra-syllabic units more apparent. Children should perform better in


9/27/2006

1:15 PM

Page 183

Writing and Phonological Awareness

183

segmentation tasks when confronted with an external written stimulus, even if they are not yet able to use an alphabetic script. The number of letters in the written word might be an important cue for children to adjust the number of oral segmentations into which they analyze the word. For children with more knowledge about the sound values that letters represent, the written word may also give clues as to the quality of the segment that should be put into correspondence with each letter. The other study we will report here examines two other tasks. The first one, a letter identification task, tries to examine the relationship between children’s level of writing and the knowledge they have of the sound values the letters represent. In relation with this, it is of interest to explore what kind of interaction exists between the ability to recognize letters, the ability to use the correct letters in a script, and phonological awareness (tested through a deletion task that will be explained later). Our hypothesis is that children may not use productively, in their writings, all the letters they may be able to recognize as pertinent in other contexts. Rather, it may be that they progress in their use of this knowledge as they advance in their understanding about the way oral segments are represented in writing. Once children have understood that letters represent some unit of speech (syllables, intra-syllabic units or phonemes), letters may become useful not only in the specification of the number of units in the word, but also as a cue about the nature of the unit itself. The second task examined in this study was an initial phoneme-deletion task, which was administered to the same children as the letter identification task. This type of task has been used extensively in previous research works (e.g. Bruce, 1964; Rosner & Simon, 1971). This is one of the most difficult tasks of phonological awareness, since it requires several cognitive processes for correct performance: hearing the stimulus, perceiving its different sounds, holding the stimulus in memory, segmenting the sounds, locating the indicated position and the sound in such a position, isolating the indicated sound and holding that sound, as well as the rest of the sounds, in memory, and blending and sequencing the sounds (Yopp, 1988). As will be explained later, the physical presence of writing was introduced as a variable in this task to explore whether it enabled more analytical responses. Children’s level of writing was also considered in this task.

12.2.

Phonological Awareness

Research over the last 30 years has shown that one of the problems a pre-literate child faces is to understand that speech can be segmented into phonemes, and that these segmented units can be put into correspondence with written letters. The problem is that speech is not naturally segmented. As Ball and Blachman (1991, p. 56) point out “although we may teach children to “hear” three sounds in cat, the three sounds are not characterized in the acoustic stimulus…Therefore, gaining access to these co-articulated or “encoded” phonemes is more a matter of abstraction than discrimination.” From the very beginning, research has shown that pre-readers have considerable difficulties in segmenting words into phonemes, but that segmentation into syllables is relatively easy for these children. Treiman (1992) has made the claim that awareness of intra-syllabic units (onset and rime) appears before awareness of phonemes. Maclean, Bryant, and Bradley (1987) have shown


184

9/27/2006

1:15 PM

Page 184

Sofia A. Vernon

that children become aware of intra-syllabic units before receiving reading instruction. Longitudinal studies show that there is a strong relationship between children’s ability in rhyme and alliteration tasks and learning to read, and that rhyming ability is a strong predictor of reading success (Ellis & Large, 1987; Goswami & Bryant, 1992). Some authors claim that rhyme may be the first step in phonological development, which ends in the awareness of phonemes (Bradley & Bryant, 1991). Although researchers agree that intra-syllabic units are easily handled by Englishspeaking children at an early age, even before formal reading instruction begins, there is no consensus about whether segmental analysis follows the onsetrime awareness in a natural way, and whether onset and rime sensitivity or phoneme sensitivity are better predictors of reading ability. For example, in a short-term longitudinal study, Hulme et al. (2002) showed that onset-rime makes no contribution as a predictor of reading skill, once phonemic skills are accounted for. The nature of phonological awareness development as an important skill for literacy, therefore, remains unclear.

12.3.

Phonological Awareness Studies Concerning Writing

Most phonological awareness studies present oral stimuli for children to perform operations such as identification, blending, deletion and segmentation of sound units. However, a few studies have explored the relationship between writing and phonological awareness. Goswami and Bryant (1992) advance an interactive analogy model of learning to read. In accordance with this model, Goswami (1994) has defended the idea that phonological and orthographic knowledge influence one another throughout development. She claimed that awareness of rhyme enables children to learn about spelling sound relationships: children seem to be able to predict the pronunciation of an unknown word such as beak when they know a word with a similar spelling sequence, such as peak. Also, results from training in phonemic awareness improve when letters are used as an aid (Ball & Blanchman, 1991; Bradley & Bryant, 1983; McGuinness, McGuinness, & Donohue, 1995). These studies have shown that writing and phonological awareness interact during development. Several studies have pointed out that spelling affects people’s ideas about oral language. For example, Ehri (1984) showed that American fourth graders, who knew how to spell words such as interesting, found one more syllable in the oral word than children, who did not spell the word correctly. Scholes (1995) asked undergraduate students to delete a phoneme out of words. In the first type of stimuli, students had to delete a sound that was directly represented by a single letter in the word (e.g., to take away the “f” letter in raft). In the second type, the phoneme to be deleted was not represented by a letter in the spelling. For example, he asked them to delete the “s” sound in taxed. In the third type, two different responses were possible, depending on whether the students responded on the basis of sound, or whether they responded with a visual mental orthographic representation. For example, subjects had to delete “t” in thought. The possible answers in this case were thaw or though. Correct responses for the first type were given in 77% of the trials. For the second type, correct responses dropped to 35%. For the third type, responses that dealt with the sound were 46%, and 37% were visually based answers.


9/27/2006

1:15 PM

Page 185


185

Treiman, Bowey and Bourassa (2002) found that spelling did not influence oral syllabification in 6- and 7-years olds with words such as habit and rabbit, but that spelling did influence syllabification in older children and adults. This suggests that subjects must have a broad knowledge of spelling to affect oral syllabification. These studies have shown that writing does indeed influence the capabilities to think about sound units in speech. Publications refer to older children and young literate adults. The question is whether the influence of writing over phonological awareness starts even before children become competent readers and writers, and even in the pre-alphabetic stages of writing and reading development. As stated before, my hypothesis in these studies was that even pre-writers would benefit from writing because scripts give clues to the child concerning the types of segmentation that are best suited to represent writing.

12.4.

Some Phonological Differences between English and Spanish

Finally, it is important to consider some linguistic differences between Spanish and English. Phonological awareness is probably language-oriented, and some differences may be found in the development of phonological awareness between languages. Syllables in Spanish are more regular than in English. English has a broader dispersion of syllable types (Bradley, Sánchez-Casas, & García-Albea, 1993) and has less defined syllabic boundaries (Sampson, 1985). This is probably because English is a stressed-time language, and any English syllable can be weak or strong. Syllables get contracted or expanded in order to fit in the required rhythmic pattern (Halliday, 1985). Also, in English, the categories describing consonants are clearer than those describing vowels. As Ladefoged (1982) pointed out, “Part of the problem in describing vowels [in English] is that there are no distinct boundaries between one type of vowel and another” (p. 72). This author also states that part of the problem in describing vowels is that there are more vowel sounds than letters in the alphabet. Finally, English has a “deep” or irregular orthography in which lettersound correspondences are inconsistent. Spanish, on the other hand, has a more regular syllabic structure and clearer syllabic boundaries, although this language “shares with English the property of lexical stress. It does not, however, share its widespread stress-conditioned reductions, i.e. the tendency in English for centralization of unstressed vowels” (Bradley et al., 1993, p. 203). Spanish rhythm is syllable-timed. This means that all syllables in Spanish have the same duration. Unlike English, which has a large number of monosyllabic content words, Spanish monosyllabic content words are rare. Alvarez, Carreiras, and de Vega (1992) studied syllabic frequency in written texts from a variety of sources. These authors found that most Spanish content words have three syllables, followed by bi-syllabic and tetra-syllabic words. Most monosyllables are function words. Even if Spanish syllables may have an onset, a vocalic nucleus and a coda, only the nucleus is obligatory. Although the data from the above-mentioned study were taken from a written corpus, it is still significant because Spanish orthography is quite regular. Much of the regularity of Spanish orthography is due to the fact that there are five vowel letters for five vowel sounds. Although lettersound correspondences for consonants are more or less regular also, there are variations, especially if dialectal differences are taken into account.


186

9/27/2006

1:15 PM

Page 186

Sofia A. Vernon

Although some authors have claimed (Carrillo, 1994; Jiménez, 1992) that the development of phonological awareness in Spanish follows the same patterns as in English, other authors give no evidence of children having onset-rime sensitivity. For example, Tolchinsky and Teberosky (1997) reported a study comparing Spanish- and Hebrew-speaking kindergartners and first and second graders. They asked children to segment words orally (without a model). In both languages, the syllable was the preferred unit for kindergartners and first-graders, and 35% of Spanish second-graders still used it. Very few children segmented words phonologically. Furthermore, the examples these authors provide usually show responses where the coda was segmented from the rest of the syllable, or where one of the syllables (in bi-syllabic words) had been phonologically divided, but the other syllable remained unsegmented.

12.5.

Study 1

This first study was designed to examine the relationship between variations of writing ability and children’s capabilities to segment words or to blend different types of segmentations into words. Another objective was to test whether the physical presence of writing in the segmentation task would enhance more analytical responses than in the purely oral task. The participants were 54 working-class kindergartners from a public school in Queretaro, Mexico. All of the children were monolingual, Spanish speakers, with an average age of 5 years, 8 months (SD 3.43 months). Children did not receive any explicit reading or writing instruction, and the teacher rarely read aloud to them. Data were collected through individual interviews at the school premises. Each child was interviewed on two succeeding days. Each interview lasted for approximately 30 min. Children were administered an initial writing task, and then asked to perform three phonological awareness tasks: blending, oral segmentation (guessing game), and oral segmentation with written stimuli. 12.5.1.

Writing Task

In order to test whether children’s level of writing contributes to their segmentation capacities, an initial writing task was included in the studies presented here. Its main purpose was to classify children in different writing levels. The task was originally used by Ferreiro and Teberosky (1979,1982), and later by other authors (Pontecorvo & Zucchermaglio, 1988; Tolchinsky & Teberosky, 1997). Each child was individually asked to write, one at a time, several words he/she had not been taught to write, “The best way you can.” In all the tasks, words were common nouns with a regular syllabic structure (CV-CV-CV-CV, CV-CV-CV, CV-CV or CVC). In this particular study, the words children wrote were mariposa (butterfly), gusano (worm), venado (deer), perico (parrot), sapo (toad), toro (bull), and pan (bread). When the child finished writing each word, the researcher asked the child to “read it to me slowly, pointing with your finger”. All verbalizations during the writing process, as well as during the finger-reading process were carefully registered, as well as the way the child matched his/her verbalizations to the letters or groups of letters when writing or reading.


9/27/2006

1:15 PM

Page 187


187

Children were classified into the following writing levels: 1. Pre-syllabic (P): Children do not attempt to make one-to-one correspondences between letters and units of speech smaller than the word. That is, each string of letters is interpreted (or “read”) as saying a whole name (or “word”1). In addition, children use any set of letters to write any given word: There is no systematic attempt to use pertinent letters. 2. Initial Syllabic (S1): Children attempt making lettersound correspondences in a very inconsistent way. The letters they use are not pertinent in the word context. Children do not anticipate the number of letters they need, so they write and, once they have finished writing, they try to adjust their reading to the written string. The result is that they can make some correspondences between one letter and one unit of sound (generally a syllable), but usually interpret one sound unit for several letters. 3. Syllabic, without the use of pertinent letters (S2): At this level, children write one letter for each spoken syllable in a systematic way. However, the letters they use are not usually pertinent. That is, they may choose any given letter to represent any syllable. Children can write, for example, “mes” for “perico.” 4. Syllabic with the use of pertinent letters (S3): Children write one letter per syllable. The letter they use for each syllable is one of the letters that are conventionally used for that syllable. Due to the salience of vowels as syllable nuclei, children usually represent each syllable with the corresponding vowel letter. Consonants are not excluded, especially when the letter’s name corresponds to the syllable. For example, perico (parrot) could be written as pio or as eio. Children systematically control the number of letters they use either counting the oral syllables before they write, or saying each syllable before writing the corresponding letter. After they write, children read their production in a syllabic way (that is, one letter per oral syllable). 5. Syllabic-Alphabetic (SA): Children usually alternate between the use of letters to represent a phoneme or a syllable. A typical example would be writing the word mariposa (butterfly) as maiosa or perico (parrot) as pico. 6. Alphabetic (A): Children have understood that the alphabetic script represents phoneme-letter correspondences. Although their spelling may not be conventional, these children write one letter per phoneme. The classification of children’s written productions was checked by two graduate students. The only two disagreements in the whole population we have considered here were resolved by the main researcher. Children were selected at random, and sorted according to their writing level. 12.5.2.

Phonological Awareness Tasks

Three different tasks were administered: blending, oral segmentation (guessing game), and oral segmentation with written stimuli.

1

Although the children may not know what a word is, in the same way as adults do, we use this term to refer to the fact that children match a string of letters with an oral, meaningful noun. Children at this moment are unable to analyze the sound units in words and have no way of knowing how many letters a given written noun should have.


188

9/27/2006

1:15 PM

Page 188

Sofia A. Vernon

In the blending task, words were said by the experimenter “in little bits” (i.e. segmented), so that the child could “guess”, or reconstruct, the whole word. As will be explained later, the “little bits” varied. Our hypothesis was that children would have the least difficulty guessing words that had been segmented into syllables or those which kept the initial CV intact, but that only children with an advanced level of writing would be able to blend sequences that implied phonological segmentations or initial onset-rime segmentations. In other words, we believed that to become aware of these more conceptual units of sound, the child should have become aware of these units through the difficulties they experience when writing words. That is, it is probably not because they can “hear” these units that they later use them in writing, but on the contrary, their progress in writing makes them more aware of the existence of these units in speech. In all tasks, we expected their performance ability to correlate significantly with their writing level. Additionally, two segmentation tasks were administered. Segmentation is important because, in the process of writing a word, a child segments the word orally to decide how many letters they will need to represent the word. Written words have discrete elements (the letters), and generally each one of them represents a sound (at least in more transparent orthographies, such as Spanish). The oral word, however, is continuous, and can be segmented in many different ways. One of the problems the child faces, when learning to write, is to decide which sound unit should correspond with each letter. Our hypothesis was that children would be able to make a more analytical analysis of the word when confronted with a written stimulus, even when the child was unable to read independently, and that children at a more advanced writing level would evidence more awareness of phonemes and intra-syllabic units. 12.5.3.

Procedure for Phonological Awareness Tasks

The test words for all the tasks were monomorphemic monosyllabic words with a CVC structure (e.g. pan bread), and two-syllable words with a CV-CV structure (e.g. me-sa table). 12.5.3.1. Blending Stimuli for the blending task were designed incorporating both the types of segmentation used for blending tasks (phonological segmentation and onset-rime segmentation) and two other types of segmentation that have been observed in studies dealing with writing development in Spanish (Ferreiro & Teberosky, 1979/1982; Vernon, 1991). Four different types of segmentation were used: • Type 1 (T1): This is basically a syllabic segmentation. However, after the initial CV for monosyllabic words, the vocalic nucleus is repeated in isolation, and is followed by the vowel and final consonant (e.g. so-o-ol for sol, sun). For bi-syllabic words, both syllables are kept intact. After the first syllable, the onset of the second syllable is isolated, and then followed by the second syllable (e.g. me-s-sa for mesa, table). • Type 2 (T2): The initial CV sequence remains unsegmented, followed by the coda for monosyllabic words and a phonological segmentation of the second syllable for bi-syllabic words (e.g., so-l for sol, sun, or me-s-a for mesa, table). Types 1 and 2 have been observed in other studies, and appear mostly when pre-alphabetic children read their own productions, and try to adjust the number of spoken segments with the number of letters they have used.


9/27/2006

1:15 PM

Page 189


189

• Type 3 (T3): For monosyllabic words, this is equivalent to onset-rime segmentation (s-ol for sol, sun). For bi-syllabic words, the first syllable is segmented the same way, and the second syllable remains unsegmented (e.g. m-e-sa for mesa, table). • Type 4 (T4): Complete phonological segmentation (e.g., s-o-l for sol and m-e-s-a for mesa). To randomize the difficulty associated with both each particular word and the order in which they were presented, 8 different orders of presentation were established. Thus, every child was given the same 18 words and all the types of segmentations, but each particular word could be segmented in any of the four different ways we have mentioned (T1, T2, T3, or T4). The procedure was as follows: The experimenter first looked at the pictures with the child and asked him to name each one. Then she explained to the child that she would say the name of a picture “in little bits,” and the child had to guess which one it was, saying the complete name. The experimenter gave four examples with feedback, and then proceeded to the experimental items. The experimenter said each item one at a time. Segmented parts were made evident by stopping for approximately 2 safter each part. If the child guessed correctly, they would keep the card. If not, the experimenter would give the stimulus once again. If the answer was incorrect, the experimenter would keep the card. All the children were eager to guess. Each child was given four practice items (one for each type of segmentation of a bi-syllabic word) and then the 18 experimental items. 12.5.3.2. Segmentation After the reconstruction task, half the children proceeded with an oral segmentation task (guessing game), while the other half proceeded with an oral segmentation task with written stimuli. After that, they proceeded with the other task. In the oral segmentation task (guessing game), children were asked to produce oral segmentations, so that the researcher could guess the word. Cards were the same as in the blending task. Instructions were as follows: “Did you see how I did it? Now it’s your turn to make me guess… Remember that some of the words I said were very easy for you to guess, like lu-n-na. But others, like p-an or l-u-z, were very hard. Try to make it hard for me, so that you win the game.” Children were then given a set of drawings (naming each one of them). After that, the researcher covered her eyes, and asked a child to pick a picture up and to say it “in little bits,” the way she (the researcher) had done before. During the first three attempts, if the child produced anything but a phonological segmentation, the researcher immediately “guessed” the word and prompted the child “to make it harder”, giving examples of more analytical segmentations, using the same word the child had said previously. For example, if the child said “lu-na,” the interviewer would say “Luna. That was very easy. A harder way to say it would be lu-n-a or l-u-na. The hardest for me to guess would be l-u-n-a.” Children were only given this kind of feedback for the first three chosen items. Stimuli for this task were CV-CV and CVC common nouns: gis (chalk), pez (fish), sol (sun), pan (bread), luna (moon), foca (seal), taco (taco), mesa (table), sapo (toad), and gato (cat) In the oral segmentation with written stimuli task, children were presented with written common nouns. The experimenter verified whether the child was able to read each of the words, one at a time. As none of the children were able to read the words, the experimenter read aloud each card with normal stress and intonation. After reading each card, the experimenter then asked the child to point to each letter while saying the word “in little bits, one for each letter.” Letter naming was discouraged. The experimenter gave three examples, showing a strict correspondence between letters and sounds, then asked the child to repeat


190

9/27/2006

1:15 PM

Page 190

Sofia A. Vernon

what she had done, in exactly the same way. Feedback was provided in the three cases. After that, the experimenter proceeded to the test items. Only the verbal responses (and not the ability to point out each letter) were taken into account. Words for this task were gis (chalk), luz (light), pan (bread), luna (moon), taco (taco), sapo (toad) and gato (cat), peso (the name of the Mexican currency). 12.5.4.

Results for the Phonological Awareness Tasks

Mean percentage of responses (1= correct; 2 = incorrect)

12.5.4.1. Blending We performed a multiple regression analysis (using a stepwise method) with percentage of correct answers as the dependent variable and level of writing, type of stimulus, age, type of word (monosyllabic vs. bi-syllabic), and order of presentation as independent variables R² for the first step (stimulus) was .254 and .309 when level of writing was entered. In other words, the kind of stimulus (T1, T2, T3 or T4) explained most variance in the children’s score for correct answers (R .505, R2 .254, F 331.188, p .001). Level of writing was also significant (entered on step number 2, R .557, R2 .309, F 217.817, p .001). When all the variables (level of writing, type of stimulus, age, type of word and order of presentation) were entered in the linear regression with a stepwise method, SPSS excluded type of word, age, and order of presentation from the analysis because, in all cases, p .10). Graph 1 shows the means for the percentage of correct answers for all types of stimuli. The incorrect answers include repetitions of stimuli, refusals to give an answer or saying a word that did not correspond to the stimulus. When a child produced more than one answer, only the closest to the target answer was taken into account. All the children in the study found it easy to reconstruct words when the first syllable (for CV-CV, bi-syllabic words) or the onset plus the syllabic nucleus (for CVC monosyllabic words) remained unsegmented (types of stimuli T1 and T2). 2.0

1.8

1.6

Type of Stimuli 1.4 T1 1.2 T2 1.0

T3 T4

0.8 P

S1

S2

S3

SA

A

Writing Level

Graph 1: Mean percentage of correct answers. Blending


9/27/2006

1:15 PM

Page 191


191

However, they all found it hard to blend words when the initial consonant and vowel were segmented (T3 and T4). Even children, who produced alphabetic writings, had difficulties blending words with these types of segmentations. Correct answers for the children with alphabetic writings were 82.4% (monosyllabic words) and 82.6% (bi-syllabic words) for T3, and 72.2% (monosyllabic words) and 63.6% (bi-syllabic words) for T4. Children at the syllabic-alphabetic level performed in a similar way. Children at the pre-syllabic and syllabic levels of writing got 50% (or slightly fewer) correct answers for T3, although percentages of correct answers dropped for monosyllabic words with this same type of segmentation (T3). For T4, children at the pre-syllabic and syllabic levels (P, S1 and S2) had fewer than 15% correct responses for bi-syllabic words, whereas S3 children achieved 31.8%. Since the main difference between S3 children and S2 children was the use of pertinent letters in their writings, the knowledge of letter in terms of sound values may have been responsible for better performance in the reconstruction of phonologically segmented words. The qualitative analysis of incorrect answers showed that when children faced T3 stimuli (s-ol, or s-o-pa), they were usually unable to take into account the information provided by the onset. Incorrect answers usually included the correct rime (for T3), preceded (or followed, in fewer cases) by a syllable so that a new word was produced. For example, for “t-os,” children frequently said “oso”, and for “p-an” they answered “animal” or “Juan.” Incorrect answers for bi-syllabic words for this type of stimuli (T3) were very similar: children could usually reconstruct the first syllable’s rime and the second syllable (“eso” for “b-e-so”), and sometimes changed the first consonant in order to form a different word (e.g., for “f-o-ca,” they answered “boca,” for “t-a-co” they said “barco”). Incorrect answers for fully phonologically segmented items (T4) were very similar. All children found it easier to blend words when the first part of the word was maintained intact (the onset plus the vocalic nucleus, for monosyllables, as in “pa-n”, or the first syllable, for bi-syllabic words, as in “ta-c-o”). Summarizing, the types of segmentation we presented seem to have different levels of complexity. For all children, the main difficulty appeared to be related to the segmentation of the first part of the word. These results suggest that our hypothesis concerning writing level and phonological awareness is correct: as children advance towards a more advanced level of writing, they become more aware of the sound structure of words and they are able to blend sounds in a more efficient way. Also, results suggest that blending develops in a hierarchical way, from the capacity to blend syllabic units, to the capacity to blend phonemes. The positions in which these units appear in a word seem crucial in this process. 12.5.4.2. Segmentation tasks The purposes of segmentation tasks were to test whether writing development interacts with phonological awareness, and whether the physical presence of writing enables a more analytical analysis of words and thus contributes to a better performance in phonological awareness tasks. In order to test the importance of the physical presence of writing, two segmentation tasks were administered: a purely oral task and a task where children were presented with visual stimuli. Responses for both tasks were classified according to the types of segmentation given by the children.2

2

These tasks have been previously reported in Ferreiro and Vernon (1999), Vernon (1997), and Vernon (2002).


192

9/27/2006

1:15 PM

Page 192

Sofia A. Vernon

1. No segmentation: In spite of examples and initial feedback, the children repeated the complete word. 2. Syllabic segmentations: For bi-syllabic words, strict syllabic segmentations were considered here. For monosyllabic words, responses such as so-ol or gi-is were also considered as syllabic. 3. Partial isolation of a segment: These were syllable-based responses. However, a segment (usually a vowel), which also appeared before or after the isolation in a CV sequence, was partially segmented. For example, s-.o-ol for sol, or lu-u-na for the word luna. 4. Phonological segmentation of the last syllable or the coda (in monosyllabic words): Segmentations such as lu-n-a or so-l have been considered here. 5. Phonological segmentation of the first syllable or the onset in monosyllabic words: For example, l-u-na or s-ol. 6. Complete phonological segmentations: l-u-n-a or s-o-l. A multiple regression analysis (using a stepwise method) was used with type of segmentation (scored from 1 to 6) as the dependent variable and task (guessing game, written stimuli), age, level of writing, type of word (bi-syllabic, monosyllabic) and order of task presentation as independent variables. The analysis showed that most of the variance was explained by the level of writing. R2 for the first step (level of writing) was .571 (F 702.202, p .001). The order of tasks, in this case, was also significant (entered on step number 2, R2 .590, F = 407.895, p .001), as was the task (written stimuli vs. guessing game, entered on step number 3, R2 .602, F 277.833, p .001). All other variables (age, and type of word) were excluded from the analysis, because in both cases, p .10. More advanced responses were given as children advanced in their level of writing. Also, in all cases, more advanced responses were given when a written stimulus was shown. More analytical responses were given in the guessing game task when the writtenword task occurred first. These results show that phonological responses to both segmentation tasks could only be accomplished by children, who had the most advanced level of writing (A, SA and S3); the nature of the stimuli (written or purely oral) seemed to have a strong effect also. None of the children at levels P, S1 or S2 segmented any word phonologically. Children at the syllabic level with the use of pertinent letters (S3) started doing so, when the written word was present (3.7% of responses). Children at the syllabic-alphabetic (SA)) and alphabetic (A) levels of writing were better able to segment phonologically. The differences between the two tasks were notable: In the oral guessing game, children at SA level gave 34.5% of these responses, whereas with written stimuli they gave 55.6%. At A level, 64.4% of responses in the oral guessing game were phonological segmentations, compared to 88.9% in the written stimuli task. Table 1 shows the percentages for responses type 1 to 6. Type 5 responses (onset-rime for monosyllabic words, or the segmentation of the first syllable for bi-syllabic words) were not frequent, and were not given in a systematic way by any of the participants. Syllable-based responses appeared in all levels of writing, especially in the least advanced children. As the level of writing advanced, type 4 gained importance. Even the most advanced children gave this type of response more frequently than type 5.


9/27/2006

1:15 PM

Page 193


193

Table 1: Percentages of each type of response for each segmentation task (O = Oral stimuli, W Written stimuli). Type of segmentation response by task (oral vs. written stimuli)

1

Level of writing

O W O W O W O W O W O W

2 3 4 5 6

P

S1

S2

S3

SA

A

30.7 18.5 62.8 60.5 1.1 19.8 0 1.2 0 0 0 0

9.2 9.9 73.6 58 14.9 27.2 2.3 3.7 0 1.2 0 0

7.1 0 45.7 25.9 45.9 51.9 3.5 18.5 0 3.7 0 0

3.5 1.2 58.1 22.2 19.8 34.6 11.6 25.9 7 12.3 0 3.7

0 0 13.8 0 12.6 4.9 29.9 32.1 6.9 7.4 34.5 55.6

0 0 3.4 0 8 2.5 18.4 3.7 5.7 4.9 64.4 88.9

In conclusion, both hypotheses seem to be supported by our results. More analytical responses (type 6) appeared only in children with advanced levels of writing, and their frequency increased when a written stimulus was present. This was true even for children at the S3 level, who made no use of phonemes when making letter-sound correspondences in the oral task.

12.6.

Study 2

In this study, two tasks were included. One was a phonological task, which, in this case, dealt with initial phoneme deletion. As indicated in the introduction, this task has a greater degree of difficulty for children. In order to test the hypothesis concerning the physical presence of writing, this task included two modalities: a purely oral task, as has been previously used by other authors, and another modality, where written stimuli were presented. The oral deletion task was included to test the same hypotheses as in study 1. The other task concerned letter-identification. The purpose of this task was to compare children’s skills at identifying letters in a task that required an ability to judge the appropriateness of each proposed letter in a word. The results were compared to the ability to use letters in the initial writing task. In the letter identification task, the child received information about the position of the missing letter, the number and type of the other letters present in the written word, and was given options from which they could choose the missing letter. However, in the writing task, all they received was information about the word they


194

9/27/2006

1:15 PM

Page 194

Sofia A. Vernon

should write. The choice of how many letters or which letters to use depended only upon their knowledge of the writing system (that is, the nature of letter-sound correspondences and the sound value of each letter). The aim was to test whether children could use the information provided by letters in context (i.e. the physical presence of writing). That is, could children make the same choices when they were relying solely upon their internal knowledge of letter-sound correspondences as when they were using the written context? 12.6.1.

Participants and Procedure

The participants in this study were 100 working-class kindergartners from a public school in Querétaro, México. All the children were monolingual Spanish speakers with an average age of 5 years, 7 months (SD 6.3 months). Children did not receive any explicit reading or writing instruction, and the teacher rarely read aloud to them. Data were collected through individual interviews conducted at the school premises. Each child was interviewed on two succeeding days, andeach interview lasted for approximately 30 minutes. As in study 1, children were classified according to their writing level, which was determined through an initial writing task. The procedure for this task was the same as for study 1. In this case, 5 different writing levels were considered: pre-syllabic (PS), syllabic without the use of pertinent letters (S1, which in this case included children that in study 1 were classified as S1 and S2), syllabic with the use of pertinent letters (now S2), syllabic-alphabetic (SA) and alphabetic (A). Children were selected at random, until groups of 20 children each were completed for each level. Children wrote 6 common nouns: mariposa (butterfly), camote (sweet potato), pepino (cucumber), sopa (soup), tuna (a common cactus fruit) and pan (bread). All the children started with the letter identification task. In the second part of the interview, the children were given a task involving deletion of the initial phoneme. Half the children started with the oral stimuli for this task, and then proceeded with the written stimuli. For the others, the order was reversed. 12.6.2.

Letter Identification

This task consisted in asking the children to identify the missing letter from 9 different common nouns. The missing letter was always in an initial position. The missing letters were the 5 vowels found in Spanish scripts (A, E, I, O, U), and 4 common consonants (M, C, S, P). Children were given a card at a time, with a printed word. On each card, the initial letter had been removed, and there was a visible mark where a little card with the pertinent letter should be placed so the word would be complete. Table 2 shows words used as stimuli and letters offered as options to solve the task. The interviewer told the children what she had written on the card, pointing out that the first letter had been removed, and that the word was incomplete. She then showed the little cards, and told the children that one of the letters was the missing letter in the word, and asked them to find the correct letter, and place it where it belonged. After that, she asked the children what the complete word said, to make sure the children remembered


9/27/2006

1:15 PM

Page 195


195

Table 2: Letter identification task. Words used as stimuli and letters offered as options to solve the task. Word ATOLE PISO CARA IMAGEN SAPO MANO ELOTE OREJA UÑA

Letter options B P G B S M G i P

A U E E F E O T O

N L C i E C M U U

i E U S D i E S T

E B M U i N i O i

the item correctly. Whenever a vowel was missing, the 5 cards that were given as possible options included 3 vowels and 2 consonant letters. If a consonant was missing, 3 consonants and 2 vowels were presented. It should be noted that the options children had to choose from always included a letter that represented a sound similar to the target one (P and B, for example). In the case of vowels, 2 other vowel letters were given, plus 2 consonants. In the case of consonants, options included the correct letter, another similar consonant, a totally different consonant, plus 2 vowels. 12.6.2.1. Deletion of the initial phoneme Oral stimuli: Children were provided with an oral word, and were asked to delete the first sound. Two examples were given, and then two trial items with feedback. The experimenter explained that she would say a word, and the children had to say the word “without the first sound. For example, if I say NUBE, you say UBE. NUBE (pointing to herself)…UBE (pointing to the children).” Experimental items were 15 Spanish words. Seven had a CV-CV syllabic structure (foto, lago, cana, sopa, pila, meta, luna), three were CVC monosyllables (pan, sal, gol), and 5 had a V-CV structure (eco, osa, humo, ajo, hipo). In Spanish, H is a “silent” letter and does not, therefore, modify the syllabic structure. Written stimuli: Children were provided with written words on individual cards (upper case, Times New Roman, size 26). After checking whether the children were able to read the word by themselves, the experimenter would read the word aloud to them. The interviewer then covered the first letter and asked the children to say the word without the first sound. Two examples were given, and then two trial items with feedback. The researcher would show the children one of the words, and then say, “I am going to read this card. It says PALO (reading with the normal stress and intonation). Now I’m going to cover the first letter. Now it says ALO, because there is one letter less. PALO (uncovering item), ALO (covering the first letter).” The 15 experimental items were very similar to the oral stimuli task. Again, 7 had a CV-CV syllabic structure (foca, lata, cama, sope, pino, mesa, lupa),


196

9/27/2006

1:15 PM

Page 196

Sofia A. Vernon

3 were CVC monosyllables (paz, sol, gas), and 5 had a V-CV structure (eso, ola, uno, aro, iva). All written words had regular letter-sound correspondences.

12.7. 12.7.1.

Results Letter Identification and Use

In order to calculate scores for letter use, the writings children had produced in the initial writing task were analyzed. Scores both for letter identification and letter use were calculated dividing the number of correct responses by the total number of items (that is, the ratio). In the letter identification task, this measure was straightforward, since children chose one of the five letters to complete each item, and the choice was either correct or incorrect. In the writing task, some letters appeared in words used as items more than once. Any letter was considered correct if it appeared at least once in the child’s overall production, and if the letter was used in a syllable, which contained that letter. That is, single letters that were used to represent syllables or intra-syllabic units were considered correct, as long as they were pertinent for that syllable. For example, if an M was used for the syllable MA, it scored as correct. If both letters were used, they were both counted as correct. For children at the pre-syllabic level, who did not make any kind of correspondences between letters and sound units in language, only the initial and final letters could be considered, since these children did not make any overt attempts to make any soundletter correspondences. Feedback or examples were not given in any of the two tasks. Most children, even those at the pre-syllabic level of writing, were able to identify some of the letters we presented in the task. As Table 3 shows, pre-alphabetic children could identify vowels more easily then consonants, which is consistent with the fact that in Spanish vowels are more salient than in languages such as English. Most children were, however, able to choose the correct consonant at least once during the task. The ability to correctly identify the initial letter improved as children advanced through writing levels. A multiple regression analysis was carried out with letter identification as the dependent variable and age and level of writing as independent variables. The analysis showed that most of the variance was explained by the level of writing. R² was 0.730 for the first step (level of writing) (F 265.429, p .001). Age was not significant. Errors in the letter identification task were not haphazard. Even the children with the least advanced writings (pre-syllabic and syllabic without the use of pertinent letters) were able to distinguish vowels from consonants. More advanced children usually selected the correct letter. Usually, errors were confined to the choice of a letter that represented a sound with clear phonological similarities to the target item. Although children’s performance in the letter identification task was related to their level of writing, it did not always accurately reflect their capability to choose a letter with a pertinent sound-value in the writing task. For instance, although some of the children at the pre-syllabic and syllabic levels of writing (the least advanced) could correctly identify the vowels, they were unable to use them in their writings in pertinent contexts (Table 3). Again, the physical presence of writing seems to “open” metalinguistic awareness.


9/27/2006

1:15 PM

Page 197


197

Table 3: Percentages of correct responses for the letter identification task. Level of writing

P S1 S2 SA A

Letter identification

Letter use in writing

Vowels

Cons.

Total

Vowels

Cons.

Total

30 65 96 98 100

7.5 37.5 62.5 91.2 100

20 52.2 81.1 95 100

11 32 82 97 100

1.4 6.4 33.6 57.9 98.6

5.4 17 54.2 74.2 99.2

Performance in letter identification and letter use were significantly correlated. A Paired Samples T test with both letter identification and letter use indicated a correlation of .850, with a 95% confidence interval of the difference, t 10.022, df 99, and p (2 tailed) .001. 12.7.2.

Deletion Tasks

Responses to both deletion tasks were classified in the following way, starting with the ones we have considered more advanced. The examples that follow show the word provided as stimulus by the researcher, followed by the children’s response: (1) Correct deletion (for example Sope-ope).3 (2) Extended deletion. Children managed to delete part of the word. However, they deleted more than a phoneme (usually a whole syllable) either at the beginning or at the end of the word (for example, mesa-me, they deleted one or two consonantal units, but left the vocalic nucleus (or nuclei); for example, mesa-mea, pan-a). Finally, others deleted the final phoneme of the word (the coda in monosyllables or the final vowel). (3) Phoneme substitution. Children produced a different, but phonologically similar word with the same syllabic structure and rhythmic pattern. (e.g. Pan-paz). (4) Phoneme addition. Subjects produced another word, adding a sound unit to restore the most frequent syllabic structure and rhythmic pattern, CV-CV (e.g. Ola-sola). This occurred mostly in V-CV words. The change in the syllabic structure of these items may suggest to the children that the item is similar to the answers that have been previously expected, so that they produced a similar syllabic pattern to the other, previous items. (5) Word repetition. Children repeated the item or gave a word not related phonologically or semantically to the stimulus ( e.g. Foca (seal), big tree). When children hesitated and gave more than one answer for a word, the most “advanced” solution (in our way of interpreting their responses) was taken into account. Since none of the children gave the same type of answer throughout each task, we have considered each response in the analysis. 3

In the example, the word given by the experimenter comes first, followed in italics by the children’s response.


198

9/27/2006

1:15 PM

Page 198

Sofia A. Vernon

Results show clearly that the level of writing and the type of stimuli (oral or written) were important variables. An analysis of multiple regression with a stepwise method taking the level of response as the dependent variable and the level of writing, the task (oral or written word), order of tasks, type of word (CV-CV, CVC or V-CV) and age as independent variables, showed that the level of writing explained most variance, with R2 .353, p .01. The type of task (oral vs. written stimuli) was also significant (R2 .392, p .01). The type of word (R2 .430, p .01) was also marginally significant. Age and the order, in which the tasks were administered, were not significant. Table 4 shows percentages of correct responses for each writing level and task modality. Correct responses increased as writing level advanced. Also, incorrect responses change in the different writing levels. For example, only 10.1% of the overall responses were type 5 (repeating the whole word or saying an unrelated word or expression). However, when we consider the responses given by the pre-syllabic children, this kind of response accounted for 41.1%. In contrast, children who have begun making at least syllabic correspondences (S1), even if they did not use pertinent letters in their writings, only gave 7% of type 5 responses. Once children start using pertinent letters, percentages diminished. Only 1.3% of S2, and 0.8% of SA and A children gave this kind of response. The presence of writing seemed to enable advanced solutions for all the levels of writing, except the pre-syllabic level. At this level, children did not make any attempts to make letters correspond to any kind of sound-unit. Knowledge of conventional sound values seems to play an important part in the increase of correct answers. Children that write with pertinent letters (A, SA and S2) showed a greater difference in the responses for this task in the written stimuli modality. It seems obvious that looking at the letter that represents the sound to be deleted enhanced their performance. As in study 1, the results in these two tasks seem to confirm our hypotheses. Writing development seems to be crucial for understanding the phonological operations of children in the three phonological awareness tasks that were studied. Kindergartners benefit greatly from the physical presence of written words. Even children at low levels of development seem to be able to understand the tasks and perform in a different way when writing is involved.

Table 4: Percentage of correct responses for each writing level in deletion tasks. Task/stimuli

Written Oral

Level of writing A

SA

S2

S1

PS

99 73.7

84.7 34.4

55.7 19

25.8 18.4

1.7 1.4


9/27/2006

1:15 PM

Page 199


12.8.

199

Discussion

Results from both the studies show that writing is a variable that should not be disregarded in phonological awareness studies. Writing can be understood as both the understanding the children have acquired of the writing system (level of writing) and as the presence of written stimuli in the tasks. In all the tasks we have presented, writing level explains most variance. As children progress through writing development, they appear to face important problems. One of them is to find the appropriate sound unit that corresponds to each letter. Syllables are naturally identified by children at a very early age, before reading or writing instruction has begun (Liberman, Shankweiler, Fischer, & Carter, 1974). This is because the syllable is the most salient unit of the flow of speech (Daniels, 1992). However, it takes children some time to realize that letter-syllable correspondences can be used in writing (Vernon, 1991). In the blending and segmentation tasks, for example, presyllabic children could sometimes segment syllables, even if they did not use them when they wrote independently. For other types of segmentation, which are not “natural”, data suggest that when children started using a given sound unit in writing (even if they cannot do so systematically), they could also start using them in blending and segmentation. For example, children at the SA level started using a mixture of syllabic and phonological representations in writing, and in the oral tasks they could sometimes divide words into phonemes, even if other, less analytical responses appeared frequently. Alphabetic children, who represented phonemes in their writings in a systematic way, were more able to do phonological awareness tasks involving phonemes.



9/27/2006

3:06 PM

Page 201

Chapter 13

Developmental Trends in a Writing To Learn Task Perry D. Klein, Jennifer S. Boman and Melanie P. Prince

Students’ writing strategies show considerable development during their school years; how does this affect their ability to use writing as a tool for learning? Students in Grades 4, 6, 8, and university, completed one of two physics tasks, and then wrote explanations as they thought aloud. Some students simply recorded their previous explanations, while others constructed more complex explanations during writing. A path analysis showed that educational level contributed to pre-writing science explanations and metacognitive writing operations; general writing strategy and pre-writing science explanations contributed to post-writing science explanations. This suggests that young students can use writing for learning, but older students can do so more effectively.

13.1.

Introduction

Traditionally, cognitive research has focused on the causes of good writing: topic knowledge, strategies, working memory span, oral language ability, and so forth (e.g., Abbott & Berninger, 1993; Bereiter & Scardamalia, 1987; Butterfield, 1996; McCutchen, 2000). However, attention has recently turned to the cognitive effects of writing (e.g., Galbraith, 1999; Grabowski, this volume; Klein, 1999; Tynjälä, Mason, & Lonka, 2001; Piolat, this volume; Vernon, this volume). Considerable evidence now indicates that writing can contribute to students’ learning in content areas such as science and history (e.g., Bangert–Drowns, Hurley, & Wilkinson, 2004; Rivard, 1994; Wiley & Voss, 1999). Consequently, writing to learn activities have been implemented at every educational level, from elementary school through university (e.g., Audet, Hickman, & Dobrynina, 1996; Beins, 1993; Rosaen, 1990). This raises an important issue: during this developmental span, students’ writing changes profoundly (e.g., Beal, 1996; Bereiter & Scardamalia,


Klein, P.D., Boman, J.S., & Prince, M.P. (2007). Developmental trends in a writing to learn task. In G. Rijlaarsdam (Series Ed.), and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in Writing, Vol. 20, pp. 201–217). Amsterdam: Elsevier.


202

9/27/2006

3:06 PM

Page 202

Perry D. Klein et al.

1987; Berninger, Fuller, & Whitaker, 1996). Does this change affect students’ ability to use writing as a vehicle for learning? And if so, how? Developmental changes in writing can be broadly characterized as a gradual shift in the kind of processes that drive composing. For children in early elementary school, writing depends largely on language production processes and resources that are shared by speech. These include verbal intelligence, idea generation, oral vocabulary, and clause production (Abbott & Berninger, 1993; Berninger & Swanson, 1994; Berninger et al., 1992; Juel, 1988). Writing is also affected by mechanical skills, including handwriting and spelling (Abbott & Berninger, 1993; Berninger et al., 1992; Juel, 1988; Maki, Voeten, Vauras, & Poskiparta, 2001). In the mid-elementary grades, language-production abilities and mechanical skills continue to affect the quality of students’ writing (Berninger et al., 1994; Graham et al., 1997; McCutchen, Covil, Hoyne, & Mildes 1994). Additionally, knowledge about genre such as narrative and exposition develops and contributes to writing (Berninger & Swanson, 1994; Donovan & Smolkin, 2002; Englert, Stewart, & Hiebert, 1988). Planning occurs, but remains largely local, i.e., focused on the upcoming sentence. Some planning occurs prior to writing, and contributes to its quality, but this remains sporadic, and consists largely of producing possible content. Students’ revisions address both semantic and mechanical features, but remain largely local, i.e., focused on the current sentence; when post-writing revision occurs, it often fails to improve the quality of the final text (Beal, 1996; Berninger & Swanson, 1994; Cameron, Edmunds, Wigmore, Hunt, & Linton, 1997; van Gelderen, 1997). To a large extent the process of text production continues to resemble that of speech production, so that pre-writing plans, first drafts, and final drafts have substantially similar content and language; think-aloud protocols taken during composition are often nearly identical to final texts (Bereiter & Scardamalia, 1987). During the later elementary and secondary school years, students’ writing comes increasingly under the control of metacognitive processes (Bereiter & Scardamalia, 1987; Berninger & Swanson, 1994; Berninger, Whitaker, Feng, Swanson, & Abbott, 1996). Students supplement local planning with more extensive planning prior to writing. They also supplement local revision with post-writing revision; they become increasingly able to detect, diagnose, and correct errors, moving from surface level revisions toward text level revisions that can change the gist and quality of a composition (Beal, 1996; Berninger & Swanson, 1994). Students acquire explicit knowledge about writing and mobilize this to guide their writing behavior (Berninger, Whitaker, Feng, Swanson, & Abbott, 1996). These changes amount to a qualitative shift from writing that is dominated by language production processes and mechanics, to writing that is guided by metacognitive processes. However, this does not imply a strict stage model of writing. The late-developing metacognitive aspects of writing complement, rather than replace, the early-developing languageproduction processes, and the language-production processes themselves continue to develop (e.g., Berninger & Swanson, 1994; Flower & Hayes, 1981; Torrance, Thomas, & Robinson, 1996). Moreover, the extent to which writers engage in metacognition depends on variables such as individual level of writing ability, personal writing style, rhetorical task, and availability of instructional assistance (e.g., Beal, 1996; Bereiter & Scardamalia, 1987; De La Paz & Graham, 2002; Levy & Ransdell, 1996; van Gelderen, 1997).


9/27/2006

3:06 PM

Page 203

Writing to Learn

203

The importance of these developmental changes could be viewed very differently, depending on which theory of writing to learn one adopts. I have noted that theories of writing to learn comprise a continuum, ranging from those that emphasize spontaneous processes to those that emphasize deliberative or metacognitive processes (Klein, 1999). Metacognitive theories have traditionally dominated the cognitive psychology of writing (e.g., Flower and Hayes, 1980; Bereiter & Scardamalia, 1987). These theories differ in their details, but agree that writing allows students to structure and restructure their ideas through strategies comprising explicit problem-solving operations such as goal setting, planning, organizing, rereading, evaluation, and revising. Correlational studies have supported the notion that metacognitive processes contribute to learning during writing (e.g., Keys, 2000; Klein, 2000; Newell & Winograd, 1989). Metacognitive processes develop slowly, so these theories would imply that writing is likely to lead to learning only for relatively mature or well-educated writers. Educationally, writing to learn would be reserved primarily for secondary and university students. Alternatively, teachers might attempt writing to learn activities with younger students, but they would need to provide considerable instruction and support. The other end of the theoretical spectrum attributes learning to spontaneous writing processes (Britton, 1980/1982; Murray, 1978). This viewpoint has included the “text production” or “knowledge constituting” theory (e.g., Galbraith, 1999; Galbraith & Torrance, 1999). Authors in this tradition have held various points of view, but generally believe that tacit knowledge and spontaneous writing processes contribute to learning (Britton, 1980/1982; Galbraith, 1999). As the writer composes, this tacit knowledge is activated and translated into language. The syntax and semantics of language help to structure this knowledge, but the process of translation takes place largely outside of the writers’ awareness so that the ideas generated during writing have an “unbidden” quality. The distinctive claim of the text-production view is that non-metacognitive processes are sufficient for learning to occur. Support for this view comes from anecdotes (Murray, 1978), some correlational evidence (Copeland & May, 1987; Newell & Winograd, 1989), and some experimental evidence (Galbraith, 1992, 1999). Educationally, this class of theories implies that educators should encourage exploratory writing for learning, and reserve rhetorical planning and revision for later in the writing process. These theories imply that young writers, who generate ideas and language, but do little planning and revising, may nonetheless be capable of writing to learn. However, language-production abilities and tacit knowledge develop with age; so under these theories, writing to learn would also be expected to show some developmental changes (Galbraith, personal communication, January 13, 2005). It should also be noted that metacognitive and text production theories are not contradictory; for example, Galbraith’s (1999) theory actually includes two kinds of processes, one of which translates implicit knowledge into explicit knowledge, and the other of which operates on explicit propositions. An additional aspect of the writing to learn process is the use of sources. Writers always draw upon some source of knowledge: texts previously written by other authors, observations, or at the very least, their own prior experiences. In the course of writing, they take up ideas and language from these sources, and transform them to serve their current writing goals (McGinley & Tierney, 1989; Spivey, 1997; Wiley & Voss, 1999). Often they


204

9/27/2006

3:06 PM

Page 204


show evidence of learning as they do so (e.g., Boscolo & Borghetto, September, 2002; Wiley & Voss, 1999; see Tynjälä, 2001, for a review). The developmental aspects of writing to learn have not been extensively explored empirically. Adults show a writing superiority effect in which they display more of their knowledge in writing than speech, but children show the reverse pattern (Bourdin & Fayol, 2000; Grabowski, this volume). However, the idea that elementary students can successfully write to learn has been supported by previous empirical research (e.g., Boscolo & Mason, 2001; Konopak, Martin, & Martin, 1990). Recently, Bangert–Drowns, Hurley, and Wilkinson (2004) completed a meta-analysis of classroom-based writing to learn interventions. They found that writing had a small but significant effect on learning in Grades 1 through 5, Grades 9 through 12, and college, but not in Grades 6 through 8. This raises the question of what might mediate the effects of grade level on writing to learn. The research discussed in this chapter explores the relationships among development, writing, and learning in the context of writing about science experiments. From the viewpoint of writing to learn, science provides a valuable context for research. In most early research on writing to learn, students read textual sources and then wrote about them. This meant that in many instances, ideas, vocabulary, and text organization were largely provided by the sources rather than created by the students. In contrast, science experiments provide students with relatively novel experiences, and allow researchers to study the way in which they transform these experiences into written language. Conversely, writing is an interesting activity from the viewpoint of science education. It has become a commonplace that science learning must be not only “hands-on” but also “minds-on.” Writing appears to provide an opportunity for students to reflectively interpret science experiments. One of the two science topics explored in this study was buoyancy. Understanding buoyancy involves several essential concepts about matter, including weight, volume, and density. Both children and adults find the question, “Why do objects sink or float?” intelligible, but challenging. Initially, children attribute buoyancy to various variables, including weight, type of material, and volume. With experience and instruction, they gradually progress toward the standard scientific view that objects float in a fluid if they are less dense than the fluid, and sink if they are more dense, where density is weight per unit volume (Halford, Brown, & Thompson, 1986; Inhelder & Piaget, 1958; She, 2002). However, many adult non-scientists continue to attribute buoyancy to weight, largely because they only gradually and partially differentiate weight from other attributes of matter, such as volume (Ginns & Waters, 1995; Léoni & Mullet, 1993). The substantial research literature on buoyancy provides a valid criterion for assessing research participants’ explanations of buoyancy (see Table 1). The second science topic was forces acting on a balance scale. This topic draws on the concept of torque, which is important for understanding levers and other simple machines. The question, “what make the scale tilt in one direction or other, or balance?” is readily intelligible to both children and adults. Initially, children predict and explain balance solely in terms of the amount of weight on each side of the fulcrum. Later, for cases in which the weights on each side of the scale are equal, they predict that it will tilt toward those that are further from the fulcrum. Then, in mid to late elementary school, most participants understand that greater weight on one arm of the scale can be compensated for by a greater distance between the weights and the fulcrum on the other arm. However, only


9/27/2006

3:06 PM

Page 205

Writing to Learn

205

a minority of upper elementary students, secondary students, and adults infer that the scale balances if the product of the weight and the distance on one arm of the beam equals the product of the weight and the distance on the other arm. Assessments of research participants’ explanations of the balance scale can be validated against previous research (Hardiman, Pollatsek, & Well, 1986; Inhelder & Piaget, 1958; Siegler, 1985) (see Table 2). In order to better understand the relationships among development, writing, and learning, it is necessary to compare novice and experienced writers within a single study. Consequently, for this chapter we have combined and re-analyzed the data from two previous studies of writing to learn. In the first study, Klein (2000) asked 70 elementary students in Grades 4, 6, and 8 to carry out science experiments concerning either buoyancy, or the factors that affect a balance scale, then to write informal, journal style notes about these experiments while thinking aloud. The students provided explanations of the science phenomena before experimenting, after experimenting but before writing, and again after writing. The think-aloud protocols were transcribed and segmented into meaning units, which were classified with respect to writing operations and strategies. In a subsequent study, Klein (2004) asked 56 university non-science majors to carry out the same science experiments and think-aloud writing activity. Both data sets included counts of metacognitive writing operations (goal setting, organizing, rereading, evaluating, revising), and text-production operations (generating, transcribing, generating and transcribing simultaneously). In the present chapter, path analysis was used to examine relationships among development, writing, and science learning. We hypothesized that educational level would contribute to level of writing strategies, and to pre-writing science explanations; level of writing strategies and pre-writing science explanations would contribute to level of postwriting science explanations. We were also interested in whether the effects of some variables might be mediated by others, although we had no strongly held hypotheses about this. For example, would educational level directly affect post-writing science explanations, or would educational level act exclusively through mediating variables such as students’ writing strategies and pre-writing science explanations? We were also interested in questions about interactions. For example, did students’ writing strategies interact with level of prior knowledge, or did they act similarly across knowledge levels?

13.2. 13.2.1.

Method Participants

The 126 participants included 70 elementary school students and 56 university students. The elementary students were drawn from an urban school that served a middle socioeconomic status neighborhood, and a rural school that served a lower and middle socioeconomic status area. The school students included 36 girls and 34 boys randomly selected from Grade 4 (age 9 years, 0 months to 9 years 11 months), Grade 6 (11 years, 0 months to 11 years, 11 months), and Grade 8 (13 years, 0 months to 13 years, 11 months). The university students were recruited through a poster advertising for “students who are not science majors” at a research-intensive university with high undergraduate admissions


206

9/27/2006

3:06 PM

Page 206


criteria. They comprised 39 women and 17 men (median age = 20.00 years, IQR 19.00 to 23.50 years), majoring in the humanities, social sciences, fine arts, and non-science professions. 13.2.2.

Apparatus

The materials directly pertinent to the current study were 126 think-aloud protocols and written texts collected for two previous studies (Klein, 2000, 2004). The materials for the buoyancy task included a six-liter bucket of water and a sorting mat with three sections labeled ‘float,’ ‘sink,’ and ‘other.’ The 10 test objects were: a small wooden block (6g); a large stone (187g); a small stone (27g); a large wooden block (435g); a sealed plastic vial filled with salt (216g); a sealed plastic vial filled with wheat germ (57g); a sealed plastic vial filled with water (178g); a medium-sized wooden block (labeled 50g); a smaller, denser wooden block of equal weight (labeled 50g); and an aluminum can with a hole in the bottom (32g empty, 230g filled with water). Materials for the balance-scale task consisted of a pivot arm beam-balance scale, with four pegs on each arm, and nine disk-shaped weights that could be placed on these pegs. Materials also included a worksheet with 15 diagrams illustrating various arrangements of these weights on the beam. They were sequenced in order of difficulty, ranging from trials for which the outcome could be accurately predicted using weight alone, through to trials that required a comparison of the products of weight and distance on each arm of the beam. For those cases in which the beam balanced, the worksheet also contained a table for participants to record the number of weights and the distances of the weights from the fulcrum. 13.2.3.

Procedure

The same procedure was utilized in both studies. Participants completed the tasks individually, and their sessions were audiotaped. First, they participated in a warm-up activity to familiarize them with thinking aloud while problem solving. Next, they answered a pre-experiment question specific to the science task they would be performing. For the buoyancy task, the question was, “What makes objects float or sink?” For the balancescale task, the question was, “What makes the beam tilt in one direction or the other, or balance?” Participants then carried out the science experiment. For the buoyancy task, they predicted verbally whether each object would “float, sink, or other,” then tested each one and recorded the result by placing the object on the appropriate section of the sorting mat. For the balance-scale task, participants worked from a series of 15 diagrams representing increasingly complex arrangements of weights on the balance scale. For each, the participant predicted whether the scale would balance, tilt left, or tilt right. They then arranged the weights on the balance scale and released it. They recorded the result on the diagram. For trials on which the scale balanced, participants also recorded the number of weights and their distance from the fulcrum in a table. After the participants had completed the task, they explained the phenomenon again. Then, the participants wrote a journal-style note about the science experiment. Each received a lined page, with the question at the top appropriate to the phenomenon they had


9/27/2006

3:06 PM

Page 207

Writing to Learn

207

studied: “What makes objects float or sink? How did you learn this?” or “What makes the scale tilt in one direction or the other, or balance? How did you learn this?” Participants were asked to think aloud while writing. If they became silent, the researcher asked, “Tell me about what you are thinking.” Finally, after the participants finished writing, the appropriate questions were posed to them again.

13.3.

Analysis

For the original analysis of this data, the participants’ written texts and verbal protocols were transcribed. Participants’ pre-experiment explanations, pre-writing (post-experiment) explanations, and post-writing explanations were independently rated by two research assistants; reliability for scoring of the buoyancy task was 80% exact agreement, and for the balance-scale task was 87% exact agreement. Data from two participants were excluded from the present analysis because they rejected their pre-writing science explanations but did not articulate a post-writing explanation. Consequently, data were analyzed for 124 participants, of which 63 completed the balance-scale task and 61 completed the buoyancy task. Each verbal protocol was transcribed, and then segmented into meaning units delimited by a principal clause and any additional clauses dependent upon it. Each meaning unit was categorized according to the following writing operations, adapted from Flower and Hayes (1981): goal setting/organizing, generating, transcribing, generating and transcribing simultaneously, rereading, evaluating/revising, and “other”. The number of times each operation appeared in a participant’s protocol was counted. The level of inter-rater agreement ranged from 76% exact agreement for “other” operations to 90% exact agreement for “generating and transcribing simultaneously.” These counts were used for the current re-analysis.

