Clinical linguistics: theory and applications in speech pathology and therapy

CLINICAL LINGUISTICS

AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E. F. KONRAD KOERNER (University of Ottawa) Series IV – CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board Raimo Anttila (Los Angeles); Lyle Campbell (Christchurch, N.Z.) Sheila Embleton (Toronto); John E. Joseph (Edinburgh) Manfred Krifka (Berlin); Hans-Heinrich Lieb (Berlin) E. Wyn Roberts (Vancouver, B.C.); Hans-Jürgen Sasse (Köln)

Volume 227

Elisabetta Fava (ed.) Clinical Linguistics Theory and applications in speech pathology and therapy

CLINICAL LINGUISTICS THEORY AND APPLICATIONS IN SPEECH PATHOLOGY AND THERAPY

Edited by

ELISABETTA FAVA Università di Ferrara

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA

To

Alberto Mioni for his 35 years of dedication in the teaching of General Linguistics

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data Clinical Linguistics: theory and applications in speech pathology and therapy / edited by Elisabetta Fava. p. cm. -- (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in Linguistic theory, ISSN 0304-0763; v. 227) Includes bibliographical references and index. 1. Speech disorders. 2. Speech theraphy. 3. Applied linguistics. I. Fava, Elissabetta. II. Series. RC432 .C556 2002 616.85'506--dc21 2002025406 ISBN 90 272 4735 8 (Eur.) / 1 58811 223 3 (US) (Hb; alk. paper) © 2002 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O.Box 63224 • 1020 ME Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA

ACKNOWLEDGMENTS I am indebted to many people for making this volume possible. First of all, to the authors of the articles here included, for having presented and discussed their theses at the conference "Linguistics Theory in Speech and Language Pathology and in Speech Therapy", held in Padova in August 2000. Their help and co-operation with the editorial work has been invaluable. My intellectual debt goes to my colleagues of the Department of Linguistics of the University of Padova, Alberto Mioni and Alberto Zamboni. They helped me constantly with my project to involve the Department of Linguistics in the offering of Clinical Linguistics courses at the Faculty of Medicine. Without their unstinting support, this project would never have materialised. This Padova conference was organised with the support of a grant from Cofinanziamento Murst (1998-9810197024), whose national co-ordinator was Professor Alberto Mioni. It was important for this project to succeed that I was at the time a member of the Psychology Faculty of the University of Padova. This volume is giving me the opportunity to thank my colleagues of the Faculty of Psychology of the University of Padova, Remo Job, Alberto Mazzocco and Sergio Roncato, for having first supported the role of linguistics within the psychological sciences, for having fostered the academic co-operation between researchers from various disciplines, and for having included, already ten years ago, the teaching of General Linguistics within the Psychology curricula. Likewise, I would also like to express my gratitude to my colleagues in the Faculty of Medicine of Padova, Corrado Angelini, Anna Maria Laverda, and Carla Monciotti. To all of them, I extend my gratitude for believing that lin­ guistics can play an important role in cognitive and neurolinguistic research, and for their active interest in humanistic disciplines. The delegate of the Rector of the University of Padova, Filippo Donà dalle Rose, responsible for the Socrates exchange program of the Padova Athe­ naeum, has made incredibly tenacious and generous efforts to involve students and professors in the organisation of the Erasmus-Socrates Projects. The di­ dactic experience with Socrates students and colleagues was an important moment of this project. I owe fruitful discussions on the topics of this volume to Paola Crisma of the University of Trieste, Denis Delfitto of the University of Utrecht, Giuseppe Longobardi of the University of Trieste, Anna Thornton of

VI

ACKNOWLEDGMENTS

the University of L'Aquila, and Nicola Grandi of the University of Milano Bicocca. I also would like to thank my new colleagues of the Humanities Faculty of Ferrara, Claudio Iacobini, Marina Nespor, and Carlotta Sanfilippo. I hope to be able to collaborate with them on future projects. Tullio De Mauro of the University of Rome was generous with information on and discussion of the structuralist tradition. I would also like to express my gratitude to my colleagues of the Master's degree program in Logopedics in Padova, especially Paola Crisma, Mario D'Angelo, Sara Gesuato, Carlo Schirru, Flavia Ursini, and Alessandro Zijno. Their contribution in defending the project of a Clinical Linguistics degree at the Faculty of Medicine in Padova has been invaluable. As well, I owe an expression of thanks to Sara Gesuato for her editing of the first draft of the volume; I also remain grateful to her for the emotional support she offered me during this period. To Nadia Radovich and to Alberto Mello, of the printing desk of Palazzo Maldura of the University of Padova, goes my expression of thanks for helping in the formatting of the volume, for their technical support, for sharing their experience with me, and for their kind co-operation generally. Last but not least, I am grateful to E.F.K. Koerner, the editor of "Current Issues in Linguistic Theory", who first proposed the publication of this vol­ ume, for keeping an active interest in the process and progress in the pre­ paration of the final copy. His work as editor and his role as moderator in the linguistic sciences cannot be overestimated. He provided invaluable criticism, comments and directions in editing this volume. It has been a great privilege for me to benefit from his unequalled experience. I also would like to express my gratitude to Anke de Looper of John Benjamins Publishing Company for her patience and kindness. Ferrara, May 2002