Table 1: Buoyancy explanations, percentage by educational level. Explanation

None Weight Type of material Characteristic “weight” Weight per volume Density versus water

Pre-writing

Post-writing

Grd 4 Grd 6 Grd 8 Univ n = 11 n = 12 n = 11 n = 27


0.0 36.4 18.2

0.0 25.0 16.7

0.0 9.1 27.3

3.7 3.7 3.7

0.0 27.3 18.2

0.0 16.7 0.0

0.0 9.1 18.2

0.0 3.7 3.7

27.3

25.0

9.1

48.1

36.4

25.0

9.1

5.9

18.2

33.3

54.5

40.7

18.2

50.0

45.5

51.9

0.0

0.0

0.0

0.0

0.0

8.3

18.2

14.8


208

9/27/2006

3:06 PM

Page 208


Table 2: Balance scale explanations, percentage by educational level. Explanation

Pre-writing

Post-writing

Grd 4 Grd 6 Grd 8 Univ n = 13 n = 10 n = 11 n = 29 None None Weight only Weight and distance Distance and weight compensate Partial torque rule Torque rule


0.0 0.0 0.0

0.0 0.0 0.0

0.0 0.0 0.0

0.0 0.0 0.0

0.0 0.0 0.0

0.0 0.0 0.0

0.0 0.0 0.0

0.0 0.0 0.0

15.4

20.0

9.1

13.8

7.7

10.0

0.0

6.9

84.6

80.0

63.6

69.0

84.6

90.0

81.8

65.5

0.0 0.0

0.0 0.0

27.3 0.0

17.2 0.0

7.7 0.0

0.0 0.0

9.1 9.1

20.7 6.9

The general writing strategy of each participant was also identified. In the two original studies, different schemes were used to categorize the overall writing strategies of the children and the adults, so the protocols were completely re-coded for the current analysis. First, the educational level of each participant was masked. Then two of the researchers randomly selected 24 of the protocols, and sorted these into categories that appeared to represent holistic similarities and differences in general writing strategy. They discussed the results, settled on five categories, created descriptions of each category, and labeled them. These categories were considered to represent increasing levels of complexity, ranging from “knowledge telling” at the lowest level (Bereiter & Scardamalia, 1987), to “hypothesis testing,” at the highest level. Next, two researchers used these descriptions and the first set of exemplars to rate the remaining 100 protocols, with an intra-class correlation of .88, p < .01. Finally, the third researcher resolved disagreements. The generalwriting strategies are defined in Table 3, and described in the Results section. To create a variable representing text-production operations, the counts of generating, transcribing, and generating and transcribing simultaneously, were summed for each protocol. To create a variable representing metacognitive operations, the counts of goal setting/organizing, rereading, and evaluating/revising, were summed for each protocol. To represent use of sources, we included the number of times each participant reviewed experimental results. Path analysis was the primary tool for exploring the relationships among the two exogenous variables (educational level, type of science task), five possible mediating variables (pre-writing science explanation, text-production operations, metacognitive operations, reviewing experimental results, and general writing strategy), and the endogenous variable (post-writing science explanation). Prior to the path analysis, multivariate analysis and univariate follow-up analyses were used to test whether educational level interacted with type of science task (buoyancy vs.


9/27/2006

3:06 PM

Page 209

Writing to Learn

209

balance scale) to affect writing operations and strategies. Polynomial contrasts were used to examine whether the effects of educational level on writing strategies and science explanations changed consistently across grade levels (linearly), or inconsistently (quadratically or cubically). Bonferroni post-hoc tests were used to determine which differences in educational levels resulted in significantly different writing strategies and science explanations.

13.4.

Results

13.4.1.

Effects on Understanding of Science Concepts

The percentage of participants who presented each type of explanation of the buoyancy task appears in Table 1. Prior to writing, most of the Grade 4 and 6 students explained buoyancy in terms of the weight of the object, the type of material, or the characteristic weight of the material; most of the Grade 8 and university students explained buoyancy in terms of the characteristic weight of the material, or the weight per volume of the object. After writing, most Grade 4 students continued to explain buoyancy in terms of weight of the object, type of material, or characteristic weight of the material, and most Grade 6, 8, and university students explained buoyancy in terms of the weight per volume of the object, or the density of the object versus the density of water. Results of the balance-scale task are presented in Table 2. Before writing, most students at every educational level explained the balance scale in terms of compensation between the weight of the objects and their distance from the fulcrum, but did not relate these to one another mathematically. After writing, most students continued to offer the compensation explanations, but more students now coordinated weight and distance mathematically using a torque rule. In order to allow the results of the buoyancy and balance-scale tasks to be analyzed together, the levels of explanation were converted to Z-scores. A repeated measures multivariate analysis of variance showed that the time factor was significant, indicating that students’ post-writing explanations were significantly more advanced than their pre-writing (i.e., post-experiment) explanations, F (1, 120) 17.65, p .01, 2 .13. Educational level significantly affected the combined variable of pre-writing and post-writing science explanations, F (3, 120) 4.24, p .01, 2 .10. Bonferroni post-hoc tests indicated that the explanations of Grade 4 students differed significantly from those of university students; polynomial contrasts indicated that the linear trend across educational level was significant and other possible trends (e.g., quadratic) were not. Interaction effects between educational level and time were very small and statistically non-significant, F (3, 120) .63, p .05, 2 .02, suggesting that although students at higher educational levels showed more advanced science explanations, they did not show significantly greater changes from pre-writing to post-writing. 13.4.2.

Effects on Processes

Recall that students’ general-writing strategies were coded to group similar approaches to writing, and to represent increasing levels of complexity, such that higher-level strategies usually included the processes of lower-level strategies, plus additional ones (Table 3). The


210

9/27/2006

3:06 PM

Page 210


Table 3: General-writing strategies. Level 1: Level 2: Knowledge Rehearse and telling write

Level 3: Discursive reasoning

Overview

Student Student Student records prior generates ideas, produces knowledge transcribes some discourse, selectively explicitly reasons about it

Goal statements

None

None

Source and method of generating content

Prior knowledge

Prior knowledge

Text Text follows Same as organization sequence knowledge ideas were telling generated

Criteria for evaluation, completion

No evaluative “Out of ideas” comments “Answered question” etc.

Other indicators

Think-aloud protocol nearly identical to text

Think aloud consists of written text, plus some additional content

Level 4: Experiment based

Student generates ideas and language primarily by reviewing experimental results Goals pertain Content goals, to own e.g., “First I’m previous going to write discourse, about the ones e.g., “I will that floated… ” try to define Genre goals, that” e.g., “…like a lab report…”

Level 5: Hypothesis testing Similar to Level 4, but adds explicit hypothesis testing

Similar to Level 4, but also explicitly states intention to test ideas against record of results Operations on Reviews Generates previous experimental hypotheses, discourse: results and then comexplaining, prior pares to inferring, experiences, record of evaluating, then draws experimen etc. inferences tal results Chain-like, Text follows Text follows each sequence of sequence of statement experimental hypotheses, refers to procedures, rather than previous or types of experimenstatements results tal trials Clarity, Discussed all Has evalucoherence of experimental ated each text results, or hypothesis completed text proposed genre Text, think Think-aloud Same as aloud protocol Level 4 frequently includes refer considerable back to material in themselves addition to text


9/27/2006

3:06 PM

Page 211

Writing to Learn

211

lowest-level strategy, identified as knowledge telling (Bereiter & Scardamalia, 1987), consisted of students simply recording their previous beliefs. The rehearse and write strategy included knowledge telling, but also included students generating ideas and language, in addition to those that they transcribed, and recording them selectively. The third strategy, discursive reasoning, consisted of students producing speech or writing, then explicitly referring to this discourse; reacting to their initial discourse by elaborating upon it, drawing inferences from it, rejecting it, and so forth; then incorporating these reactions into their subsequent writing. A fourth strategy, experiment-based, was dominated by writers reviewing the results of experimental trials, using them to generate ideas and language, and organizing their text to correspond to the sequence of experimental trials or to classes of results (e.g., floating, sinking). Finally, hypothesis testing, like the experiment-based strategy, also included many references to the experiment, but also included writers initially developing tentative explanations and then explicitly testing these by comparing them to their records of the experimental results. The percentage of students who used each kind of general-writing strategy is reported in Table 4. General-writing strategies differed significantly across educational levels, F (3, 116) 5.94, p .01, 2 .13. Polynomial contrasts indicated that this trend was linear, and Bonferroni post-hoc tests showed that university students differed significantly from Grade 4 and Grade 6 students. Most Grade 4 and 6 students relied on lower-level writing strategies that reiterated previous knowledge, such as knowledge telling, or rehearsing ideas and selectively recording them. About half of university students used advanced writing strategies, such as reviewing experimental results to generate ideas and text content. Students who completed the buoyancy task showed slightly higher levels of general-writing strategies than those who completed the balance-scale task F (1, 116) 4.43, p .05, 2 .04. Quantitative features of the protocols and texts are reported in Table 5. The texts were typically about two hand-written paragraphs in length. As the first row indicates, the length of the text measured in clauses differed significantly by educational level, F (3, 120) 4.66, p .01, 2 .10; Bonferroni contrasts indicated that university students’ texts were significantly longer than those of Grade 4 students. Similarly, the second row indicates that the length of the think-aloud protocols as measured in total operations differed significantly by educational level, F (3, 120) 5.56, p .01, 2 .12, and that university students produced significantly longer protocols than both Grade 4 and 6 students.

Table 4: Level of general-writing strategies, percentage by educational level. Educational level Writing strategy Level 1: Knowledge telling Level 2: Rehearse and write Level 3: Discursive reasoning Level 4: Experiment based Level 5: Hypothesis testing

Grade 4

Grade 6

Grade 8

University

41.7 50.0 0.0 4.2 4.2

50.0 22.7 13.6 4.5 9.1

18.2 45.5 18.2 13.6 4.5

14.3 35.7 16.1 21.4 12.5


212

9/27/2006

3:06 PM

Page 212


The third through fifth rows of Table 5 provide a more fine-grained analysis of the writing process by reporting the frequency of each kind of writing operation. Prior to analysis, the frequency distributions of text-production operations, metacognitive operations, and reviewing experimental results were normalized using log transformations. A two-way multivariate analysis of variance was applied to examine the effects of educational level and type of science task on the resulting variables. Educational level produced statistically significant, medium-sized effects on the combination of writing variables, Wilk’s .86, F (12, 299.26) 3.04, p .01, 2 .10, as did type of science task, Wilk’s .86, F (4.00, 113.00) 4.72, p .01, 2 .14. Educational level did not interact significantly with type of science task to affect the combined writing variable, Wilk’s .86, F (12.00, 299.26) 1.51, p .05, 2 .05. Follow-up univariate analyses of variance showed that the frequency of text-production operations differed significantly by educational level, F (3, 116) 6.44, p .01, 2 .14. Polynomial contrasts indicated that this trend was linear, and Bonferroni post-hoc tests indicated that university students employed more text-production operations than Grade 4 and Grade 6 students. Students showed slightly more text-production operations in the buoyancy task than the balance-scale task, F (1, 116) 5.45, p .05, 2 .05. Metacognitive operations also differed significantly across educational levels, F (3, 120) 11.49, p .01, 2 .23. Polynomial contrasts showed that this trend was linear, and Bonferroni tests indicated that university students employed more metacognitive operations than Grade 4 and Grade 6 students. Metacognitive operations did not differ significantly by science task, F (1, 116) .14, p .05, 2 .00. Reviewing experimental results differed significantly by educational level, F (3, 116) 5.52, p .01, 2 .13. This trend was linear, and Bonferroni tests indicated that university students reviewed experimental results significantly more often than Grade 4 and Grade 6 students. Students showed significantly more reviewing in the buoyancy task than the balance-scale task, F (1, 116) 4.43, p .01, 2 .09.

Table 5: Mean frequency of writing operations (with standard deviation), by educational level. Grade 4 (n 24) Total operations 23.17a (11.64) Total text clauses 7.83a (3.75) Text-production operations 12.00a (5.79) Metacognitive operations 3.67a (4.14) Reviewing experimental results 2.21a (2.48)

Grade 6 (n 22)

Grade 8 (n 22)

University (n 56)

28.73a (24.22) 12.64ab (8.39)

33.95ab (23.32) 48.73b (36.60) 13.00ab (8.31) 14.50b (7.61)

14.00a (9.96)

17.05ab (11.70) 22.11b (16.07)

7.05a (9.61)

6.27ab (5.55)

12.57b (12.01)

3.14ab (3.62)

4.55ab (4.75)

7.20b (9.68)

Note: Means sharing a common subscript do not differ significantly at the p .05 level.


9/27/2006

3:06 PM

Page 213

Writing to Learn 13.4.3.

213

Relationships among Exogenous Variables, Processes, and Understanding

Prior to the path analysis, the data were examined to check statistical assumptions. In addition to the transformations reported above, the level of general-writing strategy was transformed by taking its square root. For the science-task variable, the balance-scale task was arbitrarily coded as “0” and the buoyancy task was coded as “1.” Three cases were identified as outliers by calculating Mahalanobis distance and comparing it to the chi-square criterion of 26.13 (df = 8). However, removal of these outliers did not change the relationships among the variables, so to preserve fidelity of the empirical data all cases were included in the calculation of the final path model. The creation of path models was subject to the following constraints for logical and temporal reasons: (1) type of science task and educational level were considered to be exogenous variables, unaffected by the writing process, so were placed at the beginning of the path model; and (2) post-writing science explanation was the endogenous variable of interest and so served as the final variable in the path. The remaining variables (pre-writing science explanation, text-production operations, metacognitive operations, reviewing experimental results, general-writing strategy) were ordered according to the best-fitting model using a stepwise regression heuristic. In addition, five other path models were generated and tested. The goodness of fit of the models was assessed based on the proportion of variance that they predicted in post-writing science explanations, and the number of variables for which the reproduced correlations (those implied by the model) and the empirical (observed) correlations corresponded closely, i.e., within r = .05 of one another. Figure 1 presents the standardized parameter estimates for the best path model. Overall, this model fit the data well: it accounted for 65% of the variance in post-writing science explanations; for 24 of the 28 possible pairs of variables, the reproduced correlations

Figure 1: Path-analytic model: Influence of task, educational level, pre-writing science explanation, writing strategies, and writing operations on post-writing science explanation.


214

9/27/2006

3:06 PM

Page 214


implied by the model were within r = .05 of the empirical correlations; for the remaining 4 possible pairs of variables, the reproduced correlations were within r = .10 of the empirical correlations (Table 6). In this model, pre-writing (post-experiment) science explanation, level of general-writing strategy, and frequency of reviewing the experimental results were direct, independent predictors of post-writing science explanations. General-writing strategy was a stronger and more direct predictor of post-writing science explanations than any of the three specific kinds of writing operations (i.e., reviewing the experiment, textproduction operations, or metacognitive operations). Together, these three kinds of writing operations accounted for 63% of the variance in level of general-writing strategy. The metacognitive operations variable was a direct predictor of both reviewing the experimental results and text-production operations, while reviewing experimental results was a direct predictor of text-production operations. Educational level directly predicted metacognitive operations and pre-writing science explanation. Type of science task directly predicted frequency of reviewing the experiment, reflecting the fact that students who completed the buoyancy task reviewed the results somewhat more often. This path model was chosen over two other good path models. The second best model was similar to the accepted one, but omitted level of general-writing strategy and used other writing operations (metacognitive operations, text productions, and reviewing experimental results) to directly predict post-writing science explanations. The total fit was slightly less

Table 6: Observed and reproduced correlations for path model. 1. 1. Science task 2. Education level 3. Reviewing experiment 4. Metacognitive operations 5. Text-production ops 6. Pre-writing explanation 7. General-writing strategy 8. Post-writing explanation

–.01 .19 –.07 .12 –.09 .13 .01

2.

.33 .45 .36 .22 .34 .28

3.

.56 .70 .10 .72 .47

4.

.66 .14 .66 .41

5.

.07 .70 .40

6.

.07 .67

7.

.46

Reproduced correlations 1. 1. Science task 2. Education level 3. Reviewing experiment 4. Metacognitive operations 5. Text-production ops 6. Pre-writing explanation 7. General-writing strategy 8. Post-writing explanation

.00 .23 .00 .11 .00 .12 .08

2.

.26 .45 .31 .22 .30 .28

3.

.58 .71 .06 .71 .45

4.

.67 .10 .66 .37

5.

.07 .70 .36

6.

.07 .65

7.

.44


9/27/2006

3:06 PM

Page 215

Writing to Learn

215

good than the selected model, accounting for 63% of the variance in post-writing science explanations, but the reproduced correlations fit the observed correlations poorly: six pairs of variables differed by more than r = .05; another six pairs differed by .10 or more. In another path model, the number of total operations was tested as a predictor. However, the stepwise regression heuristic deleted this variable in favor of other variables (e.g., general writing strategy, reviewing experimental results) that accounted for more variance in postwriting explanations. Finally, given that the immediate predictors of final science explanations were prewriting science explanation and level of general-writing strategy, we wanted to test whether the effects of these variables combined additively or interactively to affect understanding. To do so with reasonable statistical power, both variables were split into three categorical levels. The resulting 3 3 analysis of variance confirmed that level of generalwriting strategy and pre-writing science explanation both affected post-writing science explanation; the interaction between these variables was small and not statistically significant, F (4, 115) 1.50, p .05, 2 .05, implying that they combined additively.

13.5.

Discussion

In this study, the general pattern of findings was that students’ educational level predicted their pre-writing science explanations and writing process; these in turn predicted students’ post-writing science explanations. Therefore, our first proposition is that in content-area writing, the effects of educational level on learning are mediated by students’ writing strategies and prior knowledge. Moreover, this mediation was complete; there was no direct path from educational level to post-writing science explanations. Methodologically, this suggests that the mediating variables were validly conceptualized and measured. The fact that educational level, per se, had a statistically significant but modest effect on science explanations may be surprising, but can be accounted for by the fact that adults vary widely in their prior knowledge of physics, with most non-science majors showing conceptions similar to those of upper elementary students (e.g., Table 1 and 2; Ginns & Waters, 1995; Léoni & Mullet, 1993). Our second proposition is that writing which contributes to learning can be conceptualized as a strategic gestalt. Recall that among the writing variables, the best single predictor of post-writing science explanations was general-writing strategy. The relationship between general-writing strategy and specific kinds of writing operations (text production, metacognitive operations, reviewing experimental results) is a whole to part relationship; the arrows in the path analysis from each type of writing operation to general writing strategy should be read as indicating constituency, rather than cause-effect relationships. At the same time, the general-writing strategy as a whole appears to go somewhat beyond the sum of its parts in explaining learning; the path model which included it was substantially more coherent that the model that included the three kinds of component writing operations alone. This writing gestalt appears to be driven by metacognition. Metacognitive operations predicted students’ general-writing strategies, and their use of the two other kinds of writing operations (text production, reviewing experimental results). In turn, metacognitive


216

9/27/2006

3:06 PM

Page 216


operations were most strongly predicted by educational level. This pattern of findings is consistent with metacognitive theories of writing development (e.g., Bereiter & Scardamalia, 1987; Flower & Hayes, 1980). However, text-production operations (generating, transcribing) also contributed significantly to the level of students’ general writing strategies, which is consistent with text-production theories (e.g., Galbraith, 1999). So overall, these results support aspects of both metacognitive and text-production theories of writing to learn. Our third proposition is that writing strategies and prior knowledge contribute independently to learning. In the present study, level of pre-writing explanation and level of general-writing strategy combined additively, rather than interactively, to predict level of post-writing explanation. This means that a given general-writing strategy contributed approximately equally to learning across various levels of prior science knowledge. Moreover, the relationship between writing strategies and post-writing science explanations was direct; it was not the case, for example, that higher levels of prior science explanations facilitated advanced writing strategies, or that lack of prior science knowledge hampered good writing strategies. The fourth proposition is that non-text sources contribute to writing and learning in several qualitatively different ways. The protocols and path analysis showed that the experimental results provided a source of content for writing. Additionally, the protocols featuring Level 4 and 5 general strategies indicated that many students used the experimental results as the primary guide for organizing their compositional work and structuring their texts, either by following the sequence of experimental trials (trial 1, trial 2, trial 3, etc.), or by organizing their text according to types of experimental outcomes (e.g., trials in which the scales balanced, trials in which they did not). Finally, reviewing experimental results contributed directly to post-writing explanations, independently of students’ general-writing strategy. In this respect, writing served as an occasion for learning, rather than as a direct cause of learning. The role of sources has been noted by some authors in the writing to learn literature (e.g., Tynjälä, 2001), but has often been overlooked. This may be because the striking aspect of writing is that it can be a way of “telling ourselves something that we never knew before,” and this is most salient when no external sources of information are available to the author. However, realistically, most school-writing tasks involve sources, either textual or non-textual. This returns us to the question of whether or not writing is an effective learning strategy for elementary students. The results of our study, taken together with previous research in this area, tell a “good news, bad news” story. The bad news is that students in earlier grades typically use less sophisticated writing strategies, hold simpler initial explanations, and finish with simpler final explanations, compared to students in later grades. However, the good news is that the effect of educational level on writing strategies is moderate, with considerable variation within all grades, and the effects of good writing strategies, once in play, are consistent across grades. This finding is consistent with the generally positive effects of writing to learn that Bangert–Drowns et al. (2004) reported for elementary students, but somewhat inconsistent with their finding of a Grade 6–8 slump. It may be relevant that many students show diminishing motivation in late elementary school (Anderman & Maehr, 1994). Possibly, the Bangert–Drowns review reflects the diminishing motivation that these students show in classroom settings, while our study reflects the


9/27/2006

3:06 PM

Page 217

Writing to Learn

217

growing competence that they can display in an individual setting with a researcher. In summary, this study suggests that younger students can make gains during writing to learn activities, although these will be somewhat smaller or less frequent than gains for older students. Before concluding, we wish to acknowledge some of the limitations of this study. First, we note that the term “development” in the title of this chapter is not an explanatory construct in itself; it is a proxy for various processes, including instruction, maturation, and practice. Second, we have focused on a journal-style writing activity. This is a common form of writing in progressive science pedagogy, but many other forms of writing could also have been chosen that might have yielded different results. Third, we have focused on physics, which is only one of the many disciplines that use writing as a learning tool. Fourth, we relied largely on think-aloud protocols, which capture explicit, verbal thinking processes, but not implicit and imagistic processes. Fifth, the results of a path analysis do not represent fixed relationships, but depend on the effects of other variables in the model; introducing new variables, such as attitude toward writing (Lavelle, this volume), might change the regression coefficients observed here. More generally, path analysis is a correlational method, albeit one that helps to disentangle the variance accounted for by multiple predictor variables, so firm causal inferences cannot be based on it. The constructs used here invite a more fine-grained analysis. Our data comprised only the counts of various kinds of writing operations. Galbraith (personal communication, January 13, 2005) has pointed out that students’ text-production abilities develop with age. For example, vocabulary, sentence structure, and local coherence all increase. These developments affect the quality and quantity of students’ writing, particularly in elementary school (Abbott & Berninger, 1993), so it is reasonable to hypothesize that they also affect learning through writing. Similarly, this analysis focuses on relatively observable writing behaviors; this raises the question of how underlying psychological abilities, such as working memory (McCutchen, 2000) and approach to writing (Lavelle, this volume), might affect these observable behaviors. Including these variables in a path analysis could both enrich our understanding of the role of development in writing to learn, and test alternative “third variable” interpretations of the present findings. With these considerations in mind, what strategies might teachers encourage students to use in their writing? The descriptions of Level 4 and 5 protocols, and the path analysis of relationships among writing operations, suggest three actions. First, students could be encouraged to use writing to develop new ideas, rather than to simply record the ideas that they already have. Second, teachers could encourage students to review sources frequently, and to use them as a resource for generating ideas and language. Third, teachers could encourage students to reread their texts and evaluate them critically, with particular attention to anomalous source information, and to revise their ideas in light of this information. Expository writing is a “growth area” for students from elementary school through university, suggesting that it is a worthwhile activity at all educational levels. Given that university students in the humanities and social sciences often show good writing skills, combined with some anxiety about science learning, writing could be a particularly useful learning tool in courses on science for non-science majors.



9/27/2006

3:14 PM

Page 219

Chapter 14

Approaches to Writing Ellen Lavelle

To advance understanding of writing processes at the university level, a series of investigations were conducted to define a model of writing, approaches-to-writing, and to fully validate a related questionnaire, Inventory of Processes in College Composition. Psychometric methods served to yield five factors, Elaborative, Low Self-Efficacy, Reflective-Revision, Spontaneous-Impulsive, and Procedural, as representative of the interrelationship between students’ beliefs and strategies in academic writing. Validity studies encompassed a full range of methodologies and demonstrated support for the model. The discussion concludes with consideration of current applications of the model and inventory and with suggestions for further research.

14.1.

Introduction

While evidence suggests that writing is a valuable educational task demanding focus, expression, and rigor, what university students do when facing a writing assignment, or how they think about writing remains elusive. Writing is cognitively complex, involving multiple attentional demands, strategies, and processes, yet it is also affective involving intentionality and self-expression. It is, perhaps, both an art and a science, inspired yet routine, reflective yet directive. It is the mysterious and very personal nature of writing that has prompted me to conduct a series of investigations focused on how university students think about and engage in their craft. In this chapter, I provide an overview of the development of the approaches-to-writing model, discuss several applications, and offer some ideas for future directions. 14.1.1.

Theoretical Background

In the area of university learning, researchers have described students’ approaches to learning as reflective of the relationship between the student and the task (cf. Biggs, 1999). The constructs of deep and surface approaches have become common in the literature based on Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Lavelle, E. (2007). Approaches to writing. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes, & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 219–230). Amsterdam: Elsevier.


220

9/27/2006

3:14 PM

Page 220

Ellen Lavelle

both qualitative studies (Hounsell, 1997; Marton & Saljo, 1976; Van Rossum & Schenk, 1984) and on psychometric analyses (Biggs, 1987; Entwistle & Ramsden, 1983; Kember & Leung, 1998; Schmeck, 1983; Schmeck, Geisler-Brenstien, & Cercy, 1991). In a landmark study, Marton and Saljo (1976) queried students regarding their processes when studying an expository text and concerning the meanings that they constructed in doing so, focusing on what is learned, or how it is that students structure and understand, rather than on how much is learned. Two basic categories of description evolved. Students using a deep-level process focused on what is “signified” by the text, or the implications and intentions, and those employing a surface level process focused on the “sign,” or literal meaning (cf. Marton, 1988). In extending the deep and surface paradigm, researchers used psychometric methods to analyze students’ responses to survey items, thus advancing the distinction between deep learning, involving the intention to understand or create a meaning, self-referencing, and surface learning as marked by literal translation and the intention to reproduce or memorize information (Entwistle & Ramsden, 1983; Schmeck, 1983; Schmeck et al., 1991). Deep and surface approaches represent a modifiable dimension reflective of the interaction between the student and the learning environment. Students’ intentions and strategies are “framed” by the situation of learning and its related cues. It is the cues and affordances that instructional climates provide which impact the approaches that students take (cf. Biggs, Lai, Tang, & Lavelle, 1999). The deep and surface model had been linked to specific academic tasks such as reading (Marton & Saljo, 1976), studying (Schmeck, 1983), computer programming (Marton & Booth, 1997), and writing (Biggs, 1988; Hounsell, 1997). 14.1.2.

A Model and a Measure

There are parallels between writing assignments and other academic tasks such as reading, or presenting (e.g., vocabulary, genre or domain familiarity, and problem-solving skills), suggesting that the approaches framework might be well suited to adapt to writing. Also, there are differences but these too support the extension of the model. For one thing, in writing the interaction between learner and task is largely reciprocal because both editing and revising for meaning demand response to one’s own product, and to one’s own thinking. This is not to say that reading or studying are not reflective but rather to argue that reflection in writing is necessarily more self-referencing. In the revision process, it is as though writers continually grapple to refine and clarify their own creations as they move in successive iterations from the task. Perhaps no other instructional task mandates such dynamic and personal interaction. Along the same line writing is ill-defined with no right answers or specific rules for success, and genre often provides a very sketchy framework. Other tasks are likely to have procedures, rules, and specific desired outcomes. In writing, organization, skills, and following rules alone may be insufficient to create meaning in the deep sense. It is intentionality and beliefs that are integral (cf. Biggs, 1988; Lavelle & Zuercher, 2001). An approaches-to-writing framework takes beliefs and intentions into consideration as well as the strategic processes (cf. Biggs, 1988). Deep writing goes beyond the literal or technical level. It is as though the meaning becomes greater than the sum of the parts (cf. Marton, 1988). Biggs and Collis (1982) refer


9/27/2006

3:14 PM

Page 221

Approaches to Writing

221

to this phenomenon in their extended abstract level of the Structure of Learning Taxonomy. Along the same line, in studies involving examination of the strategies of children, Scardamalia and Bereiter (1982) differentiated knowledge transforming as opposed to knowledge telling strategies, Graves (1973) argued for reflective vs. reactive writing, and Dyson (1987) suggested a similar difference between socializers and symbolizers. In a landmark study working with twelfth graders, Emig (1971) supported an extensive — reflective distinction, and I have drawn on these ideas in my own work. A comprehensive model was needed — one that accounted for the intentions of the writer as well as the strategies of writing. Too often examination had focused on writing strategies as divorced from writing beliefs. Along the same line, writing processes had been separated into discrete components: planning, translating, and editing (cf. Hayes & Flower, 1980), doing violence to the assumption of writing as a tool of integration (cf. Vygotsky, 1962), and fostering cohesion often at the expense of coherence (cf. Witte & Faigley, 1988). Writing involves processes working together and instruction that is based on teaching discrete processes brings a technical or reductionistic emphasis. The approaches to learning model seemed an ideal framework with which to better understand writing. My goal was to develop a comprehensive model and inventory, the Inventory of Processes in College Composition, based on psychometric methods (Lavelle, 1993). A list of 212 true and false statements regarding beliefs and strategies in writing was developed by adapting items from the Inventory of Learning Processes (Schmeck, Ribich, & Ramanaiah, 1977) and the Approaches to Studying Inventory (Biggs, 1988). Items were also based on Biggs’ (1988) theoretical extension of his learning model to composition and on Hounsell’s (1997) interview study as well as on composition theory. A large sample of undergraduate students (423) completed the inventory. Factor analysis, based on the scree criterion, and orthogonal rotation, common in the student learning survey literature, yielded five independent factors reflective of a total of 72 items (see the Appendix). Two factors, Reflective-Revision and Elaborative, suggested a deep writing approach as they reflected the intention to make meaning and awareness of writing as a tool of learning. The other three factors, Low Self-Efficacy, Spontaneous-Impulsive, and Procedural, were interpreted as surface approaches-to-writing in their strong focus on micro skills, listing or repetition and organization strategies and more passive orientation. Elaborative is marked by a search for personal meaning, self-investment, and by viewing writing as symbolic. The emphasis is on active, personal engagement and on adeptly managing macro constraints such as audience and voice. Reflective-Revision, the third factor, describes a deep writing approach based on a sophisticated understanding of the revision process as a remaking or rebuilding of one’s thinking, “I re-examine and restate my thoughts in revision,” Reflective-Revision strategies are thesis-driven, involving taking charge to make meaning in writing. On the other hand, Low Self-Efficacy, Procedural, and Spontaneous-Impulsive are readily interpreted as surface approaches. Low Self-Efficacy describes a writing approach based on thinking about writing as a painful task. “Writing is always a slow process,” writers scoring high on this scale are virtually without a strategy and see the acquisition of micro skills and teacher encouragement as necessary for progress. The fourth factor, Spontaneous-Impulsive, profiles an impulsive and non-planful approach. “My writing just happens with little planning or preparation,” this approach is linked to viewing writing as


222

9/27/2006

3:14 PM

Page 222

Ellen Lavelle

a one-step procedure. The emphasis on minimal involvement and sticking to the rules is suggestive of a surface approach. The Procedural approach represents a method-oriented approach based on adherence to rules and a minimal amount of involvement: “When writing an essay, I stick to the rules.” I actually had a student who, in response to being asked to write a 500 word essay, quit in the middle of a sentence when the word count indicated 500 words! The scales were found to be fairly independent, interscale correlations ranged from – 0.01 to 0.32, and internal consistencies ranged from 0.83 (Elaborative) to 0.53 (Reflective-Revision). The 0.32 correlation was between Elaborative and ReflectiveRevision. An initial validity study using the new 72-item inventory, Inventory of Processes in College Composition, was conducted to test for the predictive power of the scales as linked to expository writing outcomes, and to examine the relationship of the scale scores to learning approach as measured by the Inventory of Learning Processes (Schmeck et al., 1977). Regression analysis supported that Reflective-Revision scale scores were strongly predictive of grade in composition (B ⫽ 0.30, p ⬍ .01) with Low Self-Efficacy serving as a negative predictor (B ⫽ ⫺0.28, p ⬍ .01). It is perhaps important to note that the original interpretation of the factors used the term writing styles but the scales had since been reinterpreted to represent writing approaches. The construct of style assumed consistency although research supported that variation in scale scores was linked to instruction (Biggs et al., 1999). The notion of approach provided a much more flexible interpretation and one more “instructionally flavored”. Furthermore, Biggs et al. (1999) had argued for flexibility in interpretation dependent on research or practical consideration. Table 1 presents the relationship of motives and strategies as linked to writing approaches. The initial investigation had been promising but questions remained. In order to more fully validate the inventory it was important to examine the relationship of the scale scores to various types of writing, and to emotional reactions to writing such as writing anxiety. Also, looking more fully at students’ beliefs about writing, and establishing the validity of the scales across populations were important issues. It also seemed important to test for second-order deep and surface factors as suggested in the literature.

Table 1: Approaches-to-writing. Approach Elaborative Low Self-Efficacy Reflective-Revision SpontaneousImpulsive Just like talking Procedural

Motive

Strategy

To express onself To acquire skills and/or avoid pain To make meaning To get done

Visualization, audience, voice, self-reference Study grammar, collaborate, find encouragement Revision, reshaping, multiple drafts Last minute, no planning

Please the teacher

Observe the rules, organize, manage writing


9/27/2006

3:14 PM

Page 223


223

In examining the relationship of scale scores to narrative writing performance, Lavelle (1997) hypothesized that Elaborative scale scores, which reflected the need for selfexpression, would predict narrative essay outcomes in terms of both complexity and degree of personal investment. Also of interest was the relationship of scale scores to writing apprehension as measured by the Daly-Miller Writing Apprehension Survey (as cited in Lavelle, 1997). In a study involving 74 students enrolled in a mandatory composition course, Elaborative scale scores were moderately but significantly correlated with degree of personal investment in narrative essay outcomes (r ⫽ .40) and served as negative predictors of writing apprehension (B ⫽ ⫺.54, p ⬍ .00). Low Self-Efficacy was correlated with both writing apprehension and writing complexity. Then Biggs et al. (1999) conducted an experimental study working with graduate students writing in English as a second language. Here the Inventory of Processes in College Composition was given as a pre-test/post-test measure for students attending a writing skills workshop. Significantly lower Procedural and Spontaneous-Impulsive scores, and significantly higher Elaborative scores were found after the workshop. Open-ended feedback supported the view that positive change had occurred. Nancy Zuercher and I investigated university students’ beliefs about themselves as writers and about the experience of learning in writing as related to writing approaches as measured by the Inventory of Processes in College Composition (Lavelle & Zuercher, 2001). Interview data included support for the deep and surface paradigm and as well as variation in students’ conceptions of writing, in their attitudes about themselves as writers, and in their felt need for personal expression in writing. Specifically, students scoring high on the Elaborative approach expressed a more personal and affective dimension involving self-reference and feeling in writing whereas, students scoring high on Reflective-Revision expressed a more critical, structural emphasis, incorporating awareness of process and an appreciation for writing as a tool of learning. It also seemed important to examine development in writing. Would scale scores be different for younger students/students in secondary school? Would a different factor structure more adequately explain secondary writing processes? Would scale scores predict writing competence for secondary students? Lavelle, Smith, and O’Ryan (2002) conducted an analysis involving administration of the Inventory of Processes in College Composition to 398 junior-level (third year) secondary students. Data were factor analyzed using the orthogonal rotation, and examination of the scree plot suggested three process factors. The first factor, Elaborative-Expressive, described a writing strategy based on personal investment and audience concern. The second factor, Planful–Procedural, denotes sticking to a plan, following rules, and “preparing” for writing. Achieving-Competitive, a third factor, reflects a teacher pleasing attitude or doing only what needs to be done to get a good grade. Two factors from the college model, Elaborative and Procedural, were replicated but two were not, Reflective-Revision and Low Self-Efficacy, suggesting a different pattern and the possibility of a developmental trend in writing. Reliabilities for the scales were acceptable: Elaborative-Expressive, r ⫽ .77, Planful-Procedural, r ⫽ .67, and AchievingCompetitive, r ⫽.62. Both Planful-Procedural and a measure of self-regulatory efficacy were predictive of grade in language arts class but not of the quality of writing under a timed condition. Table 2 reflects the motive and strategy components of each of the approaches to writing for secondary students.


224

9/27/2006

3:14 PM

Page 224

Ellen Lavelle

Table 2: Approaches-to-writing for secondary students. Approach Elaborative-Expressive Planful-Procedural Achieving-Competitive

Motive To express oneself Learn to write, learn skills To manifest competence

Strategy Organization and description Planning and observing rules Managing time

The Inventory of Processes in College Composition provided a valid measure of writing approaches, as well as a useful model for teaching and research. The initial sample size was large (N ⫽ 423), and validity studies involved diverse and rigorous methods; both statistical and qualitative. The interview research, in particular, served to support and expand the interpretation suggested by the psychometric investigations. Here, students’ degree of awareness of process in writing differentiated both the Elaborative and Reflective-Revision approaches from the surface approaches, as did feelings of satisfaction and wholeness upon completion of writing assignments. Both Elaborative and Reflective-Revision encompassed the idea that process was critically linked to learning in writing and to writing outcomes. The Reflective-Revision dimension suggested a more analytic, critical, and perhaps “detached” or covert dimension, whereas students adopting an Elaborative approach consistently acknowledged writing selfhood, ownership, and attachment to writing. On the other hand, those students adopting surface approaches tended to consistently mention a dislike for writing and had no firm conception of themselves as an author-agents. They tended to maintain an exclusive focus on micro-level phenomena such as grammar, spelling, and syntax and to see outside support as critical for development. In a recent study, the approaches-to-writing model served in development of a valid rubric to reflect the quality of undergraduate writing across the four years of college (Lavelle, 2003). A preliminary study was conducted to test for differences in writing based on writing samples as part of a university portfolio collection. Writing samples were evaluated using both a traditional analytic measure based on organization, integration, fluency, audience, voice, and word usage, and a deep and surface rubric based on level of integration, refection, structure (hierarchical vs. linear), and an assessment of overall meaningfulness. Results supported the validity of the deep and surface rubric as evidenced by the correlation between results based on that rubric and the deep and surface measure (r ⫽ 0.53). Interestingly, evaluation of both early (first and second year) and late (third and fourth year) writings found no significant differences in the quality of writing as measured by either rubric, calling into question the sequencing and content of freshman writing courses. 14.1.3.

Confirmation

In re-assessing the psychometric dimensionality of the Learning and Study Process Questionnaires, Biggs and Rhin (1984) carried out a confirmatory factor analysis involving six subscales of the SPQ, and identified two higher-order factors, deep and surface. More recently, Kember and Leung (1998) supported a two-factor model based on deep/meaning and surface/reproducing factors as linked to both strategy and motivational


9/27/2006

3:14 PM

Page 225


225

indicators, which lead to a recent revision of the SPQ as a two-factor, deep, and surface scale for use by teachers (Biggs, Kember, & Leung, 2001). Tony Guarino and I (2003) wanted to examine the relationships among scales and hypothesized second-order deep and surface factors. We (Lavelle & Guraino, 2003) conducted a confirmatory analysis involving 517 undergraduate students. Specifically, we hypothesized, based on the original study (Lavelle, 1993), that the scales were independent indices that would load on two independent, latent factors. Specifically we thought that Reflective-Revision and Elaborative would be indices of the Deep factor, with Procedural, Low Self-Efficacy and Spontaneous-Impulsive indicating an independent, Surface factor. Using a two-step structural equation modeling strategy to estimate parameters, and employing aggregated scale scores as observed variables, the hypotheses were tested. A series of confirmatory analyses revealed that the items representing the five scales were valid indicators of their respective factors. Although Chi square values were significant (x ⫽ (5) ⫽ 29.91, p ⬍ .01), the model yeilded acceptable goodness of fit indices (.998 and .994 for the CFI and TLI, respectively, and .09 for the RMSEA). All measured variables loaded significantly on their respective factors, and these loaded significantly on the latent factors (see Lavelle & Guarino, 2003 for a full description). In sum, deep writing is strongly indicated by constructive revision as reflected in the Reflective-Revision scale. Students adopting this strategy take an agentic position, see themselves as makers of meaning and are aware of the powerful role of revision as a tool of transformation (cf. Lavelle & Zuercher, 2001). Thus, deep writing may be similar to what Segev-Miller refers to as transformational; based on deliberate intention and on consideration of macroproposition in the synthesis of text (this volume). The Elaborative approach encompasses personal investment and ownership in writing and is also indicative of deep writing. Researchers had consistently linked dimensions of selfhood to writing skills under the frames of self-regulation (Zimmerman & Bandura, 1994) and self-efficacy (Meier, McCarthy, & Schmeck, 1984; Shell, Colvin, & Bruning, 1995). Table 3 reflects deep and surface writing criteria based on the full spectrum of writing research. 14.1.4.

Applications and Suggestions for Further Research

The Inventory of Processes in College Composition serves as a popular tool for diagnosis and remediation in developmental education. At Owens Community College (Ohio, USA), Table 3: Chi-square and Goodness of Fit Indices for Confirmatory Factor Models Factor Model Elaborationist Low Self-Efficacy Reflective-Revision Spontaneous-Impulsive Procedural * p ⬍ .05.

χ2

df

CFI

TLI

RMSEA

555.24* 202.04* 93.23* 201.19* 255.75*

230 44 35 65 54

.988 .986 .994 .990 .986

.985 .979 .991 .985 .980

.059 .094 .064 .072 .096


226

9/27/2006

3:14 PM

Page 226

Ellen Lavelle

Kay Blue used the IPIC to measure developmental education students’ approaches-to-writing before and after a course in writing (personal communication, September 7, 2004). The post-test analysis revealed that students scored higher in Elaborative and ReflectiveRevisionist and lower on the Low Self-Efficacy scale and Spontaneous-Impulsive scale. Students responded well to the inventory and were interested in both pre-test and post-test results. This fall she is using the questionnaire in all of the developmental writing classes to bolster traditional assessment measures. Similarly, the Organizational Leadership Program at Greenville College (Illinois, USA) is using the IPIC to acquaint adult reentry students with their writing strengths and weaknesses. The curriculum is also undergoing change to prompt both students and faculty to use deep learning activities, and to develop reflective writing approaches (Dave Holden, personal communication, September 15, 2004). Dave Holden (2004) used the IPIC to test for differences in writing approaches and in the quality of writing of adult, reentry students enrolled in the two-year program. While significant changes were not found, adult learners scored significantly higher on the Elaborative scale and low on the Reflective-Revision scale than traditional age college students. This finding was congruent with other research with adult or nontraditional age college students. In my own teaching of educational psychology at both the graduate and undergraduate levels, I have consistently relied the deep and surface model as a guide for instruction, delivery, and evaluation with good success. My focus has been on what Biggs (1999) calls “constructive alignment” in designing integrated tasks and assessments that foster meaning. Also, I have not been hesitant to encourage self-referencing activities such as journaling, and I use the deep and surface rubric to evaluate writing, presentation, and discussion activities. Based on the observations of colleagues, and on my own experience, I offer instructional strategies linked to each of the characterisitcs of deep writing (Table 4).

Table 4: Deep and surface writing approach characteristics. Deep approach Metacognitive, reflective High or alternating focus Hieraricical organization Enagagement Self-referencing, agentic Audience concern Thesis-driven Revision Coherence Transforming Autonomous Feelings of satisfaction and connectedness

Surface approach Redundant, reproductive Focus at the micro-level Linear, sequential organization Detachment Passive Data concern Data or teacher-driven Editing Cohesion Telling Rule-bound Just get done


9/27/2006

3:14 PM

Page 227


227

Table 5: Instructional strategies. Deep approach characteristic Metacognitive, reflective High or alternating focus Hierarichical organization Engagement Authorship, agency Audience Thinks about essay as a whole Thesis-driven Revision Transforming, going beyond the assignment Autonomous Teacher-independent

Instructional strategy Relevant tasks, free writing, modeling Genre familiarity Simple to complex tasks, mapping Task options, providing choice Voice, relevance Perspective taking, peer review Grading rubrics, task integration Modeling, task options, genre familiarity Emphasis on full, integrated revision Collaboration and modeling Task choice, journaling Teacher role is facilitator

In particular, it is important to design an integrated writing environment that fosters both deep beliefs and strategies. Instructors might consider providing choices for students such as choices of topics, types of writing, and timing of assignments, to empower students as makers of meaning. Along the same lines, providing relevant tasks, encouraging perspective taking, and modeling writing are effective instructional tools. Strategies such as genre familiarity and mapping may serve to scaffold developing writers. Implications for future research are many. It is important to examine the validity of the Inventory of Processes in College Composition for use with a range of populations. While cross-cultural validity is suggested (Biggs et al., 1999), more studies need to fully examine that issue. It would also be useful to fully develop the inventory for use with a secondary or even elementary population. Preliminary results are promising but full validation is important. Longitudinal and cross-sectional studies are also needed to trace development in writing across time, and experimental studies need to be conducted to test for the effects of interventions based on the model. Finally, it is important for college instructors and developmental educators to use the inventory as a teaching tool. The Inventory of Processes in College Composition and the approaches-to-writing model suggest alternative ways to think about writing and about what writers believe and do when faced with writing tasks. In particular, emphasis on creating a deep writing climate is critical. Too often teachers of writing give verbal support to/for this notion but fail to translate maxims into instructional activities. A recent informal review of composition syllabi from the internet reflects piecemeal instruction as evidenced by syllabi lacking teaching/learning objectives and containing disparate assignments.Along the same lines, administering the Inventory of Processes in College Composition will raise students’ awareness of themselves as writers and of strategic options.


228

9/27/2006

3:14 PM

Page 228

Ellen Lavelle

Appendix: Inventory of Processes in College Composition FACTOR I Elaborative 53. 68. 2. 11. 20. 30. 33. 39. 22. 67. 36. 4. 57. 7. 17. 34. 64. 65. 12. 73. 46. 121. 6.

Writing makes me feel good I tend to give a lot of description and detail. I put a lot of myself in writing. I use written assignments as learning experiences. Writing an essay or paper is making a new meaning. At times, my writing has given me deep personal satisfaction. Writing is like a journey. It’s important to me to like what I’ve written. I think about how I come across in my writing. I often think about my essay when I’m not writing (e.g., late at night). I sometimes get sudden inspirations in writing. Writing helps me organize information in my mind. I cue the reader by giving a hint of what’s to come. I often use analogy and metaphor in my writing. I imagine the reaction that my readers might have to my paper. When writing a paper, I often get ideas for other papers. I compare and contrast ideas to make my writing clear. I visualize what I’m writing about. Writing reminds me of other things that I do. Writing is symbolic. Originality in writing is highly important. I try to entertain, inform, or impress my audience. I use a lot of definitions and examples to make things clear.

.62 .56 .54 .51 .49 .49 .48 .47 .45 .44 .43 .42 .41 .41 .40 .38 .38 .37 .36 .35 .33 .33 .31

FACTOR II Low Self-Efficacy 66. 24. 69. 10. 21. 60. 47. 50. 62. 23. 8.

I can write a term paper. Writing an essay or paper is always a slow process. Studying grammar and punctuation would greatly improve my writing Having my writing evaluated scares me. I expect good grades on essays and papers. I need special encouragement to do my best writing. I do well on essay tests. I can write simple, compound, and complex sentences. My writing rarely expresses what I really think. I like to work in small groups to discuss ideas or to do revision in writing. The most important thing in writing is observing the rules of grammar, punctuation, and organization. 72. I often do written assignment/s at the last minute and still get a good grade. 18. I can’t revise my own writing because I can’t see my own mistakes. 38. If the assignment calls for 1000 words, I try to write just about that many.

−.54 .52 .47 .41 −.41 .39 −.38 −.37 .36 .35 .35 −.33 .29 .26


9/27/2006

3:14 PM

Page 229


229

FACTOR III Reflective-Revision 27. 70. 42. 5. 59. 40. 3. 43. 46. 32. 39. 16. 35.

I re-examine and restate my thoughts in revision. There is one best way to write a written assignment. I complete each sentence and revise it before going onto the next. The reason for writing an essay really doesn’t bother me. Often my first draft is my finished product. Revision is a one time process at the end. When given an assignment calling for an argument or viewpoint, I immediately know which side I’ll take. My prewriting notes are always a mess. I plan out my writing and stick to the plan. In my writing, I use a\some ideas to support other, larger ideas. It’s important to me to like what I’ve written. Revision is the process of finding the shape of my writing. The question dictates the type of essay called for.

.52 −.45 −.41 −.39 −.39 −.39 −.39 .36 −.35 .33 .33 .32 .31

FACTOR IV Spontaneous-Impulsive 15. 72. 51. 59. 9. 29. 48. 25. 41. 52. 31. 18. 45. 40. 19.

My writing ‘just happens’ with little planning or preparation. .51 I often do written assignments at the last minute and still get a good grade. .47 I never think about how I go about writing. .45 Often my first draft is my finished product. .45 I usually write several paragraphs before rereading. .42 I just write ‘off the top of my head’ and then go back and rework the whole thing. .41 I start with a fairly detailed outline. −.40 I plan, write and revise all the same time. .37 I am my own audience. .35 When I begin to write, I have only a vague idea of how my essay would come out. .35 Revision is making minor alterations — just touching things up and rewording. .34 I can’t revise my own writing because I can’t see my own mistakes. .33 When writing an essay or paper, I just write out what I would say if I were talking. .32 Revision is a one time process at the end. .31 I set aside specific time to do written assignments. .29

FACTOR V Procedural 54. 63. 62. 49.

When writing an essay, I stick to the rules. I closely examine what the essay calls for. I keep my theme or topic clearly in mind as I write. I can usually find one main sentence that tells the theme of my essay.

.54 .52 .43 .41

(Appendix: Cond.)


230

9/27/2006

Page 230

Ellen Lavelle

FACTOR V Procedural 58. 14. 71. 28. 1. 55.

3:14 PM

Cond.

The teacher is the most important audience. I like written assignments to be well-specified with details included. My intention in writing papers or essays is just to answer the question. The main reason for writing an essay or paper is to get a good grade on it. An essay is primarily a sequence of ideas, an orderly arrangement. I worry about how much time my essay will take.

.40 .34 .33 .31 .29 .28


9/29/2006

2:23 PM

Page 231

Chapter 15

Cognitive Processes in Discourse Synthesis: The Case of Intertextual Processing Strategies Rachel Segev-Miller

Discourse synthesis, or writing-from-sources, is a common but cognitively demanding reading-writing task requiring students to select, organize, and connect content from multiple source texts as they compose their own new texts. The purpose of the present study was to investigate the intertextual processing strategies underlying the performance of the task in an authentic academic context. The subjects were 12 teachers’ college seniors required to submit a discourse synthesis task (a literature review) in partial fulfillment of their teaching methodology course requirements. Data were collected over two semesters using a variety of verbal reporting techniques: process logs, think-aloud protocols, and interviews. Data analysis yielded a taxonomy of intertextual processing strategies, which represent distinct discourse synthesis processes. Two of the major implications of the present study are the validity of the process log as an instrument to collect data otherwise unavailable, and the need for explicit instruction of these strategies.

15.1.

Introduction

Discourse synthesis, or writing-from-sources, is a common academic task, or rather a range of tasks (e.g., a literature review, a critique), requiring students to select, organize, and connect content from multiple source texts as they compose their own new texts (Spivey, 1997). Discourse synthesis is a task involving both reading and writing. Although the early 1980s evidenced the surge of cognitive research of reading and writing processes and the relationship between the two became a focus of research interest (for a review see Nelson & Calfee, 1998), discourse synthesis was largely neglected until the early 1990s. The only reading-writing task that had been extensively investigated involved summarizing a single text (for a review see Kirkland & Saunders, 1991). Most studies of summarizing draw on the theoretical model of Kintsch and van Dijk (1978), which postulates that textual Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Segev-Miller, R. (2007). Cognitive processes in discourse synthesis: The case of intertextual processing strategies. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 231–250). Amsterdam: Elsevier.


232

9/29/2006

2:23 PM

Page 232

Rachel Segev-Miller

processing and construction of meaning involve producing a mental summary of the text by means of three major operations: deletion of redundant propositions; substitution of a sequence of propositions by a more general one; and selection of the macroproposition of the text, or the construction of a macroproposition when one is not explicitly stated. The discourse synthesis task is similar to the summary, but it is cognitively more demanding: when synthesizing students are required to construct their own macroproposition, or rather “superproposition” (henceforth the Macroproposition), from different or even sometimes contradictory propositions and macropropositions of multiple source texts, and to organize these in a previously non-existent conceptual structure. Conceptual restructuring or transforming, therefore, requires a higher order intertextual processing (Flower, 1989), and the production of personal and creative perspectives on the part of students (Schumacher & Gradwohl, 1991). Discourse synthesis is thus an act of literacy in line with recent definitions of learning (e.g., Tynjälä, 1998), which emphasize the significance of these cognitive abilities. However, school does not emphasize these abilities that may develop by means of explicit instruction and practice of such learning tasks (Lohman, 1993; Sternberg, 1998). Rather, school teachers’ questions more often require students to make intratextual rather than intertextual connections — although significant learning is believed to depend on the latter, possibly because professional materials and teacher training programs contain very few suggestions for teachers to promote intertextual processing (Segev-Miller, 2004a). One purpose of the present study, then, was to investigate the intertextual processing strategies underlying the performance of a discourse synthesis task in the context of a teachers’ college. 15.1.1.

Review of the Literature

Most discourse synthesis studies investigated the effect of different variables (e.g., reading proficiency, the nature of the task, or the sources) on the quality of the written product, which they defined in terms of the selection of relevant information from the sources, the organization of this information in an appropriate rhetorical structure, etc. (e.g., Spivey, 1997). However, the studies, which investigated the process underlying the product, indicated that most of the strategies used by the subjects in these studies (e.g., elaborating, paraphrasing) were not unique to the synthesis process, but rather characteristic of reading and writing processes in general. On the other hand, the strategies conceived of as crucial to the successful performance of the discourse synthesis task, namely, strategies of intertextual processing, particularly strategies of conceptual transforming such as comparing the sources or constructing a Macroproposition, were rarely used by school learners (Mani, Fyfe, Lewis, & Mitchell, 1996; Raphael & Boyd, 1991) and even by college learners (Kantz, 1989; Kennedy, 1985; McGinley, 1992). In fact, some of these process studies did not investigate these strategies at all. For example, Ackerman (1991) investigated the effect of his subjects’ disciplinary knowledge on their use of the strategies of elaborating and rhetorical evaluating, which accounted for 31% of their think-aloud protocols. He found that the subjects used elaborating consistently across reading and writing episodes; that rhetorical evaluating appeared more often


9/29/2006

2:23 PM

Page 233

Discourse Synthesis

233

in writing; and that unlike the low-knowledge subjects, the high-knowledge subjects relied less on explicit, text-based information. However, they all made organizational decisions, i.e., rhetorical planning based primarily on the order of ideas in the sources; that is, they hardly used reorganizing or rhetorical transforming strategies. Likewise, Nelson (1992) investigated only the strategies of rhetorical planning and revising. She found a significant correlation (r = .59; p = .005) between her subjects’ elaborateness of rhetorical planning, as evidenced from their written notes, and the quality of their products. However, she did not investigate the rhetorical strategies they used. Other process studies (e.g., Flower, 1987, 1989; Kantz, 1989; Nelson, 1990; Slattery, 1990; Sternglass, 1988) also did not investigate their subjects’ rhetorical strategies, but rather their rhetorical purposes, which, they argued, were determined by the subjects’ task representations. For example, Kantz (1989) attributed the differences among the processes of her subjects to their rhetorical stance, that is, to the role they assumed vis-à-vis their readers. This role determined how they presented the information (e.g., as a summary, an explanation), and to what extent they reproduced or interpreted the sources. Similarly, Nelson (1990) and Sternglass (1988) found that their subjects’ processing strategies were determined by two different task representations — content-driven and issue-driven, or teacher-designed and student self-defined, which resulted, respectively, in low investment strategies (e.g., reproduction) and high investment strategies (e.g., interpretation). The only taxonomy, proposed to date of the strategies used by college students in the process of performing a discourse synthesis task as an authentic course requirement, is Yang’s (2002). Her subjects, six university students in an introductory course on Greek culture were required to write an interpretive essay using multiple sources from Perseus — a hypermedia database. However, her taxonomy includes only three intertextual processing strategies: (1) making connections (i.e., comparing the sources in order to select relevant information); (2) generating relationships (i.e., comparing the sources for similarities and differences and demonstrating or generalizing these relationships); and (3) integrating prior knowledge with source information, which is more frequently referred to as elaborating in the literature on reading and writing (see Ackerman, 1991, p. 232). Little, then, is known about college students’ intertextual processing strategies. 15.1.2. Purpose of the Present Study The purpose of the present study was to extend the research framework in order to gain a more comprehensive view of the cognitive processes underlying the performance of a common discourse synthesis task — the literature review, which has not yet been investigated by discourse synthesis studies, and particularly to investigate more closely the intertextual processing strategies used by college students in an authentic academic context. With only few exceptions (e.g., Nelson, 1990, 1992), the discourse synthesis process has not yet been investigated in such a context but rather under laboratory conditions, which may partly account for the scant evidence of intertextual processing in previous discourse synthesis studies. Moreover, several studies (e.g., Strømsø & Bråten, 2002), although undertaken in such a context, did not investigate discourse synthesis as a reading-writing task, but rather as a reading task involving purposes other than writing (e.g., preparation for exam) which may call for different intertextual processing.


234

9/29/2006

2:23 PM

Page 234

Rachel Segev-Miller

Also, the discourse synthesis process has not yet been investigated in the academic context of a teachers’ college. This seems worthy of investigation in light of the repeated criticism leveled by teachers’ college graduates as well as university education majors at the lack of explicit instruction of reading-writing tasks (Segev-Miller, 1989, 1994, 2004a). A better understanding of the processes underlying the performance of the task may promote the preparation of relevant instruction programs at the college level as well as for students adopting an intertextual approach in their teaching, which, as mentioned earlier, is not prevalent at school. The research question that the present study set out to answer, then, was: What intertextual processing strategies characterize the discourse synthesis processes of teachers’ college students?

15.2.

Methodology

15.2.1.

Subjects

The subjects, 12 college elementary-education majors at a teacher’s college, all female, between 23–25 years old, were selected on the basis of their scores on the ALT — a test of academic literacy composed in light of Sarig and Folman’s (1993) recommendations and validated elsewhere (Segev-Miller, 1990). The test was administered to all 150 elementary-education majors in college at the beginning of their third year of studies. The subjects were then randomly selected from the three distinct proficiency groups that emerged — high (above 80%), medium, and low (below 60%) — four from each group. 15.2.2.

Task

Elementary education majors in different disciplines (e.g., language arts, special education) are required to perform a discourse synthesis of the literature review type for their final research paper in their respective teaching methodology courses. The task in the present study differed from tasks in previous discourse synthesis studies in three major respects: (1) Authenticity of task and context: Unlike the contrived tasks in previous studies, the task in the present study was an authentic course requirement related to their teaching practicum, and the subjects were at liberty to select the topic and the sources themselves. The task was, therefore, expected to engage the subjects and motivate them to invest in its performance. (2) Number and nature of source texts: Unlike previous studies where the subjects were usually assigned two or three short, informative (sometimes contrived) texts, the sources selected by the subjects in this study — articles or book chapters — were numerous, longer, and conceptually and rhetorically more complex. (3) Time on task: Unlike previous studies where the performance of the task took place over a brief period (usually 1–2 sessions under laboratory conditions), performance in the present study took place over two academic semesters (an average of 111.25 days for all the subjects together), starting with the subjects selecting a topic for their research and seeking relevant information in the library, and ending with the subjects submitting their final products to their teaching methodology instructors for feedback.


9/29/2006

2:23 PM

Page 235

Discourse Synthesis 15.2.3.

235

Instruments

Data were collected using a variety of verbal reporting techniques, both retrospective (process logs and interviews) and concurrent (think-aloud protocols), in order to capture the following different aspects of the performance: 15.2.3.1. Process log With the advent of the Writing-Across-the-Curriculum movement in the mid 70s and the Writing-to-Learn movement in the 80s, the course journal has become a standard component of writing instruction programs at many US colleges (for a review see Segev-Miller, 2005). However, very few discourse synthesis studies used it as a research instrument requiring their subjects to retrospectively report the strategies they used in the process (Nelson, 1992; Segev-Miller, 1997; Sternglass, 1988), or the time they devoted in performing the task (Greene, 1993). The two major advantages of the process log over thinking-aloud are that (1) it allows the subjects to report without interfering with their performance of the task, and thus make them free from the cognitive load involved in thinking-aloud, and (2) it allows the researcher to trace the subjects’ writing processes over a long period of time (Greene & Higgins, 1994). 15.2.3.2. Think-aloud protocols Think-aloud data, that is, verbalized thought processes elicited by the subjects while performing the task (Ericsson & Simon, 1980) were collected from six subjects (two randomly selected from each ALT level) in an initial writing session. 15.2.3.3. Interviews The purpose of the first two or three interviews, which were “informant” or unstructured (Powney & Watts, 1987), was to gain the subjects’ trust and cooperation. The interviews that followed served three major purposes: (1) to clarify data in the process logs, (2) to allow for weekly submission of data to the researcher, and (3) to encourage the subjects to persist in the study. The process logs were systematically compared with the subjects by-products (summaries of the sources, notes, and drafts), and, in the case of the six who also elicited thinkaloud protocols, with the data obtained from their protocols, for the purpose of establishing the reliability of the process log data. A randomly selected 10% of the process logs and think-aloud protocols were also analyzed by another rater who was an expert in academic literacy. 15.2.4. Data Collection Procedures The subjects were individually approached by the researcher and volunteered to participate in the study. The subjects were informed of the purpose of the study and provided with instructions with regard to conducting the process log (e.g., to document on a daily basis thoughts and actions related to the performance of the task, to be candid and elaborate). The subjects were also required to attach all the sources they used and by-products they wrote, and to submit these once a week to the researcher for photocopying. Over 3000 pages (including 769 log pages) were photocopied and documented.


236

9/29/2006

2:23 PM

Page 236

Rachel Segev-Miller

The subjects also met the researcher, individually, once a week for an interview. All interviews were tape-recorded and transcribed. Only the sections directly related to the process logs were marked. Each interview lasted approximately 25 min. About 60 h. of interviews were recorded. The dates for the think-aloud sessions were individually initiated by the six subjects when they decided it was time to start writing the review. They individually met the researcher and were trained in the think-aloud technique following Cohen’s (1987) and Sarig’s (1987) operationalizations of Ericsson and Simon’s (1980) model. The six subjects elicited think-aloud protocols of their first attempts at writing the review. No time limit was set. Each session lasted for approximately 3.5 h. 21 hours of protocols were taperecorded, transcribed, and coded using conventional notations (Swarts, Flower, & Hayes, 1984; Smagorinsky, 1994). 15.2.5.

Data Analysis

Conventional think-aloud protocol analysis makes a distinction between reading and writing operations. Such a distinction was not made in the analysis of the process log data due to the complex nature of reporting. The analysis unit, however, was similarly defined in terms of a “move” (Deegan, 1995), or the use of a strategy, which could be distinguished from the other strategies by its purpose. This purpose was often reflected by the explicit use of verbs: e.g., “I put off”, “I selected”, “I concluded that”. Often a move was accompanied by another move (divided by a double slash), which served as an explanation or reason for it: e.g., “[1] I incorporated the (…) // [2] because I realized that (…)”. Unlike the move in the think-aloud protocol, a move in the process log may often extend over a longer period of time, and would, therefore, be more appropriately conceived of as an “episode” (Greene & Higgins, 1994). The following is an illustration of the analysis of a section from a subject’s process log: January 16: (…)// [70] I was very efficient. It is so much better for me to write in the morning when I’m not tired (…) [evaluating task performance] // [71] I thought that it was not such a good idea to synthesize all this information: [planning task management] // [72] The thing is the focus of the writers on different and varied issues [evaluating sources ] // [73] So I decided to present the views one after another [planning rhetorical transforming] // [74] I noticed the fact that I’ve collected info from very different writers from different domains (…) and it is interesting how this affects their views [evaluating sources] // [75] Moses argues that absorbing a work of art is similar to educational absorbing, to the ability to develop in the process of learning and experience; Lavi-also on the development of the ability to absorb, but he emphasizes the ability to evaluate others’ as well as one’s own works of art; Ma’or–the basic approach and the purpose are similar to the latter; Borowits on the other hand emphasizes the learning environment in the same process [conceptual transforming ] // [76] So I wrote an introduction to the chapter: “… Each of these writers emphasizes a different perspective of art education, etc.” [conceptual transforming] // [77] and then I rewrote


9/29/2006

2:23 PM

Page 237

Discourse Synthesis

237

the summaries [linguistic revising] // [78] I thought it was very interesting to deal not only with technical info as I had to in the previous chapter but also with ideas, and summarizing was most interesting [evaluating self] // [79] because I was “faithful” to the ideas and still able to pass them on [evaluating self] // [80] Then I read what I wrote. I was quite satisfied with its organization [evaluating product text] // Analysis of the process log data followed Strauss and Corbin’s (1990) grounded theory model, often referred to in the literature as the constant comparative method of analysis. According to this model, the data are parsed into discrete analysis units, which are constantly compared for similarities and differences. The data are processed and reprocessed cyclically as more data are collected. The analysis units are further conceptualized (i.e., named) and grouped in categories and subcategories, which emerge from the data. All the subjects but one separated their “reading” (i.e., information seeking and processing) from their “writing” (i.e., synthesizing), as they referred to these processes. Their reading was characterized by the summarizing of one source at a time, with hardly any consideration for the conceptual connections among the sources. Even when they did detect some connection, they did not relate to it beyond their report in the process log, nor was their further reading affected as a result in any significant way. Selecting of information was, then, carried out at an early stage in the performance of the task, and was textual rather than intertextual as required by the task. The sources they used in the writing process were the summaries they had written. The present study focused on data collected from the subjects’ “writing”, or synthesizing, processes.

15.3.

Findings and Discussion

Analysis of the process log data yielded a comprehensive taxonomy of the cognitive and metacognitive strategies employed by the subjects in their performance of the discourse synthesis task investigated (see the Appendix A). The three-part division of the major categories of planning, evaluating, and executing identified are consistent with the research literature on reading and writing (e.g., Pressley & Afflerbach, 1995). The sub-categories under executing, however, are different from those suggested by Spivey (1997), who distinguished among selecting, organizing, and connecting. In the present study, selecting is a sub-category of executing; connecting and organizing are conceived of as foci — conceptual and rhetorical, respectively — of the sub-categories of selecting, transforming, and revising. A third focus, on linguistic processing, was also suggested by the data. Previous discourse synthesis research with college students has indicated minimal, and mostly conceptual, intertextual processing strategies crucial to the performance of the discourse synthesis task. The present study, however, indicated a variety of conceptual, rhetorical, and linguistic intertextual transforming strategies, which represent distinct processes of discourse synthesis. The findings pertaining to these strategies are presented and discussed with illustrative quotes from the subjects’ process logs.


238

9/29/2006

2:23 PM

Page 238

Rachel Segev-Miller

15.3.1.

Transforming Strategies

15.3.1.1.

Conceptual transforming

Conceptual transforming requires (1) intertextual processing (e.g., comparing) of the sources for the purpose of detecting the conceptual connections among them, and (2) formulating these connections by constructing a Macroproposition for the different propositions and macropropositions of the sources. These roughly correspond to Yang’s (2002) second intertextual processing strategy — generating relationships (see p. 233): comparing the sources for similarities and differences and demonstrating or generalizing these relationships. The findings of the present study also indicated two such connections among the sources: (1) The information in one source is similar or identical to information in another source (underlining is not in the original quotes): All — Salomon, Kubabi, Rogers — are very similar [Anna, 16]. Azulai actually repeats Weingarten and Katzenelson with the same words [Goldie, 35]. (2) The information in one source is different from (e.g., presents a different approach or aspect), or even contrary to, information in another source: I summarized many sources for this chapter — “the self-image”. The information very much repeats itself, but different sources emphasize different aspects of the issue. The two sources I summarized today; a book by Open University — “Not alone” — emphasizes the factors affecting changes in the self-image; and the article “Educational and social implications of the selfimage” — the educational implications of a low self-image [Ossie, 122–124]. Levine (1977) expresses an opinion contrary to the one I read before: He believes that the assumption, that imitation and originality are factors affecting the child’s psychological development and not the result of this development, is wrong [Lee, 112–113]. The subjects, who were identified as either successful or unsuccessful synthesizers on the basis of their final product scores, reported difficulties in detecting the differences among the sources. Two of the unsuccessful synthesizers were led by their knowledgetelling task representations at the beginning of their writing processes to “simply” detect information in the sources that was conceptually similar. They did not bother to look for differences among the sources when these contained mainly similar information. This finding is in line with both discourse synthesis (e.g., Slattery, 1990) and psychological (e.g., Schmeck, 1988) studies, which indicated the ease with which subjects detected the similarities among sources of information. Similarly, school learners’ products (e.g., Spivey & King, 1989; Stahl, Hynd, Britton, McNish, & Bosquet, 1996) included mostly similar information that was repeated across the sources, although these also contained information that was different or even contradictory.


9/29/2006

2:23 PM

Page 239

Discourse Synthesis

239

However, some of the subjects displayed a tendency for cognitive flexibility (Spiro, Coulson, Feltovich, & Anderson, 1988), that is, for going beyond the simplistic distinction of “similar or different”, and focusing instead on the complexity of the conceptual connections among the sources: I decided to begin with “the instinct” (…). While writing I decided to introduce the contradiction I found earlier in Yankovska in order to deepen the problemacity of the issue (…). I wanted to show how it is still possible to make connections between what seem to be different theories [Lee, 70–76]. Unlike previous studies, which indicated that even college students rarely used the strategy of constructing a Macroproposition, in the present study both successful and unsuccessful synthesizers used this strategy (albeit with different rates of frequency and success; see Segev-Miller, 2004b). This is illustrated in the following quote: The quote from Gardner p. 44 refers to the ease or difficulty of seeing the painting. That is, art. Her point of departure is that art is a language that one can learn (…). He, Roy Lichtenste in the quote from the Encyclopedia of art, vol. 1, p. 95 – adds, in fact emphasizes, that this language, contrary to what we may have understood from Gardner with regard to its universal components, also reflects tradition and conventions. I believe that what connects them is that understanding this language is a condition for understanding art [Julie, 1–4]. All the subjects reported difficulties in implementing this highly demanding strategy, and some often resorted to cognitively less demanding strategies that they referred to as “survival” or “substitute” strategies. Three such strategies were: (1) Using a proposition that appeared across all sources and imposing it as a Macroproposition on the other sources: I will also take the characteristic common to all the texts, indeed I found it — that the need to play is innate [Eli, 40–41]. With regard to the symptoms I found in all the books the same thing — that LD students (…) usually develop emotional-social disorders as a result of frustrations [Lucy, 19–20]. (2) Using a proposition from one source and imposing it as a Macroproposition on the other sources: I took out all the articles about the importance of playing (…) and I selected one sentence from Akavia I thought would be appropriate for all: “The different opinions with regard to the importance of playing must be understood as complementing one another” [Eli, 16–22].


240

9/29/2006

2:23 PM

Page 240

Rachel Segev-Miller

(3) Using a macroproposition from one source and imposing it as a Macroproposition on the other sources: Here I brought Cohen’s major argument about using literature as a psychological instrument (…). I also attributed it to Regev, Goldberg, and Roth [Ossie, 43–44]. Detecting the conceptual connections among the sources was very often the outcome of “sudden”, “unexpected”, or “surprising” discovery rather than a planned move: Suddenly it was clear to me what the main issue of the texts was and why the classification must be like this and not like that (…). I love these moments. It is like some kind of enlightment, never deliberately planned, never with the intention of thinking about it. It is simply suddenly there in the head [Julie, 54–56]. This finding is in line with McGinley’s (1992) and Sternglass’s (1988) successful synthesizers’ processes, which were determined to a large extent by such discoveries while writing. This finding is also in line with the prevalent conception of writing as a process of discovery (Galbraith, 1992, 1996). 15.3.1.2. Rhetorical Transforming The subjects also used five distinct rhetorical transforming strategies, which they translated to their product text structures. These strategies were evaluated for their cognitive demand by (1) the number of sources used (one or more); (2) the purpose for using the sources (to translate the subjects’ task representation or to cope with difficulties translating it); and (3) the manner of their use as described by the subjects. The rhetorical transforming strategies can thus be located along a continuum (see Figure 1) to represent their increasing cognitive demands: As in the case of conceptual transforming strategies, the more demanding rhetorical transforming strategy — synthesizing — was often substituted by the other, less demanding strategies. The following are the rhetorical transforming strategies: 15.3.1.2.1. Summarizing one source text Summarizing one source text (i.e., using only one source rather than multiple sources as required by the task) was not always a reflection

1

2

3

summarizing

listing

incorporating

one source text

4

5

Decomposing & recomposing synthesizing several source texts

Figure 1: Rhetorical transforming strategies.


9/29/2006

2:23 PM

Page 241

Discourse Synthesis

241

of the subjects’ difficulties in processing and synthesizing the multiple sources, as was the case in discourse synthesis studies of school learners (e.g., Mani, 1996; Mani et al., 1996): sometimes it was the result of evaluating the quality (e.g., authority) of a source text, or the result of misleading instructions, as the following two quotes, respectively, illustrate: I copied several sections from Kidron’s famous article (see what I wrote about it earlier), and when I finished it looked to me like some kind of summary [Eli, 119–120]. I summarized the social conception according to Glanz and did not attempt to synthesize the sources because I was told by my instructor that here I can use one book only [Lucy, 33–34]. Two of the subjects made use of sources that, in fact, were “ready-made” syntheses of several sources, but they did not use the appropriate citation conventions in their products. In their process logs, however, they admitted to having committed plagiarism: I copied from Weingarten that “Researchers argue that, etc.” I added these from his list of sources. My instructor of course will never know [Goldie, 53–55]. 15.3.1.2.2. Listing sources Listing the sources one after the other by author, rarely by chronology — two structures commonly used by college students (Swales, personal communication, February 1992) — is a reproduction or knowledge-telling strategy requiring little intertextual processing or transforming. In this sense it is similar to the common linear summary of a single text, which does not require any significant intratextual processing (see Segev-Miller, 1989). No wonder one of the subjects (Lee, a successful synthesizer) used the metaphor of “a shopping list” to express her frustration with it. I copied from the articles of Ben-Brit, Tal-Shir, Efrat, and Biberman (…) parts that looked to me like a continuation of one another. But in spite of my many efforts I was unable to create a connecting frame for them [Eli, 76–80]. Each book presented different methods of treatment and rehabilitation. I tried but could not make a general summary [i.e., synthesis] of all the methods. It was too difficult. So instead, I summarized the main information from each book at a time [Lucy, 22–25]. Listing has been found to be the most frequently used structure in the previous process discourse synthesis studies (e.g., Ackerman, 1991). McCarthy, Young, and Leinhardt’s (1998) product study, for example, indicated that their subjects — college freshmen in an advanced history course — used listing in 85% of their products. Listing was also the most frequent structure (80%) in 25 research papers randomly selected from the teachers’ college library and analyzed prior to the present study. In the present study, both successful and unsuccessful synthesizers used the listing strategy more frequently than any of the


242

9/29/2006

2:23 PM

Page 242

Rachel Segev-Miller

other rhetorical transforming strategies — 34.7% of the rhetorical transforming strategies used by all the subjects together (cf. incorporating – 21.5%; decomposing and recomposing – 14.7; summarizing – 16.8%; and synthesizing – 12.6%). 15.3.1.2.3. Incorporating sources in one source Incorporating sources in one source involved using one source as a “frame” or “skeleton” (the metaphors used by some subjects), for incorporating, sometimes as a quote, information from other sources. That is, no attempt was made at synthesizing the sources by constructing a new conceptual and rhetorical structure but rather, as in the case of summarizing a single text, by replicating the structure of one source, for example: I decided to use Schmeck — on growth climate — as a frame or skeleton, and to incorporate the other texts in it [Anna, 7–8]. I chose to concentrate on Glantz as a basis and to add other studies and theories by Ericsson, Mahler, etc. [Edith, 74–75]. Often, the information incorporated from the other sources was similar to the original information deliberately deleted from the main source: I made up some sort of a puzzle by using other articles: I read one part of Kidron [i.e., the main text] at a time and connected it to other sources by deleting sentences from Kidron and inserting sentences from the other sources instead. I cited the sources [Eli, 121–124]. A special case was that of a subject who used two sources rather than one as a frame: I decided to base this part on two sources — Cohen and Schartzwald — they’re very similar, and then incorporate reinforcements and reservations from the other five sources: (…). I find this part very difficult [Ossie, 138–140]. 15.3.1.2.4. Decomposing and recomposing sources Unlike listing, decomposing and recomposing sources involved organizing source information “by ideas rather than by authors”, as one subject explains in the following interview: Q: Ossie, in your process log you wrote “I took the Ofek and Goldberg texts and I simply rearranged them differently like I was playing with Lego blocks, breaking up the texts and putting them together again (…). That’s what I did.” How exactly did you do it? A: I made some sort of plan: I more or less organized by ideas rather than by authors. For example, if Ofek spoke about “developing a sense for the aesthetic” and she, that is — Goldberg, she also talked about it (…), then I made a list of these ideas and then in my synthesis I incorporated them.


9/29/2006

2:23 PM

Page 243

Discourse Synthesis

243

Q: How did you incorporate them? A: (…). For example, Goldberg talked about the fact that literature develops the child’s vocabulary. So I later wrote: “Ofek also realized the significance of enriching the child’s vocabulary, and explained how (…). He also realized the significance of the imaginary element” in connection with what she talked about earlier, about the imaginary element. And so I continued, as if (…).The element of the imaginary which appeared at the end of Ofek I put at the beginning to make connections between similar ideas of Ofek and Goldberg [Ossie, interview, June 19]. The strategy of decomposing and recomposing sources, then, represents an attempt to connect propositions from different authors or sources: the subject begins to process one source; when she realizes, however, that it contains more than one proposition, she stops in order to search for similar, or even contradictory, propositions in the other sources, before she returns to her first source, to deal with the next proposition in a similar fashion. The subject in the interview used the metaphor of “playing with Lego blocks” to refer to this strategy. Another subject used the metaphor of an “assembly line”. Decomposing and recomposing is also different from incorporating (the frame). Although both use one source as a reference point, the frame is established by one source from the very beginning and the subject merely fills it in with information from the other sources; whereas, in decomposing and recomposing the subject attempts to create novel connections among the sources, usually two. The surface structures of the resulting product texts of these two strategies may look similar, but they are the products of different underlying processes. The subjects who used decomposing and recomposing sources explained that it helped them to cope with the difficulty of processing sources that contained “many different” propositions, some of which were repeated across the sources. This strategy, then, is an intermediate stage between the strategy of listing sources by author, which does not require intertextual processing, and the next strategy, synthesizing, which does. 15.3.1.2.5. Synthesizing sources Unlike decomposing and recomposing, the strategy of synthesizing sources involved more than two sources, and the subjects quite appropriately used metaphors such as the “lattice”, “fabric”, and “complex weaving” to refer to it. Also, unlike decomposing and recomposing, this strategy depended on prior successful operations of conceptual transforming, which it was expected to translate to an appropriate rhetorical structure (e.g., contrast), for example: I started synthesizing the chapter “What is bibliotherapy?” when surprisingly the sources I summarized – Cohen, Laniado, and Kramer – complemented one another in terms of their contents. Cohen suggested a general definition (…). Laniado listed the advantages of the method (…), and Kramer added to the general definition of bibliotherapy the explanation for using literature (…). So what the three have in common is that the three believe that literature can be used as a therapeutic instrument. What I have to do now is present at the beginning what the three have in common — that


244

9/29/2006

2:23 PM

Page 244

Rachel Segev-Miller the three believe bibliotherapy to be so and so, but there are differences in their focus. And then write about each separately but emphasize how they are different. So I wrote: [Ossie, 107–112].

This strategy was translated to her product text as follows: Cohen (1990), Kramer (1994), and Laniado (1983) all deal with the advantages of “bibliotherapy” (therapeutic literature), especially the raising of suppressed contents from the subconscious to the conscious and the processing of these conceptually-verbally, etc. However, there are differences in the emphasis they put on this or that advantage: Cohen (1990) argues that in the simple sense bibliotherapy means (…), and points out that one of the purposes of bibliotherapy is “Know thyself”; Kramer (1984) explains this in that there is in bibliotherapy a combination of the affective and the cognitive but unlike Cohen she emphasizes that the consequent empathy may be indirect, that is, impersonal. Laniado (1983) reinforces this purpose and adds that the next stage would be solving problems by means of changing attitudes [Ossie, product, p. 22]. Very few discourse synthesis studies have investigated the rhetorical structure of the product text, which they conceived of as affected either by the structure of the sources (e.g., Gradwohl-Nash, 1993), or, more often, by the subjects’ task representations (e.g., Greene, 1993, 1999). That is, a task representation of knowledge-telling would be translated to listing the sources and replicating the original structures of the sources, whereas a task representation of knowledge-transforming would result in constructing a new structure. However, analysis of the processes underlying the products in the present study indicated that the connections between the subjects’ task representations and the structures of their products were not that simple or direct. The task representations of most of the subjects, both successful and unsuccessful synthesizers, can be characterized as knowledge-transforming; the initial knowledge-telling representations of some of the subjects in the reading process evolved over time into one of knowledge-transforming. Although all the subjects reported difficulties in translating their task representations to appropriate rhetorical transforming strategies and structures, their reasons for using one or another strategy were not always determined by these difficulties. For example, in the sample process log analysis (see p. 236), the subject (a successful synthesizer) decides to use the strategy of listing, rather than synthesizing. This is an unsuccessful plan, in spite of her successfully evaluating the sources. On the other hand both the successful and unsuccessful synthesizers often planned to use successful rhetorical transforming strategies, but the difficulties they encountered in executing their plans often resulted in postponing these plans or in substituting them with less demanding ones (and later, especially in the case of the successful synthesizers, in revising them). 15.3.1.3.

Linguistic Transforming

Linguistic transforming in the present study refers to the use of intertextual linguistic devices, rather than to the more frequent textual ones (e.g., copying, paraphrasing; see


9/29/2006

2:23 PM

Page 245

Discourse Synthesis

245

linguistic selecting Appendix A), which have been dealt with extensively in previous discourse synthesis studies. The subjects in the present study used two linguistic devices to connect the sources: speech acts and lexical repetition. 15.3.1.3.1. Speech acts The rhetorical strategy of listing was often accompanied by the use of speech acts. Speech acts were used by experts in previous research (e.g., Higgins, 1992) to translate their intertextual reading to their writing. However, the analysis of the subjects’ processes in the present study indicated that these were used to make up for the lack of significant intertextual processing of the sources, or, to quote one of the subjects, to obtain the “illusion of some deep (i.e., conceptual) connection” among the sources, in line with her task representation that “the product must somehow be connected”. The subjects had a very limited repertoire of relevant speech acts: e.g., “adds”, “agrees with”, “also argues”, and “disagrees”. One of them became aware of this while thinking aloud and writing: He, i.e., the psychologist, also adds. Again “adds”! These connectors! I think I’m repeating myself. OK. Until I find something better. [He also adds] that the game (…). There must be other words I could use. I want to say now that she agrees with the psychologist (…). Which words? That is, (…). Are there other words I could use? [Eli, protocol, 262–266].

15.3.1.3.2. Lexical repetition The rhetorical strategy of incorporating (rarely listing) was often accompanied by the use of lexical repetition. Lexical repetition is a common device of cohesion (Halliday & Hasan, 1976). However, the analysis of the subjects’ processes indicated that some of them used lexical repetition to make up for the lack of intertextual processing of the sources, that is, as in the case of speech acts, to obtain the effect of conceptual transforming and synthesis, for example: Erickson’s first developmental stage ended with “age 2”. So I succeeded in connecting it with “from age 2” in Carroll, who argues that “age 2 is the stage of struggling for independence” and then I came back to Ericsson: “the age of independence is characterized by, etc.” [Ronnie, 16–17]. I was listing the advantages of playing (…). I wrote what Siegel said, and then what Akavia said: Her last sentence ended with “learning” so I was able to get to Ben-Brit who argues that “learning is the major goal of playing” [Eli, 14–15]. Two other subjects used the metaphors of the “chain” and “glue” to refer to this strategy. 15.3.2.

Summary

Previous studies that investigated the processes underlying the performance of a discourse synthesis task by college students indicated few, and mostly conceptual, intertextual processing strategies crucial to the performance of the task (Yang, 2002). These strategies were


246

9/29/2006

2:23 PM

Page 246

Rachel Segev-Miller

also rarely used by college students (Kantz, 1989; Kennedy, 1985; McGinley, 1992). These findings may partly be accounted for in terms of the contextual conditions under which the data were collected — laboratory conditions rather than an authentic academic context. Yang’s (2002) study, too, is based on data collected under similar laboratory conditions, although the discourse synthesis task investigated was an authentic course requirement. The findings of the present study, however, indicated that the subjects used a variety of conceptual, rhetorical, and linguistic intertextual transforming strategies, which were largely not indicated by previous discourse synthesis studies. These findings may be accounted for in terms of the authentic context of the study: the long period over which the performance of the task extended, and the subjects’ motivation, which they often referred to in their process logs, to invest time and cognitive effort in the performance of a task, which they considered to be significant and relevant both to their learning and future teaching. Contrary to previous studies, which argued that the subjects’ unsuccessful use of strategies was the result of a knowledge-telling task representation and consequent low investment in the performance of the task (e.g., Nelson, 1990; Sternglass, 1988; Lavelle, this volume), the unsuccessful intertextual transforming strategies, which the subjects frequently used in the present study, should be understood as default strategies. They are substitutes for the more demanding, or “successful”, strategies required for successful performance. Successful conceptual and rhetorical transforming strategies were often substituted with unsuccessful conceptual or rhetorical as well as unsuccessful linguistic transforming strategies by different subjects. It has been argued (e.g., Spivey & King, 1989; Afflerbach & van Sledright, 2001) that discourse synthesis is a cognitively demanding task especially for school-aged learners. Studies of summarizing (e.g., Brown, Campione, & Day, 1982) also indicated that the ability to transform knowledge is a metacognitively developmental one. Although Mani et al. (1996) and Raphael and Boyd (1991) argued that their young subjects’ knowledge-telling strategies (e.g., listing, copying, paraphrasing) were the result of their knowledge-telling task representations, rather than lack of motivation or strategic knowledge, Mani et al. found that the subjects who were characterized by a knowledge-transforming task representation nonetheless often synthesized the sources using a “cut-and-paste” strategy. In other words, they did lack strategic knowledge relevant to the performance of the task. The findings of the present study are in line with the findings of those discourse synthesis studies that indicated discourse synthesis as a demanding task for college students as well. The metaphors, which the subjects used to refer to their strategies (e.g., shopping list, skeleton), are quite revealing with regard to their awareness of their difficulties and the different strategies they creatively “invented” to cope with those difficulties. The quotes from the subjects’ process logs — “But in spite of my efforts I was unable to” and “I tried but could not” (see p. 241) — explicitly illustrate the subjects’ difficulties in synthesizing the sources. Two successful synthesizers also reported putting off their writing as a result: December 18: Maybe I have to start synthesizing the sources, but I find it very difficult (…). February 1: I obviously have a problem with the task. I’m so stressed, but now that I finished reading and summarizing everything, I must concentrate on the writing [Sam].


9/29/2006

2:23 PM

Page 247

Discourse Synthesis

247

December 19: I’ll start with the more technical parts — those that require less of a mental effort [i.e., listing the variables] (…). The parts requiring synthesis I put off (Julie]. According to the construct of academic expertise (Geisler, 1994), knowledge and experience are crucial factors in the successful performance of different academic reading-writing tasks. Indeed, all the subjects in the present study accounted for their difficulties, as did students in the past (see p. 234), in terms of lack of relevant knowledge (they had not been provided with explicit instruction of the task), and lack of experience (they had not been required to perform a similar task before).

15.4.

Implications

The present study has two major implications: (1) the significance of investigating the discourse synthesis processes underlying the discourse synthesis product in an authentic context over time, in particular the significance of the process log as an instrument to collect data pertaining to these processes, and (2) the need for explicit instruction of the performance of the discourse synthesis task. 15.4.1.

The Process Log

The process log proved to be an invaluable instrument to collect significant process data pertaining to the performance of an authentic discourse synthesis task in the academic context of a teachers’ college. The process log not only allowed the researcher to trace the processes underlying the subjects’ performance of the task over a long period of time, but also to gain insights into these processes otherwise unobtainable. As a rule, students in the academy are assessed on the basis of their written products. A more valid assessment in line with current trends of alternative assessment (Birenbaum & Dochy, 1996) should take into account process measures as well, such as their use of strategies. Such an assessment could be carried out by means of a process log similar to the one used by the subjects in the present study. Unlike the portfolio, which consists of the student’s best work and serves to assess the student (Smagorinsky, 1997), the process log in the present study served as a means for the subjects to document the cognitive processes underlying their performance of the discourse synthesis task. The process log can, however, be used for instructional purposes as well. It would facilitate the instruction of the task by providing the instructor with data relating to the students’ difficulties where intervention was needed. The process log, if properly used, may also promote the students’ awareness of their own processes (see Segev-Miller, 2005). However, the discourse synthesis tasks that students are required to perform must be authentic in order to engage the students as was the case in the present study. 15.4.2.

Explicit Instruction

The subjects in the present study, both successful and unsuccessful synthesizers, were motivated to invest time and cognitive effort in the performance of the task. However, they


248

9/29/2006

2:23 PM

Page 248

Rachel Segev-Miller

all encountered difficulties, which they attributed to lack of relevant knowledge and experience. Intervention studies of discourse synthesis (for a review see Segev-Miller, 2004a, 2004c) indicated the significant effect of explicit instruction of the task on either process or product quality. Since all knowledge and learning are grounded in context (Brown, Collins, & Duguid, 1989), explicit instruction must take place in an authentic context. The failures of present introductory writing courses may be accounted for in terms of the lack of such a context. With regard to the performance of the discourse synthesis task in the context of the teachers’ college, in which the present study was carried out, this requirement would mean that explicit instruction of the performance of the task should take place in the students’ respective disciplinary courses (e.g., language arts, special education). This would also be in line with the tenet of Writing-Across-the-Curriculum movement that all instructors in all disciplines are equally responsible for the instruction of writing-to-learn. Disciplinary courses that provide no explicit writing instruction (e.g., McCarthy et al., 1998), or very little explicit writing instruction (e.g., Greene, 2001) with an emphasis on the product rather than on the process, have not been very helpful. However, since not all the students come to college with the basic abilities of academic reading and writing, and since the instruction of these abilities in content-loaded and overcrowded courses is almost impossible, an introductory course in academic reading and writing should be offered to these students, with an emphasis on summarizing and discourse synthesis strategies in the students’ common discipline (teacher education). The students will then practice these strategies further in their respective disciplinary courses when required to perform relevant discourse synthesis tasks. Sternglass’s (1993) suggestion that “supportive” guidance throughout the college years be offered to students whose reading and writing processes need to develop, should also be adopted. This guidance could be provided by the college writing center, where the students will be individually assisted in meeting their course requirements.

Appendix A: A Taxonomy of Discourse Synthesis Strategies 1. 1.1. 1.1.1. 1.1.2. 1.1.3. 1.2. 2. 2.1. 2.1.1. 2.1.1.1. 2.1.1.2. 2.1.1.3. 2.1.1.4. 2.1.1.4.1.

Planning (Task Management) Planning use of strategy Selecting strategy Putting off use of strategy Abstaining from using strategy Planning work schedule, time on task, incubation Evaluating Evaluating source texts Conceptual evaluating Relevance Significance Comprehensibility Intertextuality Presentation of similar information


9/29/2006

2:23 PM

Page 249

Discourse Synthesis 2.1.1.4.2. 2.1.2. 2.1.2.1. 2.1.2.2. 2.1.2.3. 2.1.2.4. 2.1.3. 2.1.4. 2.1.4.1. 2.1.4.2. 2.1.4.3. 2.1.4.4. 2.2. 2.2.1. 2.2.1.1. 2.2.1.2. 2.2.1.2.1. 2.2.1.2.2. 2.2.1.2.3. 2.2.1.2.4. 2.2.2. 2.2.2.1. 2.2.2.2. 2.2.2.3. 2.2.3. 2.3. 2.3.1. 2.3.2. 2.4. 2.4.1. 2.4.2. 2.4.3. 2.4.4. 2.4.5. 2.5. 2.5.1. 2.5.2. 3. 3.1. 3.1.1. 3.1.2. 3.1.2.1. 3.1.2.2. 3.1.3. 3.1.3.1.

Presentation of diverse aspects Rhetorical evaluating Level of informativeness Format Rhetorical selecting Rhetorical function Linguistic evaluating Evaluating quality of source texts Authority Updateness Professionalism Primary or secondary source Evaluating product text Conceptual evaluating Relevance Conceptual organization Intertextuality of source texts included in product Presentation of diverse aspects Accuracy Detection of contradiction Rhetorical evaluating Evaluating rhetorical organization Evaluating level of informativeness Evaluating rhetorical function Linguistic evaluating Evaluating task (“task representation”) Knowledge-telling Knowledge-transforming Evaluating oneself Ability to perform task (self efficacy) Motivation Interest or pleasure in performance Connection to practicum Comprehension of source texts Evaluating task management Evaluating performance Evaluating progress Executing Selecting Conceptual selecting Rhetorical selecting To introduce or summarize To elaborate Linguistic selecting Copying

249


250

9/29/2006

2:23 PM

Page 250

Rachel Segev-Miller

3.1.3.2. 3.1.3.2.1. 3.1.3.2.2. 3.1.3.3. 3.1.3.4. 3.2. 3.2.1. 3.2.1.1. 3.2.1.2. 3.2.1.2.1. 3.2.1.2.2. 3.2.1.2.2.1. 3.2.1.2.2.2. 3.2.1.2.2.3. 3.2.1.3. 3.2.1.3.1. 3.2.1.3.2. 3.2.2. 3.2.2.1. 3.2.2.2. 3.2.2.3. 3.2.2.4. 3.2.2.5. 3.2.3. 3.2.3.1. 3.2.3.2. 3.3. 3.3.1. 3.3.1.1. 3.3.1.2. 3.3.2. 3.3.2.1. 3.3.2.2. 3.3.2.3. 3.3.3.

Quoting To support an argument To substitute own text Paraphrasing Summarizing Transforming Conceptual transforming Detecting the conceptual connection among source texts Formulating the conceptual connection Creating (“inventing”) a Macroproposition Using an existing proposition Detecting the macroproposition in one text and imposing it as a Macroproposition on other texts Selecting a proposition that is not a macropoposition in one text and imposing it as a Macroproposition on other texts Selecting a proposition that is not a macropoposition but appears across all texts and imposing it as a Macroproposition on all texts Categorizing Creating categories previously non-existent in source texts Collapsing existing categories Rhetorical transforming Summarizing one source text Incorporating source texts in one source Listing source texts Decomposing and recomposing source texts Synthesizing source texts Linguistic transforming Speech acts Lexical repetition Revising Conceptual revising Revising conceptual selecting Revising conceptual transforming Rhetorical revising Revising rhetorical transforming Revising level of informativeness Revising rhetorical selecting Linguistic revising


9/27/2006

3:32 PM

Page 251

Chapter 16

Preformulation in Press Releases: What the Writing Process Tells Us about Product Characteristics Kim Sleurs

Existing product research revealed that press releases are preformulated in such a way that they allow for easier copying in newspapers or magazines. Features of preformulation are, for example, headlines, the use of a lead paragraph and the insertion of (pseudo-)quotations. In this paper we discuss a single press release written by a PR professional. The writing process was recorded and the writer thought aloud throughout; afterwards the writer also produced a retrospective protocol through stimulated recall. Pause analysis revealed that two elements of the press release were especially difficult to write: quotations and headlines. We focused on the latter and turned to the concurrent and retrospective protocols for more information. From the protocols it became clear that, although preformulation is certainly an issue in the writing of press releases as a whole, the function of headlines is different. Rather than being meant for copying, headlines are aimed at attracting the journalist’s attention and pleasing the client.

16.1.

Introduction

This paper fits in the context of the study of communication in and around the workplace, more specifically, the way in which organisations interact with the media. Most companies and organisations have long since started to realise that media communication needs to be proactive, rather than reactive. That is, it is no longer sufficient to be available to the press in times of crisis. Rather, it is in a company’s best interest to send out information to the press on a regular basis as press coverage can provide free and indirect (and therefore often considered to be more objective) publicity.


Sleurs, K. (2007). Preformulation in press releases: What the writing process tells us about product characteristics. In G. Rijlaarsdam (Series Ed.), and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in Writing, Vol. 20, pp. 251–263). Amsterdam: Elsevier.


252

9/27/2006

3:32 PM

Page 252

Kim Sleurs

One of the most routinised ways for businesses to try and gain access to the media is the use of press releases. The underlying aim is to see their views and words reproduced in news reporting. Press releases are meant to be retold as accurately as possible, preferably even verbatim. To allow for easier copying, press releases are “already prefabricated in an appropriate news style and therefore require the minimum of reworking” (Bell, 1991). These so-called preformulating devices include the use of headlines followed by a lead paragraph as well as a number of special metapragmatic features such as third-person self-reference (e.g. ‘Siemens’ and ‘the company’ instead of ‘we’) and (pseudo-) quotations (e.g. ‘The CEO, Mr. X, added: “we are delighted that our product Y can contribute to a better and more efficient means of doing Z”) (see Verschueren, 1995, for a comprehensive definition of metapragmatics). The question we address in our research is to what extent preformulating concerns play a role in the actual process of writing, or rather constructing, press releases. We take a diachronic view of the news with an in-depth look at the complex writing and rewriting process of press releases.

16.2. 16.2.1.

Methodology Data

In this paper we present a case study that is part of a larger research project aimed at offering a process view of preformulation in press releases. In order to gain insight into the full context and dynamics in which press releases are written, we chose to apply a combination of methods in a variety of settings. We conducted ethnographic and cognitivepsychological writing research in both a large bank with its own in-house public relations (PR) department and in two small-scale PR agencies. In this paper we report on the research that we carried out with a professional writer of press releases working for one of the PR agencies. The agency is specialised in providing PR services for the information technology (IT) and telecommunications industries. It employs 13 people, 10 of whom are account managers. Each account manager has a portfolio of clients and is responsible for, among other things, online news services, the organisation of workshops, press relations, launching special events, media training and press release and speaker programmes. At the time of our research our writer, whom we call Mark (26 years old), had been with the company for 14 months. Before that he worked as research assistant at a university for a little over a year. He has a degree in Political and Social Sciences and did an MBA. The research that we report on here is based on extensive fieldwork that we did in the fall of 2001. We used some of the methods of ethnography (including interviews and direct observation) and combined this approach with methods from cognitive-psychological writing research. This should allow for a close, micro-level reconstruction, analysis and interpretation of PR routines (Rijlaarsdam et al., 1996; Levy & Ransdell, 1996); in particular, we set out to combine concurrent and retrospective protocols (Smagorinsky, 1994) with online registration of the total writing process (Eklundh and Kollberg, 1996; van Waes, 2000; Van Waes & Schellens, 2003). In this paper, we focus on the results from our cognitive-psychological writing research.


9/27/2006

3:32 PM

Page 253

Writing Press Releases 16.2.2.

253

Combined Methodology

To gain more insight into the details of the actual writing process, we made a recording of the writing process for a press release, using an online camera (CamtasiaTM). The entire writing process took 92 min. This recording enabled us to reconstruct the writing process accurately, letter by letter and revision by revision. While he was writing, we asked Mark to produce a concurrent think-aloud protocol, which involved verbalising whatever thoughts went through his head during the writing process. Immediately afterwards, we asked Mark to produce a retrospective protocol to offer more explanation on certain choices he had made in the process of writing the press release. Here the online registration was used to help him reflect on a higher level, providing comments on selected strategic choices in the writing process. This method in which the writer is prompted by a visual or oral stimulus to recall what he or she thought while writing, is known as stimulated recall (Gass & Mackey, 2000). It is important to remark at this point that our methodology can of course not be used to explore the caverns of Mark’s mind or uncover his deepest thoughts. Rather, the concurrent protocol offers a good starting point for an analysis of the writing process because it provides information that is readily available to Mark while he is working. Some things will of course be left unsaid, because they are either too obvious to be mentioned or because they are too complicated to reflect upon while working, i.e. reflection would distract too much from the writing process. In both cases, the retrospective protocol offers a way into some of the missing information. On the one hand, it helps to make explicit certain choices that are the result of automatisms. On the other hand, Mark’s attention is drawn to decisions that are made on the basis of intricate underlying thought processes, which are too complex to be elaborated upon in the concurrent protocol. In this sense, the retrospective protocol can be considered complementary and more in-depth than the concurrent protocol.

16.3.

Research Focus

To discuss the extent to which preformulation is apparent and plays a real role in the writing process of press releases, we started from product research (Jacobs, 1999) that lists features of preformulation such as third-person self-reference, headlines, the use of a lead paragraph and quotations. Rather than choosing one of these elements randomly for discussion in our paper, we turned to pause analysis as a means of determining which features were most interesting for discussion here. In our analysis of the technical logging we looked at the development of the writing process and at pauses. At this point we did not yet look at Mark’s comments in the concurrent protocol, but focused on the pauses only. In our interpretation of these pauses it is important to bear in mind that the production of protocols slows down the writing process. For example, pause length is seriously influenced by the need to verbalise thoughts in simultaneous reflection, especially at complex cognitive junctions in the text (Janssen & van Waes, 1996). To avoid distortions our analysis of the technical logging focuses on process-internal comparisons, i.e., we did not compare Mark’s writing of this particular press release with other writing


254

9/27/2006

3:32 PM

Page 254

Kim Sleurs

processes, but made comparisons only between different parts of this one writing process. In this particular case, the production of the protocol proved to be a constant factor in the process: i.e., Mark produced a very steady concurrent protocol. Average pause length over the total writing process is 12 s. We needed to check whether fatigue was a factor, i.e., we wanted to know whether Mark got tired towards the end of the writing process, especially given the mentally strenuous task of having to produce a concurrent protocol in addition to the writing itself. Therefore we divided the writing process in two halves and looked at average pause length. We found there was no significant difference in average pause length between the first and second part of the writing process (ANOVA: F (1, 249) ⫽ 0.084, p ⫽ 0.773 ⬎ 0.05). There was also an equal distribution of the number of pauses: 46.4% in the first part; 53.6% in the second part (n ⫽ 250). Long pauses of more than one minute, which typically occur in parts of the writing process featuring high cognitive complexity (Schilperoord, 1995), can clearly be traced back to two places; to the writing of the quotations and the phrasing and revision of the title and subtitle. This demonstrates that the writing of both the quotations and the headlines involves a serious cognitive load. In this paper we discuss the role of headlines in more detail (for a discussion of the role of quotations, see Sleurs, Jacobs, & van Waes, 2003). Specifically, in this paper we explore what information our combined methodology can offer on the design and function of headlines.

16.4. 16.4.1.

Analysis Headlines: General Description

First, let us take a broad look at the process of writing the title. This consists of four stages. First, the title and subtitle are constructed. This construction is followed by three revision rounds (see Appendix for an overview of the title from the first to the final version). The first revision round is simply the addition of the product name to the subtitle. It starts after Mark has written the title and subtitle and has started on the first paragraph. Having written the first line of the first paragraph (this is usually not difficult in press releases as the first line has a fairly fixed format, i.e. Company X, market leader in Y announces Z), he starts to write the second line: ‘The agreement’. He then deletes these two words and says: [CON 09.48]1 I’m just thinking what really is the essence of the story. So the essence of the story is that erm (..)2 |Media company| erm (..) will be using the product |Product X| (.) to to send newsletters (moves cursor to end of |Magazine 2| (.) erm which also means that I erm my client will probably appreciate this that I also mention the product in title or subtitle erm weekly internet so now I’m just going to (.) in the subtitle weekly internet newsletters |Magazine 1| and |Magazine 2| (..) erm (.) make sure it doesn’t sound too commercial either title and subtitle (.) erm (..) yes I’m just going to keep it very basic for a bit sent with |Product X| (types). 1 We use the abbreviation “CON” to refer to excerpts taken from the concurrent protocol and “RETRO” to refer to excerpts taken from the retrospective protocol. “TL” refers to the technical logging. 2 (.) indicates a very short pause (not timed, less than 1 s); (..) indicates a longer pause.


9/27/2006

3:32 PM

Page 255

Writing Press Releases

255

The second revision round is mainly content revision and involves the correction of practical information (numbers and timing). This revision round is triggered by Mark’s trouble in writing up a quotation. As mentioned earlier we found longer pauses in two parts of the writing process, i.e. in the writing up of the headlines and in the writing up of quotations. Here, Mark has almost finished the press release and all he needs to do is think of an appropriate quotation for the CEO of the software company. First he looks at one of the company’s earlier press releases for inspiration. Then he thinks about something original to say. He says: [CON 53.17] I’m also going to what is also very important what receives a lot of attention too is the aspect that the mails and newsletters are personalised (.) so that you get them as (.) mister [last name Mark] (moves cursor to beginning of the document) erm (.) the question of course is where I can put this best (while he is speaking, Mark is scrolling through the document and he reaches the title at 53.42) in the (moves cursor through title) now I’m just skipping from one subject to another now I’m just looking from the beginning to the end how the message now already looks as a whole it will probably still look very crude (.) but perhaps I can already start to streamline it a little bit now (goes on to read and revise the title). Finally, revision round three is mainly strategic revision and shows a clear cognitive knot. After changing the words ‘takes care of sending’ to ‘secures smooth sending’, Mark moves his cursor to the subtitle in front of the words ‘on time’ (TL 01.21.33). After 31 s Mark puts the cursor in front of the word ‘time’ (TL 01.22.04). Then he seems distracted and makes a revision elsewhere, placing the word ‘personalised’ in front of the word ‘stock-updates’ (TL 01.22.10), only to return his cursor to its position in front of ‘on time’ a few seconds later (TL 01.22.30). After a 35-s pause Mark finally takes action and changes the subtitle: ‘to subscribers on time’ becomes ‘brought to subscribers with’ (TL 01.23.05). Thirty seconds later Mark moves his cursor away from the subtitle into the first paragraph. What is interesting to note at this point is the very strong emphasis on Mark’s part of the preliminary nature of the headlines when he first starts writing them. He says, for example: [CON 00.01.20] ‘a preliminary working title for now’ and: [CON 00.04.58] ‘with as a consequence of course that the title is too long now but it will have to do for now’ Later on, Mark says: [CON 00.06.29]: ‘these titles will probably have to be slightly edited later on erm (pause) for now I’ll start with ‘weekly…’ (starts to write)’ The reason for this emphasis on the headlines’ preliminary nature is linked to the question of what a title should look like.


256

9/27/2006

3:32 PM

Page 256

Kim Sleurs

16.4.2.

Design: What Should Headlines Look Like?

16.4.2.1. Title ⫽ essence From Mark’s comments in both the concurrent and retrospective protocols, it is clear that the title needs to tell the essence of the story. In the concurrent protocol, for example, Mark literally says: [CON 01.20.21]: ‘the essence should really be in the title’ In the retrospective protocol, this is elaborated upon in the following statement: [RETRO]: ‘No I had the impression that there was really still vital information that should have been in the title or subtitle in one shape or form (.), I didn’t think the full information was actually in there’. As was clear from our description above, this emphasis on the headline as a carrier of the most essential information is also apparent in the triggers for revisions. While writing the press release, Mark comes across information that he thinks is important enough to be in the headline, thus triggering a revision. He is clearly aware of the fact that the writing up of the full story may unfold new elements that deserve inclusion in the title, hence his emphasis on its preliminary character. The following statement from the concurrent protocol confirms this: [CON 00.00.49]: ‘So now I am trying to think a little what the essence of the story is to turn it into a title erm it is fairly likely that this will be a preliminary title because as the story progresses erm I will probably have to edit a bit here and there’. This observation that a writer often discovers new elements to discuss in the course of writing his first draft, is in fact often made in writing research. 16.4.2.2. Title ⫽ objective? Another criterion for the design of a ‘good’ title seems to be its objectivity. Objectivity is mentioned on more than one occasion in both the retrospective and the concurrent protocols as a requirement for press releases in general. We also see some indications from our protocol data that this requirement applies equally to titles and subtitles. For example, in the concurrent protocol, we hear Mark say: [CON 00.10.30]: ‘have to make sure it doesn’t sound too commercial either, title and subtitle’ And in the retrospective protocol, he says: [RETRO]: ‘the language, precisely because we are in a B-to-B environment, shouldn’t be too jubilant, too hurrah’ Despite these indications that objectivity is one of the elements that determine the design of a title, we see Mark make an interesting revision in revision round 3. The title originally read: |Software company| takes care of sending of newsletters and stock information to 150.000 on-line subscribers to |Media company|


9/27/2006

3:32 PM

Page 257


257

Mark then changes it to: |Software company secures smooth sending of newsletters and stock information to 150.000 on-line subscribers to |Media company| Clearly, this revision makes the title more ‘flashy’ and this seems to go against preformulating concerns. Surely a journalist would never copy a title that is ‘flashy’, ‘commercial’ and ‘subjective’. Let us have a look at what Mark has to say about this: [RETRO]: ‘‘secures’ is a much stronger word […] ‘takes care of’ is a weaker term than ‘secures’ (.) in the title I thought we could use a strong term (.) ‘smooth’ yes I mean the fact that it is an alliteration certainly plays a part here (.) I couldn’t have simply inserted any other old adjective I feel but this offers an extra in the sense that actually further down in the press release I think it becomes clear why it has to be ‘smooth (.) because it is about investors for whom a day an hour a minute can make a difference’. So why does Mark say it is okay to use a ‘strong term’ in the title, but not in the rest of the press release (quotations are another exception, see Sleurs et al., 2003)? To make sense of this dilemma, we once again turn to the protocols for information, in this case on the function of titles in press releases. 16.4.3.

Function: What Purpose Do Headlines Serve?

As headlines are considered to be a feature of preformulation (Jacobs, 1999), it makes sense to assume they are meant to be copied. However, is this indeed the case? In fact, Bernaers, Jacobs, and van Waes (1996) looked at a corpus of press releases and checked whether their titles were copied in newspaper articles. The research could not find any real evidence to support the hypothesis that titles were meant to be copied. So what did we find in our (process) research? Interestingly, we found absolutely no mention either in the concurrent or the retrospective protocol of titles being meant for copying. Nor did we find any mention of the target audience, i.e. the readers of the newspapers or magazines this press release would be sent to. So if titles are not meant to be copied and thus not necessarily meant to be read by the end-reader, who are they for? And more importantly, what purpose do they serve? 16.4.3.1. Getting the journalist’s attention Looking in more detail at the retrospective protocol, the first function of the title and subtitle seems to be attracting the journalist’s attention: [RETRO]: ‘What I also did here (.) in concrete terms name |Magazine 1| and |Magazine 2| because it gives more information on what the title is about first of all but secondly also because when the journalist reads it (.) even if he doesn’t get it all from the title the main title there is still a serious chance that for example the journalist is subscribed to |Magazine 1| or |Magazine 2| (.) if he sees that in the subtitle his attention will be focused more easily’.


258

9/27/2006

3:32 PM

Page 258

Kim Sleurs

It is interesting to see that Mark does not say anything about the journalist copying the title or subtitle of his press release. He wants the journalist’s attention to be ‘focused more easily’. We see the same motivation in the next excerpt from the retrospective protocol. Again Mark talks about words in the title ‘standing out’ or being ‘highlighters’, but makes no mention of the headlines being suitable for copying. [RETRO]: ‘I also had my doubts a bit (.) if I just (.) I don’t really doubt it but I’d just rather not take the risk if it just says ‘online subscribers to |the Media company| or |of Media company| whether that is enough to stand out (.) if it says Generale Bank before or Dexia or another strong brand name Coca Cola so to speak then that immediately stands out (.) |Media company| is a short word it was at the end of the line (.) I’d rather not take the risk that perhaps they don’t see it at a glance so that’s why I put in a few erm yeah (.) highlighters by putting the name |Magazine 1| |Magazine 2| in the subtitle’.

16.4.3.2. Pleasing the client and the client’s client Another concern for Mark in the writing up of the headlines seems to be whether or not his client will approve of what he has written. After all, the client is the one who orders the press release in the first place and thus future business for the PR agency depends on him. So it is only natural for Mark to have his client first and foremost on his mind: [CON 00.10.12]: ‘which also means that I (.) my client will probably appreciate this that I also mention the product in the title or or subtitle’. Mark expresses the same concern in more detail in the retrospective protocol: [RETRO]: ‘important that I put the product |Product X| in the subtitle because it is a further link directly to the company (.) what I write in the title namely that [Software company] will be working for [Media company] is kind of thinly rehashed in the subtitle but in more concrete terms (.) “|Product X|, of [Software company] will be handling the sending of |Magazine 1| and |Magazine 2|” (.) so then it actually gets more personal it says more in-depth what the story really is and that is also more interesting for the client I think (.) that there is a clear (.) that the journalist clearly sees look an important magazine (.) like |Magazine 1| and |Magazine 2| is linked to this product of the client which also instantly boosts the client’s credibility’. Of course, Mark is in a very difficult position here as he not only has to convince the journalist to take a closer look at his press release, but at the same time he also has to keep his own client satisfied. The fact that the journalist’s and the client’s expectations are more often than not at odds with each other makes this a difficult task indeed. Things get even more complicated because interestingly, Mark’s concern is not only with his own client, but also with his client’s client. And this is where we return to the cognitive knot in revision round 3.


9/27/2006

3:32 PM

Page 259


259

Considering the fact that average pause length over the writing process was 12 s, we clearly see a number of longer pauses in this part of the writing process. Some of the pauses actually reflect an even longer process of thought as the cursor moves but no changes are made initially. So what problem is Mark dealing with here? In order to find out more about this particular cognitive knot, we need to turn to the concurrent and retrospective protocols. The concurrent protocol clearly reveals that Mark is worried he might offend his client’s client by writing that his own client now manages to bring information to the media company’s subscribers ‘on time’ (thus possibly implying the media company did not manage to do so before): [CON 01.21.14]: ‘Perhaps not a clever move either because that would that implies more or less that those poor sods at |Media company| before hardly ever succeeded in time’ and: [CON 01.22.30]: (sighs) ‘I do worry a bit that when I say that erm (.) that now this is done on time or that just that it happens on time or smoothly or efficiently that this sh would show that before at |Media company| that actually wasn’t the case (.) and since I actually know nothing about that but can assume that they wouldn’t like that erm (.) so I’ll just write ‘brought to subscribers by, no, with |Product X|’ When asked for more information about this particular episode in the retrospective protocol, Mark stated: [RETRO]: ‘That then I made it look as if before the poor sods at [Media company] couldn’t send it out “on time” themselves’. In other words, even though the fact that information from the media company now reaches their subscribers on time is clearly a sales argument for his own client, Mark decides it is not worth the risk of offending his client’s client. Clearly, strategic considerations weigh more heavily on the writing process here than do concerns of preformulation. Interestingly, we also found this to be the case in our research in the in-house PR department of a large bank.

16.5. 16.5.1.

Conclusion Design and Function of Headlines

Despite the fact that titles are traditionally considered to be a feature of preformulation, in the sense that press releases mimic newspaper articles to allow for easier copying, it seems that preformulation is not on Mark’s mind when he is writing them. He is mainly concerned with a title that offers the journalist the essence of the story. Based on the title, the


260

9/27/2006

3:32 PM

Page 260

Kim Sleurs

journalist will decide whether or not the rest of the press release is worth reading. Of course, this means that the title needs to be an eye-catcher first and foremost. Objective and business-like phrasing is less important in the design of titles as they are not usually copied anyway. In addition, we found that many more parties are present in the background when a press release is being written. Mark does not only have to persuade the journalist but he also needs the approval of his client and his client’s client. Of course, we can wonder here whether this situation is perhaps not typical of a PR consultant working for an independent PR agency. Possibly, a press officer working in a company’s in-house PR department need not be as concerned with client relations as does our PR consultant. However, some early research we did in precisely such an in-house PR department revealed that a press officer faces similar concerns. He needs to impress not a client, but his own employer. Moreover, especially in larger companies, a company’s separate departments may almost be looked upon as clients themselves, with their own demands, expectations and sensitivities. So here too we found that concern for the reactions of partners or other departments at the bank often overrode preformulating concerns. This finding is of course a strong argument for studying press releases in the context in which they are written. 16.5.2.

Methodology

In choosing to combine pause analysis with both concurrent and retrospective protocols, we opted for a fairly complex methodology. However, we feel the data gathered from these three methods went a long way toward making up for each other’s disadvantages and led to a fuller understanding of our author’s writing process. First, pause analysis offered an objective way of determining where and when our author experienced difficulties. As mentioned before, longer pauses indicate more cognitive activity and therefore groups of longer pauses reflect cognitive knots. We found these knots to coincide with the writing up of two elements of the press release, i.e., quotations and headlines. Although pause analysis offered some insight into where exactly problems could be located, it went no further in explaining where those difficulties came from. In this respect, the information we gathered from the concurrent protocol, was certainly complementary. Here, Mark indicated why he was struggling and what kinds of considerations went through his mind in the course of his writing process. While the advantage of a concurrent protocol is its immediacy, comments are usually limited to lower-level processes (e.g. spelling) and when higher-level processes are verbalised (e.g. a client demands), not much detail is offered as this would interrupt the writing process. Here, the retrospective protocol can offer more information. Statements from the concurrent protocol could be commented upon, i.e. we often heard statements such as “my client will be happy to hear/read that”, but could only guess as to the reason why. This reason was then usually offered in the retrospective protocol. To make sure the writer did not have to rely on memory too much, we used stimulated recall, showing the writer parts of his writing process and then asking him to comment. The added advantage of this method was that we ran no risk of leading questions. We simply showed our writer a particular episode from the technical logging and asked him to comment.


9/27/2006

3:32 PM

Page 261


261

For our purposes, it is also interesting to compare what was said on, for example, clientconsultant relationships in the interview we conducted beforehand (and which was not discussed here), and what was said in the retrospective protocol about that relationship, starting from concrete examples in the writing process. However, the analysis of our data from interview, split-run and direct observation were beyond the scope of the chapter at hand. Of course, the retrospective protocol will always, to some extent, be a reconstruction; the writer tries to explain why he did and said certain things and is sometimes actually surprised when he is confronted after the writing process with his own words (CON) or actions (TL). Precisely because the retrospective protocol is a reconstruction, it is important that we continually go back to the technical logging and the concurrent protocol to check for inconsistencies, as these can be very revealing. Our combined methodology does of course mean we are faced with a large number of data and this can be quite overwhelming. That is why it was important to start from the framework on preformulation that had been provided by product research and that readily lent itself to process research.

Acknowledgement We thank Professor Dr. L. van Waes for his help on the pause analysis.

Appendix: Construction

Version 1

|software company| takes care of sending of newsletters and stock information to 1 million on line subscribers to |the media company| Version 2 – 03.14 |software company| takes care of sending of business news to 1 million on line subscribers to |the media company| Version 3 – 04.57 |software company| takes care of sending of business news to 1 million on line subscribers to |official name media company| Version 4 – 06.35 |software company| takes care of sending of business news to 1 million on line subscribers to |official name media company| Weekly Internet newsletters |magazine 1 and magazine 2| Revision round 1 Version 5 – 10.50 |software company| takes care of sending of newsletters and stock information to 1 million on line subscribers to |official name media company|


262

9/27/2006

3:32 PM

Page 262

Kim Sleurs

Weekly Internet newsletters |magazine 1 and magazine 2| sent with |product X| Revision round 2 Version 6 – 54.58 |software company| takes care of sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Weekly Internet newsletters |magazine 1 and magazine 2| sent with |product X| Version 7 - 55.24 |software company| takes care of sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Internet newsletters |magazine 1 and magazine 2| sent weekly with |product X| Version 8 – 55.40 |software company| takes care of sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Stock updates and internet newsletters |magazine 1 and magazine 2| sent weekly with |product X| Version 9 – 56.23 |software company| takes care of sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Stock-updates and internet newsletters |magazine 1 and magazine 2| sent weekly with |product X| Version 10 – 56.49 |software company| takes care of sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Stock-updates and internet newsletters |magazine 1 and magazine 2| reach subscribers on time via |product X| Revision round 3

Version 11 – 01.19.24 |software company| takes care of sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Stock-updates and internet newsletters |magazine 1 and magazine 2| reach subscribers on time via |product X|


9/27/2006

3:32 PM

Page 263


263

Version 12 – 01.20.53 |software company| secures smooth sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Stock-updates and internet newsletters |magazine 1 and magazine 2| reach subscribers on time via |product X| Version 13 – 01.22.18 |software company| secures smooth sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Personalised stock-updates and internet newsletters |magazine 1 and magazine 2| reach subscribers on time via |product X| Version 14 – 01.23.05 |software company| secures smooth sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Personalised stock-updates and internet newsletters |magazine 1 and magazine 2| brought to subscribers by |product X| Version 15 – 01.23.23 |software company| secures smooth sending of newsletters and stock information to 150.000 on line subscribers to |the media company| Personalised stock-updates and internet newsletters |magazine 1 and magazine 2| brought to subscribers with |product X|



9/27/2006

SECTION 3: WRITING MEDIA

3:37 PM

Page 265



9/27/2006

3:37 PM

Page 267

Chapter 17

Talking to Write: Investigating the Practical Impact and Theoretical Implications of Speech Recognition (SR) Software on Real Writing Tasks Noel Williams, Peter Hartley and Vanessa Pittard

Speech Recognition (SR) technologies are beginning to enter education to assist writing, but we yet have little knowledge of their potential impact on writers. This paper outlines some of the expectations we may have about SR, if we apply well-known cognitive and writing approaches to writers’ use of SR. It then briefly reports a small exploratory study aiming to establish if such insights are helpful in practice. Our research supports the view that the variety of writers’ contexts entails a mix of positive and negative impacts of SR, but does not support the claims that SR is a simple alternative to conventional writing, except perhaps for some tasks, for some writers. We suggest that technological limitations still prevent many writers from finding any serious value in SR and that novice writers largely find it frustrating. We also suggest that no existing theoretical perspective is well-placed to account for all the potential impacts of SR on writing.

17.1.

Introduction

Early advertisements for speech recognition (SR) software promised an exciting and liberating future. For example, ‘Picture your words flowing quickly and automatically onto the screen. With fewer typos and mis-spellings, and no writer’s block!’ (from an advertisement for Dragon Point and Click). Unfortunately for many, if not most, early users the reality did not live up to these expectations. However, the software has made significant progress over the last 5 years, and has also become much more affordable. Coupled with


Williams, N., Hartley, P., & Pittard, V. (2007). Talking to write: Investigating the practical impact and theoretical implications of speech recognition (SR) software on real writing tasks. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes, & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 267–278). Amsterdam: Elsevier.


268

9/27/2006

3:37 PM

Page 268

Noel Williams et al.

the speed and processing power of current PCs, the software is at last living up to the promotional claims: “Voice Recognition Software is fundamentally changing the way we compute. It enables you to produce accurate documents at the speed of a touch typist, and control you (sic) computer by voice alone. It also reduces your risk of developing repetitive strain injury. … Enrol your voice in just 5 min, and watch as Dragon NaturallySpeaking Professional Version 7.3 immediately turns your speech into text at up to 160 words-per-minute!” (Voicepower, 2004). “IBM ViaVoice Pro USB Edition, Release 10, is the latest edition to the UK English IBM ViaVoice family of products for Windows. It is designed for people who view speech recognition as a powerful productivity tool, and offers an improved speech engine, and background noise adjustment which can provide exceptional accuracy and fewer corrections” (Viavoice, 2004). Independent reviewers now confirm that these claims are reasonable, reporting ease of use and accuracy rates of over 90% (e.g. Brewer, 2003; Witchalls, 2003). We are certainly at the stage where we can argue that SR technology is here to stay, and the growing number of ‘support’ websites reinforces this view (see, for example, Fulton, 2004). The future promises further technological development of SR, including integration into standard PC interfaces and its development as a tool for writing short messages, emails and other text on 3G mobile phones. However, SR has been slow to establish itself as a mainstream educational technology and there is relatively little research to enable us to predict the extent of its eventual usage or its likely range of applications. Business analysts have argued for some time that SR can have an important impact on organisations (e.g. Janal, 1999) and there are now examples of significant benefits in specific contexts (e.g. Corces, Garcia, & Gonzalez, 2004). The impact of SR on disabled users has received some attention (e.g. Goette, 2000), and there is also some research highlighting possible health issues (e.g. Williams, 2003). Previous research also suggests that SR can have an interesting impact on the way we write. For example, Quinlan (2004b) found that SR had particular outcomes for less fluent young writers (aged 11 to 14), increasing the length of, and decreasing errors in, their narratives. However, despite interesting examples like these, the existing research base is very limited. We do not have much systematic research on SR’s role in enhancing or facilitating written communication with a wide range of users and over a range of educational contexts. In two articles, which are essential reading for anyone interested in this area, Lee Honeycutt (2003, 2004) argues that: we have very limited evidence of the long-term effects on users; we need to investigate the software’s use with a much wider range of users; we need more systematic investigation on how SR affects different forms of writing; and we need to take account of a wider range of variables, including cultural influences, to assess the impact of SR. He also argues that we need to incorporate and revisit insights from research into previous technological developments, such as early research on the development of dictation. For example, he notes that business communication textbooks and popular guides routinely advocate the importance of formal planning for successful dictation. He can find only one study that tests this proposition in the context of speech recognition technology


9/27/2006

3:37 PM

Page 269

Talking to Write

269

(and this study stimulated SR to overcome the technological limitations of the software at the time). This study did not find that planning led to improved dictation, whereas it did improve performance in written and conventionally dictated texts (Honeycutt, 2003, p. 91). This example reinforces our need to better understand what talking has to offer writing. As Leijten argues in the present volume, speech recognition can be seen as a hybrid writing mode, somewhere between dictation and computer-based writing. So we need to know more about this hybrid, the relationship between the different ways people write and the experience of ‘talking to write,’ and its impact on the existing practices and products of writers. Leijten’s own study compares the adaptation strategies of writers who were either used to dictation as a means of writing, or who were unfamiliar with it, to see whether the strategies used by these two types of writer to adapt to the new method had different characteristics. Leijten concludes that these writers barely adapt their existing strategies to the new medium. Hartley, also in this volume, presents a study with similar intent, though different design, comparing through case studies the impact of SR on the practices and products of experienced writers, and concluding that impact is not great. In both cases, therefore, we might judge that SR has relatively little effect on the writing process. In this chapter, we draw on existing knowledge and theory about speech production and about writing processes to generate related hypotheses about talking to write, with a view to taking the first step along a road towards a well-developed theory of ‘talking to write.’ We suggest that different theoretical perspectives entail rather different possibilities for the impact of SR on writers. Furthermore, as writers are of differing dispositions, practices and experience, the nature of the impact of this technology may vary. Our central view, however, is that writers’ preferred or most commonly adopted approaches to writing will be critical to the experience of using SR in writing, as most prevalent theories of writing suggest SR would have detrimental effect. We examined this idea through a small pilot study, exploring the impact of SR on two small groups of writers: one a group of student writers, and the second a group of experienced academic and professional writers. Both groups were asked to adopt SR technology while engaged in normal written work. We were as interested in exploring the many possible impacts of SR on writers as we were in discovering any particular effect, so our study aimed to see if there were indications that any given theoretical insight was likely to be more useful than any other.

17.2.

The Problem of Student Writing in Academic Settings

SR is already present in academic environments as a means of addressing barriers to academic writing experienced by students who have specific difficulties with conventional word processing, e.g. students with disabilities such as severe arthritis or repetitive strain injury. Many colleges and universities are now providing SR software for these students, and there are major projects in Europe and the USA exploring the benefit of speech technologies for learners with particular disabilities. For example, in the US the federally funded Speaking to Write project explores “the use of speech recognition software … for writing by secondary students with disabilities” (U.S. Department of Education, 2000). In the UK the JISC have undertaken a review of the value of SR for Higher and Further Education (HE/FE), concluding that there is “a lack of formal analytical research into how effective speech recognition systems have been in UK HE/FE situations and environments”


270

9/27/2006

3:37 PM

Page 270


(Kirriemuir, 2003), while BECTA’s website maintains an overview of educational SR technologies, primarily servicing the needs of schools (BECTA, 2004). As SR technology advances, however, it may increasingly represent a tool for learners beyond those with specific difficulties with typing. Many students in fact find academic writing a challenge. There are also increasing concerns from universities and employers about the levels of students’ and graduates’ skills in written communication. We are faced with an academic writing problem which may (at least partly) have a technological remedy. Perhaps SR can ease some of the burdens of academic writing and even help students produce more readable text, make better, more coherent, arguments, or just write more efficiently, leaving more time for other learning. Alternatively, it may be that SR is a Trojan horse, promising much, but serving as a diversion or distraction in the academic writing process. Theoretical perspectives on speech production and writing processes may offer some clues. The following sections present selected positions derived from speech and writing research. These are by no means comprehensive, but offer tentative predictions about the impact of SR on writing, including some predictions about what kinds of writers or writing may benefit from SR. Having identified some of these, we report on what our study exploring the impact of SR on practicing writers tells us about them.

17.3.

Theoretical Positions

17.3.1.

Basic Writing Processes

The most enduring cognitive model of writing process has been applied in many studies of school and college writing (e.g. Carey, Flower, Hayes, Schriver, & Haas, 1989; Kellogg, 1990). The Flower and Hayes model (1980), and subsequent revisions and related theoretical developments (e.g. Hayes, 1996; Bereiter and Scardamalia, 1987), allow us to understand why students’ experience of academic writing can often prove difficult. According to the cognitivist approach, mature writing should be treated as a cognitively demanding problem-solving activity (e.g. in Bereiter and Scardamalia, 1987). Various processes occupy a central monitor (working memory) as we progress in solving a writing problem, or moving towards producing coherent finished text. These processes include organising, goal setting, translating (producing new text) and evaluating. Because writing processes are recursive rather than linear, they regularly compete with each other for limited cognitiveprocessing capacity, and when working-memory capacity is overburdened, writing gets very difficult. This is more likely to be the case with inherently challenging writing tasks (e.g. those in unfamiliar genres and those which require knowledge or idea generation, rather than simple memory search) and for relatively inexperienced writers, who require working-memory capacity to fulfill ‘low-level’ functions more experienced writers process automatically, like spelling, word and grammar formulation or the mechanics of typing or handwriting. Evidence on speech production from psycholinguistics, on the other hand, (e.g. Garrett, 1982) demonstrates that speech articulation processes and error monitoring occur automatically. This theoretical position predicts that talking to write may be effective for writers, as (in principle, at least) it liberates them from burdensome writing-specific tasks — typing and


9/27/2006

3:37 PM

Page 271

Talking to Write

271

spelling. It possibly liberates writers from some syntactic and stylistic decision making, too, as this kind of formulation is more ‘natural’ during speech. In liberating writers from these, there is more capacity for higher-level processes, thus affording greater scope for idea generation and strategic approach to writing as a result. Taken at its simplest, it can be argued that this theoretical position predicts that: P1: Learners who have limited writing skills, are new to academic writing, are new to a topic, and who have a challenging writing task are likely to benefit most from SR. P2: Writers who work ‘more’ recursively (juggling more writing processes at the same time) may benefit from using SR. 17.3.2.

Writing as Social Interaction

Sociolinguistics can also contribute to our predictions about the impact of SR on academic writing. Nystrand (1989) maintains that writing is rather like a conversation. He observes: “A major postulate in composition theory [...] has been that writing significantly differs from speech in the sense that writers, unlike speakers, cannot interact with their addressees” (Nystrand, 1989, p. 70). He claims that the distinction has been over-stated and objects to the Flower and Hayes (1980) characterisation of writing as goal-oriented problem solving, with the production of written text as the end result. He prefers to view the process as extending to the reader/audience and as based in the establishment and maintenance of a social relationship, with text as a means for that process. According to Nystrand (1989), all writers, including academic writers, enter a ‘temporarily shared social reality,’ a TSSR (Rommetveit, 1974) while writing. Social inputs are realised by the writer’s references to shared understandings or anticipated reader responses. So meaningful text arises out of the writer’s ongoing negotiation with a (notional) reader/audience — effectively an internalised conversation. This drives the writing process, directing the ‘black box’ of the Flower and Hayes (1980) monitor. Nystrand (1989) maintains that effective writers successfully maintain a TSSR, effectively being highly ‘reader-focused,’ and poor writers fail. However, it is difficult to apply this theory of effective writing to SR, until two further, related, theoretical approaches are considered. The first of these, the Vygotskian perspective, has influenced Bereiter and Scardamalia’s (1987) developmental theory of writing. This perspective explicitly links speech and writing development. Writing is derived from speaking. Early writers require explicit conversational inputs to support their writing; thus, early writing is, in social interaction terms, very like conversation. As writers mature, they shift away from a dependence on conversational input, gradually adopting strategies for generating coherent text without it (e.g. memory search, attention to the whole). So writing is really an extension of talking, but is simply talk written down and enhanced by processes, which enable the discourse to develop without the ‘scaffolding’ of social input. This perspective, again at its simplest, suggests that once writers have become ‘mature,’ talking to write is feasible. All the requisite cognitive paraphernalia is present for this to be the case.


272

9/27/2006

3:37 PM

Page 272


This is, of course, a simplification of the position, as Vygotsky would actually argue that writing becomes functionally separate from speech. Conversation Analysis (CA), however, tells us more about the conditions under which talking to write is most likely to be successful (see Nofsinger, 1991; Psathas, 1995). In most everyday contexts, talk is very ‘interactive’ — it is usually ‘guided’ by a respondent as well as a talker, and the meanings established in talk are dependent upon real conversational practices. These practices are effortless in adult speakers and include alignment devices, like ‘continuers’ (‘uh huh’), adjacency responses, ‘formulations’ (summary responses) and repairs and collaborative completions. Conversation is a very co-operative enterprise, with genres of its own — very different from writing. This perspective suggests that talking without conversational support (e.g. talking to write, recitation) does not come naturally. That is why we often need to be trained to do oral presentations and require ‘scaffolding,’ like OHPs and mnemonic devices, to help us through. The overall messages coming from these three sociolinguistic and social-developmental positions is ambiguous. The most pessimistic prediction would be that writers talking to write will struggle to align the practice of speaking with the conditions and requirements of writing. However, some tentative further predictions are also possible: P3: A technology (perhaps SR?) which, even marginally, affords more ‘conversation-like’ approaches to writing (encouraging writers to think ‘through the page’ to the reader) will promote more effective writing. That SR is this kind of technology is an issue, of course. P4: Writers who have the right ‘scaffolding’ as they talk to write (e.g. those who have a detailed enough structure prepared) will find the writing task more comfortable and more efficient. P5: Writers who are generally explicitly reader-focused–especially those who have imaginary ‘conversations’ with the reader while translating–will find the process easier than those who do not (all other things being equal). 17.3.3.

Monitoring and Feedback in Talk and Writing

The Flower and Hayes (1980) model of writing posits a central monitor, akin to working memory (Baddeley, 1986, 1990). Since the development of the Flower and Hayes model, we have learnt more about the nature of working memory, its components and the constraints on the information it can deal with. These findings have implications for SR, particularly in respect of the role of feedback during writing and how this relates to feedback during talking, which we briefly describe below. There is surprisingly little research on the relationship between speech-production processes and working memory, but evidence suggests that when we talk, as well as possibly utilising certain components of working memory (e.g. the central executive) for things like idea generation, we also monitor speech for errors by maintaining rolling clause and sentence articulation (utilizing the ‘phonological store’ — Baddeley, 1990). We are effectively geared up to monitor speech acoustically and we do this without deliberation,


9/27/2006

3:37 PM

Page 273

Talking to Write

273

so we can do things like correct errors as we go along, or even correct them before actually speaking. So, though speech production is a highly automatic process, often feeling less demanding than writing, it does in fact occupy some working-memory capacity fleetingly as we talk — hence, for example, competing capacity demands from speaking and listening at the same time. This kind of speech monitoring or feedback function is different from that of writing. With writing, of course, the primary monitoring is based on visual information — what we have just written, or previously created text. Though visual, written words are translated into a phonological representation via articulatory control processes (again, a workingmemory function). Emerging text on the screen provides an additional source of feedback and an additional demand on working memory other than that which would be the case when simply speaking. Talkers for writing are in fact engaged in a double-monitoring process which (albeit marginally) is likely to increase attentional demands. There is an argument that writers do this anyway (i.e. ‘listen’ to what they translate, as well as monitor the screen), but the fact of vocal articulation is the key difference here. Garrett’s (1982) work on speech hesitation tells us that it can occur as a result of environmental cues. It is likely that visual feedback will feed this category of speech hesitation on occasion. We can predict that: P6: The greater the visually based monitoring task is attentionally, during talking to write, the less effective SR will be. This could be the case if the writer is forced to monitor for word errors, for example. A robustness ‘threshold’ is required before SR can impact positively on writing in terms of cognitive burden. 17.3.4.

Speech Recognition and the Writing Task — What are the Issues to Investigate?

The main aims of our exploratory project were to see if the predictions outlined above were in any sense meaningful, and thereby to help refine areas for future SR research. More concretely, we were interested in knowing: • Does speech recognition software actually favour, or disadvantage, certain ways of writing? • In practical terms, would writers engaged in tasks such as typical academic writing find SR a benefit? • Does any theoretical perspective seem most likely to apply most readily to SR as writing?

17.4.

The Pilot Study

We saw our study as a pilot exploration of these issues aiming to examine some of the predictions above, to see whether there was any consensus in users’ views of SR concerning the impact of SR on their way of writing, and whether such consensus, if it existed, aligned with any of these particular theoretical positions. However, we were partly driven by pragmatic constraints (the number of users available to us, the wide variation in their writing needs and


274

9/27/2006

3:37 PM

Page 274


so on) so were not free to construct an experimental or statistically robust design. We initially recruited 25 subjects. These were 8 professionals, working either as academics or members of a small multimedia development company, who described themselves as “confident” writers, and 17 students, who described themselves as less confident. Technical problems1 reduced us to only 12 subjects actively taking part, reducing our chance of firm conclusions. Our starting point was to assess the writing styles of the participants, through an initial questionnaire (which was framed in such a way that it reflected the distinguishing characteristics of different styles of writing), equip participants to use SR in their normal writing activities, and, after six months’ use, ask them to report on the perceived values of SR. Comparison of their views with the initial writing-style assessment should have allowed us to judge whether their experiences of SR correlated with their preferred ways of writing. Two questionnaires were to be used as pre- and post-assessments (along with logs of use to be maintained by student writers). In addition, the post-assessment questionnaire was to ask users for their self-assessment of the value and impact of the technology on their writing practices. 17.4.1.

Results of the Study

From the 12 subjects who did generate data, our second problem was that our approach to modelling writing styles was hard to apply with any firm conviction. Although we asked questions, which would identify particular writing styles if consistently selected, we had not anticipated that subjects would indicate tendencies towards several styles. Our questions should have been designed to discriminate degrees of accord with particular styles, so that cumulative assessment could have been made, but they were not. Consequently no strong summative judgements could be made, and this element of the study was therefore inconclusive, even though there appeared to be some indicative clusters of behaviour associated with particular styles of writing. Simply put, it appeared that writers in the less confident or novice group (the students) found the technology unhelpful or even a barrier if they were used to an incremental, linear, sentence by sentence style of writing. Conversely, confident, experienced writers who used such an incremental linear approach to their writing had much less of a problem. In addition, our data suggested that, overall, discovery writers seemed less comfortable with SR than planners. Where our study could report more confidently was in examining the consensus across all subjects for the second, post-assessment, impact questionnaire. This asked respondents to rate 24 statements about SR on a 5-point scale, ranging from “Strongly Agree” to “Strongly Disagree.” These responses were scored as 0 for “Neither agree nor disagree,” –1 and –2 for degrees of disagreement, with 1 and 2 for degrees of agreement. Table 1 shows these statements and their cumulative scores. 1 In the event, the project hit two difficulties: one technological and the other analytical. Initially users reported many hardware deficiencies. The software could be installed, and did run, on low-specification computers (within the specification claimed as adequate by the software), but in two cases it crashed the computer, and in others ran so slowly that users found it too frustrating to use in any serious way. In addition, many users noted recognition problems due to the poor microphone supplied with the software, and sometimes the result of ambient noise in the background. Ten users reported serious technology weaknesses. A total of 11 learners and 2 professionals were too frustrated with the software and hardware problems to use SR after initial attempts. This reduced us to only 12 subjects actively taking part, reducing our chance of firm conclusions.


9/27/2006

3:37 PM

Page 275

Talking to Write

275

Table 1: Responses to second questionnaire for 12 subjects. Statement

Score

Agree/disagree

1 When using SR, I don’t go back over my writing

–19

0/12

2 SR helps me plan my writing

–16

0/11

3 SR helps me find my way around my text

–15

0/11

4 Using SR makes me a more confident writer

–13

0/10

5 When using SR, I do not alter my writing as much as with conventional word processing

–12

2/8

6 SR interferes with my writing plans

–12

0/8

7 I focus more on language using SR than without it

–12

2/10

8 SR makes editing easier

–12

0/11

–9

0/5

10 If I use SR, I don’t print out a draft

–8

4/8

11 SR helps me organise ideas whenever I need to

–7

2/6

12 SR helps me review my text on screen

–7

2/6

13 SR makes it difficult to structure my writing

–6

0/4

14 SR speeds the writing process

–5

4/8

15 SR helps me create a complete draft

–4

4/4

16 SR makes writing harder

1

5/7

17 SR helps me write text in any order I feel like

3

5/4

18 SR is more restrictive than keyboard word processing

4

7/5

19 I find it hard to correct text using SR

7

8/2

20 I find it hard to make changes using SR

8

8/0

21 I use SR to correct errors as I go along

7

10/2

22 Sometimes SR interferes with my thinking as I write

9

10/2

23 SR helps me jot down ideas as they occur

10

10/2

24 SR adds more errors to my writing

13

11/1

9 I make fewer errors using SR than conventional word processing


276

9/27/2006

3:37 PM

Page 276


With 12 respondents the maximum score in either direction would be 24, and a score of 12 can be regarded as majority opinion, as it represents a score somewhere between half the respondents feeling strong (dis)agreement and all of them feeling some (dis)agreement. The greater the difference from 0, the greater the strength of feeling. However, a linear calculation like this may not always clearly represent divergence, as negative and positive scores can simply cancel each other out, tending towards an overall score of 0, an apparent consensus. The final column in Table 1 therefore also records the number of individuals Agreeing/Disagreeing with each statement. Clearly statements 11, 12, 15, 16, 17 and 18 represent neither strong nor consensual views of the group as a whole, and so must be ignored. Statements 9 and 13 form one interesting pair. For these two, no cumulative strong opinion is voiced, as most people had no real opinion. However, those who did have an opinion all disagreed with the statement. A second pair, statements 10 and 14, both show that every respondent has a clear opinion, but there is substantial dissent between them, with one third agreeing and two thirds disagreeing with those statements. Responses to 10 and 14, therefore, may suggest that there are two “types” of writer in relation to SR, but our work cannot characterise these types beyond these responses. For all other statements the difference between Agree and Disagree is at least 6 respondents, and the total score given by those respondents at least 7 above or below 0. We took these two conditions as an operational definition of consensus. That is, we saw any statement where the total score was 7 above 0 (for “Agree”), or 7 below 0 (for “Disagree”), and for which there were also at least 6 more people voting for one of the opinions than for its opposite, as indicative of a consensual statement. In other words, our respondents disagreed with statements 1–8, and agreed with 19–24. These therefore suggest the following opinions of SR, for our small number of writers. (“S” numbers in brackets are the supporting statements from Table 1). Our respondents believe speech recognition systems: • • • • • • •

Do not force a linear approach to writing (S1) Neither help planning nor interfere with writing plans (S2, S6) Do not help navigation around text (S3) Do not aid confidence (S4) Lead to at least as much re-writing as in conventional word processing (S5) Do not lead to increased focus on language (S7) Do not aid editing, and can make it harder, though it can be used to correct errors actively during composition (S8, S19, S21) • Interfere with thinking during composition (S22) • Help with idea generation (S23) • Add more errors to writing (S24) Across the data, therefore, this lack of overall agreement, even within such a small group, suggests that different people find SR software both a help and a hindrance. So, the answer to our first research question is that SR does both favour and disadvantage different ways of writing. The obvious follow-up for future research will be to clarify those writing tasks or ways of writing which SR best accords with, and the studies by Leijten and by Hartley in this volume begin to open up such avenues.


9/27/2006

3:37 PM

Page 277

Talking to Write 17.4.2.

277

Relating Results to Theory

How, then, does the data we obtained relate to the predictions that emerge from different theoretical perspectives? Below we briefly relate our data to relevant predictions from the discussion above, annotated as P1 to P6. 17.4.2.1. P1: Learners who have limited writing skills, or are novices in the writing task should benefit most from SR Collectively, our project suggests this is not the case. Eleven of our 17 novice learners dropped out of the project. The impact questionnaire suggests confidence is not aided by SR, that it can make some processes harder and interfere with others. The data also suggested that confident writers liked SR whilst the less confident felt frustrated. This suggests that for learner writers, at least those who prefer to work one incremental sentence at a time, SR interferes with the writing process. 17.4.2.2. P2: Writers who work ‘more’ recursively may benefit from using SR There is strong agreement amongst our writers that SR does not force a linear approach. All 12 of our writers felt strongly that they do “go back over” their text using SR, 8 felt that they do alter their text as much as with conventional WP, 10 that they find it helps incidental idea generation and 10 that they can correct errors on the fly. 17.4.2.3. P3 and P5: If SR affords ‘conversation-like’ and reader-focused approaches to writing, it will promote more effective writing The consensus of our writers is that SR does not promote more effective writing, and, although it does appear to support some writing subtasks for some writers, reported negative impact is much stronger than positive. 17.4.2.4. P4: Writers who have a prepared structure will find the writing task more comfortable It seems from our data that planners are more positive about SR than discoverers. In other words, our study would suggest that those who already know what they are going to write, and therefore largely use SR to generate and edit text to that plan, find it a more useful technology than those who write to find out what they are going to say. 17.4.2.5. P6: The greater the visually-based monitoring task is, the less effective SR will be because of the cognitive burden Judgements from our group of writers suggest that this view has some force, notably, the feeling of 11 that SR causes more errors than in WP, of 6 that it does not really help the review of text on screen, and of 10 that SR can interfere with writing. However, an interesting exception was one student who, on completing the project felt that SR had a very definite and positive effect on his writing, claiming that his normal text was improved because SR conditioned him to think more about structure and style during the process of constructing his essay. This suggested that perhaps the reported need to monitor the output from SR as being used was beneficial in his case, as it caused him to reflect more than he might otherwise do.


278

17.5.

9/27/2006

3:37 PM

Page 278


General Conclusions

The experience of all our users was mixed. Even those who were most positive about the technology (generally experienced writers who were also experienced users of computers) reported mixed feelings about using SR, with the most positive respondents still reporting more negative than positive answers. We have to conclude that SR does not give immediate benefits. It needs investment in time and energy in order to work positively, and it needs good quality hardware. Our work suggests that different types of writers may benefit from the use of SR, if they are prepared to make this investment, but it will not suit everyone, and discovery writers in particular may find it problematic. Using SR certainly adds to the writing process, as additional monitoring and skills are needed, and users probably need training to make best use of the software. We think our data shows that the benefits may depend critically on context, in terms of the experience of the writer and the physical context of use, and these may be more important factors in successful use than any habits of writing style. The data also suggest that there is no obvious theoretical perspective which is clearly best positioned to examine the impact of SR on writing, as the predictions we extracted from the theoretical perspectives are in no cases well supported (either the prediction is contradicted by our evidence, or some predictions consonant with a theory are supported whilst others are not). In sum, therefore, we probably need to evolve a theory of ‘talking to write,’ which adapts rather than adopts existing theoretical perspectives on the writing process.


9/27/2006

4:34 PM

Page 279

Chapter 18

How do Writers Adapt to Speech Recognition Software? The Influence of Learning Styles on Writing Processes in Speech Technology Environments Mariëlle Leijten

This paper describes the adaptation and learning processes of writers who have started using speech recognition systems for writing business texts. We observed writers during five set moments in their daily work. The data from these observation sessions were used to describe the adaptation strategies during the learning process. In a case study we analyzed the learning processes of two writers with similar writing experience (classical dictating), but with a different learning style: accommodator versus diverger style (based on a taxonomy of Kolb). The participants of the case study differed mainly on (a) the amount of time they spent in the speech recognition mode, and (b) the mode they used to solve ‘technical problems’ caused by the speech recognition software. The results show that both participants have a different adaptation process, which evolves differently and seems to be driven by the learning styles of the participants.

18.1.

Introduction

Speech recognition software enables writers to write their texts by ‘talking to the computer.’ By adding a toolbar to the normal on-screen working environment (see Figure 1), the software enables normal keyboard and mouse commands to be voice-activated, e.g., dictating text, navigating through programs or texts, emboldening words, and opening or closing files. (Williams et al., this volume, provide an extensive description of the practical impact of speech recognition software; Honeycutt, 2003, gives a historical overview of speech recognition.)


Leijten, M. (2007). How do writers adapt to speech recognition software? The influence of learning styles on writing processes in speech technology environments. In G. Rijlaarsdam (Series Ed.), and M. Torrance, L. van Waes, & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in Writing, Vol. 20, pp. 279–291). Amsterdam: Elsevier.


280

9/27/2006

4:34 PM

Page 280

Mariëlle Leijten

Figure 1: Speech recognition toolbar (Lernout & Hauspie Voice Xpress). Until a few years ago research on speech recognition applications focused on qualitative (technical) improvement. More recent research has dealt with speech recognition as a writing tool (e.g., Halverson, Hall, Karat, & Karat 1999; Hartley, this volume; Hartley, Sotto, & Pennemaker, 2003; Karat, Halverson, Horn, & Karat, 1999; Karat, Horn, Halverson, & Karat, 2000; Quinlan, 2004a). These studies mainly focus on the usability of speech recognition as a writing mode. In the study by Karat et al. (1999), initial and professional users tested the usability of three speech recognition systems. The initial users were given two kinds of writing tasks, one they had to perform with speech recognition, and the other with keyboard and mouse. The results showed that users struggled with speech-driven error correction and that they spent more time making corrections than the researchers had expected. Perhaps this can be explained by (i) the goal of the study (usability and error correction), (ii) the fact that the participants were observed all the time, and (iii) the fact that they had just received extensive training in speech-based correction methods. The use of speech recognition software by the learning- and physically-disabled has also received some attention. Quinlan (2004a) studied the writing processes and products of 40 children of 11–14-year-old age. Of these 20 children had writing difficulties. The children composed a series of four narratives in one of two main writing conditions, handwriting and speech recognition, with a further two conditions for each mode, with or without advance planning. Quinlan (2004a) hypothesized that speech recognition would provide cognitive benefits to children with writing difficulties, and that advance planning would be supportive for the real-time planning process. The results showed that the children with writing difficulties produced longer texts when using speech recognition than when using handwriting. Speech recognition seemed to reduce transcription-related interference, enabling the children to produce more fluently written — and hence longer — texts. For the fluent writers, however, composing with speech recognition did not lead to improved narratives. Advance planning had a significant positive effect on text quality for both groups. Hartley et al. (2003) compared the academic correspondence produced by an experienced writer, initially using keyboard and mouse and then using speech recognition. The writing products differed only slightly in their average sentence lengths, the use of especially long sentences and the first pronoun (see Hartley, this volume). In a previous study we described a similar case study (Leijten & Van Waes, 2003) in which we focused on writers with dissimilar writing experiences, i.e., those with previous classical dictating experience versus those without previous dictating experience. This study showed that speech recognition does not impose a specific writing style on the user, contrary to previous findings stating that pen and paper writers had to adapt to the keyboard and mouse environment first. Speech recognition can be described as a hybrid-writing mode, because it combines elements of dictating and computer-aided writing. Traditional dictating is characterized by a high degree of linear text production (Gould, 1978; Gould & Alfaro, 1984; Schilperoord, 1996b), whereas writers using speech technology receive immediate visual feedback from their computer screen. This allows them to review the text at all stages of the writing


9/27/2006

4:34 PM

Page 281

Speech Recognition and Learning Style

281

process, thereby opening the gates to non-linearity. Indeed, a high degree of non-linearity is typical of computer writing processes (Severinson–Eklundh, 1994; Severinson–Eklundh & Kollberg, 2003; Van Waes & Schellens, 2003). The constant feedback from the screen allows writers to revise continuously, without losing an overview of the final text. 18.1.1.

Learning

An important characteristic of this study is that the participants had to learn how to produce texts in a different way. Therefore, we will pay attention to different learning styles. In this study1 we used Kolb’s definition of learning, viz. “the process whereby knowledge is created through the transformation of experience” (Kolb, 1984, p. 38). In this definition we would like to emphasize two aspects of the learning process that are especially important from our perspective. The first one is the emphasis on the processes of adaptation and learning as opposed to content or outcomes. The second one is that knowledge is a transformation process; it is continuously created and recreated. The knowledge people have of working with speech recognition is constantly transformed by the experience they build up while working with the program. Kolb (1984) describes the learning process as a four-stage cycle involving four adaptive learning modes: concrete experience, reflective observation, abstract conceptualization, and active experimentation. In his Learning Style Inventory he evaluates the relative preferences an individual holds for each of these four learning modes. The Learning Style Inventory is an objective, selfscoring instrument that reveals four statistically prevalent learning styles: diverger, converger, assimilator, and accommodator (Figure 2). People with a converger or assimilator learning type prefer a higher level of abstract conceptualization. Convergers prefer to learn via the direct application of ideas and theories and have been described as somewhat unemotional, preferring to work on their own. Assimilators are good at taking in a wide range of information and reducing it to a more

Figure 2: Learning styles (Kolb, 1984). 1

This study is part of a PhD, supervised by Professor Mr. Dr. P. Van den Hoven, Dr. D. Janssen, and Professor Dr. L. Van Waes. The study is part of an NRI project on the effect of speech recognition on writing processes (research grant of the University of Antwerp, 2000–2002).


282

9/27/2006

4:34 PM

Page 282

Mariëlle Leijten

logical form. They tend to prefer theoretical models and deductive reasoning; this leads to a greater interest in abstract concepts and ideas than in interaction with other people. Divergers and accommodators rely heavily on concrete experiences when learning. Because they also prefer active experimentation, accommodators are described as having the ability to carry out plans and get things done. They get involved quickly in new material by way of trial and error learning. Divergers are identified by their ability to look at a learning situation from multiple points of view. They often have a hard time making a decision and mostly prefer to observe rather than to participate (Kolb, 1984; Terrell, 2002). In this study we have opted to observe writers in their professional environments. The texts they were writing were part of their ‘normal’ work; the only difference was that we asked them to use speech recognition as a writing medium. Kolb predicts that people in technology and information science careers generally benefit from a converger or an assimilator learning style. Consequently, we expected that those participants who were characterized by either converger or assimilator learning styles, would outperform other participants. But of course, participants whose ‘preferred’ learning style is less suitable for learning to work with a new writing mode, will also be able to adapt their learning style to meet the requirements of this new mode, perhaps less efficiently. In this study we will take a closer look at the concrete strategies participants develop in learning to work with speech recognition. In this chapter we focus on the adaptation and learning processes of writers learning to write their business texts in the new medium of speech technology. We are mainly interested in the general cognitive processes that characterize the writing process and will become more visible by focusing on a new writing medium: How will the writers adapt their writing to the new medium? Will their writing strategy change over time? What is the influence of learning style on the initial use of speech recognition?

18.2.

Description of the Study

In this section we will first describe the participants involved in this study, then we will elaborate on the design and procedure of the study, and finally, we will briefly describe the writing materials that the participants produced. In the case study presented here, we will focus on the writing and adaptation processes of two participants, Frederik and Bart. Both participants only had to adapt to the speech software because they were experienced in working with computers and dictating devices. Frederik, an associate in a large law firm, had 5 years of work experience and about 6 years of experience working with computers and classical dictating devices. Bart had been working as a lawyer for 9 years, but had a comparable level of experience in working with computers and dictating devices. However, the reason to select them for this case study was that they were characterized by a different learning style (cf. Learning Style Inventory of Kolb, 1984, as described in Figure 2). According to Kolb’s taxonomy, Frederik is an accommodator. He scored high on active experimentation and concrete experience. Kolb’s model states that people with an accommodative orientation tend to solve problems in a trial-and-error manner. They also rely heavily on other people for information rather than trust their own analytic ability. People


9/27/2006

4:34 PM

Page 283


283

with this learning style are sometimes seen as impatient and ‘pushy.’ As Kolb predicted, Frederik’s learning style is determined by his job to the extent that he has to make a lot of decisions in uncertain and unpredictable circumstances. Bart is a so-called diverger. He scored high on concrete experience and reflective observation. In this learning style the emphasis is on adaptation by observation rather than by action. A person of this type performs better in situations that call for generation of alternative ideas and implications. Furthermore, they tend to be imaginative and they prefer to reflect on situations. Characteristics of Bart’s job that relate to his learning style are personal relationships and effective communication with other people. Both participants have different learning style than what Kolb predicted to be most adequate for learning to deal with (new) technology. However, since learning styles are adaptable to various tasks, we can predict that both participants will be — in some way — influenced by the task requirements of writing in a new medium. The active orientation of both participants will guide them to learn certain tasks and to improve active skills. 18.2.1.

Design and Procedure

The participants were required to complete a Dutch translation of Kolb’s Learning Style Inventory (Kolb, 1984), which provided information about differences in learning styles. In a post-hoc analysis we used this information to determine the learning styles of the participants. Before the participants started using speech recognition for the first time they watched an introductory video on the use of the speech technology program. This video was provided by the software company. The participants were then informed about the procedures of the study in more detail. They were asked to use the speech recognition system during their day-to-day work for at least three hours a week. The participants could decide for themselves how to use the software and they were not restricted to the exclusive use of speech input. Keyboard and mouse could also be used as complementary input devices. In total we observed the participants five times while they were writing in their own environment, respectively after 1, 3, 6, 9 and 12 h of working with speech recognition. The data were collected using an online camera (CamtasiaTM) and a sound recorder (QuickrecordTM). Because of the combination of the different input modes (keyboard, mouse, and speech) we were not able to use existing logging programs. We observed the participants during each writing session and noted down the specific writing circumstances that could not be registered in any other way (see Figure 3). These recordings and notes enabled us to reconstruct the writing process in detail.

18.3.

Results

To describe the product and process data of the study a categorization model was developed that takes the complexity of the hybrid-writing mode into account. This model also made the enormous amount of process data accessible for further research (for a full description see Leijten & Van Waes, 2003; Severinson–Eklundh, 1994, 1996; Kollberg, 1998; Van Waes & Schellens, 2003).


284

9/27/2006

4:34 PM

Page 284

Mariëlle Leijten

Figure 3: Observation setting. First we describe the product data — the material — to contextualize the writing activity. In the second section we describe the use of the speech recognition writing mode and focus on the input of the three possible writing modes: namely speech, keyboard and mouse. Finally, we describe the repairs. We have adapted the concept of repairs from conversation analysis, because it provides a valuable addition to the traditional concept of reviewing. Moreover, it enabled us to incorporate a broader view of reviewing, taking into account the specificity of the reviewing process in the speech recognition mode. Repairs could either refer to ‘technical problems’ or to ‘revisions.’ ‘Technical problems’ in this study are things that go wrong because of the speech mode (Karat et al., 1999). For example: • misrecognitions (dictated text is misinterpreted and a non-intended text appears, e.g., ‘eye’ instead of ‘I’), • command misrecognition (e.g., instead of executing the command ‘end of sentence,’ the text ‘end of sentence’ appears on the screen), • zero recognition (the computer does not respond). In the repair analysis we distinguished these technical problems from ‘revisions.’ Revisions are changes in the text after an evaluation of the previously written text. These changes are not initiated by technical problems but by the writer. They are designed to change the content, formulation or appearance of the text. 18.3.1.

Material

The product data informed about the length of the texts, and the duration of the observations. The different tasks that the participants carried out were job-related and part of their normal writing activities, e.g., letters, e-mails, or reports. Because they were familiar with these kinds of texts, the tasks could be performed structurally and routinely.


9/27/2006

4:34 PM

Page 285


285

Table 1: Characterization of writing product in five observations. Task

Length observation

No. of words in final text

Mean no. of words per minute

Frederik 1. Letter 2. Petition 3. Conclusion 4. Conclusion 5. Letter

04⬘08⬘⬘ 22⬘17⬘⬘ 21⬘23⬘⬘ 23⬘33⬘⬘ 19⬘46⬘⬘

92 449 419 247 305

22.27 20.15 19.59 10.49 15.43

Bart 1. E-mail 2. Conclusion 3. Petition 4. Report 5. Petition

06⬘00⬘⬘ 24⬘33⬘⬘ 22⬘13⬘⬘ 17⬘25⬘⬘ 37⬘00⬘⬘

62 586 520 419 685

10.33 23.86 23.42 24.05 18.52

The observation session differed in length because the participants were observed while working on a task they had chosen themselves. The first observation of both participants lasted about 5 min and the mean time of the remaining observations was 23⬘31⬘⬘ (Frederik: mean ⫽ 18⬘05⬘⬘, Bart: mean ⫽ 21⬘18⬘⬘). During this time Frederik produced an average of 17.76 words per minute whereas Bart produced 20.33 words per minute. Because of the variable length of the observation sessions, the texts also differed in length. In the first observation the participants explored the program and wrote a very short text; during the subsequent observations their texts were between 247 and 685 words per session (Frederik: mean ⫽ 355 words, Bart: mean ⫽ 552.5 words). Table 1 gives a description of the five observation sessions. 18.3.2.

Writing Modes

The mode analysis shows that both participants use the potential of speech recognition quite differently (Table 2). Frederik uses the speech input and the keyboard and mouse input almost just as much. The part of the speech mode drops gradually from the first observation session (47%) to the final session (34%). On the other hand, Bart uses the speech input almost twice as much as Frederik does. For over 80% of his writing tasks, he uses the speech mode. This way of writing adheres to Bart’s traditional dictating habits and he hardly changes this strategy over the different observation sessions. If we compare the use of writing modes in the first and second part of the writing process, it can be seen that Bart prefers to write his texts with speech input and that his choice of writing mode stays the same over the two writing parts. Speech is his preferred writing mode, both in the formulation and revision writing subprocesses. On the contrary, Frederik uses speech input in the first part of the writing process for more than 50% of the


286

9/27/2006

4:34 PM

Page 286

Mariëlle Leijten Table 2: Mean use of writing modes in percentages (five observation sessions). Frederik Mean (%) Speech Keyboard Mouse

42.5 44.4 13.2

Bart SD

Mean (%)

SD

8.9 13.9 5.1

81.1 13.8 2.2

7.6 7.2

time. In the second part he clearly prefers to write with keyboard and mouse. The use of speech input to dictate text segments drops to a mean of 32.7%. In the first observation session Frederik used speech for 40% of the process time in the second part of the writing process. However, in the last session, the speech input almost completely disappeared from the writing process. He hardly uses it to finish his text (cf. infra). In Bart’s writing process, changes in the use of writing modes are hardly seen. His use of speech recognition remains relatively constant (80%) in the different writing sessions and in both parts of the writing process. In summary, it can be said that the mode analyses confirm to a large extent the learning styles of both participants. Frederik, the accommodator, actively explores in the first sessions the possibilities of the speech recognition mode, but gives in after some trial and error. His use of speech recognition is rather restricted. Bart, on the other hand, as a diverger, uses the speech recognition system much more intensively and he explores the possibilities in several phases of the writing process. Table 1 shows that these explorations do not have a negative effect on the writing ‘performance.’ Bart produces even more words per minute than Frederik [mean per session: Bart 20.33 (SD 5.66) vs. Frederik 17.76 (SD 4.72) words per minute]. 18.3.3.

Repairs

In total we observed more than 500 repairs (technical problems and revisions). Frederik interrupts the linearity of his writing process 306 times in five observation sessions and Bart has 223 recursive actions (Frederik, mean ⫽ 61.20 per session vs. Bart, mean ⫽ 44.60). If we correct these data for the time differences between the observation sessions, this results in 3.45 repairs per minute for Frederik, as opposed to 2.09 repairs per minute for Bart. These numbers may seem high, but they are in line with Levy and Ransdell (2002). In their study of ‘concurrent memory loads,’ they found that frequent concurrent task presentation (comparable to interruptions of, e.g., technical problems) may lead the research participant to devote his or her full attention to the secondary task, thereby reducing the opportunities for proficient text generation. They concluded that critical events for secondary tasks should occur about four times a minute, a conclusion which compares favorably to our observation of the number of times Frederik and Bart are willing to interrupt their writing process.


9/27/2006

4:34 PM

Page 287


287

The data in Table 3 also show that both participants deal with repairs quite differently. As we have seen in the mode analyses, Frederik hardly uses the speech mode to solve technical problems or to revise (6.6%), while Bart tries to use speech recognition in 55% of the cases for recursive text interventions. If we add the repairs that are made with a combination of writing modes, including speech, the difference is even more significant. Frederik does 90% of the repairs exclusively with keyboard and mouse. In Bart’s case, this percentage drops to 30. He chooses to explore the possibilities of speech recognition. In 30 to 40% of the cases during the first observation sessions, Bart needs more than one attempt to solve a (technical or textual) problem. During the course of the observation session the solutions taking more than one attempt decline to less then 10%. Frederik, however, needs no more than one attempt to solve a problem in most cases, because he prefers to use keyboard and mouse for text repairs. As such, he seems to be less eager to explore the possibilities of speech recognition. 18.3.3.1. Technical problems If we compare the proportion of technical problems then it becomes clear that the first category is the most extensive one (Frederik: 81.4% technical problems vs. Bart: 91.1% technical problems). This pattern is typical of the adaptation process, as proved by the low number of technical problems in the last observation session (Frederik: 70.7% vs. Bart: 79.8%). The analyses of the technical problems show a different pattern for both writers. Frederik struggles more with the new technology than Bart does. If we compare the technical problems per 100 words, we see that Frederik has to deal with almost one-third as many technical problems than dealt by Bart (Frederik: mean ⫽ 16.75 problems per 100 words vs. Bart: mean ⫽ 10.43). We can again relate this result to the learning styles of both participants. Frederik is fairly impatient and he is primarily focused on productivity. As a result, he keeps writing in his traditional way: dictating larger text segments with only short interruptions. He verifies the result of his dictations on the computer screen rather superficially. Consequently, he is not able, or not willing, to adjust his use of the speech recognition tool. He repairs his technical problems mostly in the second writing phase when he rereads the produced text, again mostly without the use of speech recognition. On the other hand, Bart uses the text that appears on the screen more consistently during the formulating phase. He reflects on the possible causes of the problems that occur (misrecognitions) and consequently tries to anticipate and to avoid these problems. This strategy results in a lower problem ratio.

Table 3: Writing mode used before a repair in percentages (five observation sessions). Frederik

Speech Keyboard and mouse Combination

Bart

%

SD

%

SD

6.63 88.13 5.24

5.24 8.62 3.75

55.01 31.48 13.51

29.98 22.74 11.04


288

9/27/2006

4:34 PM

Page 288

Mariëlle Leijten

18.3.3.2. Revisions It is striking that both professional writers revise very little in the five observation sessions. This may be partly explained by the fact that both writers performed routine tasks. Furthermore, the writing strategy of dictating writers is also relevant: these writers mentally preformulate and therefore already partly revise their text before one word is dictated. A large part of the dictators’ revisions takes place in the writers’ mind. This limits the on screen revisions and places them higher in the text hierarchy (Frederik: 43.21% of the revisions are above the word level vs. Bart: 73.87%; see Table 4). Both writers partly maintain this writing style in the new writing mode, but we still see that the representation of the text starts to play an active role in the writing process. Contrary to the writing process that is typical for traditional dictators (Gould & Alforo, 1984; Schilperoord, 1996b) we now observe that recursivity becomes more apparent in the writing process of these writers. Nevertheless, both writers also show a different revision behavior (Frederik revises 3.41 words per 100 words vs. Bart: 0.93). If we assume that the mental revision is also reflected in the pausing behavior of the participants, then we can incorporate this factor in our explanation. Bart’s pausing time takes up about 60% of his writing process, while for Frederik this amount drops to 30% of the total time. This distribution of pausing and writing remains constant for both writers over the five observation sessions. The difference in pausing behavior is also in line with the learning styles of both participants: experimenting vs. reflective. Moreover, the distribution of the revisions in the two writing phases differs for both participants (Table 4). Frederik is a writer who — in accordance with his previous dictating experience — mainly revises in the second phase of the writing process, after he has finished a first draft of his text. Approximately 60% of his revisions are in the second phase. Bart, however, hardly revises in the second writing phase (only 5%). This kind of revising behavior closely matches the writing profile of the professional writer who uses a word processor to write his texts. These writers also prefer to revise their texts during the writing process of the first draft, leaving only few revisions for the rereading phase.

Table 4: Distribution of the revisions per level and per writing phase (five observation sessions). Frederik

Bart

%

SD

%

SD

Level Word Sentence Higher

56.79 35.90 7.31

13.52 5.73 11.08

26.13 35.80 18.07

26.94 44.67 31.35

Writing phase Phase 1 Phase 2

42.95 57.05

30.01 30.01

75.29 4.71

43.31 10.52


9/27/2006

4:34 PM

Page 289


18.4.

289

Conclusion

In this case study we have proposed an analyses of two professional writers who, while equally experienced in classical dictating, started working for the first time with speech recognition as a dictating device. Both writers are characterized by a different learning profile (accommodator vs. diverger). We wanted to learn more about the cognitive aspects of learning to write with a new writing mode. How do writers adapt to a new writing mode? And how do learning styles influence the initial use of speech recognition? The mode analysis illustrated that Frederik explores the possibilities of the new writing mode actively in the first writing sessions, but that he opts rather quickly for a selective and modest use of the speech mode. He uses speech recognition mainly to dictate his text and does not use it to solve technical problems or to revise his text. This pattern confirms the behavior characteristic of accommodators (trial-and-error and get things done). Moreover, during the completion of the text Frederik ‘evolves’ in his use of speech recognition to a writing pattern that is comparable to the writing processes of writers using classical dictating devices. He dictates his text and then adjusts it in the second writing phase by using keyboard and mouse. The text on the screen has a more passive, monitoring function and not a guiding function. Bart explores the different possibilities of the new writing mode more systematically and more patiently. His writing process is very reflective and is characterized by long pauses. During these pauses Bart actively rereads the text produced so far on the screen and when necessary he repairs it. When possible he tries to work with speech input. Bart uses the speech mode during 80% of his writing time, twice as much used by Frederik. In Bart’s case speech recognition is definitely not an elevated dictating device, but more a hybrid writing tool that combines elements of classical dictating devices with the classical computer mode (keyboard and mouse). In Bart’s adaptation process, quantitative movements between the writing sessions were rare. His profile remained stable over the different sessions. As a ‘diverger’ he explores the possibilities of speech recognition and develops a satisfactory way of working with speech recognition (22 words per minute) through reflection and observation. In studies by Karat (2000) and Halverson et al., (1999), the initial users produced 13.6 words per minute (keyboard and mouse ⫽ 32.5 words per minute) and the four experienced users who worked for 30 h with speech recognition produced 25.1 words per minute. These descriptions show that both participants have a different adaptation process, which evolves differently over time and seems to be driven by the learning styles of the participants. Consequently, the speech input mode has a different effect on the writing behavior of both writers. The characteristics of both writers are summarized in Table 5.

18.5.

Discussion and Further Research

In this chapter we described a case study and found that the participants hardly adapted their personal writing strategies to speech recognition. To obtain a more complete picture it would be interesting to compare the writing modes of speech recognition to keyboard and mouse in a more controlled experimental setting.


290

9/27/2006

4:34 PM

Page 290

Mariëlle Leijten

Table 5: Characteristics of the adaptation and writing process of Frederik and Bart.

Learning style Use of speech mode

Adaptation process

Revision behavior

Frederik

Bart

Accommodator Active exploration (initially) Selective use (mostly formulating) Gradual evolution of hybrid use to a more traditional dictating process Mainly in the second writing phase

Diverger Systematic exploration Dominant use (formulating and revising/repairs) Continue adaptation and hybrid use

Mainly in the first writing phase

In the present study the participants produced different text types, varying from a short routine letter to a complex petition. This observation was not considered problematic, because we wanted this study to have high ethnographic validity; the participants worked in their own environments and developed their own strategies. However, to compare the writing and adaptation strategies of writers in both the keyboard and mouse mode and the speech recognition mode, a follow-up experiment with a Latin square design with counterbalanced texts and modes of writing would have to be carried out. In such a study we would also like to include classical dictating devices as a third writing mode. Because, as was mentioned earlier, the focus of this overall study was not on the speech recognition itself but rather on its effect on different aspects of writing processes. The methods of analysis used in this study show that focusing on a new writing mode enables us to gain an insight into fundamental cognitive aspects of the writing processes. Or as Haas stated: “As is often the case, new situations – in this case new technological contexts and situations for writing – bring to the fore aspects of writing that may have been there all along, but that have not been previously noticed. […] – something with which writers have presumably been operating all along – becomes obvious in such a new situation or context: the changed technological context of composing with word processing.” (Haas, 1996, p. 117.) So, by observing and contrasting the writing process in the speech recognition, keyboard and mouse, and classical dictating mode, we would also hope to gain insights into certain aspects of the writing process in general. From a more methodological point of view we would like to raise two points. Firstly, for the purpose of this study we designed a categorization model analyzing writing modes, technical problems, and revisions (for a full description see Leijten & Van Waes, 2003). The variables of this categorization model were useful in studying the writing process at case level. However, if we want to analyze a larger set of participants these categorization models are too time-consuming to work with. Thus we need a more automated way of collecting and analyzing writing process data. Consequently, a point of particular interest with this kind of research lies in the automated logging of the writing processes. For this study


9/27/2006

4:34 PM

Page 291


291

we did not make use of any of the existing logging tools. Software like Trace-it (Severinsson, 1994, 1996b; Kollberg, 1998) and Scriptlog (Holmqvist, Johansson, Strömqvist, & Wengelin, 2002) make it possible to register, reconstruct, and analyze detailed online writing processes on several levels (including pause analyses). However, these existing logging tools have been developed for specific applications, and are hardly adapted to the current Windows environment and commercial word processors. Moreover, none of these logging tools have integrated the logging of the speech recognition mode. As such we did not use any of the above-mentioned software. Because of the speech input we were forced to broaden the existing methods. The manual way of analyzing the data was very time-consuming. In the near future we would like to integrate the use of speech recognition software with the logging program Inputlog that is currently being developed.2 This would facilitate data collection, description and analysis.

Acknowledgments We thank both participants for their willingness to learn how to work with speech recognition and for letting us observe them while doing so. We are grateful to Lernout and Hauspie for the free license to VoiceXpress LegalTM during the course of this study. We also thank Tom Van Hout for proofreading this chapter.

2

More information on the development of the logging tool Inputlog can be found on the website of the author: http://www.ua.ac.be/marielle.leijten & http://www.inputlog.net



9/27/2006

4:00 PM

Page 293

Chapter 19

Longitudinal Studies of the Effects of New Technologies on Writing: Two Case Studies James Hartley

This chapter contributes to the discussion about writing processes by discussing whether or not people’s writing styles change when new technologies are introduced. In Study 1 this was done by assessing samples of texts written by three prolific authors using different technologies over a thirty-year period. In Study 2 samples of typed word-processed letters were compared with dictated word-processed letters after their author had changed to using a voice recognition system. Thus, in both studies, there were great changes in the modes that the authors used to write. However, the results indicated in both cases that the authors’ individual styles of writing remained remarkably consistent over time. These findings suggest that, although new technologies change the ways in which experienced writers work, they do not necessarily alter their writing styles. The case for inexperienced writers, however, is likely to be different.

19.1.

Introduction

There is an ongoing debate in the field of writing research about whether or not computeraided writing programs will change the ways that people write and think. Undoubtedly, the new technology has changed — and will continue to change — the methods by which people write. But is this change only a cosmetic one? Is what we do with new technology basically much the same as what we did with the old technology? Or are the changes more fundamental than this? Well over one hundred studies must have been reported by now that examine the effects of new technology on writing (see references to six reviews in Hartley & Tynjala, 2001). Here, for the purpose of illustrating the various techniques used by different investigators, I outline the results from five separate enquiries. Kellogg and Mueller (1993) carried out two studies, the first with 16 college students who were relative novices at using a word-processor and the second with 69 more experienced Writing and Cognition: Research and Applications Copyright © 2007 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-045094-6

Hartley, J. (2007). Longitudinal studies of the effects of new technologies on writing: Two case studies. In G. Rijlaarsdam (Series Ed.), and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and cognition: Research and applications (Studies in Writing, Vol. 20, pp. 293–305). Amsterdam: Elsevier.


294

9/27/2006

4:00 PM

Page 294

James Hartley

users. In both experiments the participants were trained to classify their thoughts about their writing processes while they were writing. Then, when they were asked to do so by an experimenter, they had to press one of four keys to indicate what they were doing — planning, writing, revising, or some other process. After training and practicing for 30 min, the participants wrote an essay (within a 30 min time limit), half using a word-processor and half writing in long-hand. They were interrupted approximately every 30 s and they responded each time by pressing one of the four keys. Two judges rated the resulting essays using various measures of content and style. The results of the first study showed that there were no significant differences between the essays in terms of content, but that the hand-written essays were judged superior in style. It appeared with long-hand that the three processes of planning, writing and revising required similar amounts of time, but that with the word-processor more time was spent on planning and revising, and less time on the actual writing. The results of the second study were very similar, with the more experienced users of word-processors showing the effects slightly more clearly than the less experienced users who participated in the first study. Since the time spent on writing the essays was the same for both studies, Kellogg and Mueller (1993) concluded that using word-processors restructured the process of writing, but failed to improve writing performance. (More recent studies, developing Kellogg’s methodology, are described in Olive & Levy, 2002). Haas (1996, pp. 77–115) reported findings that were very different from those of Kellogg and Mueller (1993). Haas again worked with university students but in this study she used a more complex word-processing system. Haas (1996) found that her students spent less time planning when using a word-processor than they did when writing in longhand. The difference was particularly marked at the initial planning stage. More detailed analyses of the different kinds of planning used by the students in this study showed that there was less conceptual planning and more sequential planning with word-processing than with handwriting. Haas commented that a likely explanation for these results was that the students began writing sooner and spent less time planning with the word-processor because making changes was easier with the word-processor. Oliver and Kerr (1993) studied the grades obtained by 240 student teachers who submitted essays in either a handwritten, typewritten or word-processed form. In addition, these students completed a questionnaire on their writing strategies. The results indicated that the students who used a word-processor obtained higher marks than did those who used a typewriter or wrote their essays in longhand. Subsequent analyses revealed, however, that this was mainly an artefact of the amount of revision that the students did when they were writing their essays. Students who revised more, irrespective of which group they were in, obtained higher marks. But, of course, as word-processors facilitate revision, it is possible that they helped some students to achieve higher marks than they might have achieved without them. Zellermayer, Salomon, Globerson and Givon (1991) used a specially designed computing tool, called The Writing Partner, in their study. This presented learners with a suite of programs that were based on work carried out by Bereiter and Scardamalia (1987). The programs provided three kinds of help during writing: (a) memory support; (b) advice about evaluation and coherence; and (c) guides to remind the writer of rhetorical elements, writing goals and the writer’s audience. All three kinds of help were largely contingent


9/27/2006

4:00 PM

Page 295

Effects of New Writing Technologies

295

upon the writer’s input — that is to say the program ‘chose’ when to intervene, depending on what the students wrote. Zellermayer et al. (1991) studied 60 thirteen–fifteen-year-old pupils who were divided into three groups. Group 1, the control group, wrote five essays with a conventional wordprocessor. Group 2 wrote five essays using The Writing Partner. Group 3 wrote five essays with a version of The Writing Partner that provided the same guidance, but this time this guidance only appeared at the writer’s request. Two trained judges assessed ‘blind’ the resulting essays. The findings showed that the students who used the original Writing Partner (Group 2) produced significantly better essays overall than did either the control group or the group who called on the programs when they wished (Groups 1 and 3). To test whether or not the computer-aided writing programs had changed writing performance, members of all three groups completed another essay, using pen and paper, two weeks later. The assumption was that if writing performance had been changed then this would show itself on such a post-experimental writing task. The results were very similar to those described above. The essays written by Group 2 were again significantly better than the essays written by Groups 1 and 3. Crinon and Legros (2002) studied three groups of 10-year-old children (with 18 participants in each). These children were asked to write a story and then to improve it after looking at eight model texts. Group 1 studied the model texts individually using a computer database, Scripertexte, a tool designed to help re-writing, Group 2 read the texts on paper, and Group 3 were not provided with any model texts. The experiment took place in three periods, each of an hour and 15 min, several days apart (after a practice session). Analyses of the original and the final versions of the texts showed that the children in Group 1 increased the number of propositions used in their texts significantly more than did the children in Groups 2 and 3, who did not differ significantly from each other. Crinon and Legros (2002) attributed this finding largely to the nature of Scripertexte, rather than to novelty or experimenter effects. The results from these five enquiries are thus mixed. There is some evidence that the new technology can facilitate various aspects of writing, but there is limited evidence to suggest that the nature of writing itself changes. Much appears to depend upon the particular participants, the writing tasks, the technology used and the research methods employed. Furthermore, each of the five studies described above took place over a relatively short time period. What might the evidence suggest if we analysed the writing styles of authors over much longer periods? Suppose, for example we compared pieces written by the same authors over lengthy periods using different technologies? It is this question that this chapter addresses.

19.2.

Study 1: Hartley, Howe and McKeachie (2001)

In this study, each of the three authors (Hartley, Howe and McKeachie), all prolific academic writers, wrote a brief biographical passage describing how their methods of writing had changed over the previous 30 years. They then chose samples from their work for analysis spread over this time period. These samples were chosen within particular genres — e.g.,


296

9/27/2006

4:00 PM

Page 296

James Hartley

research reviews, book chapters or journal articles. Three or four printed pages from each of these samples were scanned into a computer and converted into computer-based text. These were then checked for accuracy and certain irrelevant features were removed (such as titles and preliminaries, abstracts, subheadings, tables and lengthy quotations from other authors). The resulting texts were then analysed using the readability software provided with the standard Microsoft package ‘Office ’97’. 19.2.1.

Results: The Experience

Panel 1 provides a summary of the autobiographical accounts of the three authors in terms of their modes of writing. All three authors have published extensively in the field of educational psychology. The first author (myself) has published over 300 papers and several books in the field. The second author (Howe) likewise published several articles and books before his untimely death in 2002. The third author (McKeachie) has been a distinguished contributor to the educational psychology for over 60 years. A past president of the American Psychological Association, his Teaching Tips is currently in its 12th edition. The accounts given in Panel 1 indicate that each of these authors has experienced considerable changes in modes of writing over time.

Panel 1: Changes in methods of writing over time for the three authors in Study 1. James Hartley 1980. I wrote my first draft in longhand — very roughly — and then I re-wrote it again in longhand maybe once or twice more. My secretary then typed it out double-spaced (on a manual typewriter). I then edited this ‘printed’ version. This usually entailed moving paragraphs about, or even deleting them, and re-phrasing ideas and sentences. My secretary then re-typed what I had written, and we continued this process until I was satisfied or I dare not give it back to my secretary again! 1990. I wrote my first draft in longhand — very roughly — and then I re-wrote it again more neatly. My secretary then word-processed the manuscript and printed it out doublespaced. I then edited the print-out fairly drastically by hand. My secretary then made the changes for me on the word-processor and then re-printed the document. This procedure then re-iterated, perhaps several times. 1993. I wrote my first draft in longhand and then I word-processed it, edited it on screen and printed it out double-spaced. I then edited the print-out by hand, and then re-processed it, and so the process continued until I was satisfied. 2000. I now sketch a few preliminary notes or headings on the back of an envelope or rough sheet of paper, and then I compose directly ‘on screen’. I word-process the text, and I edit it as I go along, and again after completing a section, article or chapter. I incorporate ideas from concurrent reading on the topic as I am writing. I print out initial versions to help with the editing process, and I keep re-iterating this process of composing on screen, editing, adding information, printing and revising until I am satisfied.


9/27/2006

4:00 PM

Page 297


297

Michael Howe Up to 1980. Chapters were usually written out in note form or in a preliminary draft and I would type out the first, roughly grammatical, version. Next I did a lot of cutting and pasting. Then, the material would be typed out again, usually by someone else. This would effectively be the final draft in most cases. 1980–1990. From the early eighties some or other word-processing method would be used. This was slow and primitive at first. 1990. From about 1990 onwards the only major changes have been in speed. Once wordprocessors became available, I usually used them from just about the earliest drafts, although in some cases these would be preceded by written notes, and occasional sentences would be tried out in longhand. 2000. With today’s word-processing facilities almost all my writing is done directly on the computer. A particular piece of writing is likely to be re-read and changed on up to a dozen occasions. Compared with 20 years ago something written by myself nowadays will typically have gone through considerably more drafts/versions before it appears in print. Wilbert McKeachie 1940–1960. I originally wrote everything out by hand, and my secretary typed successive drafts as I interlined, crossed out, added inserts, etc. 1960–1980. I began to dictate some things into a dictaphone and the secretary typed the first draft. I used dictation in the 1960s and continued with it until the 1970s and 1980s, but this was more likely to be the case with speeches than with articles, which I still wrote out by hand. 1990–2000. In the last decade I’ve been doing everything on the computer. Now I do the first draft on the computer and do a good deal of revision on the computer, but I still like to go over a hard copy and revise it in long-hand before making the revisions on the computer. Sometimes I do two or three drafts with long-hand corrections before the final draft. Almost everything I do now is co-authored, even if I do the first draft and most of the revising.

19.2.2.

Results: The Products

Table 1 presents the results of the analyses of the writing samples for each author and shows the results across time for average sentence lengths, percentage of passive sentences and two readability scores. (Note that for the Flesch scores, the higher the score, the easier the text, whereas for the Flesch-Kincaid score, the lower the grade, the easier the text. Flesch scores are also banded in terms of difficulty: see Hartley, 1994, for details.) The samples analysed in Table 1 for the first author were all drawn from review chapters that I wrote between 1972 and 2000 and which were published in edited collections (see Hartley et al., 2001, for details). The table shows that there is a remarkable consistency in average sentence lengths and in the Flesch-Kincaid Grade Levels over time. The percentage of passive sentences is initially high (35%) in the 1972 chapter, but this then levels off to an


298

9/27/2006

4:00 PM

Page 298

James Hartley

average of 18% if this first chapter is discounted. There is thus a temptation to suggest that the passives start high and end low, but this is reading too much into the data — for it can be seen that the second (1978) has the lowest percentage of passives (13%). The last chapter (2000) has the highest Flesch score, which is encouraging, but again a score as high as this was obtained earlier (in 1987). Whatever the case, these data do not suggest any marked changes in the products of writing over time, despite the changes in my writing methods documented in Panel 1.

Table 1: Readability statistics for samples from materials written over many years for the three authors. Hartley Chapter date 1972 1978 1982 1987 1995 No. of words sampled 1032 978 990 1125 1014 Average no. of sentences 45 45 38 48 42 Average no. of words/sentences 23 22 26 23 24 Average % passive sentences 35 13 21 14 21 Flesch indexa 42 36 39 50 47 Flesch–Kincaid grade levelb 12 12 12 11 12

1999 2000 1026 931 42 41 24 23 19 17 44 53 11 10

Average 1014 43 24 20 44 11

Howe Chapter date 1972 1977 1980 1984 1989 1994c 1998 Averaged No. of words sampled 1377 1687 849 1449 1479 1209 1320 1360 Average no. of sentences 47 61 37 61 56 51 58 53 Average no. of words/sentences 29 28 23 24 26 24 23 26 Average % passive sentences 27 21 32 21 25 17 24 25 Flesch indexa 33 33 32 44 38 55 52 39 Flesch–Kincaid US grade levelb 12 12 12 12 12 11 11 12 McKeachie Article date 1958 1968 1976 1984 1987 1997 Average No. of words sampled 1378 1397 1422 1579 1458 1448 1447 No. of sentences 60 63 64 76 63 68 66 Average no. of words/sentences 23 22 22 21 23 21 22 % passive sentences 15 12 9 15 1 11 11 Flesch indexa 38 31 37 49 34 32 37 Flesch–Kincaid US grade levelb 12 12 11 12 12 12 12 a Flesch scores range from 0 to 100. The higher the score the more readable the text. Scores below 30 are designated ‘very difficult’ to read, and scores between 49 and 30 ‘difficult’ (see Hartley, 1964). b Add 5 to obtain an approximate reading age. c This chapter differs from the others in that it comes from a text specifically written for parents. This extract contained no academic references. The chapter was also co-authored, unlike the others. d This overall average excludes the data from the 1994 chapter (see Note 3).


9/27/2006

4:00 PM

Page 299


299

The samples analysed in Table 1 for the second author (Howe) were all drawn from chapters in books written by Michael Howe between 1972 and 1998, with one co-authored chapter in 1994 (see Hartley et al., 2001, for details). The results show that there are some variations in the average sentence lengths but that these are not time-related. The FleschKincaid scores are consistent for all the samples. The jointly-authored chapter is clearly more readable (according to the Flesch) than the other chapters and this is of interest because the book from which this chapter came was specifically written for parents, unlike the other books which were more standard academic texts. These data then do show changes according to the target audience, but they do not suggest marked changes in writing styles over time — despite the changes in Howe’s writing methods documented in Panel 1. The samples analysed in Table 1 for the third author (McKeachie) were drawn from articles written by Wilbert McKeachie and published between 1958 and 1997 (see Hartley et al., 2001, for details). Three of these articles took the form of presidential addresses (in 1958, 1968 and 1976) and two were general research overviews originally delivered as talks (1984, 1987). The last paper (1997) was a critique of a set of papers written by others and, in this sense, is rather different. The data show that there is an amazing consistency in terms of average sentence lengths and Flesch-Kincaid grade levels over time. Furthermore (with the exception of the 1984 paper) the percentage of passive sentences is extremely low for academic papers of this kind and this is clearly a distinguishing feature of this author. These data again do not suggest any marked changes in writing style over time, despite the changes in McKeachie’s writing methods documented in Panel 1.

19.3. Study 2: Hartley, Sotto and Pennebaker (2003) In this second study that examines the effects of new technology on writing I focus on yet another longitudinal study, but this time I compare products written over a historically shorter period of time. For some years now my colleague Eric Sotto (ES) and I have maintained a regular correspondence largely about academic matters. About six years ago, in one of his letters, ES noted with a mixture of frustration and triumph that he was using a new method of writing. He had bought a new computer, and had switched from typing at a keyboard to using a voice-recognition system (Dragon NaturallySpeaking, Version 5, ‘Preferred’). In my reply to ES, I remarked that I had not noticed any difference in the style and form of that particular letter, and indeed, that I had been surprised by ES’s information. Much discussion of ES’s use of this new writing technology then followed in subsequent letters. Here then were the seeds of a natural experiment. I had kept copies of ES’s earlier and later correspondence. Hence, once a suitable number of the dictated letters had been written and received (but not before), I proposed to ES that we carry out a study on the possible differences between the typed letters and the dictated ones. Accordingly, the last 14 of the earlier letters and the first 14 of the later ones were scanned into a computer, and then re-keyed to achieve a standard format. This involved, for example, stripping all the letters of their various opening and closing salutations, Americanising UK spellings and standardising features like dates of the month, and the


300

9/27/2006

4:00 PM

Page 300

James Hartley

spelling of words like ‘program’ and ‘word-processor’. I then used two text-analysis programs to assess the writing styles in the two groups of letters. These were Microsoft’s Office ’97, and Pennebaker’s Linguistic Inquiry and Word Count (LIWC) (Pennebaker, Francis & Booth, 2001). The LIWC calculates the percentages of words used in any one text in any one of the 74 different linguistic categories. Some of these separate categories are grouped — for example into emotional words (e.g., ‘happy’, ‘sad’, ‘angry’), self-references (e.g., ‘I’, ‘we’), and cognitive words (e.g., ‘realise’, ‘think’, understand’). However, for my study to be complete, I also needed to examine the effects of this change on the writer. After all, ES had bought the software — and gone through the timeconsuming and often frustrating process of teaching himself how to use it — not because he thought it would have any effects on his writing style but because he thought it would make his writing physically easier. 19.3.1.

Results: The Experience

An examination of the dictated letters showed that ES encountered many difficulties when he first tried to use his voice-recognition system. Some of these are listed in Panel 2. ES clearly found that writing with a voice-recognition system was very different from writing with a keyboard. Learning how to use the voice-recognition system required a great deal of effort (Letter 1); and then, even after much practice, ES found that the methods used for constructing longer sentences were very different in the two systems. More exactly, changing the structure of a sentence in ‘mid-flight’ was experienced as relatively easy when typing, but much more difficult when dictating (Letter 4). On the other hand, ES found that dictating was physically much easier than typing at a keyboard; and he also found dictating was more pleasant because it felt ‘more like chatting’ (Letter 7). Panel 2: Excerpts from the correspondence between ES and JH on the experience of using a voice-recognition system. Letter 1 It might interest you to know that I am dictating the contents of this e-mail. I am using a program called Dragon Naturally Speaking. I have done quite a lot of work with it, and it now catches about 85% of what I dictate. I then have to go over what I have done and do quite a bit of correcting. That’s far easier than typing everything out with two fingers! I’d have to correct anyway, especially when dictating as I am not used to it, and it is a medium that requires quite a different cast of mind compared with tapping away at a keyboard. Letter 2 I was surprised to read that you did not notice any change in my style of writing. I experience dictating as very different from writing. It feels far more contrived, in fact stilted, and I am often surprised to see my sentences coming out reasonably correct grammatically speaking. At the same time, I find dictating physically far easier than tapping at a keyboard with two fingers.


9/27/2006

4:00 PM

Page 301


301

Letter 4 On reflection I see that when I compose while typing, I have no fears that I shall not be able to complete a sentence in a grammatically acceptable way. This is because, as I type, I frequently change things. Indeed, it might be worth adding that when I am writing something which I consider important, I sometimes go over a paragraph a great many times. Most of these semi-conscious feelings are absent when dictating. I find that I concentrate a very great deal more, and attempt to get the sentence structure and wording correct right from the beginning. In part this is probably due to the fact that I do not yet know how to do all the correcting through dictating. This is a matter of learning a considerable number of commands and procedures by heart, and I am bad at such things, and loath to spend the time required for it. … Here is another matter which I think you might find intriguing. It so happens that I have a poor visual memory, and only an average ear; my brain seems happiest with the abstract. Perhaps this explains why I have always been a poor speller. Whatever the case, I find it highly pleasurable that, when I dictate with this programme, I make no spelling mistakes! I would not bother to mention this were it only a personal matter, but I imagine that you will immediately sense that this may well have an implication in a school setting. Letter 5 Now that you have drawn my attention to the matter it’s obvious that dictating is a skill that one has to learn. Even so there are times when I just cannot get a sentence right via dictating, even if I do lots of corrections. I sense that there is here a substantial difference, but don’t have the linguistic tools to analyse it. … Letter 6 I don’t think that the move from typing to dictating — when using a word-processor for both — is anywhere near as big as the leap from using a typewriter to using a word-processor. It is the latter which is the huge leap, and, as I see it, the main difference lies in what is involved in correcting. Letter 7 No, I don’t think e-mail has changed my writing habits. If I now write more, I think it’s partly because dictating is far easier than typing, and perhaps because it is more like chatting to someone. Letter 17 I am struck by your finding that there are more long sentences in the typed letters. I do hope you will find that there is a significant difference here because I think it highlights a difference I experience in using keyboard or microphone. That is, and as I think I have mentioned, when typing there is no need to try and hold the whole of a sentence in one’s mind, whereas in dictating, because one is attempting to avoid the need for correcting, one does seek to do that. And as holding a short sentence in one’s mind is easier than holding


302

9/27/2006

4:00 PM

Page 302

James Hartley

a long one, it wouldn’t be surprising if the sentences in dictated letters was (were) shorter. Incidentally, I had to concentrate very hard when dictating the previous sentences! (And I left ‘was’ as it is what the program picked up.) In thinking about the above, I found myself ruminating that youngsters may well find dictating more difficult than using a keyboard. That is, dictating strikes me as a higher order skill than typing, and this may have important consequences. Letter 19 You may be amused if I tell you that I am dictating this with my new dictation program, the one I bought when I was recently in the UK, the one made by IBM called ViaVoice. I installed it and tried to get the hang of it when I got back, and must have spent several days with it. I gave up eventually in total disgust because I found it awkward to use and it kept crashing. A few days ago I decided I would have another go, and I’m slowly getting on top of it, and this e-mail is one result. I still find it cumbersome, but I’m beginning to suspect that this may be the result of comparing it with a different program with which I am now pretty familiar, perhaps a common phenomenon. However, its recognition accuracy appears to be somewhat better than the previous program, so I am persisting. Of course, I mention these mundane details to put our current paper into perhaps better perspective. The materials presented in Panel 2 indicate that moving to the voice-recognition system had marked effects on ES’s experience of writing. The task was found to be difficult at first, and to require a great deal of practice. Then, even with practice, it appeared that it was still difficult to construct long or complex sentences. Nonetheless, ES gave no hint of wishing to revert to the use of a keyboard at any point. Indeed, after a lengthy paragraph of complaint about the difficulties of using the new voice-recognition technology when it kept crashing, and the difficulty of mastering the intricacies of correcting a text by voice, ES concluded: ‘But I mustn’t grouse. I couldn’t have used a dictation program on my old computer, and I really am pleased to be able to use one now’ (Letter 5). Later ES even purchased another, more sophisticated voice-recognition system (Letter 19). 19.3.2.

Results: The Products

The main results of the statistical enquiries on the products are presented in Table 2. The data show that there were no significant differences in the average length, the number of paragraphs and the number of sentences in the typed and the dictated letters. Nor were there any significant differences between the products in readability, or the number of typographical and grammatical errors made. However, the dictated letters contained significantly shorter sentences on average, and had significantly fewer especially long sentences (i.e. those containing more than 50 words). The dictated letters also contained more instances of the use of the first-person pronoun and certain other textual features (as shown in Table 2). These data suggest that using the new technology did have some effect on ES’s actual writing. However, on the whole, these effects were small. The data from the LIWC program suggested that there was a significantly greater use in the dictated letters of first-person pronouns,


9/27/2006

4:00 PM

Page 303


303

Table 2: Summary statistics for various measures on the keyboarded and the dictated letters. Keyboarded letters (N ⫽ 14)

Dictated letters (N ⫽ 14)

1165 (537) 13.6 (5.7) 56 (28) 21.3 (3.3) 64.2 (3.7) 9.2 (1.2)

1051 (782) 12.9 (7.7) 54 (36) 19.1 (1.7) 64.9 (2.7) 8.7 (0.7)

Use of 1st person singular References to self Use of causal expressions References to the past

5.2 (0.8) 5.7 (0.8) 1.2 (0.8) 2.5 (1.1)

6.6 (1) 6.9 (1) 1.7 (0.3) 3.7 (0.8)

Number of ‘long sentences’c Number of typos Number of grammatical errors

2 (0–5) 0 (0–5) 0 (–)

0.5 (0–5) 0 (0–5) 0 (0–3)

Number of words Number of paragraphs Number of sentences Number of words/sentence Flesch scorea Kincaid Grade Levelb

t ⫽ 0.43 t ⫽ 0.27 t ⫽ 0.19 t ⫽ 2.07, p ⬍ .05 t ⫽ 0.56 t ⫽ 1.29 t ⫽ 3.80, p ⬍ .001 t ⫽ 3.51, p ⬍ .005 t ⫽ 2.37, p ⬍ .03 t ⫽ 3.16, p ⬍ .004 U ⫽ 39, p ⬍ .02 U ⫽ 92 U ⫽ 77

Note: Values are for means, with standard deviations in parenthesis, except in last three rows where they are for median values, with ranges in parenthesis. a The higher the Flesch Reading Ease score the more readable the text. b The lower the Grade Level the more readable the text. c Defined as sentences containing more than 50 words.

references to the self and to the past and of causal expressions. In addition, not shown in Table 2, but possibly of interest, is the fact that there were significantly more uses in the typed letters of words relating to numbers, humans, occupations and school. But in all of these cases, the differences between the mean scores were very small, and there were no overall differences in pooled categories of cognitive and emotional words. Perhaps of most relevance, however, is the fact that both sets of letters were highly readable and, despite the fact that ES felt that his dictated letters were more chatty (Letter 7), there was no significant difference in reading ease between them. The fact that the dictated letters had significantly shorter sentences overall is interesting, especially in the light of comments by ES quoted in Panel 2. However, the mean difference between the sentence lengths in the two groups of letters was only that of two words. And furthermore, although there were a greater number of longer sentences in the typed letters, the median difference of only two longer sentences per letter suggests that there were very few excessively long sentences in any one letter (as each letter contained on average a total of 56 sentences). The two measures used in this study (Office ’97 and LIWC) thus suggest that there were some small differences in the surface features of the letters. However, it is more than likely that these differences resulted from the topics under discussion rather than from the writing technology used. For example, in the letters that ES dictated, a good many of the sentences were about his experiences with the new voice-recognition system because he was aware that I was interested in this matter, so it is perhaps hardly surprising that there were more self-related words here. Programs like Office ’97 are not really sophisticated enough


304

9/27/2006

4:00 PM

Page 304

James Hartley

to detect subtle differences in style and content, although LIWC is better (Pennebaker & King, 1999). What is required in future studies of this kind are more complex measures than those currently available — measures that depend more on the actual words and sentences used in a particular context, and on the linguistic structures of the assessed texts. Such measures are available but not all of them (at the time of writing) are fully computerised (e.g. see Vidal-Abarca et al., 2002; Whissell, 1999).

19.4. Discussion Limitations on space prevent me from discussing many of the issues raised by these two studies. Readers can find more extended discussions in the original papers on which this chapter is based (Hartley et al., 2001; 2003). Here I consider just four issues: (i) limitations of case studies; (ii) problems of measurement; (iii) issues arising from studying experienced writers; and (iv) implications for further research with less experienced ones. 19.4.1.

Limitations of Case Studies

In both studies the findings reported above are limited by the fact that they are based upon data derived from single-case studies, and it may not be wise to generalise from them. (Similar criticisms apply to the chapter by Leijten, this volume.) Different individuals may produce different results. Chandler (1995), for instance, reports that commentators noted that Henry James wrote in a much briefer and crisper style before he took to dictating; and that James himself said, “I’m too diffuse when dictating”. At this stage we do not know how, or even if, differences between writers affect the ways in which they use new technology (but see van Waes & Schellens, 2003, and the chapter by Williams, Hartley & Pittard, this volume, for broader discussions). 19.4.2.

Problems of Measurement

It will not have escaped the reader’s critical eye that the measures used in the studies cited above to assess the effects of changes in modes of writing have been somewhat limited, if not to say crude. Investigators have a choice here. Either one can opt for small sample sizes and detailed analyses of text, or for larger samples and more reliable methods of counting (for this is what computer-based methods currently supply). Perhaps it would be more realistic to say in the studies reported above that the measures used describe stylistic variances in the texts (or lack of them) rather than stylistic quality (as is sometimes suggested in this paper). It might, in future, be interesting to combine approaches — using the blunter methods to discover major effects, and more detailed qualitative studies to tease these out in more detail. 19.4.3.

Issues Arising from Using Experienced Writers

The data given above — in both studies — show that, although the authors differ from each other as writers, they are remarkably consistent over time. They each display what Levy and Ransdell (1996) intriguingly call ‘writing signatures’. Investigators of other anonymous


9/27/2006

4:00 PM

Page 305


305

similar works written by them could possibly tell who wrote what, especially if more sensitive measures were used. Gudjonsson and Haward (1998) provide an interesting discussion of these issues in context of detecting authorship in the legal domain — an area they call ‘forensic linguistics’. The issue that interests me more here, however, is where do these ‘writing signatures’ come from? If writing styles are set early on, when and how is this achieved? Studies of undergraduates by Levy and Ransdell (1996), Lavelle (2003) and — to a lesser extent — by Torrance, Thomas and Robinson (1999, 2000) suggest that writing signatures are well established at university level. However, other investigators draw attention to how the written language of undergraduates changes as they progress through their degree courses and become more skilled participants in the language communities that make up their disciplines (e.g., Lea & Street, 2000). New technologies might influence this ‘hardening’ of style at this stage. 19.4.4.

Implications for Further Research with Inexperienced Writers

Although the data presented above support the ideas that new technologies do not change the written output of experienced writers, it is unlikely that this will be true for young children learning to write. But there is little research to support such a claim. As noted earlier, the findings in this respect are inconclusive. Most of the studies reviewed, of course, were done some years ago, and things have moved on — some would say dramatically — since then. And none have worked yet with children who have been using new technology from birth, as it were. But, even today, we have voice recognition and text messaging systems (almost) freely available to our children. This means that the kinds of errors that children will make in their writing, and that the kinds of feedback that they will receive will be different from that received with conventional word-processors (e.g., see O’Hare & McTear, 1999). With text messages from a thirteen-year-old like this: “My smmr hols wr CWOT. B4, we used 2go2 NY 2C my bro, his GF & thr 3 kids FTF. ILNY, it’s a gr8 plc.” (“My summer holidays were a complete waste of time. Before, we used to go to New York to see my brother, his girlfriend and their three kids face to face. I love New York. It’s a great place.”) I don’t think that it takes much arguing to agree that our children’s ways of writing (and reading) are going to change, and that new technology will play a prominent role here….

Acknowledgements I am indebted to Chris Woods and Andrew Knipe for technical support and to colleagues and referees for helpful comments on earlier versions of this paper.



9/28/2006

6:49 AM

Page 307

Chapter 20

Learning by Hypertext Writing: Effects of Considering a Single Audience versus Multiple Audiences on Knowledge Acquisition Elmar Stahl, Rainer Bromme, Marc Stadtler and Rafael Jaron

We examine the process of writing hypertexts by learners and its impact on the acquisition of knowledge. Our main research question relates to how students deal with the multi-linear structure of hypertexts in a way that promotes learning. We present a study in which students were asked to construct a hypertext document by linking prepared nodes in two subsequent sessions. One group of students had to change the anticipated audience perspective between the two sessions, while the other group had to consider one single perspective in both sessions. Cognitive processes were investigated using a computer-based assessment system (CEKOS). The resulting hypertexts were analyzed with regard to the link structure of the documents. Learning effects were measured through five knowledge tests. The results confirm that considering multiple audience perspectives is an appropriate instruction to deal with the nonlinear structure of hypertexts in a way that promotes knowledge acquisition.

20.1.

Introduction

Today, the Internet plays an important role in educational contexts. In a growing number of projects in schools and universities, students are asked to publish their own documents on the Internet, e.g., in the form of short hypertexts. Accordingly, it is worth considering learning effects of hypertext writing1 by students (see also Braaksma, Rijlaarsdam, Couzijn, & Van den Bergh, 2002; Talamo & Fasulo, 2002). In our research project we examine effects of different instructions on the process of hypertext writing and on the resulting knowledge about the subject matter processed (Bromme & Stahl, 1999, 2001, 2002; Stahl, 2001).


Stahl, E., Bromme, R., Stadtler, M., & Jaron, R. (2007). Learning by hypertext writing: Effects of considering a single audience versus multiple audiences on knowledge acquisition. In Rijlaarsdam, G. (Series Ed.); M. Torrance, L. van Waes, & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 307–321). Amsterdam: Elsevier.


308

9/28/2006

6:49 AM

Page 308

Elmar Stahl et al.

We consider that writing hypertexts can be seen as a design act (Dillon, 2002) that places special constraints on the design of documents that are due to features of this text format: nodes, links, and multi-linear structure. We assume that these constraints could support a writing process that promotes learning and can be compared with the “knowledge-transforming model” described by Bereiter and Scardamalia (1987). In their model, they propose that writing can only contribute to knowledge acquisition when a text is formulated within a continuous interaction between content-related knowledge (on the topic addressed in the text) and rhetorical knowledge (on the design of the text and, among others, its structure). This problem-oriented procedure (see also, Hayes, 1996; Hayes & Flower, 1980, 1986; Kellogg, 1994) requires hypertext producers to reflect on and to extend their own knowledge. This view on learning by writing is a helpful heuristic for analyzing conditions and processes of learning by writing hypertext. In terms of this model, the features of hypertext might support the knowledge-transforming process as follows: 1. Writing nodes require decisions on how to discriminate between concepts so that they can be presented as separately comprehensible text units. This contributes to deeper understanding of concepts and conceptual differences within a subject matter. 2. Links present a possibility to indicate semantic relations between concepts. To set links in such a mindful way, the author has to become aware of semantic relations existing between main concepts presented in different nodes. 3. In order to plan the hypertext’s structure, the author has to comprehend the structure of the contents. Owing to their multi-linearity, hypertexts can be read in a variety of ways. Hence, an author has to anticipate possible audience2 perspectives in order to create flexible reading paths. This may increase not only comprehension of semantic structures of the content but also a flexible use of the new knowledge. The last expectation is based on Cognitive Flexibility Theory (CFT) (see, e.g., Jacobson & Spiro, 1995; Spiro, Feltovich, Jacobson, & Coulson, 1991). CFT deals with how knowledge about a complex (ill-structured) content domain can be acquired in a way that ensures its flexible use. The goal is to stimulate learning transfer and to avoid “inert knowledge,” that is, knowledge a learner can reproduce but fails to apply in new situations (Bereiter & Scardamalia, 1987). Cognitive flexibility refers to this transfer of knowledge and is defined correspondingly as the ability to structure one’s own knowledge in a variety of ways in adaptation to changing situational demands (Spiro & Jehng, 1990). In a previous experiment we adopted the idea of CFT in the writing of hypertexts (Bromme & Stahl, 2002). Asking the learner to construct a hypertext from different thematic perspectives would correspond to the core assumption of CFT about the importance of learners’ activities. Thinking about different audience perspectives should contribute to the construction of multiple mental representations of the domain. Our experiment examined the effects of taking into account two audience perspectives on processes, products, and the writers’ acquisition of knowledge. We investigated this issue by focusing on the process of setting links. Participants were not asked to write the nodes themselves, but to construct hypertexts by linking prepared nodes. Node writing was not included for methodological reasons: It would increase the variance in the constructed hypertexts and in the construction processes. This might make it hard to detect the effects we were interested in.


9/28/2006

6:49 AM

Page 309

Learning by Hypertext Writing

309

The topic chosen for the study was in the domain of Internet. We manipulated the adoption of audience perspectives: Group I had to link the nodes from two anticipated audience perspectives in succession. In the first session, they were asked to write a hypertext for an audience interested particularly in the history of the Internet; in the second session, for an audience interested particularly in the services offered by the Internet. Members of Group II, in contrast, were told to write hypertexts in both sessions without any specific audience perspective being pointed out to them. They were asked to orient themselves toward the subject area as such and develop what they thought would be an optimal structure for fictitious readers who had to learn with the hypertexts. In the second sessions, participants in Group II were told they were getting a second chance to produce an improved hypertext starting with the same nodes as before. Our results showed significant effects on the process of writing hypertext as well as on knowledge acquisition. The demand to create hypertexts for readers with different interests led Group I to consider the structure of the hypertext more intensively. Evidence for this conclusion was found in the hypertext structures as well as in the process data. For example, participants in Group I expressed a greater number of different semantic relations between the nodes with their links. Additional support for our interpretation came from participants’ statements on their decisions that we recorded every 2 min during the construction process. Being asked to adopt certain audience perspectives made participants reflect a lot more on the structure of both the hypertexts and the contents they were dealing with. The requirement to take anticipated audience perspectives into account apparently supported the process of linking the semantic and the rhetorical structure. In line with the assumptions of CFT, Group I gained significantly higher scores in knowledge test about relations and in transfer tests.

20.2.

Research Questions

The study reported here is a replication of this experiment. In the previous study we had decided to orientate the control group toward the subject matter as such. As a result of this decision, it remained an open question whether it is actually the provision of several perspectives that makes the critical difference. It is also conceivable that adopting an audience perspective per se (i.e., requiring participants to work from only one perspective) would already have facilitated learning in a similar way. Furthermore, we aimed to replicate the results within the new domain medicine. Finally, we wanted to test a new instrument for assessing the cognitive processes and decisions during the construction process (CEKOS3). In the previous experiment we had asked the participants every 2 min, what they were thinking about. So data collection had to be carried out in individual sessions. In the study reported here we used a computer-assisted assessment that allowed us to run the experiment in group sessions. In this study students were also asked to write two hypertexts by linking prepared nodes in two sessions. The nodes dealt with the topic “causes, diagnosis, and therapy of cancer.” Group I had to change the anticipated audience perspective between the two sessions: interest in the causes of cancer in the first session and diagnosis and therapy of cancer in the second session. Group II was asked to consider one single perspective in both sessions. They had to construct a hypertext for an audience that was especially interested in diagnosis and


310

9/28/2006

6:49 AM

Page 310

Elmar Stahl et al.

therapy of cancer. In the second session, participants in Group II were told they were getting a chance to produce an improved hypertext constructed for the same audience by starting with the same nodes as before. In correspondence with the results of our previous study we formulated the following hypotheses: Hypothesis 1 Concerning the overall number of links we expect an interaction effect between group and session. No differences within their pairs of hypertexts should be found in Group I. Participants should set a similar number of links in both hypertexts. In Group II significant differences should be found instead. Students should set significantly more links in their second hypertext compared to their first one. Rationale: Group I had to figure out two different hypertext structures to take the different audience perspectives into account. Assuming that both perspectives are comparable with regard to their semantic content, it can be expected that this task leads to a similar number of links within their pairs of hypertexts. Group II had to structure their hypertexts from the same audience perspective and to improve its structure in the second session. One interpretation of “improving its structure” is to set more links, because setting more links can be seen as a possibility to show more semantic relations between the nodes. Hypothesis 2 Concerning the distribution of unique links the participants are setting in their pairs of hypertext we also expect to find an interaction effect: no significant differences between the pairs of hypertexts should be found in Group I. In contrast to this, in Group II we expect to find a higher proportion of unique links in their second hypertexts. Figure 1 presents a visualization of the expected proportions of unique links for the Group I and Group II. Rationale: We distinguished between identical links, defined as links that were identical in both hypertexts of one participant and unique links, defined as links that could only be found in one of the two hypertexts of one participant. For example, if a participant set a link from the node “carcinogen” to the target node “cell” in the first hypertext and repeated the same link in the second hypertext, it was counted as “identical”. If this link was set in one of the hypertexts but not repeated in the other one, it was counted as “unique.” For each hypertext we determined the proportion of unique and identical links. Comparing the pairs of hypertext in Group II, we expected to find more unique links in their second hypertexts. If the participants in this group used the structure of their first hypertext as an orientation and tried to optimize it with more links, it could be expected that they would structure their hypertexts with identical links they had used on the previous day and then would differentiate it with more links. So they should set more unique links in the second hypertext compared to the first one. In Group I the participants had to consider two different audience perspectives. This means that the participants had to emphasize the relations that are important from the respective perspective in both hypertexts. So in both hypertexts there should be a similar number of unique links. Accordingly, we expected to find no differences in the proportion of unique links between these pairs of hypertexts.


9/28/2006

6:49 AM

Page 311


311

Group I

unique links Session I

identical links

unique links Session II

Group II

unique links Session I

unique links Session II identical links

Figure 1: Expected proportions of unique links in Group I (upper part of the figure) and Group II (lower part of the figure). Hypothesis 3 Adapting two different audience perspectives should lead to more statements on planning the structures of the hypertexts and their contents within the second session. Rationale: After every 2 min during the whole writing process, all participants had to give statements on what they were actually doing at that moment (see Section 20.3). We expected no differences in the number of statements on the structure of the hypertexts and its contents between the two groups for the first session. In this session both groups had to anticipate an audience perspective and to plan their hypertexts accordingly. For the second session, we assumed that we would find more statements on planning the structures of the hypertexts and their contents within Group I: if Group II “adapted” their structure from the first session it should be easier for them to plan the second hypertexts. Group I had to anticipate a new audience perspective and to find a different hypertext structure instead. So they had to plan it more carefully again. Hypothesis 4 Group I should gain higher scores in the knowledge tests than Group II. Rationale: The reflections about two different audience perspectives foster a deeper understanding of the subject matter. The students have to reflect different relations and structures between the nodes in the two sessions. This contributes to gaining more


312

9/28/2006

6:49 AM

Page 312

Elmar Stahl et al.

knowledge about the concepts, their relations and structures, and also leads to more flexibly applicable new knowledge.

20.3.

Method

20.3.1.

Participants

Fifty college students (36 female and 14 male) of different subject areas (42 students of psychology and 8 students of economics, law, geography, or pedagogy) participated in the study. Participation was voluntary and rewarded financially with 20 Euro. We wanted to examine knowledge acquisition processes of learners without prior knowledge of the domain. Therefore, a criterion for participation was no prior in-depth knowledge on the topic of cancer, examined with a pretest. Six subjects were excluded post hoc from all analyses, since they scored more than 20% in this test. Thus, 44 subjects were left, 20 of them in Group II, 24 of them in Group I. The mean age of these participants was 23. 20.3.2.

Materials

20.3.2.1. Instruments Data were collected in a computer laboratory equipped with 15 Windows 95© PCs connected to 17-in monitors. The program used for writing the hypertexts was Microsoft Frontpage Express 2.0©, an easy-to-learn graphics editor. A specially developed computer program CEKOS was used to assess the participants’ cognitive processes as they were working on the task of hypertext writing. The standardized assessment of cognitive processes by means of a computer program is a major difference compared to our previous experiments in which participants were asked, every 2 min, to report on what they had been thinking about just then. Thus, we shall provide more information about CEKOS in what follows. CEKOS: The computer program CEKOS is a Microsoft Access© application which combines a database with a form that pops up at regular intervals and asks users to enter information into the form. Using Microsoft Access©, the concrete appearance of the application with respect to the number and content of categories can be tailored to the individual needs of the researcher. In our case we used a category system that combines the more general categories of planning, translation, and revision derived from the models of Hayes and Flower (Hayes, 1996; Hayes & Flower, 1980, 1986) and Kellogg (1994) with the hypertext-specific classification system developed by Stahl (2001, 2002). Table 1 provides an overview of the categories. By introducing this combination we sought to achieve a finer granularity with regard to the assessment of cognitive processes. In agreement with van den Bergh and Rijlaarsdam (this volume) we argue that this is a necessary step to make process-data-analysis more effective. Stahl (2001) found that activities related to cognitive processes during hypertext writing varied over time. To make sense of these activities it was necessary to introduce fine-grained subcategories. CEKOS uses directed retrospection to assess the cognitive processes of the participants. This method was applied by Kellogg (1988, 1994) in his experiments on text production. He trained his participants to handle a category system, and then asked them to use it to


9/28/2006

6:49 AM

Page 313


313

Table 1: Contents of the category system used in CEKOS. Main category A. Planning

B. Translation

C. Revision

Specification Reading and comprehending the contents Planning relations between contents Planning the hypertext structure Setting a link Adding new contents Making notes and sketches Revising the comprehension of contents Revising relations between contents Revising the hypertext structure

D. Other

Figure 2: Screenshot of the window for categorizing the construction process. report their activities at set time intervals while working on the task. Thus, participants categorized their cognitive processes themselves rather than having them categorized by subsequent raters. CEKOS enables participants to report on their activities by choosing between the predefined categories presented in Table 1.


314

9/28/2006

6:49 AM

Page 314

Elmar Stahl et al.

To this end, a window appears on the screen at a regular 2-min interval (see Figure 2). This window overrides all the other programs used at the time and blocks their use. Participants are asked to categorize the decisions made at this point in time in terms of the predefined category system. The first step for participants is to report whether they are currently engaged in planning, translation, or revision, or whether they are thinking about something else. To help them decide, short definitions of these categories are given in a text window. Once participants have selected one of the main categories, a submenu opens in which the choice has to be specified more precisely. Three subcategories — planning, translation, and revision — are available for each of the categories. At all times, participants can activate only one of the subcategories with the mouse. Once participants are satisfied with their choice, they have to click an “OK” button that appears only after one of the subcategories has been selected. This closes the window and re-enables the other programs. Students can now continue to work on their hypertexts. This procedure starts off the next 2-min interval. The results of the self-rating are entered directly into a Microsoft Access database so that they are available for immediate analysis. 20.3.2.2. Introductions to hypertexts and to the task Participants were familiarized with the text format “hypertext” using a written explanation in a hypertext format. This explanation employs a metaphor of virtual information spaces; nodes, as individual information locations; and links, as pathways between these locations. This introduction had been used in previous experiments and was shown to be suitable to explain the features of hypertexts. The hypertext also contained the task instructions, which differed for Group I and Group II. Participants in Group I were asked to construct a hypertext for readers who were mainly interested in the causes of cancer in the first session and to construct a hypertext for readers who particularly wanted information about the diagnosis and therapy of cancer in the second session. Participants in Group II were told to construct their hypertext for readers particularly interested in the diagnosis and therapy of cancer in both sessions. It was argued that students in previous experiments often liked to rewrite their hypertexts, and that they now had an opportunity to improve the structure of their hypertext by linking the nodes again in a second session. Nodes on the topic of cancer. One of the main goals of the present study was to examine whether the results of our previous experiments could be replicated within a new domain. Therefore, we chose the topic cancer, since it provides a large number of semantically rich concepts that can be viewed from different perspectives. In total, 16 nodes of about 100 words were written using medical specialist literature throughout. The 16-node texts, each explaining one concept in more detail, were presented on the computer as HTML files and as printouts on 16 file cards. 20.3.3.

Procedure

Data were collected in group settings in two sessions on two consecutive days. Each session lasted 2 h. On day one, participants started by completing pretests on their knowledge of cancer. After having read the introduction about hypertext, they were trained in the use of the HTML editor until the commands necessary to link the nodes had been mastered. Then they received the 16 nodes on file cards and had 15 min to read the texts to gain an


9/28/2006

6:49 AM

Page 315


315

overview of their content. Afterward students also received the nodes as computer files and had 60 min to create their first hypertext. In the second session, participants started by completing a test on their knowledge on computers (Kühne, 2000). After that, they once again had 1 h to link the nodes to a hypertext in line with the given instructions. In both sessions participants had to categorize their cognitive processes with CEKOS at a regular 2-min interval. They were not given the hypertext they had constructed on the previous day. In both hypertexts all 16 nodes had to be included. The session ended with the knowledge tests. In both sessions, participants had to work for the whole 60 min. 20.3.4.

Dependent Variables

The effects of the different instructions on hypertext production and on knowledge acquisition about cancer were examined on different levels: Structure of the constructed hypertexts: The students’ hypertexts were analyzed in terms of the set links. This included the total number of links and the number of unique links within the pairs of hypertexts (see Hypotheses 1 and 2). Process data: Students’ cognitive processes were recorded via CEKOS as explained above. The frequency of categories chosen was scored for each of the subjects (see Hypothesis 3). Knowledge acquisition: We assessed whether considering different audience perspectives had any effects on learning. In five subtests, knowledge about the contents of individual nodes, about relations between nodes, and transfer knowledge was assessed. Content knowledge was assessed with 10 multiple-choice questions on the contents of individual nodes. This test was also used as the pretest to assess the prior knowledge of cancer. Knowledge about relations was assessed with four items presenting a logical relation (e.g. “is a”, “belong to the same class”). Students had to select from a number of given concepts, five pairs, to which this kind of relation could apply. Furthermore, 10 multiplechoice questions on the relations between concepts had to be answered. Transfer knowledge was assessed with two further subtests. Drawing on Jonassen (1993), we designed 10 items asking students to complete analogies between concepts. The missing concept could be chosen from a list of five. The second transfer test consisted of an open question on the prevention of cancer that had to be answered in the form of a short essay.

20.4.

Results

Results are presented for the pretests and each of the hypotheses mentioned above. We defined an -level of .05 as significant for all statistical tests. Whenever we hypothesized to find no differences we defined an -level of ⬎.20. 20.4.1.

Pretest

The subtest used to examine prior knowledge about cancer revealed no significant differences between the groups (Mann-Whitney U test: z ⫽ ⫺.32, p ⫽ .75). From a maximum


316

9/28/2006

6:49 AM

Page 316

Elmar Stahl et al.

of 10 points, Group I scored an average of .88 (SD ⫽ .80); Group II .80 (SD ⫽ .83). Hence, after the exclusion of all students that gained more than 20% correct answers in the pretest, there were no differences in prior knowledge between the groups. 20.4.2 Structures of the Hypertexts Hypothesis 1 Overall number of links: In the first session, Group I students set an average of 1.85 (SD ⫽ .69) links per node compared with 1.51 (SD ⫽ .41) links per node in Group II. In the second session, students in Group I placed an average of 1.88 (SD ⫽ .64) links per node; those in Group II, 2.05 (SD ⫽ .98). An ANOVA with repeated measurements revealed no significant differences between the two groups, F(1,42) ⫽ 1.97, p ⫽ .66. However, significantly more links were set in the second session compared with the first, F(1,1) ⫽ 8.44, p ⫽ .005 and the expected interaction was significant as well, F(1,42) ⫽ 1.38, p ⫽ .013. The number of total links in the pairs of hypertexts differed for participants Group II, t(19) ⫽ ⫺3.05, p ⫽ .007, but not for students in Group I, t(23) ⫽ ⫺0.31; p ⫽ .76. Therefore, Hypothesis 1 could be confirmed. Hypothesis 2 Distribution of unique links: Group I set an average of 62.2% (SD ⫽ 19.11) unique links per node in the first session compared with 62.6% (SD ⫽ 16.11) unique links per node in the second session. Group II set an average of 48.5% (SD ⫽ 17.99) unique links per node in the first session compared with 61.2% (SD ⫽ 13.74) unique links per node in second session. These results are summarized in Figure 3. An ANOVA with repeated measurements revealed no significant differences between the groups, F(1,42) ⫽ 2.64, p ⫽ .11. Yet, significantly, more links were set in the second 65

60

55 %

group I group II

50

45

1

Session

2

Figure 3: Percentage of unique links set by Group I and Group II in sessions I and II.


9/28/2006

6:49 AM

Page 317


317

session compared with the first, F(1,1) ⫽ 9.11, p ⫽ .002 and the expected interaction was significant as well, F(1,42) ⫽ 8.21, p ⫽ .007. The number of unique links in the pairs of hypertext differed for students in Group II, t(19) ⫽ ⫺.127, p ⫽ .002, but not for students in Group I, t(23) ⫽ ⫺.003; p ⫽ .90. Therefore, Hypothesis 2 could be confirmed. Summarized, the adoption of different audience perspectives had effects on the hypertext structures that were similar to the effects found in the previous study. Group I set a similar number of links in both hypertexts, but they used a high proportion of unique links. Group II set significantly more links in their second hypertext compared to their first one. They seemed to orientate themselves toward their first hypertext and to optimize the structure by including more links. This led to a higher proportion of unique links in the second hypertext. The results indicate that students in Group I tried to structure both hypertexts in line with the demands of the respective audience perspective. Group II in contrast seemed to try to optimize their second hypertext by a more differentiated link structure compared to their first hypertext. 20.4.3.

Cognitive Processes During Hypertext Construction

Students’ statements on their decisions and strategies were classified by a system of 10 categories (see description of CEKOS). Table 2 gives an overview of the mean number of statements in the 10 categories. Hypothesis 3 Statements on planning the structure of the hypertext and their contents: In CEKOS we have two categories that are relevant to test this hypothesis: planning the relations between contents and planning the hypertext structure. For each session we calculated the sums of the statements in these categories. Table 2: Mean number of statements in 10 categories of CEKOS. Category

Mean (SD) Group 1

Comprehension of contents Planning relations Planning hypertext structure Setting a link Adding new contents Making note Revision of content Revising relations Revising structure Other

Group 2

Session 1

Session 2

Session 1

Session 2

2.21 (2.50) 5.42 (3.64) 2.29 (2.14) 5.96 (2.49) 5.79 (3.64) 0.58 (1.18) 0.58 (1.02) 1.08 (1.59) 1.75 (2.95) 1.25 (2.03)

1.58 (1.82) 4.89 (2.88) 3.18 (1.90) 7.58 (5.83) 4.20 (3.78) 0.80 (1.51) 0.60 (1.19) 0.85 (1.09) 0.90 (1.48) 2.20 (2.51)

3.50 (3.20) 5.65 (4.00) 2.50 (3.02) 5.70 (3.10) 6.17 (4.73) 1.12 (1.65) 0.29 (0.55) 0.79 (1.18) 0.63 (1.10) 1.29 (1.37)

3.00 (3.81) 3.95 (3.17) 1.60 (2.01) 6.75 (3.16) 6.05 (4.54) 1.20 (2.12) 0.70 (1.26) 1.15 (1.81) 1.10 (1.48) 1.45 (2.26)


318

9/28/2006

6:49 AM

Page 318

Elmar Stahl et al. 8.5 8.0 7.5 7.0

group I

6.5

group II

6.0 5.5 1

Session

2

Figure 4: Number of cognitive processes related to planning in Group I and Group II. In the first session, Group I made an average of 7.71 (SD ⫽ 4.70) statements in these categories compared with 8.15 (SD ⫽ 4.74) statements in Group II. In the second session, Group I made an average of 8.04 (SD ⫽ 2.99) statements; those in Group II 5.55 (SD ⫽ 3.85). Figure 4 summarizes these results. An ANOVA with repeated measurements revealed no significant differences between the two groups, F(1, 42) ⫽ .86, p ⫽ .36. There was a tendency towards more statements in the second session compared to the first, F(1, 1) ⫽ 3.91, p ⫽ .055. The interaction between session and group was significant, F(1, 42) ⫽ 6.55, p ⫽ .014. As expected, the groups did not differ in the first session, t(42) ⫽ 0.31, p ⫽ .76. But in the second session Group I made significantly more statements than Group II, t(42) ⫽ ⫺2.42; p ⫽ .02. Therefore, Hypothesis 3 could be confirmed. The task of constructing a hypertext from two different audience perspectives therefore involved an intense dealing with the structures of the hypertext and the content during both sessions. 20.4.4.

Knowledge Acquisition

To examine the effects of taking into account different audience perspectives on knowledge acquisition we calculated 10% trimmed means (see Wilcox, 1998). This is a method to counter the heterogeneity of the results that are typical of learning experiments. Thus, we excluded the two best and two worst participants in each group. Table 3 reports the distribution of results separately for both groups regarding content knowledge, knowledge about relations, and transfer knowledge as well as their overall scores. Hypothesis 4 Scores in the knowledge tests: A MANOVA across the three areas revealed significant differences between the groups, F(3, 32) ⫽ 5.70, p ⫽ .002 (Hotelling-Lawley Trace). Subsequent ANOVAs revealed a significant difference for content knowledge, F(1, 34) ⫽ 6.64, p ⫽ .015, and for knowledge about relations, F(1, 34) ⫽ 13.31, p ⫽ .0009.


9/28/2006

6:49 AM

Page 319


319

Table 3: Ten-percent trimmed mean scores on the knowledge tests. Mean (SD)

Contents Relations Transfer

Group 1

Group 2

7.80 (1.11) 24.25 (2.79) 12.45 (2.61)

6.75 (1.34) 21.06 (2.35) 11.13 (1.96)

Furthermore, Group I showed a trend to higher transfer knowledge as well, F(1, 32) ⫽ 2.84, p ⫺ .10. General computer knowledge, which was controlled as a covariable, showed no systematic relations to knowledge acquisition. Therefore, as anticipated in Hypothesis 4, the intervention of structuring the hypertext from two different perspectives supported knowledge acquisition. The groups differed in their knowledge about the contents of single nodes and about semantic relations. We also found a tendency toward differences in transfer knowledge.

20.5.

Discussion

The present experiment analyzed how far taking into account multiple audienceperspectives help participants to deal with the demands of hypertext design arising from its multilinear structure. In accordance with CFT and the results of a previous study, we anticipated that adapting multiple perspectives would support the acquisition of deeper and more flexibly applicable knowledge about the subject matter. As in the previous study, we asked students to write hypertexts by linking prepared nodes. The main difference in this study was the task of Group II. In our previous experiment they had to produce hypertexts oriented toward the subject matter as such. In this experiment they had to construct the hypertexts for an anticipated audience with one special perspective. This design was chosen to test whether the results of the previous experiment could be explained by the different audience perspectives the learners had to consider, or just by taking into account an audience perspective as such. Thus Group I had to consider two different audience perspectives and Group II one audience perspective. Furthermore, we changed the domain of the nodes from Internet to cancer to see whether our results were domain-independent. Finally, we tested a new method for ascertaining the cognitive processes during the writing process. Our results show that taking into account two audience perspectives impacts significantly on the process of hypertext writing as well as on knowledge acquisition. The demand of creating hypertexts for readers with different interests led Group I to consider the structure of the hypertext more intensely in both sessions. Group II, in contrast, used a strategy for their task of optimizing their second hypertexts by developing a more differentiated link structure compared to their first hypertext. This strategy led to less intensive reflections on the structures of the hypertexts and their contents during the second session.


320

9/28/2006

6:49 AM

Page 320

Elmar Stahl et al.

Evidence for this conclusion can be found in the hypertext structures as well as in the process data. The analysis of the hypertexts reveals that Group II set significantly more links in their second hypertext compared to the first one. Also, the proportion of unique links was significantly higher in the second hypertext. These results show that Group II used a similar hypertext structure as in their first session and differentiated it with more links. Thus, their strategy of optimizing the second hypertext was to (simply) set more links. In Group I the total number of links and the proportion of unique links did not differ within the pairs of hypertexts. These results indicate that students in Group I tried to structure both hypertexts in line with the demands of the respective audience perspective. This indicates that Group I thought more carefully about the hypertext structure in both sessions. Additional support for our interpretation comes from students’ statements on their decisions: in the first session, both groups did not differ in their number of statements about planning the structure of their hypertext and the contents. In the second session, Group I made significantly more statements about planning the structures than Group II. The task of constructing a hypertext from two different audience perspectives therefore involved a more intensive dealing with the structures of the hypertexts and its contents during both sessions. In Group II, planning activities decreased in the second session. Merely setting more links does not seem to require intensive reflections on the hypertext structure. In our research project, we argue that writing hypertexts may promote learning if the learner relates the content structures to the rhetorical structure of the respective text format (knowledge transforming, according to Bereiter & Scardamalia, 1987). The requirement to take into account multiple audience perspectives apparently supports the process of linking the semantic and the rhetorical structure. Accordingly, considering multiple audience perspectives also fostered knowledge acquisition. Group I gained significantly more knowledge about contents of single nodes and their relations and showed a tendency for more knowledge in the transfer tests. These results are in line with the results of our previous study and with the assumptions of CFT. It is interesting that we have found only a trend toward superior transfer knowledge in Group I. This is remarkable since Group I, in our previous study, outperformed Group II especially with regard to transfer knowledge (Bromme & Stahl, 2002). A possible explanation for these results might be that the tests used for assessing the transfer knowledge were not sensitive enough to uncover existing differences due to the differing subject matter (medicine vs. Internet). All students had schemata about medicine. Therefore, answering the transfer questions might have been a combination of the new knowledge and the pre-existing schemata. In the previous experiment, it is unlikely that the students — who had little knowledge about the technical and historical background of the Internet — had any schemata about Internet that could have helped them in such a way. Another aspect of the study was to test the computer-supported tool CEKOS. CEKOS allowed us to collect data in group experiments. Furthermore, it was not necessary to transcribe and rate verbal data, as we have done before. This first application of CEKOS successfully showed that it is possible to ask the learners to assess their decisions by themselves. In further studies it will be necessary to test the effects of this method on the decision processes of the participants. It can be assumed that each method for assessing cognitive processes, such as thinking aloud (e.g. Jansen, van Waes, & van den Bergh, 1996),


9/28/2006

6:49 AM

Page 321


321

or methods of directed retrospection (e.g. Kellogg, 1988), such as CEKOS, can have an impact on coping with the task. Writing processes can undergo major assessment-induced changes, because what are usually routine mental processes are made continuously explicit and conscious (Becker-Mrotzek, 1997). It might be interesting to conduct further studies to shed some light on the effects of assessment methods on the resulting data. Summarizing, we replicated our results that prescribing multiple audience perspectives proves to be an appropriate instruction that employs the nonlinear structure of hypertexts in a way that can promote learning. This instruction was more effective than the orientation toward the subject matter as such (shown in the previous study) as well as toward a single audience perspective. Accordingly, writing hypertexts may well be a useful means of supporting the acquisition of knowledge.

Endnotes 1 The notion of writing hypertext is used here in a broad sense. It is defined as encompassing all activities of editing a hypertext, including writing a hypertext from scratch as well as rearranging existing hypertexts by setting new links, integrating new nodes, or transforming traditional texts into hypertexts. 2 With audience perspectives we mean thematic perspectives of readers determining how they read (hyper-)texts. 3 The German abbreviation stands for “Computer-based assessment of cognitive processes while writing hypertexts”.



9/27/2006

4:07 PM

Page 323

Chapter 21

Supporting Individual Views and Mutual Awareness in a Collaborative Writing Task: The Case of Col•laboració Henrry Rodriguez and Kerstin Severinson Eklundh

It is common that co-authors divide the responsibility for a document according to its sections, expecting to optimize the collaborative process. Doing so they can concentrate their attention on certain sections, such as those in which they have more expertise. This strategy could lead to other problems, however. For example, co-authors’ activities could be isolated from the teamwork and they could perceive the writing task in different ways. Col·laboració is a Web-based system that supports collaborative writing activities. The system provides co-authors with different mechanisms (email notification, modularity of a document, and rapid access to a document’s sections) that enable them to have different views of the writing task. Nevertheless, Col·laboració does not detach the co-authors from the teamwork, avoiding to isolate his/her activities from the team, but supports a global perspective of the task. The system has been used in ten different writing tasks. Evidence that co-authors might need different views of the writing task was found. How these mechanisms were used and evaluated by the users is also reported.

21.1.

Introduction

To agree upon which strategy should be used to produce a written document successfully is essential (Allen, Atkinson, Morgan, Moore, & Snow, 1987). Members have to coordinate their efforts according to different strategies: parallel, sequential, and reciprocal (Sharples, Goodlet, Beck, Wood, Easterbrook, & Plowman, 1993). A common practice is to partition a document into sections and make each co-author responsible for a sub-set of them. A co-author would be expected to put more effort into, and would work more closely


Rodriguez, H., & Severinson Eklundh, K. (2007). Supporting individual views and mutual awareness in a collaborative writing task: the case of Collaboració. In Rijlaarsdam, G. (Series Ed.) and M. Torrance, L. van Waes & D. Galbraith (Volume Eds.), Writing and Cognition: Research and Applications (Studies in Writing, Vol. 20, pp. 323–334). Amsterdam: Elsevier.


324

9/27/2006

4:07 PM

Page 324

Henrry Rodriguez and Kerstin Severinson Eklundh

on, those sections he/she is responsible for. To put it another way, co-authors might be more coupled to some sections than to other sections in the same document. In principle, co-authors can handle a document in two different ways: (1) As isolated sections: those that he/she is responsible for, leaving out the rest of the document. This might correspond with a parallel strategy. (2) As an unbreakable unit with all the sections, even those that he/she is not responsible for. This might correspond with a sequential strategy. Both practices present some problems. In the first a co-author works more independently of the team. As a result, other co-authors might be poorly informed about his/her activities, and interaction in the group may decrease. Problems with regard to language style, flow, and establishing clear relationships between ideas may occur when a co-author integrates a particular section into the rest of the text. In the second practice the co-authors are required to handle all the sections in the same way, regardless of which are the sections he/she is responsible for, demanding more effort to accomplish the task. This problem is more severe if the co-author is responsible for sections that are not physically adjacent in the structure of the document (e.g., the Introduction and the Conclusion sections in a scientific paper). In general, the problem presented in this chapter is related to how a collaborative writing tool helps each co-author to focus on his/her work without detaching him/herself from the teamwork, and from the document as a whole. To achieve a global perspective is crucial during any writing process (Severinson-Eklundh, 1992), and presents particular challenges in a collaborative writing task. The approach taken here is that every co-author should be able to have an individual view of the writing task and at the same time should be able to get effortless access to the whole panorama of the writing task. It is important to point out that the strategies indicated here are not the only ones that can be used. During a writing task, co-authors can move from one strategy to another, or they could design a particular strategy fitting their particular needs. A basic problem in collaborative writing is that the communication process among coauthors is generally detached from the document itself (e.g., by using email or phone). In fact, an interview study with academic writers showed that co-authors did not have a good network infrastructure in common, which could support their collaborative writing (Kim & Severinson Eklundh, 2001). Our approach for developing Web-based tools for collaborative writing has focused on the communication needs of co-authors, sharing documents and materials, and awareness of the ongoing work. Col·laboració is a prototype that is designed to alleviate the problems co-authors might have while handling a document, using the strategies presented above. Col·laboració helps users to focus on the sections that they are responsible for, but does not isolate them from the rest of the document. The system offers a novel way of interaction with Web documents, viewing a document as a set of independent modules (sections) but keeping their relation as a whole unit, that is, as a document. In the next section we will present the system, focusing on the features that contribute to tackle these problems.


9/27/2006

4:07 PM

Page 325

Supporting Collaborative Writing

21.2.

325

Col·laboració: Overview of the System

In this section the system Col·laboració is presented. We explain the basic functions of the system and the mechanisms that the system uses to support collaboration. At the end of this section, we present some other systems that use the Web as infrastructure for collaborative writing and compare them generally with our system. 21.2.1.

Col·laboració in Brief

Col·laboració is a Web-based collaborative writing tool that takes into account the need for a dialogue among co-authors during the writing process, regardless of the strategy selected by participants. The system enables users to make text comments on shared document sections, and to access them in a shared dialogue space. The system can present different views of the shared document according to individual co-authors’ needs. Email notification is sent to co-authors for relevant events, such as when a comment or a new section has been added. It is important to bear in mind that in Col·laboració the term “document” is treated in a particular way. The approach that we have selected to define a document is related to the structure of documents. We assume that documents are divided into sections (e.g., defined or separated by headings or titles). A document is defined here as follows: A document is a set of HTML files related to each other. Each HTML file represents a distinct self-contained portion of the document called a section (module). A section, being a separate file, can be added, changed, or deleted independently of the other sections. Merging the sections produces the document as a whole. Col·laboració presents the document’s structure to co-authors displaying the titles of its sections. The titles are hypertext links that once activated show the section’s content and its comments. The system is primarily designed for collaborative reviewing of a document being produced in a group, and not for actual composition. In particular, it supports sharing of the document and the discussion among co-authors that the production of the document might demand. However, it is also possible to edit the content of the sections through the system. When the user enters the system, the Web browser window is divided into four frames as follows (see Figure 1): The left frame displays the list of links to the document sections (index-frame). The right frame is divided into three frames — the top-right frame displays the content of the currently selected section (content-frame); the middle-right frame displays the comments made so far on the section that is presented in the content frame (comment-frame); the bottom-right frame displays a set of buttons that represent the commands that can be performed with the system (command-frame). The index-frame is divided into three parts: The first is the title of the document; the only item of the list that is not a hypertext link. The second is the standard areas part (cf. infra). Col·laboració automatically creates the items of these two parts when a document is created. Finally, the third part is the document sections part, made up from the different sections’ titles. Items in the document’s sections will be added by coauthors as the work progresses. In other words, the index-frame reflects the structure of


326

9/27/2006

4:07 PM

Page 326


Figure 1: The layout of the system Col·laboració in a Web browser.

the document. Also, every index item shows the number of comments a section has received so far. From our experience using an early version of the system, it was clear that there were some common areas that should be included by default for every document that was going to be produced using this system. We have called this the standard area part and it consists of three titles, each being a hypertext link. Each title presents a document in the contentframe and they are labelled as follows: “About this prototype”. This title shows what we have called the “About this prototype” (ATP) document. The ATP document briefly summarizes the features of the system and how to use it. Its dialogue space is intended to receive co-authors’ comments on the system. In this way we can keep the comments relating to the system and those relating to the content of the document in separate spaces. “Versions of this document”. The activation of this link displays a document consisting of a list of the versions made by the co-authors. Users can select one or more sections of the document and make a version of it. Observe that by selecting all the sections in the document section area, the user can create a version of the whole document. “Ideas for this paper”. This space is designated to be a dialogue space for a meta-level discussion about the writing task during the planning stages. The objective of this section is primarily to offer co-authors a shared space designated for idea generation and co-ordination of the writing task.


9/27/2006

4:07 PM

Page 327

Supporting Collaborative Writing 21.2.2.

327

How Col·laboració Supports Collaboration

Co-authors can create the sections that will constitute the document and include them in the system. The system will update the index-frame, appending the new section to the document section list. The co-author that creates a section has to indicate from a list who are the coauthors that are responsible for the section just created. Who is responsible for a particular section, can be discussed face-to-face or remotely using the Ideas for this paper’s commenting space. This decision can be based, for example, on core competences of the co-authors. Comments can be added on every item of the standard sections part and document sections part. The comment is immediately shown appended to the comment space for that section and it is then available to all co-authors. In addition, the content of the comment is automatically sent by email to the co-authors responsible for the section. The order in which the document sections are presented in the index can be changed at any time by co-authors. It could be, for example, that the Abstract is appended to the document’s structure toward the end of the writing process. Co-authors could then locate the Abstract section at the very beginning of the document structure or to any other place in the section list. The system has one command to edit the content of a section. Co-authors can make changes to the HTML source in a separate window and then submit them. Any change made to a particular section will be available to the rest of co-authors immediately. Any section can be deleted from the document together with its comments. Col·laboració provides a novel function that allows users to create personalized views of the document: the Overview function. Co-authors can select different sections and their comments to obtain an individual view of the document, or to create a version of the document. The order of the sections will be kept in the same sequence that they have in the index-frame. Note that the user could select two or more sections that do not necessarily have to be adjacent in the index-frame. The “Ideas for this paper” section can also be included here. This overview presents the information in a format suitable for printing, e.g., it provides a margin for annotation in the hard copy. The system has a logging feature that registers every action that changes the content of any of the frames in the system. Mouse movement, frame scrolling, and frame resizing are not registered. Once a command is executed in the system, information about who did it, when, and what is registered in the log files. Col·laboració is a prototype for collaborative writing and not a finished application ready for the market. There are points that need to be expanded, improved, and examined more closely. However, Col·laboració is a working collaborative writing tool despite all its shortcomings. The most important weakness is related to the translation phase from ideas to text. In other words, editing the document text is clumsy and laborious. Version management is also very limited in the system. So far, the system allows a user to make a version of the document but automatic recovery is not supported. 21.2.3.

Systems that Use the Web as Infrastructure for Collaborative Writing

In the last decade, a few other systems have been developed to support collaborative writing through the infrastructure of the Web. As the Web is extensively used in organizations,


328

9/27/2006

4:07 PM

Page 328


it is natural to use the same environment for the common practice of collaborative writing. The possibility to access the Web from different places makes Web-based collaborative systems important. An additional point is that Web formats for documents like HTML and XML are widely used and supported, which facilitates the exchange of information among co-authors. The communication needs that co-authors might have or could also be mediated by the Web. Furthermore, the Web is a vast information space that co-authors might use to help reach their goals in writing. In this section, we shortly review some Web-based collaborative writing tools from a functional perspective. We mainly compare their approach to document sharing, communication, and annotation with the approach that we have taken in the design of Col·laboració. 21.2.3.1. Document sharing A key factor for collaboration is the possibility to share the object of interest by members of the team. This aspect is supported by all Web-based collaborative writing tools. An example of such tool is the widely used Basic Support for Collaborative Work (BSCW) system (Appelt and Busbach, 1996). Access rights in BSCW can be set on a per-object basis to control the operations available to different users. However, the complexity of a collaborative writing task requires more than sharing and setting access rights. Col·laboració does not only share the document among co-authors, but also supports handling the document through planning and reviewing (these two stages in particular) and translation. These are the three basic processes of writing according to Flower and Hayes’ (1980) cognitive model. Assigning access rights to a document implies that role definitions have to be put into action. Previous studies have shown that collaborative writing can benefit from role definition among co-authors (Beck & Bellotti, 1993; Neuwirth, Kaufer, Chandhok, & Morris, 1990). Alliance (Romero & Decouchant, 1997) defines roles that are related to the document parts (also called fragments in the system). In contrast, in Col·laboració the control of the whole document is shared with all the co-authors. This means that co-authors’ roles in the writing task are not prescribed by the system, but are expected to be defined socially. One reason for this choice is related to our research interest. If all co-authors have the same rights, observations about the specific functions of Col·laboració that support coupled and uncoupled awareness (see below Section 21.4) do not have to confront role issues. Furthermore, there is always a social structure that participants are aware of, for example, in relation to supervisor/student in our case studies. Alliance exploits special knowledge about the internal hierarchical structure of its documents to provide concurrence control on document sub-trees. Using the log file of Col·laboració, it can be detected when a user is working on a particular section and alert co-authors about the situation. 21.2.3.2. Communication Another fundamental factor in collaborative work is the way in which communication is supported by collaborative writing systems on the Web. Most of them use text-based communication. A very common feature that text-based communication systems support is threading of conversations. Threading refers to arranging the sequence of contributions according to their subject matter, by using the “reply-to” relationship as an ordering principle. In newsgroup readers, for example, users can select to read a particular thread, and the threading of discussions is emphasized by graphical means such as indentation.


9/27/2006

4:07 PM

Page 329


329

Ceilidh (Hughes, Jake, & Okelberry 1998) uses threaded messages for communication. The content of each comment is presented one by one and a navigation bar through the comments is available. The document is a file attached to the comments. Ceilidh automatically converts text, including carriage returns, URLs, and email addresses in the body of the message into HTML code. Threading requires an initiating post with a subject or topic that users consider as a representative one. In Col·laboració, the section of the document that is commented on gives the subject of the discussion. By creating another subject, the discussion deviates from the original purpose. Moreover, in a threaded discussion, participants cannot see at the same time the conversation as a whole and the message that originated the discussion. The lack of an overview of the replies and the original message might cause the next contributor to concentrate more on the last read reply than in the original message. As a result, the focus of the discussion may be lost. Therefore, we decided to present the comments in a chronological order and no threading is supported for presenting the comments in the dialogue space. 21.2.3.3. Annotations and comments EquiText presents the document as a table in which each row is a paragraph. Each column gives information about the content of the paragraph, who included it, when, and the possibility to add an annotation. It does not show the structure of the document as a whole. It gives the possibility to place a paragraph according to the position the author considers is needed. In Col·laboració, the structure of the document is always present and easy to access, presented in the index-frame. Other systems like CoNote (Davis & Huttenlocher, 1995), Annotea (Kahan, Koivunen, Prud’Hommeaux, & Swick 2001), the Anchored Conversations system presented by Churchill, Trevor, Bly, Nelson, & Cubranic, 2000, Microsoft Office 2000 that uses Web annotations (Cadiz, Gupta, & Grudin 2000) and WebAnn (Brush, Bargeron, Grudin, Borning, & Gupta, 2002) also support annotation of Web documents. An interesting design issue for theses systems is that annotations can be anchored at pre-designated spots in the document. The annotations are shown embedded in the document, at the point that they were made. The Anchored Conversation system provides a synchronous text chat window that can be anchored to a specific point within a document, moved around like a post-it note, and searched via a database. The possibility to link a comment to a particular part of the document allows users to easily associate comments with a particular paragraph, phrase, or word in the document being discussed. A common problem of linking an annotation to a “picky” part of the document is that these annotations can become, what has been called, an orphan annotation. Orphan annotations, one of the most frequent complaints in the study made by Cadiz et al., 2000, appear when the text to which the annotation is linked changes, e.g., the text is deleted. The annotation is out of place in the document. In our system, the comments are not connected to a specific part of the section, e.g., a paragraph or a word. Instead the comment is related to the whole section. This approach has its benefits. For example, the discussion might be more oriented to high level aspects. However, we have observed that comments often presented information that helped the reader to locate to which part of the section the comment was referring to (Rodriguez & Brunsberg, 2004). This indicates that it is feasible to be able to link a comment to a particular part of the section directly.


330

9/27/2006

4:07 PM

Page 330


It is natural that the content of a section changes during the writing task. This situation might cause, so-called, orphan annotations to appear. A comment can refer to a text that is later changed or deleted, say, following the instructions or suggestions in that comment. Once the change has been made to the text in the section, the comment will not make sense. Although orphan annotations may occur in Col·laboració, the comments are not read isolated but in a dialogue context. Therefore, a comment is not out of place in the section; instead, it could be out of date. The chronological sequence might indicate that the comment referred to a previous version of the section. Also, having the possibility to follow the comments from a historical perspective, might help a new member of the team to understand the transformation process that a section has gone through.. It is important to note that the approach that we have taken in the design of Col·laboració focuses on the possibility to allow co-authors to easily concentrate on the sections he/she is responsible for, but not detaching him/her from the whole document. Furthermore, as Col·laboració supports a transparent all-to-all communication among co-authors, all comments can be read by any of the co-authors regardless whether he/she is responsible for that particular section. Therefore, we consider role definitions to be less relevant.

21.3.

Case Studies

Col·laboració has been used in ten real collaborative writing tasks, all in an academic setting, in which two to nine co-authors took part. The type of writing task varied, including, e.g., a conference poster, short and full scientific papers, technical reports, and one Masters’ thesis proposal. A total of 457 comments holding 34,087 words were submitted by the co-authors on the document sections (including 146 comments made to the ATP document). Four of the case studies were carried out while some co-authors were located in different cities or countries. The time spent on the writing tasks varied between 6 and 42 days. Co-authors were asked to try to avoid, as far as possible, using other channels of communication during these writing tasks. A detailed report of the case studies is given in Rodriguez (2001, 2003). Kim (2002) performed a qualitative study of the dialogue between co-authors in one of the writing tasks of the Col·laboració case studies. A special coding scheme was developed for analysis of dialogue structure, and adapted to the activity of collaborative writing. The discussion of the results focused, among other things, on the implications of the asynchronous mode for collaboration, and the trade-offs between static annotations and textbased dialogue during a writing task. It is not possible here to go through these studies in detail. Instead, we will use the experiences gained from the case studies to discuss how the Col·laboració system supports focusing on those sections that a co-author might be responsible for and, at the same time, integrating his/her work with the teamwork. This is done by providing for both individual views of the document and mutual awareness during a writing task. 21.3.1.

Participants Who Used Col·laboració

In total there were 14 participants who used the system in real cases as co-authors of documents. All of them are experienced computer users and are familiar with the Web and


9/27/2006

4:07 PM

Page 331


331

Web-browsers. They had experience in writing and most of them had previously published a report or a paper in collaboration with others in an academic environment. Two of them were undergraduate students and one was from the industry, the others (11) had a PhD. or were graduate students. Three of the participants were part of the project that supports the development of Col·laboració. 21.3.2.

Data Collection

Two main types of data were collected and analyzed. The first type includes observations from the interaction between users (objective data). These data include the comments that they produced, and the log files that our system kept. The other main type of data is subjective, resulting from interviews and surveys. Special attention has been put on the log files. The analysis of the log files were used to disclose users’ navigation patterns, e.g., which sections of the document they visited more often, which sections they did not visit during the whole writing task, what happened after a comment was added to a section, who visited a just-commented section, or how long it took for a section to be visited by participants after a comment was added. As for the survey and interviews, among other things, participants were asked their opinion about the email notification function and the Overview function of the system. In the next section we present the results of our analysis focusing on the data mentioned here.

21.4.

Focusing on Coupled/Uncoupled Co-authors’ needs

Supporting mutual awareness is of great importance for the accomplishment of a collaborative task mediated by computers (Dourish & Bellotti, 1992). This topic has gained much attention among researchers and is one of most debated ones in the CSCW community. Fuchs, Pankoke-Babatz, and Prinz (1995) suggest different modes of awareness. In relation to “when” an event occurs we have synchronous or asynchronous awareness. In relation to the scope of work someone performs we have coupled or uncoupled awareness. Coupled awareness is provided for events closely related to the current focus of work of the user, whereas uncoupled awareness is provided for the events independent of whether they are closely related to the user’s current focus of work. Col·laboració is designed to provide both coupled and uncoupled awareness during the writing task. In the next section we describe how this is achieved. 21.4.1.

Email Notification

The Col·laboració system sends an email notification to all co-authors when a section of the document is added or deleted, or when a comment is made in the “Ideas for this paper” section. It is well known that email is one of the most used functions of the Internet and that many people read their email several times a day in a working environment. Because “Ideas for this paper” is a shared space for planning a writing task, in which meta-level discussions are carried out, it seems reasonable that all co-authors would be interested in


332

9/27/2006

4:07 PM

Page 332


these comments. Comments that are made on a document section are shared in a different way: they are sent via email only to the section’s responsible co-authors. However, the other co-authors can access these comments if needed, as they are saved in the public commenting space of the section. According to our studies, the email notification feature was perceived as a valuable reminder of the ongoing task. As one of the respondents in the survey said in relation to getting emails from the system: “That is good because it is like a trigger to go on working in the environment.” Although most of the users reported that they appreciate this function, many of them also indicated that it could involve a problem. Experiences have shown that users may feel overwhelmed by the number of emails received. One respondent wrote, “If you get Email from all the sections you get too many e-mails (I have already 30–70 emails each day) If you don’t get e-mails there is a problem of knowing what is going on.” This indicates that a balance should be found between supporting awareness and avoiding email overflow. Another problem reported was that when a comment is received by email, users indicated that it would feel natural that they could reply to it using the email program. Unfortunately, the system does not support this possibility so far. However, it appends the URL to the email so that co-authors can easily access the system and reply to the comment there. Some users indicated that an email warning that a comment has been added to a particular section would be enough, and found that including the comment in the text was redundant. They felt it to be more natural to read the comments in their discourse context, that is, in the system. The comments made to a section are presented in chronological order, often forming a dialogue. From the log file we often observed that when a comment was added to a particular section, the person responsible for it soon visited the system and “jumped” directly to his/her section. Moreover, in case the co-author was involved in more than one section, he/she jumped directly to the section that just received a comment. This suggests that the user had read the comment and entered the system to read or to reply to it. If so, the email notification was indeed “pushing” co-authors to interact with the system. Another positive aspect that the users reported was that it was not necessary to enter the system simply to see if something had happened in their section, as the email notification functioned as an indicator. What is important to point out, here, is that the system informs co-authors about the events on the “objects” (e.g., a section) that are coupled to them but does not hide the events concerning the uncoupled objects in the document. Organization of messages can be a source of problems for users, as reported by participants in a study made on electronic mail in a working environment (Bälter, 1998). One point worth noting here is that the email sent by Col·laboració is labeled so that co-authors can easily identify that the message is related to the writing task, who were the recipients of the message, what was the event that generated the notification, and, if applicable, to which section it is related and the comment itself. This information can be used to help co-authors to handle their emails. For example, emails sent by the system could be directly filtered to a certain folder and be organized by writing task or section.


9/27/2006

4:07 PM

Page 333

Supporting Collaborative Writing 21.4.2.

333

The Modular Document

In Col·laboració, as already mentioned, the document is a set of sections. For each section there is one comment space related to it. Technically, the sections and the comment space are independent files, although, of course, interrelated. These files can be manipulated and organized by the system using a “modular” approach. For example, co-authors can handle the sections as modules to create versions or to get different views of the document. This can be done by selecting which sections they want to include in his/her individual views. In one of our studies, nine co-authors produced an annual report of a project. The report was a compilation of the sub-project reports. A few respondents reported that they did not read others’ sections as they felt they were not responsible for them, nor could they change their text. The log file confirms this pattern, indicating that co-authors might need to focus only on a particular section of the document. This also depends on the role that a co-author plays in the writing task. For example, the leader of the project mentioned that it was very valuable to have all the sub-project reports in the same place, which could be easily accessible from any computer with an Internet connection. The leader of the project was clearly interested in all the sections, in contrast to those co-authors who were exclusively interested in their own sections. That is to say that co-authors might have different needs even though they are working on the same task. So the view of the document that the project leader had was, at least in this case, quite different from the one of the other co-authors.

Figure 2: An illustration of the Overview function of Col·laboració. Note that the user selects two sections that are not adjacent in the document structure (left-hand panel). However, they are presented as one unit including the comments on one of them (right-hand panel).


334

9/27/2006

4:07 PM

Page 334


Some co-authors might want to have a view of the document in which only his/her part is shown, and ignore the rest of the document. This can be achieved by the system in two ways. The first one is by just selecting a section from the index-frame. The system will show in the document-frame (the right frame in Figure 1) only the content of the section and its comments. The second is by using the “Overview” command. This function is more suitable when a co-author is responsible for or coupled to more than one section in the document. The user can select the sections and their comments as modular parts and “build up” his/her view of the document. Doing so, the co-author can exclude those sections that he/she is not coupled to, which will very likely alleviate his/her work (see Figure 2). However, the co-author is not detached from the whole document. Co-authors can still browse the rest of the sections at will, read their content, and become aware of the discussion the rest of the team has been engaged in. They might, for example, notice that others co-authors are using a certain style or spelling guidelines, which carries implications for their own work. At the same time they can notice the progress of the whole team, which is also essential in any collaborative task.

21.5.

Conclusion

The system Col·laboració uses, to some extent, the concept of coupled and uncoupled awareness, which can be very valuable for the success of a collaborative task. Considering a document as a set of independent section files can support both coupled and uncoupled awareness in a collaborative writing task. Co-authors can be involved in the document’s sections in different ways. For example, the leader of a writing team might be interested in the discussion and the work progress of all the sections of the document while a particular co-author might be interested, say, only in the section that he/she is responsible for. In Col·laboració, a section and comments on the section are saved in a file of its own. Therefore, they can be handled as modules. Users can select the modules (sections and comments on sections) and build up a view of the document and its comments if needed. The approach of seeing the document and its comments as individual files and treated as a module by Col·laboració can be used as a filter mechanism both for having an overview of the document and, also, for creating a version of the document. Co-authors can select a sub-set of sections including or not including comments to provide an overview of the common document. This subset does not necessarily have to include several contiguous sections. For example, one co-author can be interested only in two sections that could be separated in the structure by other sections, say, the Introduction and the Conclusion of a document. The system allows the user to have a view of the document that includes only those sections, breaking the linearity that ordinary word processors impose on document structure. An important feature of Col·laboració is that the communication process is not detached from the document. The comment-frame presents comments that are related to the collaborative writing task. The dialogue between co-authors, a key component for collaboration, is supported and it is linked to the document. Notification of comments and changes by email can be used to support coupled/uncoupled awareness. It has been shown that this mechanism works well as a reminder to participant in a collaborative writing task. However, it should be handled carefully as it could turn out to be a problem for the co-authors if the number of email notifications grows.

Else_SW-Torrance_REF.qxd

9/28/2006

3:00 PM

Page 335

References Abbott, R. D., & Berninger, V. W. (1993). Structural equation modelling of relationships among developmental skills in primary- and intermediate-grade writers. Journal of Educational Psychology, 85, 478–508. Ackerman, J. M. (1991). Reading, writing, and knowing: The role of disciplinary knowledge in comprehension and composing. Research in the Teaching of English, 25, 133–178. Afflerbach, P., & van Sledright, B. (2001). Hath! doth! what? middle graders reading innovative history texts. Journal of Adolescent and Adult Literacy, 44(8), 696–713. Alamargot, D., & Chanquoy, L. (2001). Studies in writing 9: Through the models of writing. Dordrecht: Kluwer Academic Publishers. Alamargot, D., & Chanquoy, L. (2002). Les modèles de rédaction de textes (Models of text composition). In M. Fayol (Ed.), Production du langage: Traité des sciences cognitives [Language production: Cognitive science approaches] (pp. 45–65). Paris: Hermes. Alamargot, D., & Chesnet, D. (2001). Etudier la dynamique des traitements rédactionnels par l’analyse synchronisée des temps de pauses et des mouvements oculaires. Note méthodologique [Studying the dynamics of writing processes by the concurrent analysis of pauses and eye movements. A methodological note]. Paper presented at the Ecole thématique CNRS “Production de texte et processus de révision: méthodes en temps réel” [Text production and revision processes: Real time methods], Poitiers, France. 9–13 July 2001. Alamargot, D., Dansac, C., & Chesnet, D. (2002). Parallel processing during writing instruction: Conjunct analysis of graphomotor movements and eye movement. Paper presented at the EARLI - SIG writing conference, Stafford, England. 9–13 July, 2001. Allen, N., Atkinson, D., Morgan, M., Moore, T., & Snow, C. (1987). What experienced collaborators say about collaborative writing. Journal of Business and Technical Communication, 1, 70–90. Allwood, J., Nivre, J., & Ahlsén, I. (1990). Speech management – on the non-written life of speech. Nordic Journal of Linguistics, 13, 3–48. Alvarez, C., Carreiras, M., & de Vega, M. (1992). Estudio estadístico de la ortografía castellana: (1) La frecuencia silábica [A statistical study of Spanish orthography: Syllabic frequency]. Cognitiva, 4(1), 75–105. Alves, R. A., & Castro, S. L. (2004, September). Writing stories from pictures: On-line effects of color and typing skill. Poster presented at the 9th International Conference of the EARLI SIG writing, Geneva. Anderman, E. M., & Maehr, M. L. (1994). Motivation and schooling in the middle grades. Review of Educational Research, 64, 287–309. Andersson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard University Press. Anderson, R. C., & Pichert, J. W. (1978). Recall of previously unrecallable information following a shift in perspective. Journal of Verbal Learning and Verbal Behavior, 17, 1–12. Anderson, J. R., & Schooler, L. J. (2000). The adaptive nature of memory. In E. Tulving, & F. Craik (Eds), The oxford handbook of memory (pp. 557–570). Oxford: Oxford University Press. Appelt, W., & Busbach, U. (1996). The BSCW system: A WWW-based application to support cooperation of distributed groups. Proceeding of the fifth workshops on enabling technologies: Infrastructure for collaborative enterprises (pp. 304–309). Stanford and Los Alamitos, CA: IEEE Computer Society Press. Arndt, V. (1987). Six writers in search of texts: A protocol-based study of L1 and L2 writing. ELT Journal, 41, 257–267. Audet, R. H., Hickman, P., & Dobrynina, G. (1996). Learning logs: A classroom practice for enhancing scientific sense making. Journal of Research in Science Teaching, 33, 205–222.


336

9/28/2006

3:00 PM

Page 336

References

Baayen, R. H., Piepenbrock, R., & van Rijn, H. (1993). The CELEX lexical database [Computer Software]. Linguistic Data Consortium, University of Pennsylvania, Philadelphia, PA. Baddeley, A. D. (1986). Working memory. Oxford: Oxford University Press. Baddeley, A. D. (1990) Human memory: Theory and practice. Hove: Lawrence Erlbaum. Baddeley, A. D. (1996). Exploring the central executive. Quarterly Journal of Experimental Psychology, 49A, 5–28. Baddeley, A. D. (2000). The episodic buffer: A new component for working memory? Trends in Cognitive Sciences, 4, 417–423. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 8, pp. 47–89). New York: Academic Press. Badecker, W., Hillis, A., & Caramazza, A. (1990). Lexical morphology and its role in the writing process: Evidence from a case of acquired dysgraphia. Cognition, 35, 205–243. Ball, E. W., & Blachman, B. A. (1991). Does phoneme awareness training in kindergarten make a difference in early word recognition and developmental spelling? Reading Research Quarterly, 26, 49–66. Bälter, O. (1998). Electronic mail in a working context. Unpublished doctoral thesis (TRITA-NA9820). Royal Institute of Technology, Sweden. Bangert-Drowns, R. L., Hurley, M. M., & Wilkinson, B. (2004). The effects of school-based writing-to-learn interventions on academic achievement: A meta-analysis. Review of Educational Research, 74, 29–58. Barbier, M. L., Faraco, M., Piolat, A., & Branca-Rosoff, S. (2004). Prise de notes et procédés de condensation en français L2 par des étudiants anglais, espagnols et japonais [Notetaking and abbreviative devices used by English, Spanish and Japanese students in French L2]. In N. Andrieux-Reix, S. Branca-Rosoff, & C. Puech (Eds), Ecriture abrégées (notes, notules, messages, codes...). L’abréviation entre pratiques spontanées, codifications, modernité et histoire [Abbreviated writing (notes, short notes, messages, codes...). Abbreviation in spontaneous practices, codings, modernity and history] (pp. 143–161). Gap: Editions Orphys. Beal, C. R. (1996). The role of comprehension monitoring in children’s revision. Educational Psychology Review, 8, 219–238. Beck, E. E., & Bellotti, V. (1993). Informed opportunism as strategy: Supporting coordination in distributed collaborative writing. In G. de Michelis, C. Simone, & K. Schmidt (Eds), Proceedings of the third European conference on computer-supported cooperative work (pp. 233–248). Dordrecht, The Netherlands: Kluwer Academic Publishers. Becker-Mrotzek, M. (1997). Schreibentwicklung und Textproduktion: Der Erwerb der Schreibfertigkeit am Beispiel der Bedienungsanleitung. [Development of writing and text production: The development of writing skills discussed on the example of operating instructions]. Opladen: Westdeutscher Verlag. BECTA (2004). Speech Recognition: Welcome to the world of speech recognition. Retrieved 24th September, 2004, from www.becta.org.uk/teachers/teachers.cfm?section=2&id=2142 Beins, B. C. (1993). Writing assignments in statistics classes encourage students to learn interpretation. Teaching of Psychology, 20, 161–164. Bell, A. (1991). The Language of news media. Blackwell: Oxford. Bereiter, C., & Scardamalia, M. (1987). The psychology of written composition. Hillsdale, NJ: Lawrence Erlbaum Associates. Berman, R. (1994). Learner’s transfer of writing skills between languages. TESL Canada Journal, 12, 29–46. Bernaers, B., Jacobs, G., & Van Waes, L. (1996). Van persbericht tot kranteartikel. Wat gebeurt er met titels? [From press release to newspaper article. What happens to titles?]. Tekstblad, 2, 16–22.


9/28/2006

3:00 PM

Page 337

References

337

Berninger, V. W., Cartwright, A. C., Yates, C. M., Swanson, H. L., & Abbott, R. D. (1994). Developmental skills related to writing and reading acquisition in the intermediate grades: Shared and unique functional systems. Reading and Writing: An Interdisciplinary Journal, 6, 161–196. Berninger, V. W., Fuller, F., & Whitaker, D. (1996). A process model of writing development across the life span. Educational Psychology Review, 8, 193–218. Berninger, V. W., & Swanson, H. L. (1994). Modifying Hayes and Flower’s model of skilled writing to explain beginning and developing writing. In E. C. Butterfield, & J. S. Carlson (Eds), Children’s writing: Toward a process theory of the development of skilled writing. Advances in cognition and educational practice (Vol. 2, pp. 57–81). Greenwich, CT: JAI Press Inc. Berninger, V. W., Whitaker, D., Feng, Y., Swanson, L., & Abbott, R. (1996). Assessment of planning, translation, and revising in junior high students. Journal of School Psychology, 34, 23–52. Berninger, V., Yates, C., Cartwright, A., Rutberg, J., Remy, E., & Abbott, R. (1992). Lower-level developmental skills in beginning writing. Reading and Writing: An Interdisciplinary Journal, 4, 257–280. Berry, J. A., & Linoff, G. (1997). Data mining techniques for marketing, sales and customer support. New York: Wiley. Biggs, J. B. (1987). Student approaches to learning and studying. Hawthorne, Victoria: Australian Council for Educational Research. Biggs, J. B. (1988). Approaches to learning and essay writing. In R. R. Schmeck (Ed.), Learning strategies and learning styles (pp. 185–226). New York: Plenum. Biggs, J. B, (1999). Teaching for quality learning at university. Buckingham, UK: Society for Research into Higher Education and Open University Press. Biggs, J. B., & Collis, K. (1982). Evaluating the quality of learning: The SOLO Taxonomy. New York: Academic Press. Biggs, J. B., Kember, D., & Leung, D. (2001). The revised two-factor study process questionnaire. British Journal of Educational Psychology, 71, 133–149. Biggs, J. B., Lai, P., Tang, C., & Lavelle, E. (1999). The effect of a graduate workshop on graduate students writing in English as a second language. British Journal of Educational Psychology, 69, 293–306. Biggs, J. B., & Rihn, B. A. (1984). The effects of intervention on deep and surface approaches to learning. In J. R. Kirby (Ed.), Cognitive strategies and educational performance (pp. 279–293). Orlando: Academic Press. Birenbaum, M., & Dochy, F. (Eds). (1996). Alternatives in assessment of achievements, learning, processes, and prior knowledge. Boston: Kluwer Academic Publishers. Bock, J. K. (1996). Language production: Methods and methodologies. Psychonomic Bulletin and Review, 3, 395–421. Bock, J. K., & Levelt, W. J. M. (1994). Language production: Grammatical encoding. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics (pp. 945–984). San Diego: Academic Press. Bonin, P., Fayol, M., & Gombert, J.-E. (1998). An experimental study of lexical access in the writing and naming of isolated words. International Journal of Psychology, 33, 269–286. Boscolo, P., & Borghetto, M. (2002, September). Writing a report from multiple sources. Paper presented at the International Conference on Ontological, Epistemological, Linguistic, and Pedagogical Considerations of Language and Scientific Literacy: Empowering Research and Informing Instruction, University of Victoria, Victoria, British Columbia, Canada. Boscolo, P., & Mason, L., (2001). Writing to learn, writing to transfer. In P. Tynjälä, L. Mason, & K. Lonka (Eds), Writing as a learning tool: Integrating theory and practice (pp. 83–104). Dordrecht, The Netherlands: Kluwer Academic Publishers. Bourdin, B., & Fayol, M. (1994). Is written language production more difficult than oral language production? A working memory approach. International Journal of Psychology, 29, 591–620.


338

9/28/2006

3:00 PM

Page 338

References

Bourdin, B., & Fayol, M. (2000). Is graphic activity cognitively costly? A developmental approach. Reading and Writing: An Interdisciplinary Journal, 13, 183–196. Bourdin, B., & Fayol, M. (2002). Even in adults, written production is still more costly than oral production. International Journal of Psychology, 37, 219–227. Boyle, J. R., & Weishaar, M. (2001). The effects of strategic note-taking on the recall and comprehension of lecture information for high school students with learning disabilities. Learning Disabilities: Research & Practice, 16, 133–141. Braaksma, M. A. H., Rijlaarsdam, G., Couzijn, M., & Van den Bergh, H. (2002). Learning to compose hypertext and linear text: Transfer or interference? In R. Bromme, & E. Stahl (Eds), Writing hypertext and learning. Conceptual and empirical approaches. Advances in learning and instruction series (pp. 15–37). Amsterdam: Pergamon. Bradley, L., & Bryant, P. E. (1983). Categorizing sounds and learning to read a causal connection. Nature, 301, 419–421. Bradley, L., & Bryant, P. (1991). Phonological skills before and after learning to read. In S. Brady, & D. Shankweiler (Eds), Phonological processes in literacy, (pp. 37–45). Hillsdale, NJ: Laurence Erlbaum Associates. Bradley, D., Sánchez-Casas, R., & García-Albea, J. (1993). The status of the syllable in the perception of Spanish and English. Language and Cognitive Processes, 8, 197–233. Branca-Rosoff, S. (1998). Abréviations et icônes dans les prises de notes des étudiants. [Abbreviations and icons in note taking by students] In M. Bilger, K. dan den Eynde, & F. Gadet (Eds), Analyse linguistique et approches de l’oral. Recueil d’études offert en hommage à ClaireBlanche-Benveniste [Speech analysis. Festchrift for Claire Blanche-Benveniste] (pp. 286–299). Leuven-Paris: Peeters. Breetvelt, I., van den Bergh, H., & Rijlaarsdam, G. (1994). Relation of writing processes and text quality: When and how? Cognition and Instruction, 12, 103–123. Breetvelt, I., van den Bergh, H., & Rijlaarsdam, G. (1996). Rereading and generating and their relation with text quality. An application of multilevel analysis to writing process data. In G. Rijlaarsdam, M. Couzijn, & H. van den Bergh (Eds), Theories, models and methodology in writing research (pp. 10–20). Amsterdam: Amsterdam University Press. Brewer, W. (2003). Speak and spell. PC Advisor, April, 107–110. Britton, J. (1982). Shaping at the point of utterance. In G. M. Pradl (Ed.), Prospect and retrospect: Selected essays of James Britton (pp. 139–145). Montclair, NJ: Boynton/Cook Publishers, Inc. [Reprinted from Freedman, A., & Pringle, I. (Eds). (1980). Reinventing the rhetorical tradition. Conway, AR: L & S Books, for the Canadian Council of Teachers of English.] Broeder, P., Extra, G., & van Hout, R. (1986). Measuring lexical richness and diversity in second language research. Polyglot, 8, 1–16. Broekkamp, H., & Van den Bergh, H. (1996). Attention Strategies in revising a foreign language text. In G. Rijlaarsdam, H. Van den Bergh, & M. Couzijn (Eds), Theories, models and methodology in writing (pp. 170–181). Amsterdam: Amsterdam University Press. Bromme, R., & Stahl, E. (1999). Spatial metaphors and writing hypertexts: Study within schools. European Journal of Psychology of Education, 14, 267–281. Bromme, R., & Stahl, E. (2001). The idea of “hypertext” and its implications on the process of hypertext writing. In W. Frindte, T. Köhler, P. Marquet, & E. Nissen (Eds), IN-TELE 99 - Internetbased teaching and learning (Vol. 3 Internet communication, pp. 302–308). Frankfurt am Main, Germany: Peter Lang. Bromme, R., & Stahl, E. (2002). Learning by producing hypertext from reader perspectives: Cognitive flexibility theory reconsidered. In R. Bromme, & E. Stahl (Eds), Writing hypertext and


9/28/2006

3:00 PM

Page 339

References

339

learning: Conceptual and empirical approaches. Advances in Learning and Instruction Series. London: Pergamon. Brown, A. L., Campione, J. C., & Day, J. D. (1982). Learning to learn: On training students to learn from texts. Educational Researcher, 10, 14–21. Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition and the culture of learning. Educational Researcher, 18, 32–41. Brown, J. S., McDonald, J. L., Brown, T. L., & Carr, T. H. (1988). Adapting to processing demands in discourse production: The case of handwriting. Journal of Experimental Psychology: Human Perception and Performance, 14, 45–59. Bruce, D. J. (1964). The analysis of word sounds. British Journal of Educational Psychology, 34, 158–170. Brush, A., Bargeron, D., Grudin, J., Borning, A., & Gupta, A. (2002). Supporting interaction outside of class. Proceedings of computer supported collaborative learning 2002, Boulder, CO (pp. 425–434). Butterfield, E. C. (1996). Environmental, cognitive and metacognitive influences on text revision: Assessing the evidence. Educational Psychology Review, 8, 239–297. Caccamise, D. J. (1987). Idea generation in writing. In A. Matsuhashi (Ed.), Writing in real time: Modelling production processes (pp. 224–253). Norwood: Ablex Publishing Corporation. Cadiz, J., Gupta, A., & Grudin, J. (2000). Using Web annotations for asynchronous collaboration around documents. Paper presented at Computer Supported Collaborative Work 2000, Philadelphia, PA. Cameron, A. C., Edmunds, G., Wigmore, B., Hunt, A. K., & Linton, M. J. (1997). Children’s revision of textual flaws. International Journal of Behavioral Development, 20, 667–680. Caplan, D., & Waters, G. S. (1999). Verbal working memory and sentence comprehension. Behavioral and Brain Sciences, 22, 77–126. Caporossi, G., Alamargot, D., & Chesnet, D. (2004). Using the computer to study the dynamics of handwriting processes. Lecture Notes in Computer Science, 3245, 242–252. Caramazza, A. (1991). Issues in reading, writing, and speaking: A neuropsychological perspective. Dordrecht, The Netherlands: Kluwer Academic Publishers. Caramazza, A., & Miceli, G. (1990). The structure of graphemic representations. Cognition, 37, 243–297. Carey, L. J., Flower, L., Hayes, J., Schriver, K. A., & Haas, C. (1989). Differences in writers’ initial task representations. Technical Report no. 35. Pittsburgh, PA: Carnegie Mellon University. Carrillo, M. (1994). Development of phonological awareness reading acquisition: A study in Spanish language. Reading and Writing: An Interdisciplinary Journal, 6, 279–298. Carroll, B. J. (1980). Testing communicative performance. Oxford: Pergamon Press. Cary, M., & Carlson, R. A. (2001). Distributing working memory resources during problem solving. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 836–848. Castello, M., & Monereo, C. (1999). Strategic knowledge in note-taking: A study in high education. Infancia y Aprendizaje, 88, 25–42. Cederlund, J., & Severinson Eklundh, K. (1989) JEdit: The Logging Text Editor for Macintosh. Stockholm: IPLab, Department of Numerical Analysis and Computing Science, Royal Institute of Technology (KTH). Chafe, W. (1982). Integration and involvement in speaking, writing and oral literature. In D. Tannen (Ed.), Advances in discourse processes (Vol. 9: Spoken and Written Language, pp. 35–53). Norwood, NJ: Ablex Publishing Corporation. Chandler, D. (1995). The act of writing. Aberystwyth: University of Wales.


340

9/28/2006

3:00 PM

Page 340

References

Chanquoy, L., & Alamargot, D. (2002). Mémoire de travail et rédaction de textes: Evolution des modèles et bilan des premiers travaux [Working memory and text composition: Evolution of models and evaluation of major studies]. L’Année Psychologique, 102, 363–398. Chanquoy, L., Foulin, J.-N., & Fayol, M. (1990). Temporal management of short text writing by children and adults. Cahiers de Psychologie (Cognitive European Bulletin of Cognitive Psychology), 10(5), 513–540. Chanquoy, L., Foulin, J.-N., & Fayol, M. (1995). Writing in adults: A real-time approach. In G. Rijlaarsdam, H. van den Bergh, & M. Couzijn (Eds), Theories, models and methodology: Current trends in research on writing (pp. 36–44). Amsterdam: Amsterdam University Press. Chanquoy, L., Foulin, J. N., & Fayol, M. (1996). Writing in adults: A real-time approach. In G. Rijlaarsdam, H. van den Bergh, & M. Couzijn (Eds), Theories, models and methodology in writing. Amsterdam: Amsterdam University Press. Chenoweth, N. A., & Hayes, J. R. (2001). Fluency in writing: Generating text in L1 and L2. Writing Communication, 18(1), 80–89. Chesnet, D., & Alamargot, D. (2005). Analyses en temps réel des activités oculaires et graphomotrices du scripteur: Intérêt du dispositif ‘Eye and Pen’ (Real-time analysis of eye and pen movements: The use of the ‘Eye and Pen’ programme). L’Année Psychologique, 105(3), 287–304. Churchill, E., Trevor, J., Bly, S., Nelson, L., & Cubranic, D. (2000). Anchored conversations: Chatting in the context of a document. Proceedings of the 2000 ACM Conference on Human Factors in Computing Systems, Hague, The Netherlands (pp. 454–461). Chuy, M., Alamargot, D., & Passerault, J. M. (2004). Content generation during text composition: Interference effect with long term memory. Paper presented at the Symposium “The Dynamic of Content Generation in Writing” organized by D. Alamargot and D. Galbraith, Writing Conference, Geneva, Switzerland. 20–22 September 2004. Cohen, J. (1968). Weighted kappa: Nominal scale agreement with provision for scaled disagreement or partial credit. Psychological Bulletin, 70, 213–220. Cohen, A. D. (1987). Using verbal reports in research on language learning. In C. Faerch, & G. Kasper (Eds), Introspection in second language research (pp. 82–95). Clevedon: Multilingual Matters. Cohn, E., Cohn, S., & Bradley, J. J. (1995). Note-taking, working memory, and learning in principles of economics. Research in Economic Education, 26(4), 291–307. Collins, A. M., & Gentner, D. (1980). A framework for cognitive theory of writing. In L. W. Gregg, & E. R. Steinberg (Eds), Cognitive processes in writing (pp. 51–52). Hillsdale, NJ: Erlbaum. Cooper, W. E. (1983). Cognitive aspects of skilled typewriting. New York, Heidelberg, Berlin: Springer. Cooper, C. R., & Matsuhashi, A. (1983). A theory of the writing process. In M. Martlew (Ed.), The psychology of written language (pp. 3–39). New York: Wiley. Copeland, K. A. (May, 1987). The effect of writing upon good and poor writers’ learning from pros (Technical Report). ERIC Document Reproduction Service no. ED 276 993. Corces, A., Garcia, M., & Gonzalez, V. (2004). Word recognition software use in a busy orthopaedic practice. Clinical Orthopaedics and Related Research, 421(1), 87–90. Cowan, N. (2000). The magical number 4 in short-term memory: Reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24, 87–185. Crinon, J., & Legros, D. (2002). The semantic effects of consulting a textual base on rewriting. Learning and Instruction, 12, 605–626. Crossman, W. (2004). VIVO [Voice-In/Voice-Out]: The coming age of talking computers. Oakland, CA: Regent Press. Cumming, A. (1989). Writing expertise and second language proficiency. Language Learning, 39(1), 81–141.


9/28/2006

3:00 PM

Page 341

References

341

Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 459–466. Daneman, M., & Green, I. (1986). Individual differences in comprehending and producing words in context. Journal of Memory and Language, 25, 1–18. Daniels, P. T. (1992). The syllabic origin of writing and the segmental origin of the alphabet. In P. Downing, S. D. Lima, & M. Noonan (Eds), The linguistics of literacy (pp. 83–110). Amsterdam: John Benjamins. Dansac, C., & Alamargot, D. (1999). Accessing referential information during text composition: When and why? In M. Torrance, & D. Galbraith (Eds), Knowing what to write: Conceptual processes in text production (pp. 79–97). Amsterdam: Amsterdam University Press. Davis, J. R., & Huttenlocher, D. P. (1995). Shared annotation for cooperative learning. Proceedings of the International Conference on Computer Supported Collaborative Learning (CSCL ‘95), Indiana, USA (pp. 84–88). Deane, P. D. (1992). Grammar in mind and brain. Explorations in cognitive syntax. Berlin: Mouton de Gruyter. Deegan, D. H. (1995). Exploring individual differences among novices reading in a specific domain: The case of law. Reading Research Quarterly, 30(2), 154–170. De La Paz, S., & Graham, S. (2002). Explicitly teaching strategies, skills, and knowledge: Writing instruction in middle school classrooms. Journal of Educational Psychology, 94, 687–698. Dell, G. S., Chang, F., & Griffin, Z. M. (2001). Connectionist models of language production: Lexical access and grammatical encoding. Cognitive Science, 23, 517–542. Desmette, D., Hupet, M., Schelstraete, M. A., & van der Linden, M. (1995). Adaptation en langue française du “reading span test” de Daneman & Carpenter (1980) [French language adaptation of the Daneman & Carpenter’s (1980) “reading span task”]. L’année Psychologique, 95, 459–482. D’Esposito, M., Detre, J. A., Alsop, D. C., Shin, R. K., Atlas, S., & Grossman, M. (1995). The neural basis of the central executive system of working memory. Science, 378, 279–281. Dillon, A. (2002). Writing as design: Hypermedia and the shape of information space. In R. Bromme, & E. Stahl (Eds), Writing hypertext and learning. Conceptual and empirical approaches. Advances in learning and instruction series (pp. 63–72). Amsterdam: Pergamon. Donovan, C. A., & Smolkin, L. B. (2002). Children’s genre knowledge: An examination of K-5 students’ performance on multiple tasks providing differing levels of scaffolding. Reading Research Quarterly, 37, 428–465. Dourish, P., & Bellotti, V. (1992). Awareness and coordination in shared workspaces. Proceedings of the conference on computer-supported cooperative work, Toronto, Canada (pp. 107–114). Dye, G. A. (2000). Graphic organizers to the rescue! Helping students link and remember information. Teaching Exceptional Children, 32(3), 72–76. Dyson, A. (1987). Individual differences. in beginning composing. Written Communication, 4, 411–442. Ehri, L. (1984). How orthography alters spoken language competencies in children learning to read and spell. In J. Downing, & R. Valtin (Eds), Language awareness and learning to read (pp. 119–147). New York: Springer-Verlag. Einstein, G. O., Morris, J., & Smith, S. (1985). Note-taking, individual differences, and memory for lecture information. Journal of Educational Psychology, 77(5), 522–532. Eklundh, K., & Kollberg, P. (1996). Computer tools for tracing the writing process: From keystroke records to S-notation. In G. Rijlaarsdam, H. Van den Bergh, M. Couzijn, (Eds), Theories, models and methodology in writing research (pp. 526–541). Amsterdam: Amsterdam University Press. Ellis, N., & Large, B. (1987). The development of reading: As you seek so shall you find. British Journal of Psychology, 78, 1–28. Emig, J. (1971). The composing processes of twelfth graders. Urbana: National Council of Teachers of English.


342

9/28/2006

3:00 PM

Page 342

References

Englert, C. S., Stewart, S. R., & Hiebert, E. H. (1988). Young writers’ use of text structure in expository text generation. Journal of Educational Psychology, 80, 143–151. Entwistle, N., & Ramsden, P. (1983). Understanding student learning. London: Croom Helm Ltd. Ericsson, K. A., & Simon, H. A. (1980). Verbal reports as data. Psychological Review, 87, 215–292. Faber, J. E., Morris, J. D., & Lieberman, M. G. (2000). The effect of note taking on ninth grade students’ comprehension. Reading Psychology, 21(3), 257–270. Faigley, L., & Witte, S. (1981). Analyzing revision. College Composition and Communication, 32, 400–414. Faraco, M., Barbier, M. L., Falaise, A., & Branca-Rosoff, S. (2003). Codage et traitement automatique de corpus pour l’étude des prises de notes en français langue première et langue seconde [Coding and automatic analysis of text in the study of note-taking in French as a first and second language]. Arob@se, 7, 1–2. Faraco, M., Barbier, M. L., & Piolat, A. (2002). A comparison between L1 and L2 note-taking in undergraduate students. In G. Rijlaarsdam (Series Ed.), Studies in writing; S. Ransdell, & M. L. Barbiers (Volume Eds), New directions in research on L2 writing (pp. 145–167). Dordrecht, The Netherlands: Kluwer Academic Publishers. Fayol, M. (1999). From on-line management problems to strategies in written composition. In M. Torrance, & G. Jeffery (Eds), The cognitive demands of writing. Processing capacity and working memory in text production (pp. 13–23). Amsterdam: Amsterdam University Press. Fayol, M., Largy, P., & Lemaire, P. (1994). Cognitive overload and orthographic errors: When cognitive overload enhances subject–verb agreement errors. A study in French written language. Quarterly Journal of Experimental Psychology, 47A, 437–464. Fayol, M., &. Monteil, J. M. (1994). Learning strategies/the learning of strategies. Revue française de pédagogie, 106, 91–110. Ferreiro, E., & Teberosky, A. (1979). Los sistemas de escritura en el desarrollo del niño [Literacy before schooling]. Mexico City: Siglo XXI (English edition, Macmillan, 1982). Flower, L. S., (1987). The role of task representation in reading-to-write. Reading-to-Write Report no. 2. The Center for the Study of Writing, Pittsburgh, PA and Berkeley, CA. Flower, L. S., (1989). Negotiating academic discourse. Reading-to-Write Report no. 29. The Center for the Study of Writing, Pittsburgh, PA and Berkeley, CA. Flower, L. S., & Hayes, J. (1980). The cognition of discovery: Defining a rhetorical problem. College Composition and Communication, 31, 21–32. Flower, L. S., & Hayes, J. R. (1981). A cognitive process theory of writing. College Composition and Communication, 32, 365–387. Foos, P. W., Mora, J. J., & Tkacz, S. (1994). Student study techniques and the generation effect. Journal of Educational Psychology, 86(4), 567–576. Ford, M., & Holmes, V. M. (1978). Planning units and syntax in sentence production. Cognition, 6, 35–53. Fotheringham, S., Brunsdon, C., & Charlton, M. (2000). Quantitative geography: Perspectives on spatial data analysis. Thousand Oaks, CA: Sage Publications. Foulin, J.-N. (1995). Pauses et débits: les indicateurs temporels de la production écrite [Pauses and fluency: Temporal indicators in writing production]. L’Année Psychologique, 95(3), 483–504. Foulin, J.-N. (1998). To what extent does pause location predict pause duration in adults’ and children’s writing. Cahiers de Psychologie Cognitive, 17(3), 601–620. Foulin, J.-N., & Fayol, M. (1988). Etude en temps réel de la production écrite chez des enfants de sept et huit ans [Studying real-time written production in seven and eight-year old children]. European Journal of Psychology of Education, 3(4), 461–475. Friedlander, A. (1990). Composing in English: Effects of a first language on writing English as a second language. In B. Kroll (Ed.), Second language writing (pp. 109–125). Cambridge: Cambridge University Press


9/28/2006

3:00 PM

Page 343

References

343

Fuchs, L., Pankoke-Babatz, U., & Prinz, W. (1995). Supporting cooperative awareness with local event mechanisms: The groupdesk system. In H. Marmolin, Y. Sundblad, & K. Schmidt (Eds), Proceedings of ECSCW’94 (pp. 247–262). Dordrecht, The Netherlands: Kluwer Academic Publishers. Fulton, S. (2004). Computing Out-Loud. Retrieved 28th September, 2004, from www.out-loud.com; https://exchange.shu.ac.uk/exchweb/bin/redir.asp?; URL=http://www.out-loud.com/ Galbraith, D. (1992). Conditions for discovery through writing. Instructional Science, 21, 45–72. Galbraith, D. (1996). Self-monitoring, discovery through writing and individual differences in drafting strategy. In G. Rijlaarsdam, H. Van den Bergh, & M. Couzijn (Eds), Theories, models, and methodology in writing research (pp. 121–141). Amsterdam: Amsterdam University Press. Galbraith, D. (1999). Writing as a knowledge-constituting process. In M. Torrance, & D. Galbraith (Eds), Knowing what to write: Conceptual processes in text production (pp. 139–159). Amsterdam: Amsterdam University Press. Galbraith, D., & Torrance, M. (1999). Conceptual processes in text production: From problem solving to text production. In M. Torrance, & D. Galbraith (Eds), Knowing what to write: Conceptual processes in text production (pp. 1–12). Amsterdam: Amsterdam University Press. Garner, R. (1988). Verbal-report data on cognitive and metacognitive strategies. In C. E. Weinstein, E. T. Goetz, & P. A. Alexander (Eds), Learning and study strategies (pp. 63–77). New York: Academic Press. Garrett, M. F. (1982). Production of speech: observations from normal and pathological language use. In A. Ellis (Ed.), Normality and pathology in cognitive functions (pp. 19–76). London: Academic Press. Gass, S. M., & Mackey, A. (2000). Stimulated recall methodology in second language research. Mahwah, NJ: Erlbaum. Geisler, C. (1994). Academic literacy and the nature of expertise: Reading, writing, and knowing in academic philosophy. Hillsdale, NJ: Lawrence Erlbaum. Gentner, D. R. (1983). The acquisition of typewriting skill. Acta Psychologica, 54, 233–248. Gentner, D. R., Larochelle, S., & Grudin, J. (1988). Lexical, sublexical, and peripheral effects in skilled typewriting. Cognitive Psychology, 20, 524–548. Gérouit, C., Piolat, A., Roussey, J. Y., & Barbier, M. L. (2001). Coût attentionnel de la recherche d’informations par des adultes sur hypertexte et sur document papier [Cognitive effort during information search via hypertext and paper documents]. In M. Mojahid, & J. Virbel (Eds), Actes du 4° Colloque International sur le Document Electronique [Proceedings of the 4th conference on Electronic Documents] (pp. 201–215). Paris: Europia Production. Ginns, I. S., & Watters, J. J. (1995). An analysis of scientific understandings of preservice elementary teacher education students. Journal of Research in Science Teaching, 32, 205–222. Glanzer, M., Fischer, B., & Dorfman, D. (1984). Short-term storage in reading. Journal of Verbal Learning and Verbal Behavior, 20, 656–670. Glenberg, A. M., Schroeder, J. L., & Robertson, D. A. (1998). Averting the gaze disengages the environment and facilitates remembering. Memory & Cognition, 26(4), 651–658. Goette, T. (2000). Keys to the adoption and use of voice recognition technology in organizations. Information Technology and People, 13(1), 67–80. Goldman-Eisler, F. (1968) Psycholinguistic experiments in spontaneous speech. London: Academic Press Goldman Eisler, F. (1972). Pauses, clauses, sentences. Language and Speech, 15(2), 103–113. Goldstein, H. (1991). Nonlinear multilevel models with an application to discrete response data. Biometrika, 78, 45–51. Goswami, U. (1994). Reading by analogy: theoretical and practical perspectives. In C. Hulme, & M. Snowling (Eds), Reading development and dyslexia (pp. 18–30). San Diego, CA: Singular. Goswami, U., & Bryant, P. E. (1992). Rhyme, analogy and children’s reading. In P. Gough, L. Ehri, & R. Treiman (Eds), Reading acquisition (pp. 49–63). Hillsdale, NJ: LEA.


344

9/28/2006

3:00 PM

Page 344

References

Gould, J. D. (1978). How experts dictate. Journal of Experimental Psychology: Human Perception and Performance, 4(4), 648–661. Gould, J. D. (1980). Experiments on composing letters: Some fact, some myths, and some observations. In L. W. Gregg, & E. R. Steinberg (Eds), Cognitive processes in writing (pp. 263–283). Hillsdale, NJ: Erlbaum. Gould, J. D., & Alfaro, L. (1984). Revising documents with text editors, hand-writing recognition systems and speech-recognition systems. Human Factors, 26(4), 91–406. Grabowski, J. (1996). Writing and speaking: Common grounds and differences. Toward a regulation theory of written language production. In C. M. Levy, & S. Ransdell (Eds), The science of writing (pp. 73–91). Mahwah, NJ: Erlbaum. Grabowski, J. (1999). Verbal modes of knowledge diagnosis: The writing superiority effect. In M.-F. Crété, & E. Espéret (Eds), Writing and learning to write at the dawn of the 21st century (pp. 11–16). Poitiers: University, LaCo–CNRS. Grabowski, J. (2005). Speaking, writing, and memory span performance: Replicating the Bourdin and Fayol results on cognitive load in German children and adults. In L. Allal, & J. Dolz (Eds), Proceedings Writing 2004, Geneva (CH): Adcom Productions. (CD-ROM) Gradwohl-Nash, J. M. (1993). Writing from sources: A structure-mapping model. Journal of Educational Psychology, 85(1), 159–170. Graham, S. (1990). The role of production factors in learning disabled student’s compositions. Journal of Educational Psychology, 82, 781–791. Graham, S., Berninger, V. W., Abbott, R. D., Abbott, S. P., & Whitaker, D. (1997). Role of mechanics in composing of elementary school students: A new methodological approach. Journal of Educational Psychology, 89, 170–182. Graham, S., & Harris, K. R. (2000). The role of self-regulation and transcription skills in writing and writing development. Educational Psychologist, 35, 3–12. Graves, D. (1973). An examination of the writing processes of seven year old children. Research in the Teaching of English, 9, 227–241. Greene, S. (1993). The role of task in the development of academic thinking through reading and writing in a college history course. Research in the Teaching of English, 27(1), 46–75. Greene, S. (1999). How beginning writing students interpret the task of writing an academic argument. In K. L. Weese, & S. L. Fox (Eds), Teaching academic literacy: The uses of teacherresearch in developing a writing program (pp. 85–118). Mahwah, NJ: Erlbaum. Greene, S. (2001). The question of authenticity: Teaching writing in a first-year college history of science class. Research in the Teaching of English, 35, 525–569. Greene, S., & Higgins, L. (1994). “Once upon a time”: The use of retrospective accounts in building theory in composition. In P. Smagorinsky (Ed.), Speaking about writing: Reflections on research methodology (Vol. 8, pp. 115–140). Sage Series in Written Communication. Thousand Oaks, CA: Sage. Gruneberg, M. M., & Mathieson, M. (1997). The perceived value of minds maps (spider diagrams) as learning and memory aids. Cognitive Technology, 2(1), 21–24. Gudjonsson, G. H., & Haward, L. R. C. (1998). Forensic psychology: A guide to practice. London: Routledge. Haarmann, J. J., Cameron, K. A., & Ruchkin, D. S. (2002). Neural synchronization mediates on-line sentence processing: EEG coherence evidence from filler-gap constructions. Psychophysiology, 39, 820–825. Haas, C. (1996). Writing technology: Studies on the materiality of literacy. Mahwah, NJ: Erlbaum. Hadwin, A. F., Kirby, J. R., & Woodhouse, R. A. (1999). Individual differences in note-taking, summarization and learning from lectures. Alberta Journal of Educational Research, 45(1), 1–17. Haenggi, D., & Perfetti, C. A. (1992). Individual differences in reprocessing of text. Journal of Educational Psychology, 84(2), 182–192.


9/28/2006

3:00 PM

Page 345

References

345

Halford, G. S., Brown, C. A., & Thompson, R. L. (1986). Children’s concepts of volume and floatation. Developmental Psychology, 22, 218–222. Halliday, M. A. K. (1985). Spoken and written language. Honk Kong: Oxford University Press. Halliday, M. A. K., & Hasan, R. (1976). Cohesion in English. London: Longman. Halverson, C. A., Horn, D. B., Karat, C., & Karat, J. (1999). The beauty of errors. Patterns of error correction in desktop speech systems, IOS Press, 1–8. Hardiman, P. T., Pollatsek, A., & Well, A. D. (1986). Learning to understand the balance beam. Cognition and Instruction, 3, 63–86. Hartley, J. (1994). Designing instructional text (3rd ed.). London: Kogan Page. Hartley, J. (2002). Note taking in non-academic settings: A review. Applied Cognitive Psychology, 16, 559–574. Hartley, J., & Davies, I. K. (1978). Notetaking: A critical review. Programmed Learning and Educational Technology, 15, 207–224. Hartley, J., Howe, M. J. A., & McKeachie, W. J. (2001). Writing through time: Longitudinal studies of the effects of new technology on writing. British Journal of Educational Technology, 32(2), 141–151. Hartley, J., Sotto, E., & Pennebaker, J. (2003). Speaking versus typing: A case-study of the effects of using voice-recognition software on academic correspondence. British Journal of Educational Technology, 34(1), 5–16. Hartley, J., & Tynjala, P. (2001). New technology and learning to learn. In P. Tynjala, L. Mason, & K. Lonka (Eds), Writing as learning tool: Integrating theory and practice (pp. 161–204). Dordrecht, The Netherlands: Kluwer. Hayes, J. R. (1996). A new framework for understanding cognition and affect in writing. In C. M. Levy, & S. E. Ransdell (Eds), The science of writing: Theories, methods, individual differences and applications (pp. 1–27). Mahwah, NJ: Lawrence Erlbaum. Hayes, J. R., & Flower, L. S. (1980). Identifying the organisation of writing processes. In L. W. Gregg, & E. R. Steinberg (Eds), Cognitive processes in writing (pp. 3–30). Mahwah, NJ: Erlbaum. Hayes, J. R., & Flower, L. S. (1986). Writing research and the writer. American Psychologist, 41, 1106–1113. Hayes, J. R., & Nash, J. G. (1996). On the nature of planning in writing. In M. C. Levy, & S. Ransdell (Eds), The science of writing (pp. 29–56). New York: Lawrence Erlbaum Associates. Herrmann, Th., & Grabowski, J. (1995). Pre-terminal levels of process in oral and written language production. In U. Quasthoff (Ed.), Aspects of oral communication (pp. 67–87). Berlin: de Gruyter. Heurley, L. (1997). Processing units in written texts: Paragraphs or information block? In J. Costermans, & M. Fayol (Eds), Processing interclausal relationships. Studies in production and comprehension of text (pp. 179–200). Hillsdale, NJ: Lawrence Erlbaum Associates. Higgins, L. (1992). Reading to argue: Helping students transform source texts. In A. Penrose, & B. Sitko (Eds), Hearing ourselves think: Cognitive research in the college writing classroom (pp. 70–101). New York: Oxford University Press. Hinsley, D., Hayes, J., & Simon, H. A. (1977). From words to equations: Meaning and representation in algebra word problems. In M. A. Just, & P. A. Carpenter (Eds), Cognitive processes in comprehension (pp. 89–105). Hillsdale, NJ: Erlbaum. Hjelmquist, E. (1984). Memory for conversations. Discourse Processes, 7, 321–335. Holden, D. (2004). Development of a processes to improve writing skills of adult learners in the GOAL program at Greenville college. Unpublished dissertation. Nova Southeastern University, Florida. Holmes, V. M. (1984). Sentence planning in a story continuation task. Language and speech, 27, 115–134. Holmes, V. M. (1988). Hesitations and sentences planning. Language and Cognitive Processes, 3(4), 323–361.


346

9/28/2006

3:00 PM

Page 346

References

Holmqvist, K., Johansson, V., Strömqvist, S., & Wengelin, Å. (2002). Studying reading and writing on-line. In S. Strömqvist (Ed.), The diversity of languages and language learning. Lund lectures in languages and literature I (pp. 103–123). Lund, Sweden: Centre for Languages and Literature, Lund University. Honeycutt, L. (2003). Researching the use of voice recognition writing software. Computers and Composition, 20, 77–95. Honeycutt, L. (2004). Literacy and the writing voice: The intersection of culture and technology. Journal of Business and Technical Communication, 18(3), 294–327. Hounsell, D. (1997). Learning and essay-writing. In F. Marton, D. Hounsell, & N. Entwistle (Eds), The experience of learning (pp. 103–123). Edinburough: Scottish Academic Press. Hughes, R., Jake, S., & Okelberry, C. (1998). Ceilidh: Collaborative writing on the Web. Proceedings of the 1998 ACM symposium on applied computing (pp. 732–736). New York: ACM Press. Hulme, C., Hatcher, P., Nation, K., Brown, A., Adams, J., & Stuart, G. (2002). Phoneme awareness is a better predictor of early reading skill than onset-rime awareness. Journal of Experimental Child Psychology, 82, 2–28. Inhelder, B., & Piaget, J. (1958). The growth of logical thinking from childhood to adolescence. New York: Basic. Inhoff, A. W., & Gordon, A. M. (1998). Eye movements and eye-hand coordination during typing. Current Directions in Psychological Science, 6(6), 153–157. Isaacs, B., & Kenni, A. T. (1973). The set test as an aid to the detection of dementia in old people. British Journal of Psychiatry, 123, 467–470. Jacobs, G. (1999). Preformulating the news: An analysis of the metapragmatics of press releases. Amsterdam/Philadelphia: John Benjamins. Jacobson, M. J., & Spiro, R. J. (1995). Hypertext learning environments, cognitive flexibility, and the transfer of complex knowledge: An empirical investigation. Journal of Educational Computing Research, 12, 301–333. Janal, D. S. (1999). Business speak: Using speech technology to streamline your Business. New York: Wiley. Jansen, D., van Waes, L., & van den Bergh, H. (1996). Effects on thinking aloud on writing processes. In C. M. Levy, & S. Ransdell (Eds), The science of writing: Theories, methods, individual differences and applications (pp. 233–250). Mahwah, NJ: Lawrence Erlbaum Associates. Jimenez, J. (1992). Metaconocimiento fonologico: Un estudio descriptivo sobre una muestra de niños prelectores en edad preescolar [Phonological metaknowledge: A descriptive study of a prereading kindergarten sample]. Infancia y Aprendizaje, 57, 49–66. Jonassen, D. H. (1993). Effects of semantically structured hypertext knowledge bases on users’ knowledge structures. In C. McKnight, A. Dillon, & J. Richardson (Eds), Hypertext: A psychological perspective (pp. 153–168). New York: Ellis Horwood. Jones, S., & Tetroe, J. (1987). Composing in a second language. In A. Matsuhashi (Ed.), Writing in real time: Modelling production process (pp. 34–37). Norwood, NJ: Ablex. Jonides, J., & Smith, E. E. (1997). The architecture of working memory. In M. S. Rugg (Ed.), Cognitive neuroscience (pp. 243–276). Cambridge, MA: MIT Press. Jou, J., & Harris, R. J. (1992). The effect of divided attention on speech production. Bulletin of the Psychonomic Society, 30, 301–304. Juel, C. (1988). Learning to read and write: A longitudinal study of 54 children from first through fourth grades. Journal of Educational Psychology, 80, 437–447. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99(1), 122–149.


9/28/2006

3:00 PM

Page 347

References

347

Kahan, J., Koivunen, M.-R. Prud’Hommeaux, E., & Swick, R. (2001). Annotea — an open RDF infrastructure for shared web annotations. Proceedings of the 10th international world wide web conference, Hong Kong, China (pp. 623–632). Kantz, M. (1989). Written rhetorical syntheses: Processes and products. Technical Report no. 1. The Center for the Study of Writing, Pittsburgh, PA and Berkeley, CA. Karat, C., Halverson, C., Horn, D., & Karat, J. (1999). Patterns of entry and correction in large vocabulary continuous speech recognition systems (pp. 568–575). New York: ACM Press. Karat, J., Horn, D. B., Halverson, C. A., & Karat, C. (2000). Overcoming unusibility. Developing strategies in speech recognition systems. Interactive Posters: in Proceedings of CHI 2000, pp. 141–142. Kellogg, R. T. (1987). Effects of topic knowledge on the allocation of processing time and cognitive effort to writing processes. Memory and Cognition, 15, 256–266. Kellogg, R. T. (1988). Attentional overload and writing performances: Effects of rough draft and outline strategies. Journal of Experimental Psychology, 14, 355–365. Kellogg, R. T., (1990). Effectiveness of prewriting strategies as a function of task demands. American Journal of Psychology, 103, 327–342. Kellogg, R. T. (1994). The psychology of writing. New York: Oxford University Press. Kellogg, R. T. (1996) A model of working memory in writing. In C. M. Levy, & S. Ransdell (Eds), The science of writing: Theories, methods, individual differences, and applications (pp. 57–72). Mahwah, NJ: Lawrence Erlbaum Associates. Kellogg, R. (1999) Components of working memory in text production, In M. Torrance, & G. Jeffery (Eds), The cognitive demands of writing. Processing capacity and working memory in text production (pp. 43–62). Amsterdam: Amsterdam University Press. Kellogg, R. T. (2001). Competition for working memory among writing processes. American Journal of Psychology, 114, 175–191. Kellogg, R. T. (2004). Working memory components in written sentence production. American Journal of Psychology, 117, 341–361. Kellogg, R. T., & Mueller, S. (1993). Performance amplification and process restructuring in computer-based writing. International Journal of Man–Machine Studies, 39, 33–49. Kember, D., & Leung, D. Y. P. (1998). The dimensionality of approaches to learning: An investigation with confirmatory factor analysis on the structure of the SPQ and LPQ. British Journal of Educational Psychology, 68, 395–407. Kennedy, M. (1985). The composing process of college students writing from sources. Written Communication, 2, 434–456. Kennedy, A., Radach, R., Heller, D., & Pynte, J. (Eds) (2000). Reading as a perceptual process. North Holland: Elsevier Sciences Publishers. Keys, C. W. (2000). Investigating the thinking processes of eighth grade writers during the composition of a scientific laboratory report. Journal of Research in Science Teaching, 37, 676–690. Kiewra, K. A. (1987). Note taking and review: The research and its implications. Journal of Instructional Science, 16, 233–249. Kiewra, K. A. (1988a). Cognitive aspects of autonomous note taking: Control processes, learning strategies, and prior knowledge. Educational Psychologist, 23, 39–56. Kiewra, K. A. (1988b). Comparing of self-questioning, summarizing, and note taking-review as strategies for learning from lectures. American Educational Research Journal, 29(2), 303–323. Kiewra, K. A. (1989). A review of note-taking: The encoding storage paradigm and beyond. Educational Psychology Review, 1, 147–172. Kiewra, K. A. (1991). Aids to lecture learning. Educational Psychologist, 26, 37–53. Kiewra, K. A., & Benton, S. L. (1988). The relationship between information-processing ability and note-taking. Contemporary Educational Psychology, 13, 33–44.


348

9/28/2006

3:00 PM

Page 348

References

Kiewra, K. A., Benton, S. L., Kim, S., Risch, N., & Christensen, M. (1995). Effects of note-taking format and study technique on recall and relational performance. Contemporary Educational Psychology, 20, 172–187. Kiewra, K. A., Benton, S. L., & Levis, L. B. (1987). Qualitative aspects of note-taking and their relationship with information-processing ability and academic achievement. Journal of Instructional Psychology, 14(3), 110–117. Kiewra, K. A., DuBois, N. F., Christensen, M., Kim, S., & Lindberg, N. (1989). A more equitable account of the note-taking functions in learning from lecture and from text. Instructional Science, 18(3), 217–232. Kiewra, K. A., DuBois, N. F., Christian, D., McShane, A., Meyerhoffer, M., & Roskelley, D. (1991). Note-taking functions and techniques. Journal of Educational Psychology, 83(2), 240–245. Kiewra, K. A., & Frank, B. M. (1988). Encoding and external storage effects of personal lecture notes, skeletal notes, and detailed notes for field-independent and field-dependent learners. Journal of Educational Research, 81, 143–148. Kiewra, K. A., Mayer, R. E., Christensen, M., Kim, S., & Risch, N. (1991). Effects of repetition on recall and note-taking: Strategies for learning from lectures. Journal of Educational Psychology, 83(1), 120–123. Kim, H.-C. (2002). From comments to dialogues: A study of asynchronous dialogue processes as part of collaborative reviewing on the Web. Proceedings of the 35th annual Hawaii international conference on system sciences (HICSS-35). Kim, H.-C., & Severinson Eklundh, K. (2001). Reviewing practices in collaborative writing. Computer supported cooperative work. The Journal of Collaborative Computing, 10(2), 247–259. Kintsch, W. (1998). Comprehension: A Paradigm for cognition. Cambridge, UK: Cambridge University Press. Kintsch, W., & van Dijk, T. A. (1978). Toward a model of discourse comprehension and production. Psychological Review, 85, 363–394. Kirkland, M. R., & Saunders, M. A. P. (1991). Maximizing student performance in summary writing: Managing the cognitive load. TESOL Quarterly, 25(1), 105–122. Kirriemuir, J. (2003). Speech recognition technologies: A Report to the JISC. Retrieved 28th September 2004, from www.jisc.ac.uk/index.cfm?name=techwatch_report_0303 Klein, P. D. (1999). Reopening inquiry into cognitive processes in writing-to-learn. Educational Psychology Review, 11, 203–270. Klein, P. D. (2000). Elementary students’ strategies for writing to learn in science. Cognition and Instruction, 18, 317–348. Klein, P. D. (2004). Constructing scientific explanations through writing. Instructional Science, 32, 191–231. Kolb, D. A. (1984). Experiental learning. Englewood Cliffs, NJ: Prentice Hall Inc. Kollberg, P. (1998). S-Notation — a computer based method for studying and representing text composition. Stockholm: Stockholm University. Kollberg, P., Lindgren, E., & Sullivan, K. P. H. (September, 2000). Coupling progression analysis with revision details to study foreign language writing. Paper presented at Eurosla-10, Krakow, Poland. Konopak, B. C., Martin, S. H., & Martin, M. A. (1990). Using a writing strategy to enhance sixthgrade students’ comprehension of content material. Journal of Reading Behavior, 22, 19–37. Kühne, C. (2000). Entwicklung eines multiple choice-fragebogens zum Computerwissen. [Development of multiple choice questionnaire for the assessment of computer knowledge]. Unveröffentlichte Diplomarbeit an der Westfälischen Wilhelms-Universität Münster. Ladefoged, P. (1982). A course in phonetics. New York: Harcourt Brace Jovanovich.


9/28/2006

3:00 PM

Page 349

References

349

Laidlaw, E. N., Skok, R. L., & McLaughlin, T. F. (1993). The effects of note-taking and self-questioning on quiz performance. Science Education, 77(1), 75–82. Largy, P., & Fayol, M. (2001). Oral cues improve subject–verb agreement in written French. International Journal of Psychology, 36, 121–132. Larochelle, S. (1983). A comparison of skilled and novice performance in discontinuous typing. In W. E. Cooper (Ed.), Cognitive aspects of skilled typewriting (pp. 67–94). New York, Heidelberg, Berlin: Springer. Lavelle, E. (1993). Development and validation of an inventory to assess processes in college composition. British Journal of Educational Psychology, 63, 489–499. Lavelle, E. (1997). Writing style and the narrative essay. British Journal of Educational Psychology, 67, 475–482. Lavelle, E. (2003). The quality of university writing: A preliminary analysis of undergraduate portfolios. Quality in Higher Education, 9(1), 87–93. Lavelle, E., & Guarino, A. (2003). A multidimensional approach to understanding college writing processes. Educational Psychology, 23, 295–304. Lavelle, E., Smith, J., & O’Ryan, L. (2002). The writing approaches of secondary students. British Journal of Educational Psychology, 72, 399–418. Lavelle, E., & Zuercher, N. (2001). The writing approaches of university students. Higher Education, 42, 373–391. Lea, J., & Levy, C. M. (1999). Working memory as a resource in the writing process. In M. Torrance, & G. C. Jeffery (Eds), The cognitive demands of writing (pp. 63–82). Amsterdam: Amsterdam University Press. Lea, M. R., & Street, B. (2000). Student writing and staff feedback in higher education: An academic literacies approach. In M. R. Lea, & B. Steirer (Eds), Student writing in higher education (pp. 32–46). Buckingham: Open University Press. Leijten, M., & Van Waes, L. (2003). The writing processes and learning strategies of initial users of speech recognition. A case study on the adaptation process of two professional writers. UFSIA Working Paper. University of Antwerp, Antwerp. Leijten, M., & Van Waes, L. (2004). Inputlog: A logging tool for the research of writing processes. Research Report. University of Antwerp, Antwerp. Léoni, V., & Mullet, E. (1993). Evolution in the intuitive mastery of the relationship between mass, volume, and density from nursery school to college. Genetic, Social, and General Psychology Monographs, 119, 391–412. Levelt, W. J. M. (1989). Speaking: From intention to articulation. Cambridge, MA: MIT Press. Levin, H., Silverman, I., & Ford, B. L. (1967). Hesitations in children’s speech during explanation and description. Journal of Verbal Learning and Verbal Behavior, 6(4), 560–564. Levy, C. M., & Marek, P. (1996). Testing components of Kellogg’s multicomponent model of working: The role of the phonological loop. M. Torrance & G. C. Jeffery (Eds), The cognitive demands of writing (pp. 25– 42). Amsterdam: Amsterdam University Press. Levy, C. M., & Ransdell, S. (1995). Is writing as difficult as it seems? Memory & Cognition, 23, 767–779. Levy, C. M., & Ransdell, S. E. (1996). Writing signatures. In C. M. Levy, & S. Ransdell (Eds), The science of writing: Theories, methods, individual differences, and applications (pp. 149–161). Mahwah, NJ: Lawrence Erlbaum Associates. Levy, C. M., & Ransdell, S. (Eds). (1996). The science of writing theories, methods, individual differences, and applications. Mahwah, New Jersey: Lawrence Erlbaum. Levy, C. M., & Ransdell, S. (2002). Writing with concurrent memory loads. In T. Olive, & C. M. Levy (Eds), Contemporary tools techniques for studying writing (pp. 9–30). Dordrecht, The Netherlands: Kluwer Academic Publishers.


350

9/28/2006

3:00 PM

Page 350

References

Liberman, I. Y., Shankweiler, D., Fischer, F. W., & Carter, B. (1974). Explicit syllable and phoneme segmentation in the young child. Journal of Experimental Child Psychology, 18, 201–212. Lindberg-Risch, N., & Kiewra, K. A. (1990). Content and form variations in note taking: Effects among junior high students. Journal of Educational Psychology, 83(6), 355–357. Lindgren, E. (2004). The uptake of peer-based intervention in the writing classroom. In G. Rijlaarsdam (Series Ed.); G. Rijlaarsdam, H. Van den Bergh, & M. Couzijn (Volume Eds), Studies in writing: Volume 14. Effective learning and teaching of writing (2nd ed., pp. 259–274). Dordrecht, The Netherlands: Kluwer Academic Publishers. Lindgren, E., & Sullivan, K. P. H. (2006). Analysing on-line revision. In G. Rijlaarsdam (Series Ed.); K. P. H. Sullivan, & E. Lindgren (Volume Eds), Studies in Writing. Computer Keystroke Logging: Methods and Applications. Oxford: Elsevier. Lindgren, E., & Sullivan, K. P. H. (2002). The LS graph: A methodology for visualising writing revision. Language Learning, 52(3), 565–595. Lindgren, E., & Sullivan, K. P. H. (2003). Stimulated recall as a trigger for increasing noticing and language awareness in the L2 writing classroom: A case study of two young female writers. Language Awareness, 12, 172–186. Logan, F. A. (1999). Errors in copy typewriting. Journal of Experimental Psychology: Human Perception and Performance, 25(6), 1760–1773. Lohman, D. F. (1993). Teaching and testing to develop fluid abilities. Educational Researcher, 22(7), 12–21. Lonka, K., Lindblom-Ylänne, S., & Maaury, S. (1994). The effect of study strategies on learning from text. Learning and Instruction, 4, 253–271. Maclay, H., & Osgood, C. E. (1959). Hesitation phenomena in spontaneous speech. Word, 15, 19–44. Maclean, M., Bryant, P., & Bradley, L. (1987). Rhymes, nursery rhymes and reading in early childhood. Merrill-Palmer Quarterly, 33(3), 255–281. MacNamara, D., Kintsch, E., Songer, N. B., & Kintsch, W. (1996). Are good texts always better? Interactions of text coherence, background knowledge, and levels of understanding in learning from text. Cognition and Instruction, 14, 1–43. MacWhinney, B. (2000). The CHILDES project (Vol. 1, 3rd ed.) Tools for analysing talk: Transcription format and programs. Mahwah, NJ: Lawrence Erlbaum Associates. Maki, H. S., Voeten, M. J. M., Vauras, M. M. S., & Poskiparta, E. H. (2001). Predicting writing development with word recognition and preschool readiness skills. Reading and Writing: An Interdisciplinary Journal, 14, 643–672. Malvern, D., & Richards, B. (2002). Investigating accommodation in language proficiency interviews using a new measure of lexical diversity. Language Testing, 19, 85–104. Mani, J. E. (1996). Patterns of selectivity in drawing on sources: Examining students’ use of intertextuality across literacy events. Reading Research and Cognition, 36, 51–63. Mani, J. E., Fyfe, R., Lewis, G., & Mitchell, E. (1996). Traversing the topical landscape: Exploring students’ self-directed reading-writing-research processes. Reading Research Quarterly, 31(1), 12–35. Martin, R. C., Shelton, J. R., & Yaffee, L. S. (1994). Language processing and working memory: Neuropsychological evidence for separate phonological and semantic capacities. Journal of Memory and Language, 33, 83–111. Martlew, M. (1983). Problems and difficulties: Cognitive and communicative aspects of writing. In M. Martlew (Ed.), The psychology of written language (pp. 295–333). New York: Wiley. Marton, F. (1988). Describing and improving learning. In R. R. Schmeck (Ed.), Learning strategies and learning styles (pp. 53–81). New York: Plenum. Marton, F., & Booth, S. A., (1997), Learning awareness. Mahwahm, NJ: Lawrence Erlbaum. Marton, F., & Saljo, R. (1976). On qualitative differences in learning. II. Outcome and process. British Journal of Education Psychology, 46, 4–11.


9/28/2006

3:00 PM

Page 351

References

351

Massaro, D. W., & Cohen, M. M. (1994). Visual, orthographic, phonological, and lexical influences in reading. Journal of Experimental Psychology: Human Perception and Performance, 20, 1107–1128. Matsuhashi, A. (1981). Pausing and planning: The tempo of written discourse production. Research in the Teaching of English, 15(2), 113–134. Matsuhashi, A. (1982). Explorations in the real time production of written discourse. In M. Nystrand (Ed.), What writers know: The language, process, and structure of written discourse. New York, NJ: Academic Press. Matsuhashi, A., & Cooper, C. (1978). A video time monitoring observational study: The transcribing behavior and composing processes of a competent high school writer. Paper presented at the meeting of the American educational research association, Toronto, Ontario, Canada. Matsumoto, K. (1995). Research paper writing strategies of professional Japanese EFL writers. TESL Canada Journal, 13(1), 17–27. Mayer, M. (1969). Frog, where are you? New York: Dial Press. McCarthy Young, K., & Leinhardt, G. (1998). Writing from primary documents: A way of knowing in history. Written Communication, 15(1), 25–68. McCloskey, M., Badecker, W., Goodman-Schulman, R. A., & Aliminosa, D. (1994). The structure of graphemic representation in spelling: Evidence from a case of acquired dysgraphia. Cognitive Neuropsychology, 11, 341–392. McCutchen, D. (1984). Writing as a linguistic problem. Educational Psychologist, 19, 226–238. McCutchen, D. (1988). Functional Automaticity in children’s writing: A problem of metacognitive control. Written Communication, 5, 206–324. McCutchen, D. (1996). A capacity theory of writing: Working memory in composition. Educational Psychology Review, 8, 299–325. McCutchen, D. (2000). Knowledge, processing, and working memory: Implications for a theory of writing. Educational Psychologist, 35, 13–23. McCutchen, D., Covil, A., Hoyne, S. H., & Mildes, K. (1994). Individual differences in writing: Implications of translating fluency. Journal of Educational Psychology, 86, 256–266. McGinley, W. J. (1992). The role of reading and writing while composing from sources. Reading Research Quarterly, 27(3), 226–249. McGinley, W., & Tierney, R. J. (1989). Traversing the topical landscape: Reading and writing as ways of knowing. Written Communication, 6, 243–269. McGuinness, D., McGuinness, C., & Donohue, J. (1995). Phonological training and the alphabet principle: Evidence for reciprocal causality. Reading Research Quarterly, 30(4), 830–852. McKee, G., Malvern, D., & Richards, B. (2000). Measuring vocabulary diversity using dedicated software. Literary and Linguistic Computing, 15, 323–337. McKeachie, W. J. (2005). Teaching tips: Strategies, research, and theory for college and university teachers (12th ed.). Boston: Houghton Mifflin. Meier, H. (1964). Deutsche sprachstatistik. Olms: Hildesheim. Meier, H. (1967) Deutsche sprachstatistik (2nd ed.). Germany: Hildesheim. Meier, S., Mc Carthy, P., & Schmeck, R. R. (1984). Validity of self-efficacy as a predictor of writing performance. Cognitive Therapy and Research, 8, 107–120. Menard, N. (1983). Mesure de la richesse lexicale [Measurment of lexical richness]. Genève-Paris: Slatkine-Champion. Miyake, A., & Shah, P. (Eds) (1999). Models of working memory: Mechanisms of active maintenance and executive control. Cambridge: Cambridge University Press. Morgan, C. H., Lilley, J. D., & Boreham, N. C. (1988). Learning from lectures: The effect of varying the detail in lecture handouts on note-taking and recall. Applied Cognitive Psychology, 2, 115–122.


352

9/28/2006

3:00 PM

Page 352

References

Murray, D. M. (1978). Internal revision: A process of discovery. In C. R. Cooper, & L. Odell (Eds), Research on composing: Points of departure (pp. 85–103). Urbana, IL: National Council of Teachers of English. Nauclér, K. (1980) Perspectives on misspelling — a phonetic, phonological and psycholinguistics study. Lund, Sweden: Gleerup Nelson, J. (1990). This was an easy assignment: Examining how students interpret academic writing tasks. Research in the Teaching of English, 24(4), 362–396. Nelson, J. (1992). Constructing a research paper: A study of students’ goals and approaches. Technical Report no. 59. The Center for the Study of Writing, Pittsburgh, PA and Berkeley, CA. Nelson, N., & Calfee, R. C. (Eds). (1998). The reading-writing connection. The ninety-seventh yearbook of the national society for the study of education. Part II. Chicago, IL: University of Chicago Press. Neuwirth, C., Kaufer, D., Chandhok, R., & Morris, J. (1990). Issues in the design of computer support for co-authoring and commenting. Proceedings of the 3rd conference on CSCW’90, ACM Press, Baltimore, MD (pp. 183–195). Newell, G. E., & Winograd, P. (1989). The effects of writing on learning from expository texts. Written Communication, 6, 196–217. Nist, S. L., & Hogrebe, M. C. (1987). The role of underlining and annotating in remembering textual information. Reading Research and Instruction, 27, 12–25. Nofsinger, R. E. (1991). Everyday conversation. London: Sage. Norton, L. S., & Hartley, J. (1986). What factors contribute to good examination marks: The role of note-taking in subsequent examination performance. Higher Education, 15, 355–371. Nottbusch, G., Grimm, A., Weingarten, R., & Will, U. (2005). Syllabic structures in typing: Evidence from hearing-impaired writers. Reading & Writing, 18, 497–526. Nottbusch, G., Sahel, S., Grimm, A., & Weingarten, R. (in preparation). Morphemes in written word production. Nottbusch, G., Weingarten, R., & Will, U. (1998). Schreiben mit der Hand und Schreiben mit dem Computer. OBST, 56, 11–27. Nystrand, M. (1989). A social-interactive model of writing. Written Communication, 6, 66–85. Oakhill, J., & Davies, A. M. (1991). The effects of test expectancy on quality of note taking and recall of text at different times of day. British Journal of Psychology, 82, 179–189. O’Hare, E. A., & McTear, M. F. (1999). Speech recognition in the secondary school classroom: An exploratory study. Computers and Education, 33(1), 37–45. Olive, T., & Kellogg, R. T. (2002). Concurrent activation of high and low-level production processes in written composition. Memory and Cognition, 30(4), 594–600. Olive, T., Kellogg, R. T., & Piolat, A. (2002). The triple task technique for studying the process of writing: Why and how? In G. Rijlaarsdam (Series Ed.), Studies in Writing & T. Olive, & C. M. Levy (Eds), Contemporary tools and techniques for studying writing (pp. 31–59). Dordrecht, The Netherlands: Kluwer Academic Publishers. Oliver, R., & Kerr, T. (1993). The impact of word processing on the preparation and submission of written essays in a tertiary course of study. Higher Education, 26, 217–226. Olive, T., & Levy, C. M. (Eds) (2002). Contemporary tools and techniques for studying writing. Dordrecht, The Netherlands: Kluwer. Olive, T., & Piolat, A. (2002). Suppressing visual feedback in written composition: Effects on processing demands and coordination of the writing processes. International Journal of Psychology, 37, 209–218. Olson, D. R. (1994). The world on paper. The conceptual and cognitive implications of writing and reading. London: Cambridge University Press. Oostdam, R., & Rijlaarsdam, G. (1995). Towards strategic language learning. Amsterdam: Amsterdam University Press.


9/28/2006

3:00 PM

Page 353

References

353

Orliaguet, J., & Boë, L. (1993). The role of linguistics in the speed of handwriting movement: Effects of spelling uncertainty. Acta Psychologica, 82, 103–113. Ostry, D. J. (1983). Determinants of interkey times in typing. In W. E. Cooper (Ed.), Cognitive aspects of skilled typewriting (pp. 225–246). New York, Heidelberg, Berlin: Springer. Passerault, J.-M. (1991). L’analyse en temps réel de l’activité de résumé: une étude des temps de pause au cours de la ‘‘re-mise’’ en texte [Real-time analysis of summarising: A study of pause times during the ‘‘re-expression’’ of text]. In M. Charolles, & A. Petitjean (Eds), Le résumé: Aspects linguistiques, psychologiques et didactiques [The summary: Linguistic, psychological and educational aspects]. Bruxelles: Klinsiek Passerault, J.-M., & Dinet, J. (2000). The role of the visuo-spatial sketchpad in the written production of descriptive and argumentative texts. Current Psychology Letters: Behavior, Brain, & Cognition, 3, 31–42. Pennebaker, J. W., Francis, M. E., & Booth, R. J. (2001). Linguistic inquiry and word count: LIWC. Mahwah, NJ: Erlbaum. Pennebaker, J. W., & King, L. A. (1999). Linguistic styles: Language use as an individual difference. Journal of Personality and Social Psychology, 77(6), 1296–1312. Penney, C. G., & Blackwood, P. A. (1989). Recall mode and recency in immediate serial recall: Computer users beware! Bulletin of the Psychonomic Society, 27, 545–547. Piolat, A. (2006). La prise de notes [Note taking]. Paris: Presses Universitaires de France, (2nd ed.). Piolat, A., & Boch, F. (2004). Apprendre en notant et apprendre à noter. [Learning by taking notes and learning to take notes] In E. Gentaz, & P. Dessus (Eds), Comprendre les apprentissages. Psychologie cognitive et éducation [Understanding learning. Cognitive psychology and education] (pp. 133–152). Paris: Dunod. Piolat, A., & Olive, T. (2000). Comment étudier le coût et le déroulement de la rédaction de textes? La méthode de la triple tâche: Un bilan méthodologique [How to study the cost and the unfolding of writing? The triple task method: A methodological review]. L’Année Psychologique, 100, 465–502. Piolat, A., Olive, T., & Kellogg, R. T. (2005). Cognitive effort of note taking. Applied Cognitive Psychology, 18, 1–22. Piolat, A., Olive, T., Roussey, J.-Y., Thunin, O., & Ziegler, J. C. (1999). Scriptkell: A tool for measuring cognitive effort and time processing in writing and other complex cognitive activities. Behavior Research Methods, Instruments, & Computer, 31, 113–121. Piolat, A., Roussey, J.-Y., & Barbier, M. L. (2003). Mesure de l’effort cognitif: Pourquoi est-il opportun de comparer la prise de notes à la rédaction, l’apprentissage et la lecture de divers documents ? [Measuring cognitive effort: Why is it relevant to compare note taking, writing, learning and reading]. Arob@se, 1–2, 118–140. Retrieved 20th May 2004 from www.arobase.to Podgorny, P., & Shepard, R. N. (1978). Functional representations common to visual perception and imagination. Journal of Experimental Psychology: Human Perception and Performance, 4, 21–35. Pontecorvo, C., & Zucchermaglio, C. (1988). Modes of differentiation in children’s writing construction. European Journal of Psychology of Education, 4, 371–385. Power, M. J. (1985). Sentence production and working memory. Quarterly Journal of Experimental Psychology, 37A, 367–386. Powney, J., & Watts, M. (1987). Interviewing in educational research. London: Routledge & Kegan Paul. Pressley, M., & Afflerbach, P. (1995). Verbal protocols of reading: The nature of constructively responsive reading. Hillsdale, NJ: Erlbaum Associates, Publishers. Psathas, G. (1995). Conversation analysis: The study of talk-in-interaction. London: Sage. Pynte, J., Courrieu, P., & Frenck, C. (1991). Evidence of repeated access to immediate verbal memory during handwriting. European Journal of Psychology of Education, 6(2), 121–125.


354

9/28/2006

3:00 PM

Page 354

References

Quinlan, T. (2004a). Speech recognition technology and students with writing difficulties: Improving fluency. Journal of Educational Psychology, 96(2), 337–346. Quinlan, T. (2004b). Technology and students with writing difficulties: Improving fluency. Journal of Educational Psychology, 96(2), 337–346. Ransdell, S.E., & Levy, C.M. (1996). Working memory constraints on writing quality and fluency. In C. M. Levy, & S. Ransdell (Eds), The science of writing: Theories, methods, individual differences and applications (pp. 93–106). Mahwah, NJ: Lawrence Erlbaum Associates. Ransdell, S. E., & Levy, C. M. (1999). Writing, reading, and speaking memory spans and the importance of resource flexibility. In M. Torrance, & G. H. Jefferey (Eds), The cognitive demands of writing: Processing capacity and working memory in text production (pp. 99–113). Amsterdam: Amsterdam University Press. Ransdell, S., Levy, M. C., & Kellogg, R. T. (2002). Effects of secondary task demands on writing. L-1-Educational Studies in Language & Literature, 2, 141–163. Raphael, T. E., & Boyd, F. B. (1991). Synthesizing information from multiple sources: A descriptive study of elementary students’ perceptions and performance of discourse synthesis. Series No. 45. The Center for the Learning and Teaching of Elementary Subjects, Michigan State University, East Lansing, MI. Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124(3), 372–422. Rickards, J. P., Fajen, B. R., Sullivan, J. F., & Gillespie, G. (1997). Signalling, note-taking, and field independence-dependence in text comprehension and recall. Journal of Educational Psychology, 89, 508–517. Rijlaarsdam, G., Couzijn, M., & Van den Bergh, H. (2004). The study of revision as a writing process and as a learning-to-write process: Two prospective research agendas. In G. Rijlaarsdam (Series Ed.); L. Allal, & L. Chanquoy (Volume Eds), Studies in writing: Volume 13. Revision: Cognitive and instructional processes (pp. 189–207). Boston: Kluwer Academic Publishers. Rijlaarsdam, G., & van den Bergh, H. (1996). Essentials for writing process studies: Many questions and some answers. In M. C. Levy, & S. Ransdell (Eds), The science of writing (pp. 107–126). New York: Lawrence Erlbaum Associates. Rijlaarsdam, G., & van den Bergh, H. (2006). Writing process theory: A functional dynamic approach. In C. McArthur, S. Graham, & J. Fitzgerald (Eds), Handbook of writing research (pp. 41–53). New York: Guildford. Rijlaarsdam, G., van de Bergh, T. H., & Couzijn, M. (Eds). (1996) Theories, models and methodology in writing research. Amsterdam: Amsterdam University Press. Rivard, L. P. (1994). A review of writing-to-learn in science: Implications for practice and research. Journal of Research in Science Teaching, 31, 969–983. Rizzi, C., Carvalho, C., Seixas, L., Costa, J., Tamusiunas, F., & Martins, A. (2000). Equitext. Retrieved 21st February 2006, from equitext.pgie.ufrgs.br Robinson, G. M. (1998). Methods and techniques in human geography. Chichester: Wiley. Robinson, D. H., Katayama, A. D. DuBois, N. F., & DeVaney, T. (1998). Interactive effects of graphic organizers and delayed review in concept acquisition. The Journal of Experimental Education, 67, 17–31. Robinson, D. H., & Kiewra, K. A. (1995). Visual argument: Graphic organizers are superior to outlines in improving learning from text. Journal of Educational Psychology, 87(3), 455–467. Roca de Larios, J., Murphy, L., & Manchon, R. (1999). The use of restructuring strategies in EFL writing: A study of Spanish learners of English as a Foreign language. Journal of Second Language Writing, 8(1), 13–44. Roca de Larios, J., Murphy, L., & Marin, J. (2002). A critical examination of L2 writing process research. In S. Ransdell, & M. L. Barbier (Eds), New direction for research in L2 writing (pp. 11– 47) Kluwer Academic Press.


9/28/2006

3:00 PM

Page 355

References

355

Rodriguez, H., (2001). Using the WWW as infrastructure for collaborative production of documents. Unpublished Licentiate thesis (TRITA-NA-0117). Sweden: The Royal Institute of Technology. Rodriguez, H., (2003). Designing, evaluating and exploring Web-based tools for collaborative annotation of documents. Unpublished Ph.D. thesis (TRITA-NA-0312). Sweden: Royal Institute of Technology. Rodriguez, H., & Brunsberg, S. (2004). Making digital annotations using the world wide web. In G. Rijlaarsdam, H. van den Bergh, & M. Couzijn (Eds), Effective learning and teaching of writing: Current trends from research (pp. 105–120). Dordrecht, The Netherlands: Kluwer Academic Publisher. Romero, M., & Decouchant, D. (1997) Structured cooperative authoring on the World Wide Web, CSCW”. The Journal of Collaborative Computing, 6(2–3), 157–174. Rommetveit, R. (1974). On message structure: A framework for the study of language and communication. London: Wiley. Rosaen, C. L. (1990). Improving writing opportunities in elementary classrooms. The Elementary School Journal, 90, 418–434. Rosenberg, S. (1977). Semantic constraints on sentence production: An experimental approach. In S. Rosenberg (Ed.), Sentence production: Developments in research and theory (pp. 195–227). Hillsdale, NJ: Lawrence Erlbaum Associates. Rosner, J., & Simon, D. (1971). The auditory analysis test: An initial report. Journal of Learning Disabilities, 4, 40–48. Roussey, J.-Y., & Piolat, A. (2003). Prendre des notes et apprendre. Effet du mode d’accès à l’information et de la méthode de prise de notes [Taking notes and learning. Effects of the access mode to information and to the note taking procedure]. Arob@se, 7, 1–2. Retrieved 20th May 2004 from www.arobase.to Ruhl, K. L., & Suritsky, S. (1995). The pause procedure and/or an outline: Effect on immediate free recall and lecture notes taken by college students with learning disabilities. Learning Disability Quarterly, 18(1), 2–11. Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed processing: Explorations in the microstructure of cognition. Vol. 1: Foundations. Cambridge, MA: MIT Press. Rummer, R., Grabowski, J., & Vorwerg, C. (1995). Kontrollprozesse beim Sprechen: Flexibilität und Determination der ereignisbezogenen Äußerungsplanung [Control processes during speech production: Flexibility and determination of the planning of event-related utterances]. Zeitschrift für Psychologie, 203, 25–51. Sachs, J. (1967). Recognition memory for syntactic and semantic aspects of connected discourse. Perception and Psychophysics, 2, 437–442. Sacks, H., Schegloff, E., & Jefferson, G. (1974). A simplest systematics for the organization of turntaking for conversation. Language, 50, 696–735. Sadoski, M., Kealy, W. A., Goetz, E. T., & Paivio, A. (1997). Concreteness and imagery effects in the written composition of definitions. Journal of Educational Psychology, 89, 518–526. Sadoski, M., & Paivio, A. (2001). Imagery and text: A dual coding theory of reading and writing. Mahwah, NJ: Lawrence Erlbaum Associates. Sahel, S., Nottbusch, G., Blanken, G., & Weingarten, R. (2005). The role of phonology and syllabic structure in the time course of typing: Evidence from aphasia. Linguistische Berichte, 201, 65–87. Sampson, G. (1985). Writing systems. London: Hutchinson. Sanchez, R. P., Lorch, E. P., & Lorch, R. F. (2001). Effects of headings on text processing strategies. Contemporary Educational Psychology, 26(3), 418–428. Sarig, G. (1987). High-level reading tasks in the first and in a foreign language: Some comparative process data. In Y. Divine, P. L. Carrell, & D. Eskey (Eds), Research in reading English as a second language (pp. 105–123). Alexandria, VA: TESOL.


356

9/28/2006

3:00 PM

Page 356

References

Sarig, G., & Folman, S. (1993). An integrative test for the assessment of academic literacy skills: A theoretical and pragmatic rationale (in Hebrew). Kibbutzim College of Education Annals, 15, 125–144. Sasaki, M. (2002). Buiding an empirically-based model of EFL Learners’ Writing Processes. In G. Rijlaarsdam (Writing Series Ed.), S. Ransdell, & M. Barbier, (Eds). New Directions for Research in L2 Writing (pp. 49–78). Dordrecht, The Netherlands: Kluwer. Sasaki, M., & Hirose, K. (1996). Explanatory variables for EFL students’ expository writing. Language Writing, 46(1), 137–174. Scardamalia, M., & Bereiter, C. (1982). Assimilated processes in composition planning. Educational Psychologist, 17, 10–24. Scardamalia, M., & Bereiter, C. (1987). Knowledge telling and knowledge transforming in written composition. In S. Rosenberg (Ed.), Advances in applied psycholinguistics (pp. 142–175). Cambridge: Cambridge University press. Scardamalia, M., & Bereiter, C. (1991). Literate expertise. In K. A. Ericsson, & J. Smith (Eds), Toward a general theory of expertise: Prospects and limits (pp. 172–194). Cambridge: Cambridge University Press. Scerbo, M. W., Warm, J. S., Dember, W. N., & Grasha, A. F. (1992). The role of time and cueing in a college lecture. Contemporary Educational Psychology, 17, 312–328. Schilperoord, J. (1995). It’s about time: Temporal aspects of cognitive processes in text production. Amsterdam/Atlanta: Rodopi. Schilperoord, J. (1996a). The distribution of pause time in written text production. In G. Rijlaarsdam, H. van den Bergh, & M. Couzijn (Eds), Theories, models and methodology in writing research (pp. 21–35). Amsterdam: Amsterdam University Press. Schilperoord, J. (1996b). It’s about time. Studies on temporal aspects of written text production. Doctoral Thesis. University of Amsterdam, Amsterdam. Schilperoord, J. (2001). On the cognitive status of pauses in discourse production. In T. Olive, & C. M. Levy (Eds), Contemporary tools and techniques for studying writing (pp. 61–87). Dordrecht, The Netherlands: Kluwer Academic Publishers. Schilperoord, J., & Sanders, T. (1999). How hierarchical structure affects retrieval processes; implications of pause and text analysis. In D. Galbraith, & M. Torrance (Eds) Knowing what to write. Conceptual processes in writing (pp. 13–30). Amsterdam: Amsterdam University Press. Schmeck, R. R. (1983). Learning styles of college students. In R. Dillon, & R. R. Schmeck (Eds), Individual differences in cognition (pp. 233–274). New York: Academic Press. Schmeck, R. R. (1988). (Eds), Perspectives on individual differences: Learning strategies and learning styles. New York: Plenum Press. Schmeck, R. R., Geisler-Brenstein, E., & Cercy, S. P. (1991). Self concept and learning: The revised inventory of learning processes. Educational Psychology, 11, 343–362. Schmeck, R. R., Ribich, F. D., & Ramanaiah, N. (1977). Development of a self-report inventory for assessing individual differences in the learning processes. Applied Psychological Measurement, 1, 413–431. Scholes, R. J. (1995). Orthography, vision and phonemic awareness. In I. Taylor, & D. R. Olson (Eds), Scripts and literacy: Reading and learning to read alphabets, syllabaries and characters (pp. 359–373). Dordrecht, The Netherlands: Kluwer. Schumacher, G., & Gradwohl, J. M. (1991). Conceptualizing and measuring knowledge change due to writing. Research in the Teaching of English, 25, 67–96. Segev-Miller, R. (1989). Linguistic processing in EFL study-summarizing. Unpublished manuscript. The School of Education, The Hebrew University of Jerusalem. Segev-Miller, R. (1990). Survey of ALT (Academic Literacy Test) assessment of college seniors. Internal Research Report. The Research Unit, Kibbutzim College of Education, Tel Aviv.


9/28/2006

3:00 PM

Page 357

References

357

Segev-Miller, R. (1994). Replanning the EFL reading course: A theoretical rationale. ISRATESOL Newsletter, 10, 3–6. Segev-Miller, R. (1997). Cognitive processes in discourse synthesis. Dissertation. The Hebrew University of Jerusalem. Segev-Miller, R. (2004a). Writing-from-sources: The effect of explicit instruction on college students’ processes and products. L1 – Educational Studies in Language and Literature, 4(1), 5–33. Segev-Miller, R. (2004b). On the differences between successful and unsuccessful synthesizers. Paper presented at the 9th international conference of the EARLI SIG of writing, 20–22 September, Geneva, Switzerland. Segev-Miller, R. (2004c). The effect of explicit instruction on college English majors’ writing-fromsources, Proceedings of The ETAI 5th International Conference, July 11–14, Renaissance Hotel, Jerusalem. Segev-Miller, R. (2005). Writing-to-learn: Conducting a process log. In G. Rijlaarsdam, H. van den Bergh, & M. Couzjin (Eds), Effective teaching and learning of writing: A handbook of writing in education (pp. 533–546). Dordrecht, The Netherlands: Kluwer. Severinson Eklundh, K. (1992). Problems in achieving a global perspective of the text in computerbased writing. Instructional Science, 21, 73–84. Severinsson Eklundh, K. (1994). Linear and non-linear strategies in computer-based writing. Computers and Composition, 11, 203–216. Severinson-Eklundh, K., & Kollberg, P. (1996a). A computer tool and framework for analyzing online revisions. In C. M. Levy, & S. Ransdell (Eds), The science of writing: Theories, methods, individual differences, and application (pp. 163–188). Mahwah, NJ: Lawrence Erlbaum Associates. Severinson-Eklundh, K., & Kollberg, P. (1996b). Computer tools for tracing the writing process: From keystroke records to S-notation. In G. Rijlaarsdam, H. Van den Bergh, & M. Couzijn (Eds), Current research in writing: Theories, models and methodology (pp. 526–541). Amsterdam: Amsterdam University Press. Severinsson-Eklundh, K., & Kollberg, P. (2003). Emerging discourse structure: Computer-assisted episode analysis as a window to global revision in university students’ writing. Journal of Pragmatics, 35, 869–891. Shaffer, L. H. (1978) Timing in the motor programming of typing. Quarterly Journal of Experimental Psychology, 30, 333–345. Sharples, M., Goodlet, J. S., Beck, E. E., Wood, C. C., Easterbrook, S. M., & Plowman, L., (1993). Research issues in the study of computer supported collaborative writing. In M. Sharples (Ed.), Computer supported collaborative writing (pp. 9–28.). New York: Springer-Verlag. She, H.-C. (2002). Concepts of a higher hierarchical level require more dual situated learning events for conceptual change: A study of air pressure and buoyancy. International Journal of Science Education, 24, 981–996. Shell, D., Colvin, C., & Bruning, R. (1995). Self-efficacy, attribution, and outcome expectancy mechanisms in reading and writing achievement: Grade-level and achievement-level differences. Journal of Educational Psychology, 87, 386–398. Siegler, R. S. (1985). Encoding and the development of problem solving. In S. F. Chipman, J. W. Segal, & R. Glaser (Eds), Thinking and learning skills, Volume 2: Research and open questions (pp. 161–185). Mahwah, NJ: Lawrence Erlbaum Associates. Silva, T. (1993). Toward an understanding of the distinct nature of L2 writing. TESOL Quarterly, 27(4), 657–677. Slattery, P. J. (1990). The argumentative multiple-source paper: College students reading, thinking, and writing about divergent points of view. Journal of Teaching Writing, 9(2), 181–199. Sleurs, K., Jacobs, G., & Van Waes, L. (2003). Constructing press releases, constructing quotations. A case study. Journal of Sociolinguistics 7(2), 192–212.


358

9/28/2006

3:00 PM

Page 358

References

Slotte, V., & Lonka, K. (1999). Review and process effects of spontaneous note-taking on text comprehension. Contemporary Educational Psychology, 24(1), 1–20. Slotte, V., & Lonka, K (2001). Note taking and essay writing. In G. Rijlaarsdam (Series Ed.); P. Tynjälä, L. Mason, & K. Lonka (Volume Eds), Studies in writing, Vol. 7, Writing as a learning tool: Integrating theory and practice (pp. 131–141). Dordrecht, The Netherlands: Kluwer Academic Publishers. Smagorinsky, P. (1994). Think-aloud protocol analysis: Beyond the black box. In P. Smagorinsky (Ed.), Speaking about writing: Reflections on research methodology (Vol. 8, pp. 3–19). Sage Series In Written Communication. Thousand Oaks, CA: Sage. Smagorinsky, P. (1997). Personal growth in social context: A high school senior’s search for meaning in and through writing. Written Communication, 14, 63–105. Snellings, P. (2003). Fluency in second language writing: The effects of enhanced speed of lexical retrieval. Unpublished doctoral thesis. University of Amsterdam, Holland. Spelman Miller, K. (2000a). Writing on-line: Temporal features of first and second language written text production. Unpublished doctoral thesis. University of Reading, Reading, UK. Spelman Miller, K. (2002b). Units of production in writing: Evidence of topic ‘framing’ in on-line writing research. SLALS Working Papers No. 4. Reading, UK: University of Reading. Spelman Miller, K. (2002c). Units of production in writing: Towards a discourse perspective. In S. Burgess (Ed.), Revista Canaria de Estudios Ingleses [The Canaries review of English studies]. Tenerife: University La Laguna. Spelman Miller, K. (2000d). Academic writers on-line: Investigating pausing in the production of text. Language Teaching Research, 4(2), 123–148. Spelman Miller, K., Lindgren, E., & Sullivan, K. P. H. (2004, September). Revising as the discourse unfolds. Paper presented at SIG Writing 2004, Geneva, Switzerland. Sperber, D. (2003). Vers une lecture sans écriture? [Towards reading without writing?] In G. Origgi, & N. Arikha (Eds), Text-e: Le texte à l’heure de l’Internet [Text-e: Text in the age of the internet] (pp. 177–190). Paris: Bibliothèque Publique d’Information (English version available from http://www.text-e.org). Spires, H. A. (1993). Learning from a lecture: Effects of comprehension monitoring. Reading Research & Instruction, 32(2), 19–30 Spiro, R. J., Coulson, R. L., Feltovich, P. J., & Anderson, D. K. (1988). Cognitive flexibility theory: Advanced knowledge acquisition in ill-structured domains. Technical Report no. 441. The Center for the Study of Reading, Urbana, IL, USA. Spiro, R. J., Feltovich, P. J., Jacobson, M. J., & Coulson, R. L. (1991). Cognitive flexibility, constructivism and hypertext: Random access instruction for advanced knowledge acquisition in illstructured domains. Educational Technology, 31, 24–33. Spiro, R. J., & Jehng, J.-C. (1990). Cognitive flexibility and hypertext: Theory and technology for the nonlinear and multidimensional traversal of complex subject matter. In D. Nix, & R. Spiro (Eds), Cognition, education and multimedia: Exploring ideas in high technology (pp. 163–205). Hillsdale, NJ: Erlbaum. Spivey, N. N. (1997). The constructivist metaphor: Reading, writing, and the making of meaning. San Diego, CA: Academic Press. Spivey, N. N., & King, J. R. (1989). Readers as writers composing from texts. Reading Research Quarterly, 24, 7–26. Stahl, E. (2001). Hyper-text-schreiben: Die Auswirkungen verschiedener Instruktionen auf Lernprozesse beim Schreiben von Hypertext [Hyper–text-writing: Effects of different instructions on learning processes during the writing of hypertext]. Münster: Waxmann. Stahl, E. (2002). Methods for assessing cognitive processes during the construction of hypertexts. In R. Bromme, & E. Stahl (Eds), Writing hypertext and learning. Conceptual and empirical approaches (pp. 177–196). Amsterdam: Pergamon.


9/28/2006

3:00 PM

Page 359

References

359

Stahl, S. A., Hynd, C. A., Britton, B. K., McNish, M. M., & Bosquet, D. (1996). What happens when students read multiple source documents in history? Reading Research Quarterly, 31(4), 430–458. Sternberg, R. (1998). Abilities are forms of developing expertise. Educational Researcher, 27(3), 11–20. Sternglass, M. (1988). The presence of thought: Introspective accounts of reading and writing. Advances in discourse processes (Vol. XXXIV). Norwood, NJ: Ablex. Sternglass, M. (1993). Writing development as seen through longitudinal research: A case study exemplar. Written Communication, 10, 235–261. Strauss, A. T., & Corbin, J. (1990). Basics of qualitative research: Grounded theory procedures and techniques. Newbury Park, CA: Sage. Strömqvist, S. (1999). Production rate profiles. In S. Strömqvist, & E. Ahlsén (Eds), The process of writing: A progress report (pp. 53–70). Gothenburg, Sweden: Department of Linguistics, University of Göteborg. Strömqvist, S., & Ahlsén, E. (Eds). (1999). The process of writing: A progress report. Gothenburg Papers in Theoretical Linguistics No. 83. Gothenburg, Sweden: Department of Linguistics, University of Göteborg. Strömqvist, S., & Holmqvist, K. (2001). The dynamics of production and perception during text writing. Paper presented at the Ecole thématique CNRS “Production de texte et processus de révision: méthodes en temps réel” [Text production and revision processes: Real time methods], 9–13 juillet 2001. Poitiers, France. Strömqvist, S., Holmqvist, K., Johansson, V., Karlsson, H., & Wengelin, Å. (2006). What keystroke logging can reveal about writing. In K. Sullivan, & E. Lindgren (Eds), Computer keystroke logging and writing: Methods and applications. Amsterdam: Elsevier. Strömqvist, S., Johansson, V., Kriz, S., Ragnarsdóttir, H., Aisenman, R., & Ravid, D. (2002). Towards a crosslinguistic comparision of lexical quanta in speech and writing. Written Language and Literacy, 5(1), 45–68. Strömqvist, S., & Malmsten, L. (1998). ScriptLog Pro 1.04 — user’s manual. Gothenburg, Sweden: Department of Linguistics, Göteborg University. Strømsø, H. I., & Bråten, I. (2002). Norwegian law students’ use of multiple sources while reading expository texts. Reading Research Quarterly, 37(2), 208–227. Swanson, H. L., & Berninger, V. W. (1996). Individual differences in children’s working memory and writing skill. Journal of Experimental Child Psychology, 63, 358–385. Swarts, H., Flower, L., & Hayes, J. R. (1984). Designing protocol studies of the writing process: An introduction. In R. Beach, & L. S. Bridwell (Eds), New directions in composition research (pp. 54–71). New York: Guildford Press. Swerts, M. (1998). Filled pauses as markers of discourse structure. Journal of Pragmatics, 30, 485–496. Talamo, A., & Fasulo, A. (2002). Opening windows in each other’s minds: Social sharing of hypertext models. In R. Bromme, & E. Stahl (Eds), Writing hypertext and learning. Conceptual and empirical approaches. Advances in learning and instruction series (pp. 99–124). Amsterdam: Pergamon. Tannenbaum, P. H., Williams, F., & Wood, B. S. (1967). Hesitation phenomena and related encoding characteristics in speech and typewriting. Language and Speech, 10, 203–215. Taube, K. (1988). Reading acquisition and self-concept. Umeå: University of Umeå. Terrel, S. R. (2002). The effect of learning style on doctoral course completion in a Web-based learning environment. Internet and Higher Education, 5, 345–352. Terzuolo, C. A., & Viviani, P. (1980). Determinants and characteristics of motor patterns used for typing. Neuroscience, 5, 1085–1103. Titsworth, B. S. (2001). The effects of teacher immediacy, use of organizational lecture cues, and students’ note-taking on cognitive learning. Communication Education, 50(4), 283–297.


360

9/28/2006

3:00 PM

Page 360

References

Titsworth, B. S., & Kiewra, K. A. (2004). Spoken organizational lecture cues and student note taking as facilitators of student learning. Contemporary Educational Psychology, 29, 447–461. Toglia, M., & Battig, W. F. (1978). Handbook of semantic word norms. Hillsdale, NJ: Lawrence Erlbaum Associates. Tolchinsky, L., & Teberosky, A. (1997). Word segmentation and writing in Hebrew and Spanish. In C. Pontecorvo (Ed.), Explicit writing development: An interdisciplinary view (pp. 77–95). Amsterdam: John Benjamins. Torrance, M., Fidalgo, R., & García, J.-N. (in press). The teachability and effectiveness of cognitive self-regulation in sixth grade writers. Learning and Instruction. Torrance, M., & Galbraith, D. (2006). The processing demands of writing. In C. Macarthur, S. Graham, & J. Fitzgerald, (Eds), Handbook of writing research. New York: Guilford Publications. Torrance, M., & Jeffery, G. C. (1999). Writing processes and cognitive demands. In G. Rijlaarsdam, & E. Espéret (Series Eds); M. Torrance, & G. Jeffery (Volume Eds), Studies in writing: Vol 3. The cognitive demands of writing (pp. 1–12). Amsterdam: Amsterdam University Press. Torrance, M., Thomas, G. V., & Robinson, E. J. (1996). Finding something to write about: Strategic and automatic processes in idea generation. In C. M. Levy, & S. Ransdell (Eds), The science of writing: Theories, methods, individual differences, and applications (pp. 189–205). Mahwah, NJ: Lawrence Erlbaum Associates. Torrance, M., Thomas, G. V., & Robinson, E. J. (1999). Individual differences in the writing behaviour of undergraduate students. British Journal of Educational Psychology, 69(2), 189–199. Torrance, M., Thomas, G. V., & Robinson, E. J. (2000). Individual differences in undergraduate essay writing strategies. Higher Education, 39(2), 181–200. Treiman, R. (1992). The role of intra-syllabic units in learning to read and spell. In P. Gough, L. Ehri, & R. Treiman (Eds), Reading acquisition. Hillsdale, NJ: LEA. Treiman, R., Bowey, J. A., & Bourassa, D. (2002). Segmentation of spoken words into syllables by English-speaking children as compared to adults. Journal of Experimental Child Psychology, 83, 213–238. Tulving, E., & Thompson, D. M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80, 352–373. Tynjälä, P. (1998). Writing as a tool for constructive learning: Students’ learning experiences during an experiment. Higher Education, 36(2), 209–230. Tynjälä, P. (2001). Writing, learning, and the development of expertise in higher education. In P. Tynjälä, L. Mason, & K. Lonka (Eds), Writing as a learning tool: Integrating theory and practice (pp. 37–56). Dordrecht, The Netherlands: Kluwer Academic Publishers. Tynjälä, P., Mason, L., & Lonka, K. (Eds) (2001). Writing as a learning tool: Integrating theory and practice. Dordrecht, The Netherlands: Kluwer Academic Publishers. U.S. Department of Education (2000). Speaking to write. Retrieved 28th September 2004, from www.edc.org/spk2wrt/ about.html Uzawa, K. (1996). Second language learners’ processes of L1 writing, L2 writing, and translation from L1 into L2. Learning and Instruction, 10(1), 107–112. Uzawa, K., & Cumming. A. (1989). Writing strategies in Japanese as a foreign language: Lowering or keeping up the standards. The Canadian Modern Language Review, 46, 178–194. van den Bergh, H., & Rijlaarsdam, G., (1996). The analysis of writing process data: A mini-longitudinal study. In M. C. Levy, & S. Ransdell (Eds), The science of writing (pp. 207–232). New York: Lawrence Erlbaum. van den Bergh, H., & Rijlaarsdam, G. (1999). The dynamics of idea generation during writing: An on-line study. In M. Torrance, & D. Galbraith (Eds), Knowing what to write: Conceptual processes in text production (pp. 99–120). Amsterdam: Amsterdam University Press.


9/28/2006

3:00 PM

Page 361

References

361

van den Bergh, H., & Rijlaarsdam, G. (2001). Changes in cognitive activities during the writing process, and relations with text quality. Educational Psychology, 21, 373–385. van den Bergh, H., Rijlaarsdam, G., & Breetvelt, I. (1993). Revision processes and text quality: An empirical study. In G. Eichler, & T. Jechle (Eds), Writing: Current trends in European research (pp. 133–147). Freiburg: Hochschul Verlag. van der Hoeven, J. (1997). Children’s composing: A study into the relationship between writing processes and text quality and cognitive linguistic skills. Unpublished doctoral thesis. University of Amsterdam, Amsterdam. van Dijk, T. (1980). Macrostructures. An interdisciplinary study of global structures in discourse, interaction, and cognition. Hillsdale, NJ: Erlbaum. Van Galen, G. P. (1991). Handwriting: Issues for a psychomotor theory. Human Movement Science, 10, 165–191. Van Galen, G. P., Meulenbroek, R. G., & Hylkema, H. (1986). On the simultaneous processing of words, letters and strokes in handwriting: Evidence for a mixed linear and parallel model. In H. Kao, G. P. van Galen, & R. Hoosain (Eds), Graphonomics: Contemporary research in handwriting (pp. 5–20). North-Holland: Elsevier Science. van Gelderen, A. (1997). Elementary students’ skills in revising. Written Communication, 14, 360–397. Van Rossum, E. J., & Schenk, S. M. (1984). The relationship between learning conception, study strategy and learning outcome. British Journal of Education Psychology, 54, 73–83. Van Waes, L. (2000). Thinking aloud as a method for testing the usability of websites: The influence of task variation on the evaluation of hypertext, IEEE Transactions in Professional Communication, 43, 279–291. Van Waes, L., & Schellens, P. J. (2003). Writing profiles: The effect of writing mode on pausing and revision patterns of experienced writers. Journal of Pragmatics, 35, 829–853. Vernon, S. (1991). El proceso de construcción de la correspondencia sonora en la escritura (en la transición entre los periodos presilábicos y el silábico). [The process of construction of sound correspondences in writing (the transition between the pre-syllabic and the syllabic levels)]. (Tesis DIE 6). Mexico: Departamento de Investigaciones Educativas, Centro de Investigacion y Estudios Avanzados. Vernon, S. (2002). Children’s analysis of oral and written words. In J. Brockmeier, M. Wang, & D. Olson (Eds), Literacy, narrative and culture (pp. 229–244). Richmond, Surrey: Curzon. Vernon, S., Calderon, G., & Castro, L. (2004). The relationship between phonological awareness and writing. Written Language and Literacy, 7(1), 101–114. Vernon, S., & Ferreiro, E. (1999). Writing development: A neglected variable in the consideration of phonological awareness. Harvard Educational Review, 69(4), 395–415. Verschueren, J. (1995). Metapragmatics. In J. Verschueren, J. Östman, & J. Blommaert (Eds), Handbook of pragmatics: manual (pp. 367–371). Amsterdam/Philadelphia: John Benjamins. Viavoice (2004). Product description. Retrieved 28th September 2004 from www.scansoft.co.uk/viavoice/ Vidal-Abarca, E., Reyes, H., Gilabert, R., Calpe, J., Soria, E., & Graesser, A. C. (2002). ETAT: Expository text analysis tool. Behavior Research Methods, Instruments, and Computers, 34(1), 93–107. Viviani, P., & Laissard, G. (1996). Motor templates in typing. Journal of Experimental Psychology: Human Perception and Performance, 22(2), 417–445. Voicepower (2004). Dragon naturally speaking professional version 7.3, Retrieved 28th September 2004 from www.voicepower.co.uk Voionmaa, K. (1993). On the semantics of adult verb acquisition. Gothenburg monographs in linguistics 12. Göteborg, Sweden: Department of Linguistics, Göteborg University.


362

9/28/2006

3:00 PM

Page 362

References

Vygotsky, L. (1962). Thought and language. Cambridge, MA: MIT Press. Ward, J., & Romani, C. (2000). Consonant–vowel encoding and ortho-syllables in a case of acquired dysgraphia. Cognitive Neuropsychology, 17(7), 641–663. Wegener, M. (2000). Spatial models and GIS. In S. Fotheringham, & M. Wegener (Eds), Spatial models and GIS — New potential and new models (pp. 3–20). London: Taylor & Francis. Weingarten, R. (1997). Wortstruktur und Dynamik in der schriftlichen Sprachproduktion. Didaktik Deutsch, 2, 10–17. Weingarten, R. (2005). Subsyllabic units in written word production. Written Language and Literacy, 8(1), 43–61. Weingarten, R., Nottbusch, G., & Will, U. (2004). Morphemes, syllables and graphemes in written word production. In T. Pechmann, & Ch. Habel (Eds), Multidisciplinary approaches to speech production (pp. 529–572). Berlin: Mouton de Gruyter. Wengelin, Å. (1998). Pauses and editings in the written language production of adults with and without severe reading and writing difficulties. In S. Strömqvist, & E. Ahlsén (Eds), The process of writing — a progress report. Gothenburg papers in theoretical linguistics 83 (pp. 71–86) Göteborg, Sweden: Department of Linguistics, Göteborg University. Wengelin, Å. (1999). Pauses and editings in the written language production of adults with and without severe reading and writing difficulties. In S. Strömqvist, & E. Ahlsén (Eds), The process of writing: A progress report (pp. 71–86). Gothenburg, Sweden: Department of Linguistics, University of Göteborg. Wengelin, Å. (2001a). From the “Dyslexia classroom” to the “classroom for reading and writing difficulties” — how do spelling difficulties influence the text production? In K. Vagle, & K. Wikberg (Eds), New directions in nordic text Linguistics and discourse analysis: Methodological issues (pp. 209–218). Oslo: Novus. Wengelin, Å. (2001b). Disfluencies in writing – are they like disfluencies in speaking? R. Lickley, & L. Shriberg (Eds), Proceedings from disfluency in spontaneous speech (DISS 2001), University of Edinburgh, UK (pp. 85–88). Wengelin, Å. (2002). Text production in adults with reading and writing difficulties. In Gothenburg monographs in linguistics (Vol. 20). Göteborg, Sweden: Department of Linguistics, Göteborg University. Wengelin, Å., & Strömqvist, S. (2000). Discourse level writing in dyslexics – methods results and implications for diagnosis. Scandinavian Journal of Logopedics, Phoniatrics and Vocology, 25, 22–28. Whissell, C. (1999). Linguistic complexity of abstracts and titles in highly cited journals. Perceptual and Motor Skills, 88, 76–86. Wilcox, R. R. (1998). How many discoveries have been lost by ignoring modern statistical methods? American Psychologist, 53, 300–314. Wilding, J., & Mohindra N. (1980). Effects of subvocal suppression, articulating aloud and noise on sequence recall. British Journal of Psychology, 71(2), 247–61. Wiley, J., & Voss, J. F., (1999). Constructing arguments from multiple sources: Tasks that promote understanding and not just memory for text. Journal of Educational Psychology, 91, 301–311. Will, U., Weingarten, R., Nottbusch, G., & Albes, C. (2005). Linguistische Rahmen und segmentale Informationen bei der Einzelwortschreibung (Linguistic frames and segmental information in single word typing. Evidence form time courses and error distributions). Evidenzen aus Zeitstrukturen und Fehlerverteilungen. In Ch. Habel, & T. Pechmann (Eds), Sprachproduktion (Speech Production). Opladen: Westdeutscher Verlag. (Preprint available at http://www.uni-bielefeld.de/lili/personen/gnottbusch/doc/Frames_Fillers.pdf) Williams, N. R. (2003). Voice recognition products – an occupational risk for users with ULDs? Occupational Medicine, 53(7), 452–455


9/28/2006

3:00 PM

Page 363

References

363

Williams, R. L., & Eggert, A. (2002). Note-taking predictors of test performance. Teaching of Psychology, 29(3), 234–237. Witchalls, C. (2003). Save your wrists, Guardian, December 18th. Inside IT supplement, p. 16. Witte, S. P., & Faigley, L. (1988). Coherence, cohesion and writing quality. College Composition and Communication, 32, 189–204. Yang, S. C. (2002). Multidimensional taxonomy of learners’ cognitive processing in discourse synthesis with hypermedia. Computers in Human Behavior, 18, 37–68. Yeung, A. S., Jin, P., & Sweller, J. (1997) Cognitive load and learner expertise: Split attention and redundancy effects in reading with explanatory notes. Contemporary Educational Psychology, 23, 1–21. Yopp, H. K. (1988). The validity and reliability of phonemic awareness tasks. Reading Research Quarterly, 23, 159–177. Zellermayer, M., Salomon, G., Globerson, T., & Givon, H. (1991). Enhancing writing related metacognitions through a computerized writing partner. American Educational Research Journal, 28, 373–391. Zesiger, P., Orliaguet, J., Boë, L., & Mounoud, P. (1994). The influence of syllabic structure in handwriting and typing production. In C. Faure, P. Keuss, G. Lorette, & A. Vinter (Eds), Advances in handwriting and drawing: A multidisciplinary approach (pp. 389–401). Paris: Europia. Zimmerman, B. J., & Bandura, A. (1994). Impact of self-regulatory influences on writing course attainment. American Educational Research Journal, 31, 845–862.


Else_SW-Torrance_authorind.qxd

10/20/2006

3:33 PM

Page 365

Author Index Abbott, R. D., 201–202, 217 Abbott, S. P., 202 Ackerman, J. M., 232–233, 241 Adams, J., 184 Afflerbach, P., 237, 246 Ahlsén, E., 58 Ahlsén, I., 69 Aisenman, R., 70 Alamargot, D., 5, 13–16, 18, 28, 152–153, 168 Albes, C., 32 Alfaro, L., 280 Aliminosa, D., 35 Allen, N., 323 Allwood, J., 69 Alsop, D. C., 99–100, 102 Alves, R. A., 6, 15, 31, 55, 63, 68, 71, 82 Alvarez, C., 185 Anderman, E. M., 216 Anderson, D. K., 239 Anderson, J. R., 167 Anderson, R. C., 167 Appelt, W., 328 Arndt, V., 151 Atkinson, D., 323 Atlas, S., 99–100, 102 Audet, R. H., 201 Baayen, R. H., 35 Baddeley, A. D., 56, 58, 97, 99, 112, 168, 272 Badecker, W., 35, 99 Ball, E. W., 183 Bälter, O., 332 Bangert-Drown, R. L., 201, 204, 216

Bandura, A., 225 Barbier, M. L., 110–113, 115, 120–121 Bargeron, D., 329 Battig, W. F., 103 Beal, C. R., 201–202 Beck, E. E., 323, 328 Becker-Mrotzek, M., 321 Beins, B. C., 201 Bell, A., 252 Bellotti, V., 328, 331 Benton, S. L., 110 Bereiter, C., 2, 4, 14, 55, 58, 68–69, 126–127, 146, 152–153, 166, 201–203, 208, 211, 216, 221, 270–271, 294, 308, 320 Berman, R., 151, 153–155 Bernaers, B., 257 Berninger, V. W., 98, 101, 201–202, 217 Berry, J. A., 84 Biggs, J. B., 219–223, 225–227 Birenbaum, M., 247 Blachman, B. A., 183 Blackwood, P. A., 55–56 Blanken, G., 34–35, 52 Bly, S., 329 Boch, F., 122 Bock, J. K., 98–99 Boë, L., 32, 34 Bonin, P., 33 Booth, R. J., 300 Booth, S. A., 220 Boreham, N. C., 110 Borghetto, M., 204 Borning, A., 329 Boscolo, P., 204


366

10/20/2006

3:33 PM

Page 366

Author Index

Bosquet, D., 238 Bourassa, D., 185 Bourdin, B., 15, 24, 55–56, 58, 63, 98, 168, 178, 204 Bowey, J. A., 185 Boyd, F. B., 232, 246 Boyle, J. R., 110 Braaksma, M. A. H., 307 Bradley, D., 185 Bradley, J. J., 111–113 Bradley, L., 183–184 Branca-Rosoff, S., 110, 121 Bråten, I., 233 Breetvelt, I., 29, 131–132 Brewer, W., 268 Britton, B. K., 238 Britton, J., 203 Broeder, P., 74 Broekkamp, H., 152 Bromme, R., 307–308, 320 Brown, A., 184, 246 Brown, C. A., 204 Brown, J. S., 14, 248 Brown, T. L., 14 Bruce, D. J., 183 Bruning, R., 225 Brunsberg, S., 329 Brunsdon, C., 87 Brush, A., 329 Bryant, P., 183–184 Busbach, U., 328 Butterfield, E. C., 201 Caccamise, D. J., 14, 152 Cadiz, J., 329 Calderon, G., 181 Calfee, R. C., 231 Calpe, J., 304 Cameron, A. C., 202 Cameron, K. A., 98, 101 Campione, J. C., 246 Caplan, D., 98 Caporossi, G., 16 Caramazza, A., 35, 98–99 Carey, L. J., 270

Carlson, R. A., 111 Carpenter, P. A., 15, 18, 24, 68, 71, 98, 122, 172 Carr, T. H., 14 Carreiras, M., 185 Carrillo, M., 186 Carroll, B. J., 154 Carter, B., 199 Cartwright, A. C., 202 Cary, M., 111 Castello, M., 110 Castro, S. L., 181 Cederlund, J., 88 Cercy, S. P., 220 Chafe, W., 70 Chandhok, R., 328 Chandler, D., 304 Chang, F., 108 Chanquoy, L., 13–15, 33, 36, 50, 64, 73, 152–153, 168 Charlton, M., 87 Chenoweth, N. A., 152–153, 157 Chesnet, D., 14, 16, 18, 28 Christensen, M., 110–111, 122 Christian, D., 110–111, 122 Churchill, E., 329 Chuy, M., 18, 28 Cohen, J., 59 Cohen, M. M., 102 Cohn, E., 111–113 Cohn, S., 111–113 Collins, A., 152, 248 Collis, K., 220 Colvin, C., 225 Cooper, C., 14 Cooper, C. R., 55 Cooper, W. E., 31 Copeland, K. A., 203 Corbin, J., 237 Corces, A., 268 Coulson, R. L., 239, 308 Courrieu, P., 32 Couzijn, M., 96, 307 Covil, A., 15, 68, 71, 81, 101, 202 Cowan, N., 100


10/20/2006

3:33 PM

Page 367

Author Index Crinon, J., 295 Crossman, W., 1 Cubranic, D., 329 Cumming, A., 151–153, 155

367

Ericsson, K. A., 235–236 Extra, G., 74

Daneman, M., 17–18, 101, 172 Daniels, P. T., 199 Dansac, C., 5, 14 Davies, A. M., 110 Davies, I. K., 111 Davis, J. R., 329 Day, J. D., 246 De La Paz, S., 4, 202 de Vega, M., 185 Deane, P. D., 127 Decouchant, D., 328 Deegan, D. H., 236 Dell, G. S., 108 Dember, W. N., 111, 114 Desmette, D., 18, 112–113, 115 D’Esposito, M., 99–100, 102 DeVaney, T., 110 Detre, J. A., 99–100, 102 Dillon, A., 308 Dinet, J., 108 Dobrynina, G., 201 Dochy, F., 247 Donohue, J., 184 Donovan, C. A., 202 Dourish, P., 331 Dorfman, D., 168 DuBois, N. F., 110–111, 122 Duguid, P., 248 Dye, G. A., 110 Dyson, A., 221

Faber, J. E., 110 Faigley, L., 94, 221 Fajen, B. R., 114 Falaise, A., 110 Faraco, M., 110, 113, 115, 121 Fasulo, A., 307 Fayol, M., 14–15, 24, 33, 36, 50, 55–56, 58, 63–64, 68, 73, 98, 101, 107, 152, 168, 178, 204 Feltovich, P. J., 239, 308 Feng, Y., 202 Ferreiro, E., 181–182, 186, 188 Fidalgo, R., 4 Fisher, B., 168 Fischer, F. W., 199 Flower, L. S., 4, 58, 126, 129, 152, 202–203, 207, 216, 221, 232–233, 236, 270–272, 308, 312, 328 Folman, S., 234 Foos, P. W., 110 Ford, B. L., 14 Ford, M., 14 Fotheringham, S., 87 Foulin, J.-N., 14, 33, 36, 50, 64, 73 Francis, M. E., 300 Frank, B. M., 110 Frenck, C., 32 Friedlander, A., 152–153 Fuchs, L., 331 Fuller, F., 202 Fulton, S., 268 Fyfe, R., 232, 241, 246

Easterbrook, S. M., 323 Edmunds, G., 202 Eggert, A., 110 Ehri, L., 184 Einstein, G. O., 110 Ellis, N., 184 Emig, J., 221 Englert, C. S., 202 Entwistle, N., 220

Galbraith, D., 5, 127, 201, 203, 216, 240 García-Albea, J., 185 Garcia, M., 268 García, J.-N., 4 Garner, R., 152 Garrett, M. F., 270 Gass, S. M., 253 Geisler, C., 247 Geisler-Brenstein, E., 220


368

10/20/2006

3:33 PM

Page 368

Author Index

Gentner, D., 31–32, 152 Gérouit, C., 111 Gilabert, R., 304 Gillespie, G., 114 Ginns, I. S., 204, 214 Givon, H., 294–295 Glanzer, M., 768 Glenberg, A. M., 29 Globerson, T., 294–295 Goette, T., 268 Goetz, E. T., 102–104, 107 Goldman-Eisler, F., 69 Goldstein, H., 134 Gombert, J.-E., 33 Gonzalez, V., 268 Goodlet, J. S., 323 Goodman-Schulman, R. A., 35 Gordon, A. M., 16 Goswami, U., 184 Gould, J. D., 152, 280 Grabowski, J., 166, 168, 173, 178 Gradwohl-Nash, J. M., 232, 244 Graesser, A. C., 304 Graham, S., 4, 55, 58, 202 Grasha, A. F., 111, 114 Graves, D., 221 Green, I., 17, 101 Greene, S., 235–236, 244, 248 Griffin, Z. M., 108 Grimm, A., 34–35, 52 Grossman, M., 99–100, 102 Grudin, J., 31, 329 Gruneberg, M. M., 110 Guarino, A., 225 Gudjonsson, G. H., 305 Gupta, A., 329 Haarmann, J. J., 98, 101 Haas, C., 270, 290, 294 Hadwin, A. F., 110 Haenggi, D., 112–113 Halford, G. S., 204 Halliday, M. A. K., 185, 245 Halverson, C., 280, 284 Halverson, C. A., 280, 289

Hardiman, P. T., 205 Harris, K. R., 55 Harris, R. J., 168 Hartley, J., 110–111, 122, 280, 293, 295, 297, 299, 304 Hasan, R., 245 Hatcher, P., 184 Haward, L. R. C., 305 Hayes, J., 4, 58, 126, 167, 203, 216, 270–272, 328 Hayes, J. R., 4, 58, 112, 126, 129, 152–153, 157, 167, 202–203, 207, 216, 221, 236, 270–272, 308, 312, 328 Heller, D., 76 Herrmann, Th., 766 Heurley, L., 113 Hickman, P., 201 Hiebert, E. H., 202 Higgins, L., 235–236, 245 Hillis, A., 99 Hinsley, D., 167 Hirose, K., 151 Hitch, G., 56, 99 Hjelmquist, E., 168 Hogrebe, M. C., 110 Holden, D., 226 Holmes, V. M., 14 Holmqvist, K., 16, 60 Honeycutt, L., 268–269, 279 Horn, D., 280, 284, 289 Hounsell, D., 220 Howe, M. J. A., 295, 297, 299, 304 Hoyne, S. H., 15, 68, 71, 81, 101, 202 Hughes, R., 329 Hulme, C., 184 Hunt, A. K., 202 Hupet, M., 18, 112–113, 115 Hurley, M. M., 201, 204 Huttenlocher, D. P., 329 Hylkema, H., 31 Hynd, C. A., 238 Inhelder, B., 204–205 Inhoff, A. W., 16 Isaacs, B., 18


10/20/2006

3:33 PM

Page 369

Author Index Jacobs, G., 253–254, 257 Jacobson, M. J., 308 Jake, S., 329 Janal, D. S., 268 Jansen, D., 73, 320 Jefferson, G., 169 Jeffery, G. C., 112 Jehng, J.-C., 308 Jimenez, J., 186 Jin, P., 111 Johansson, V., 60, 70 Jonassen, D. H., 315 Jones, S., 151, 153–155 Jonides, J., 98 Jou, J., 168 Juel, C., 202 Just, M. A., 15, 24, 68, 71, 98, 122 Kahan, J., 329 Kantz, M., 232–233, 246 Karat, C., 280, 284, 289 Karat, J., 280, 284, 289 Karlsson, H., 60 Katayama, A. D., 110 Kaufer, D., 328 Kealy, W. A., 102–104, 107 Kellogg, R. T., 5, 14–16, 27, 55–58, 63–64, 68, 71, 73, 98–102, 105, 107, 111–112, 114–115, 119, 270, 293–294, 308, 312, 321 Kember, D., 220, 224–225 Kennedy, A., 16 Kennedy, M., 232, 246 Kenni, A. T., 18 Kerr, T., 294 Keys, C. W., 203 Kiewra, K. A., 110–114, 122 Kim, H.-C., 324, 330 Kim, S., 110–111, 122 King, J. R., 238, 246 King, L. A., 304 Kintsch, E., 122 Kintsch, W., 122, 231 Kirby, J. R., 110 Kirkland, M. R., 231

369

Kirriemuir, J., 270 Klein, P. D., 201, 203, 205–206 Kovunen, M. -R., 329 Kolb, D. A., 281–283 Kollberg, P., 58, 88, 252, 283, 291 Konopak, B. C., 204 Kriz, S., 70 Kühne, C., 315 Ladefoged, P., 185 Lai, P., 220, 222–223, 227 Laidlaw, E. N., 110 Laissard, G., 31 Large, B., 184 Largy, P., 101, 107, 168 Larochelle, S., 31–32 Lavelle, E., 220–225, 227, 305 Lea, J., 68, 71, 101 Lea, M. R., 305 Legros, D., 295 Leijten, M., 88, 280, 283, 290 Leinhardt, G., 241 Lemaire, P., 101, 107, 168 Léoni, V., 204, 215 Leung, D., 220, 224–225 Levelt, W. J. M., 98–99, 168 Levin, H., 14 Levis, L. B., 110 Levy, C. M., 58, 68, 71, 101, 122, 129, 152, 175, 202, 252, 286, 304–305 Levy, M. C., 99–100, 107 Lewis, G., 232, 241, 246 Liberman, I. Y., 199 Lieberman, M. G., 110 Lilley, J. D., 110 Lindberg-Risch, N., 110 Lindblom-Ylänne, S., 110 Lindgren, E., 84–85, 87, 91, 94–95 Linoff, G., 84 Linton, M. J., 202 Logan, F. A., 35 Lohman, D. F., 232 Lonka, K., 110 Lorch, E. P., 111 Lorch, R. F., 111


370

10/20/2006

3:33 PM

Page 370

Author Index

Maaury, S., 110 Mackey, A., 253 Maclay, H., 14 Maclean, M., 183 MacNamara, D., 122 MacWhinney, B., 59 Maehr, M. L., 216 Maki, H. S., 202 Malmsten, L., 59–60, 71 Malvern, D., 59 Manchon, R., 152, 155 Mani, J. E., 232, 241, 246 Marin, J., 153 Martin, M. A., 204 Martin, R. C., 98, 100 Martin, S. H., 204 Martlew, M., 55 Marton, F., 220 Mason, L., 204 Massaro, D. W., 102 Mathieson, M., 110 Matsuhashi, A., 14, 36, 55 Matsumoto, K., 151, 153–154 Mayer, M., 59, 71 Mayer, R. E., 110–111, 122 Mc Carthy, P., 225 McCarthy Young, K., 241 McClelland, J. L., 127 McCloskey, M., 35 McCutchen, D., 15, 55, 68, 71, 81, 98, 100–101, 122, 201–202, 217 McDonald, J. L., 14 McGinley, W., 203, 232, 246 McGuinness, C., 184 McGuinness, D., 184 McKeachie, W. J., 295, 297, 299, 304 McKee, G., 59 McLaughlin, T. F., 110 McNish, M. M., 238 McShane, A., 110–111, 122 McTear, M. F., 305 Meier, H., 41 Meier, S., 225 Menard, N., 74 Meulenbroek, R. G., 31

Meyerhoffer, M., 110–111, 122 Miceli, G., 35 Mildes, K., 15, 68, 71, 81, 101, 202 Mitchell, E., 232, 241, 246 Miyake, A., 97 Mohindra, N., 34 Monereo, C., 110 Monteil, J. M., 101, 152, 168 Moore, T., 323 Mora, J. J., 110 Morgan, C. H., 110 Morgan, M., 323 Morris, J., 110, 328 Mounoud, P., 32 Mueller, S., 293–294 Mullet, E., 204, 215 Murphy, L., 152, 155 Murray, D. M., 203 Nash, J. G., 126 Nation, K., 184 Nauclér, K., 73 Nelson, J., 233, 235, 246 Nelson, L., 329 Nelson, N., 231 Neuwirth, C., 328 Newell, G. E., 203 Nist, S. L., 110 Nivre, J., 69 Nofsinger, R. E., 272 Norton, L. S., 110 Nottbusch, G., 32–35, 41, 52 Nystrand, M., 271 Oakhill, J., 110 O’Hare, E. A., 305 Okelberry, C., 329 Olive, T., 2, 14, 16, 55, 57–58, 63–64, 73, 98, 103, 111–112, 114–115, 119 Oliver, R., 294 Olson, D. R., 2–3, 182 Oostdam, R., 129 Orliaguet, J., 32, 34 O’Ryan, L., 223


10/20/2006

3:33 PM

Page 371

Author Index Osgood, C. E., 14 Ostry, D. J., 32, 52 Paivio, A., 100–104, 107 Pankoke-Babatz, U., 331 Passerault, J. M., 14, 18, 28, 108 Pennebaker, J., 280, 299–300, 304 Penney, C. G., 55–56 Perfetti, C. A., 112–113 Piaget, J., 204–205 Pichert, J. W., 167 Piepenbrock, R., 35 Piolat, A., 2, 57, 64, 103, 110–115, 119–122 Plowman, L., 323 Podgorny, P., 99 Pollatsek, A., 205 Pontecorvo, C., 186 Poskiparta, E. H., 202 Power, M. J., 99, 101, 107, 168 Powney, J., 235 Pressley, M., 237 Prinz, W., 331 Psathas, G., 272 Pynte, J., 32 Quinlan, T., 268, 280 Radach, R., 16 Ragnarsdóttir, H., 70 Ramanaiah, N., 221–222 Ramsden, P., 220 Ransdell, S., 58, 68, 99–100, 107, 122, 129, 152, 175, 202, 252, 286, 304–305 Raphael, T. E., 232, 246 Ravid, D., 70 Rayner, K., 16, 18 Remy, E., 202 Reyes, H., 304 Ribich, F. D., 221–222 Richards, B., 59 Rickards, J. P., 114 Rijlaarsdam, G., 5, 29, 96, 127–129, 131–132, 134, 148, 307, 312 Risch, N., 110–111, 122 Rivard, L. P., 201

371

Robertson, D. A., 29 Robinson, D. H., 110 Robinson, E. J., 152, 202, 305 Robinson, G. M., 84 Roca de Larios, J., 152, 155 Rodriguez, H., 329–330 Romani, C., 35 Romero, M., 328 Rommetveit, R., 271 Rosaen, C. L., 201 Rosenberg, S., 101 Roskelley, D., 110–111, 122 Rosner, J., 183 Roussey, J.-Y., 103, 110–115, 120–122 Ruchkin, D. S., 98, 101 Ruhl, K. L., 110 Rumelhart, D. E., 127 Rummer, R., 173 Rutberg, J., 202 Sachs, J., 169 Sacks, H., 169 Sadoski, M., 100–104, 107 Sahel, S., 34–35, 52 Saljo, R., 220 Salomon, G., 294–295 Sampson, G., 185 Sanchez, R. P., 111 Sánchez-Casas, R., 185 Sanders, T., 14, 127 Sarig, G., 234 Sasaki, M., 151–153, 155, 158 Saunders, M. A. P., 231 Scardamalia, M., 2, 4, 14, 55, 58, 68–69, 126–127, 146, 152–153, 166, 201–203, 208, 211, 216, 221, 270–271, 294, 308, 320 Scerbo, M. W., 111, 114 Schegloff, E., 169 Schellens, P. J., 252, 281, 283, 304 Schelstraete, M. A., 18, 112–113, 115 Schenk, S. M., 220 Schilperoord, J., 14, 31, 33, 57, 64, 126–127, 254, 280, 288 Schmeck, R. R., 220–222, 225, 238


372

10/20/2006

3:33 PM

Page 372

Author Index

Scholes, R. J., 184 Schooler, L. J., 167 Schriver, K. A., 270 Schroeder, J. L., 29 Schumacher, G., 232 Segev-Miller, R., 232, 234–235, 239, 241, 247–248 Severinson-Eklundh, K., 58, 88, 252, 324 Shaffer, L. H., 32 Shah, P., 97 Shankweiler, D., 199 Sharples, M., 323 She, H.-C., 204 Shell, D., 225 Shelton, J. R., 98, 100 Shepard, R. N., 99 Shin, R. K., 99–100, 102 Siegler, R. S., 205 Silva, T., 151, 153 Silverman, I., 14 Simon, D., 183 Simon, H. A., 167, 235–236 Skok, R. L., 110 Slattery, P. J., 233, 238 Sleurs, K., 254, 257 Slotte, V., 110 Smagorinsky, P., 236, 247, 252 Smith, E. E., 98 Smith, J., 223 Smith, S., 110 Smolkin, L. B., 202 Snellings, P., 68 Snow, C., 323 Songer, N. B., 122 Soria, E., 304 Sotto, E., 280, 299 Spelman Miller, K., 73, 91, 94 Sperber, D., 2 Spires, H. A., 110 Spiro, R. J., 239, 308 Spivey, N. N., 203, 231–232, 237–238, 246 Stahl, E., 307–308, 312, 320 Stahl, S. A., 238 Sternberg, R., 232

Sternglass, M., 233, 235, 246 Stewart, S. R., 202 Strauss, A. T., 237 Street, B., 305 Strömqvist, S., 16, 57, 59–60, 63, 70–71 Strømsø, H. I., 233 Stuart, G., 184 Sullivan, J. F., 114 Sullivan, K. P. H., 84–85, 87, 91, 94–95 Suritsky, S., 110 Swanson, H. L., 98, 101, 202 Swarts, H., 236 Sweller, J., 111 Swerts, M., 69 Swick, R., 329 Talamo, A., 307 Tang, C., 220, 222–223, 227 Tannenbaum, P. H., 14 Taube, K., 68 Teberosky, A., 182, 186, 188 Terzuolo, C. A., 31 Tetroe, J., 151, 153–155 Thomas, G. V., 152, 202, 305 Thompson, D. M., 177 Thompson, R. L., 204 Thunin, O., 103, 114 Tierney, R. J., 203 Titsworth, B. S., 114 Tkacz, S., 110 Toglia, M., 103 Tolchinsky, L., 186 Torrance, M., 5, 112, 152, 202–203, 305 Treiman, R., 183, 185 Trevor, J., 329 Tulving, E., 177 Tynjälä, P., 204, 216, 232, 293 Uzawa, K., 151 Van den Bergh, H., 5, 7–8, 29, 73, 96, 125, 127–129, 131–132, 134, 148, 152–153, 307, 312, 320 Van der Hoeven, J., 131


10/20/2006

3:33 PM

Page 373

Author Index Van der Linden, M., 18, 112–113, 115 Van Dijk, T. A., 231 Van Galen, G. P., 31–32 Van Gelderen, A., 202 Van Hout, R., 74 Van Rijn, H., 35 Van Rossum, E. J., 220 van Sledright, B., 246 Van Waes, L., 73, 88, 252, 254, 257, 280–281, 283, 290, 304, 320 Vauras, M. M. S., 202 Vernon, S., 181, 188, 191, 199 Verschueren, J., 252 Viavoice, 268 Vidal-Abarca, E., 304 Viviani, P., 31 Voeten, M. J. M., 202 Voionmaa, K., 74 Vorwerg, C., 173 Voss, J. F., 201, 203–204 Vygotsky, L., 221 Ward, J., 35 Warm, J. S., 111, 114 Waters, G. S., 98 Watters, J. J., 203, 215 Watts, M., 235 Wegener, M., 84, 86 Weingarten, R., 32–35, 41, 52 Weishaar, M., 110 Well, A. D., 205

Wengelin, A., 57–58, 60, 67, 69–71, 73, 75–76, 291 Whissell, C., 304 Whitaker, D., 202 Wigmore, B., 202 Wilcox, R. R., 318 Wilding, J., 34 Wiley, J., 201, 203–204 Wilkinson, B., 201, 204 Will, U., 32–35, 41, 52 Williams, F., 14 Williams, N. R., 268 Williams, R. L., 110 Winograd, P., 203 Witchalls, C., 268 Witte, S. P., 94, 221 Wood, B. S., 14 Wood, C. C., 323 Woodhouse, R. A., 110 Yaffee, L. S., 98, 100 Yang, S. C., 245 Yates, C., 202 Yeung, A. S., 111 Yopp, H. K., 182–183 Zellermayer, M., 294–295 Zesiger, P., 32 Ziegler, J. C., 103, 114 Zimmerman, B. J., 225 Zucchermaglio, C., 186 Zuercher, N., 220, 223, 225

373


Else_SW-Torrance_subjectind.qxd

10/20/2006

7:07 AM

Page 375

Subject Index Academic writing, 59, 219, 269–271, 273 Adaptation processes, 282 Audience perspective, 227, 307–311, 315, 317–321 Bilingual writers, 7, 151–153, 155, 159 Cognitive Flexibility Theory (CFT), 308–309, 319–320 Cognition, 1, 3, 7–10, 13, 31, 55, 67, 83, 97, 109, 125, 151, 163, 165, 167, 178, 181–182, 201, 219, 231, 251, 267, 279, 293, 307, 323 Cognitive capacity, 16, 24, 27–28, 63, 68, 71, 80–81, 169, 182 Cognitive processes, 1–2, 4, 6, 8–9, 31, 51–52, 68, 83–84, 127, 129, 178, 182–183, 202–203, 231, 233, 247, 282, 307, 309, 312–313, 315, 317–321 College learners, 232 Collaborative writing, 10, 323–325, 327–328, 330, 334 Computer–aided writing programs, 295 Concurrent protocol, 253–254, 256, 259–261 Data mining, 6, 83–85 Development, 4–5, 8, 165, 182, 184–186, 188, 191, 198–199, 201, 204–205, 216–217, 219, 223–224, 227, 236, 238, 253, 268, 271–272, 274, 331 Dictation, 268–269, 297

Discourse synthesis, 231–235, 237–238, 241, 244–248 Disfluencies, 68–70, 72–73, 75, 78–79, 82 Education, 1, 175, 177, 179, 204, 214, 225–226, 234, 248, 267, 269 Eye movements / Ocular movements, 5, 12, 16, 29, 53, 64 Framing devices, 91, 92, 93, 110 Generating strategies in writing, 159 Geographical Information Systems (GIS), 6, 83–85, 189–190 Hypertext, 10, 307–315, 317–321, 325–326 Idea generation, 125–126, 128–129, 131–132, 148, 152, 159, 202, 270–272, 276–277, 326 Imagery, 97, 101–102, 106 Intertextual processing, 231–234, 237–238, 241, 243, 245 Keystroke logging, 71–72, 84 Knowledge transforming, 5, 14, 126, 153, 221, 320 Language production, 31–32, 51, 67–69, 153, 165–166, 168–169, 174, 176, 178, 182, 202 Listening, 7, 109–122, 166, 172–173, 175–177, 273


376

10/20/2006

7:07 AM

Page 376

Subject Index

Models of writing, 5, 125–126 Motor execution, 6, 13–16, 20, 23–28, 32, 51–52, 55, 57–58, 64, 169 Note-taking, 109–111, 113, 116–122 Parallel processing, 6, 13–16, 21, 27–29 Pause analysis, 40, 73, 251, 253, 260–261 Pauses, 13–15, 18, 21, 24–25, 27–28, 31, 36, 40, 50, 55, 57–58, 60–64, 68–73, 75–77, 80–81, 182, 253–255, 259–260, 289 Planning, 4–6, 14–16, 23, 31, 36–37, 50–52, 56–57, 64, 68–70, 76, 97–98, 100–102, 107–108, 126, 129–130, 153–154, 166, 168–169, 182, 202–203, 221–222, 224, 229, 233, 236–237, 248, 268–269, 276, 280, 294, 311–314, 317–318, 320, 326, 328, 331 Preformulation, 251–253, 257, 259, 261 Press releases, 8–9, 251–257, 259–260 Process-log, 231, 235, 236, 237, 244, 247 Reading, 2–5, 7, 16–18, 28–29, 67–68, 70–72, 75, 80, 100–102, 109–123, 128–130, 132–137, 147, 152–154, 157–161, 166, 177, 182, 184–187, 189, 194–195, 199, 220, 231–234, 236–237, 244–248, 260, 268, 296, 298, 308, 313 Reading working-memory span, 109, 112, 113, 115,116, 117, 118, 119, 120, 122, 201 Retrieval, 4, 7, 28–29, 68, 70, 79, 81, 101, 126, 152–153, 166–169, 174, 178 Retrospective protocol, 251–254, 256–261 Reviewing, 4–5, 8, 15, 98, 154, 208, 210–216, 284, 325, 328 Sentence production, 6, 31–32, 36–37, 49–50, 52–53, 68, 97–100, 104, 107–108

Skilled writers, 69, 81, 151–152, 154, 159 Speech recognition, 1, 9, 267–269, 273, 276, 279–287, 289–291 Spelling, 6, 19, 56, 67–69, 71–75, 77, 79–81, 98–100, 103, 182, 184–185, 187, 202, 224, 260, 270–271, 300, 334 Stimulated recall, 83, 251, 253, 260 Strategies, 4, 7–9, 14, 63, 67, 70, 74, 77, 79–82, 110, 112, 115, 122, 126, 128, 131, 137, 151–155, 157–159, 161, 178, 182, 201, 203, 205, 208–211, 213, 215–217, 219–222, 226–227, 231–240, 242–248, 269, 271, 279, 282, 289–290, 294, 317, 323–324 Syllabic structure, 35, 52, 185–186, 195, 197 Syntactic structures, 31, 37, 39, 49 Text quality, 58, 62–64, 125–126, 128, 131–132, 137, 140–141, 143–144, 146–150, 280 Transforming strategies, 233, 237–238, 240, 242, 244, 246 Transfer of writing skills from L1 to L2, 152, 153, 154 Translating, 3–4, 7, 64, 97–98, 100, 107–108, 125–126, 129, 133, 144, 146–147, 152, 159, 221, 240, 244, 270, 272 Typing, 2, 6, 9, 31–34, 36–37, 39–43, 49–51, 53, 55–58, 60–61, 63–64, 71, 73, 76, 98, 100, 170, 174, 270, 299 Typing skill, 32, 55–58, 63 Verbal working memory, 6, 53, 97–103, 105, 107–108 Visual working memory, 6, 53, 97, 100–103, 106–108 Vocabulary, 6, 59–62, 67, 69–71, 73–74, 77–81, 100, 103, 182, 202, 204, 217, 220, 243 Voice recognition, 268, 293


10/20/2006

7:07 AM

Page 377

Subject Index Working memory, 5–6, 17, 26, 28, 53, 56, 68, 97–109, 111–112, 119, 122, 127, 168–169, 174, 201, 217, 270, 272–273 Writing across the curriculum, 235, 248 Writing, 1–11, 13–20, 24–29, 31–36, 40, 50–53, 55–61, 63–64, 67–73, 75, 78–85, 97–99, 101–102, 104–109, 112, 114, 122–123, 125–148, 150–159, 161, 163, 165–178, 181–188, 190–194, 196–199, 201–205, 207–217, 219–240, 245–248, 251–256, 258–261, 265, 267–291, 293–300, 307–309, 311–312, 319–321, 323–328, 330–334

377

Writing modes, 284–287, 289–290 Writing processes, 5, 8, 11, 13, 16, 27, 29, 56–57, 63, 68–69, 83, 127–129, 142, 144, 151, 154, 203, 219, 221, 223, 231–232, 235, 238, 248, 269–271, 279–281, 289–291, 293–294, 321 Writing strategies, 7, 9, 151–154, 182, 201, 205, 208–211, 213, 215–216, 221, 289, 294 Writing superiority effect, 7, 81, 165–166, 171–178, 204 Writing to learn, 201, 203–205, 216–217 Writing–from–sources, 132, 231


Else_SW-Torrance_lovol.qxd

10/23/2006

2:47 PM

Page 379

List of Volumes Volume 1: Theories, Models and Methodology in Writing Research Gert Rijlaarsdam, Huub van den Bergh, Michel Couzijn (Eds) 1996 558pp., Paperback ISBN 90-5356-197-8 Volume 2: Effective Teaching and Learning of Writing. Current Trends in Research Gert Rijlaarsdam, Huub van den Bergh, Michel Couzijn (Eds) 1996 388pp., Paperback ISBN 90-5356-198-6 Volume 3: The Cognitive Demands of Writing. Processing Capacity and Working Memory Effects in Text Production Mark Torrance, Gaynor Jeffery (Eds) 1999 113pp., Paperback ISBN 90-5356-308-3 Volume 4: Knowing What to Write. Conceptual Processes in Text Production Mark Torrance, David Galbraith (Eds.) 1999 190pp., Paperback ISBN 90-5356-307-5 Volume 5: Foundations of Argumentative Text Processing Pierre Coirier, Jerry Andriessen (Eds) 2000 273pp., Paperback ISBN 90-5356-340-7 Volume 6: Metalinguistic Activity in Learning to Write Anna Camps, kMarta Milian (Eds) 2000 228pp., Paperback ISBN 90-5356-341-5 Volume 7: Writing as a Learning Tool Päivi Tynjälä, Lucia Mason, Kirsti Lonka (Eds) 2001 Hardbound, ISBN 0-7923-6877-0; Paperback, ISBN 0-7923-6914-9 Volume 8: Developmental Aspects in Learning to Write Liliana Tolchinsky (Ed.) 2001 Paperback, ISBN 0-7923-7063-5; Hardbound, ISBN 0-7923-6979-3 Volume 9: Through the Models of Writing Denis Alamargot, Lucile Chanquoy (2001) Paperback, ISBN 0-7923-7159-3; Hardbound, ISBN 0-7923-6980-7 Volume 10: Contemporary Tools and Techniques for Studying Writing Thierry Olive, C. Michael Levy (Eds) 2001 Hardbound, ISBN 1-4020-0035-9; Paperback, ISBN 1-4020-0106-1 Volume 11: New Directions for Research in L2 Writing Sarah Ransdell, Marie-Laure Barbier (Eds) 2002 281pp., Paperback, ISBN 1-4020-0539-3; Hardbound, ISBN 1-4020-0538-5 Volume 12: Teaching Academic Writing in European Higher Education Lennart Björk, Gerd Bräuer, Lotte Rienecker, Peter Stray Jörgensen (Eds) 2003 240pp., Hardbound, ISBN 1-4020-1208-X; Paperback, ISBN 1-4020-1209-8 Volume 13: Revision: Cognitive and Instructional Processes Linda Allal, Lucile Chanquoy, Pierre Largy (Eds) 2004 248pp., Hardbound, ISBN 1-4020-7729-7 Volume 14: Effective Learning and Teaching of Writing. A Handbook of Writing in Education Gert Rijlaarsdam, Huub van den Bergh, Michel Couzijn, M. (Eds) 2nd ed., 2004, X, 670pp., 21 illus., Hardcover ISBN 1-4020-2724-9; Softcover ISBN 1-4020-2725-7 Volume 15: Writing in Context(s). Textual Practices and Learning Processes in Sociocultural Settings Triantafillia Kostouli (Ed.) 2005 280pp., Hardcover ISBN 0-387-24237-6; Softcover ISBN 0-387-24238-4 Volume 16: Teaching Writing in Chinese Speaking Areas Mark Shiu Kee Shum, De Lu Zhang (Eds) 2005 276pp., Hardcover ISBN 0-387-26392-6 Volume 17: Writing and Digital Media Van Waes, Leijten, Neuwirth (Eds) 2006 380pp., Hardcover ISBN 0-08-044863-1 Volume 18: Computer Key-Stroke Logging and Writing Sullivan, Lindgren (Eds) 2006 248pp., Hardcover ISBN 0-08-044934-4 Volume 19: Writing and Motivation Suzanne Hidi, Pietro Boscolo (Eds) 2007 346pp., Hardcover ISBN 0-08-045325-2


Writing and Cognition, Volume 4: Research and Applications (Studies in Writing) (Studies in Writing) (Studies in Writing)

Writing and Digital Media, Volume 17 (Studies in Writing) (Studies in Writing)

Writing and Motivation, Volume 19 (Studies in Writing) (Studies in Writing)

Writing and Digital Media, Volume 17 (Studies in Writing)

Teaching Academic Writing in European Higher Education (Studies in Writing)

Teaching Writing in Chinese Speaking Areas (Studies in Writing)

Effective Learning and Teaching of Writing: A Handbook of Writing in Education (Studies in Writing)

Creative Writing Studies: Practice, Research and Pedagogy (New Writing Viewpoints)

Persuasive Writing (Writing 4)

Computer Key-Stroke Logging and Writing, Volume 18: Methods and Applications (Studies in Writing)

Research Writing in Dentistry

Process This: Undergraduate Writing in Composition Studies

Descriptive Writing (Writing 4 Series)

Narrative Writing (Writing 4 Series)

Expository Writing (Writing 4 Series)

Writing in Context(s): Textual Practices and Learning Processes in Sociocultural Settings (Studies in Writing)

Why Writing Matters: Issues of Access and Identity in Writing Research and Pedagogy (Studies in Written Language and Literacy)

Academic Writing in Context: Implications and Applications

Embodying Technesis: Technology beyond Writing (Studies in Literature and Science)

Embodying Technesis: Technology beyond Writing (Studies in Literature and Science)

Writing and Empire in Tacitus

In Respect to Egotism: Studies in American Romantic Writing (Cambridge Studies in American Literature and Culture)

Keywords in Creative Writing

Career Opportunities In Writing

Techniques in teaching writing

Writing in Society

Writing in Paragraphs

Writing Successfully in Science

Making Progress in Writing

Writing (English in Context)

Keywords in Creative Writing

Writing and Cognition, Volume 4: Research and Applications (Studies in Writing) (Studies in Writing) (Studies in Writing)

Writing and Digital Media, Volume 17 (Studies in Writing) (Studies in Writing)

Writing and Motivation, Volume 19 (Studies in Writing) (Studies in Writing)

Writing and Digital Media, Volume 17 (Studies in Writing)

Teaching Academic Writing in European Higher Education (Studies in Writing)

Teaching Writing in Chinese Speaking Areas (Studies in Writing)

Effective Learning and Teaching of Writing: A Handbook of Writing in Education (Studies in Writing)

Creative Writing Studies: Practice, Research and Pedagogy (New Writing Viewpoints)

Persuasive Writing (Writing 4)

Computer Key-Stroke Logging and Writing, Volume 18: Methods and Applications (Studies in Writing)

Research Writing in Dentistry

Process This: Undergraduate Writing in Composition Studies

Descriptive Writing (Writing 4 Series)

Narrative Writing (Writing 4 Series)

Expository Writing (Writing 4 Series)

Writing in Context(s): Textual Practices and Learning Processes in Sociocultural Settings (Studies in Writing)

Why Writing Matters: Issues of Access and Identity in Writing Research and Pedagogy (Studies in Written Language and Literacy)

Academic Writing in Context: Implications and Applications

Embodying Technesis: Technology beyond Writing (Studies in Literature and Science)

Embodying Technesis: Technology beyond Writing (Studies in Literature and Science)

Writing and Empire in Tacitus

In Respect to Egotism: Studies in American Romantic Writing (Cambridge Studies in American Literature and Culture)

Keywords in Creative Writing

Career Opportunities In Writing

Techniques in teaching writing

Writing in Society

Writing in Paragraphs

Writing Successfully in Science

Making Progress in Writing

Writing (English in Context)

Keywords in Creative Writing

Recommend Documents