Elisabetta Fava

CONTENTS Acknowledgments Editor's Introduction I . PHONOLOGY IN CLINICAL APPLICATIONS Yishai TOBIN Phonology as human behavior: Theoretical implications and cognitive and clinical applications

v ix

3

Dirk-Bart D E N OUDEN

Segmental vs syllable markedness: Deletion errors in the paraphasias of fluent and non-fluent aphasics

23

II WORDS IN DEAFNESS AND STUTTERING Roberto AJELLO, Giovanna MAROTTA, Laura MAZZONI & Florida NICOLAI

Morphosyntactic fragility in the spoken and written Italian of the deaf 49 Peter HOWELL & James Au-YEUNG The EXPLAN theory of fluency control applied to the diagnosis of stuttering 75 Peter HOWELL The EXPLAN theory of fluency control applied to the treatment of stuttering 95

III. MORPHOLOGY AND SYNTAX IN CHILD LANGUAGE DISORDERS Roelien BASTIAANSE, Gerard BOL, Sofie VAN MOL & Shalom ZUCKERMAN Verb movement and finiteness in language impairment and language development Stavroula STAVRAKAKI A-bar movement constructions in Greek children with SLI: Evidence for deficits in the syntactic component of language Susan M. SUZMAN Morphological accessibility in Zulu

119

131 155

vin

TABLE OF CONTENTS

Yumiko TANAKA WELTY, Jun WATANABE & Lise MENN Language production in Japanese preschoolers with SLI: Testing theories

175

IV. ISSUES ON GRAMMAR AND COGNITION Leah PALTIEL-GEDALYOVICH Testing linguistic concepts: Are we testing semantics, syntax or pragmatics ? Dusana RYBÁROVÁ SLI and modularity: Linguistic and non-linguistic explanations Vesna STOJANOVIK, Mick PERKINS & Sara HOWARD The language/cognition interface: Lessons from SLI and Williams Syndrome

197 213

229

V. GRAMMATICAL STRUCTURE IN APHASIA Susan EDWARDS Grammar and fluent aphasia Anna GAVARRO Failure to agree in agrammatism Judith RISPENS, Roelien BASTIAANSE & Susan EDWARDS The verb and sentence test: Assessing verb and sentence compre­ hension and production in aphasia Esther RUIGENDIJK Case assignment as an explanation f or determiner omission in German agrammatic speech Kyrana TSAPKINI, Gonia JAREMA & Eva KEHAYIA The role of verbal morphology in aphasia during lexical access: Evidence from Greek

249

Index of Subjects

337

List of Contributors

267

279

299

315

EDITOR'S INTRODUCTION The present volume deals with different aspects of speech and language pathology and identifies and re-examines, from various perspectives, a number of standard assumptions in clinical linguistics and cognitive science. It encom­ passes issues concerning deafness, stuttering, child language acquisition, Spe­ cific Language Impairment (SLI), William's Syndromes deficit, fluent aphasia, and agrammatism. Different levels of linguistic analysis are considered: phonetics, phonology, syntax, semantics, and pragmatics. Some of their prop­ erties, their formal representations and their interfaces with other levels, are examined closely and clarified. Researchers typically summarise their individual results in individual con­ tributions, not only in the format of an article (subjects, materials, languages, scores, etc.), but also through critical discussion of the relevant literature. Their work is located in different lines of research: structuralist tenets, genera­ tive approaches, and frameworks such as Optimality Theory. The topics dis­ cussed are intricate and complex and, at the same time, wide-ranging in scope, but the chapters in the present book offer a fairly comprehensive overview of the complexity and the emerging importance of the field of Clinical Linguistics. The analyses presented here reflect various aspects of the at times fierce theo­ retical and empirical debates currently ranging over almost every issue dis­ cussed, with respect to linguistics and the other cognitive disciplines. Although it is very difficult to do justice to all of the subtleties of argumentation that each of these lines of investigation require, the rather technical issues discussed in this volume have a bearing on questions of considerable interest. They presup­ pose or imply assumptions about the internal architecture of the language fac­ ulty, whose location among other systems of the mind/brain is not at all obvi­ ous. The interdisciplinary complexity of the language/cognition interface is also explored by focusing on empirical data of different languages: among them, Germanic languages (such as Dutch, English, German), Greek, Hebrew, Japa­ nese, Romance languages (such as Catalan, Italian, and Spanish) and Sohsho, a Zulu language. Despite the different approaches and the variety of problems posed in this volume, there is a common aim to the chapters. The authors piece together various fragments of clinical linguistic research, trying to bring them into a

X

ELISABETTAFAVA

more cohesive whole, and offer a sense of some of the technical problems that lie at the forefront of research and suggest the kind of answers that their work may provide. The aim of this volume is to stress the growing importance of the theoreti­ cal and methodological tools developed in Clinical Linguistics; to put under scrutiny assumptions taken for granted in previous analyses, which may not be as obvious as they seem; to investigate how even apparently minimal choices in the description of phenomena may affect the form and complexity of the lan­ guage/cognition interface. What is called 'clinical work' is not separable from the scientific work done by linguists; instead, it should be considered as a com­ ponent part of General Linguistics. It is at the same time concerned with the actual foundation of the study of linguistic pathologies, together with neurobi­ ology, psychology and neurology and the outer reaches of scientific specula­ tion about the nature of the mind/brain complex with regard to language. In this perspective, it is very important to defend the continuity of linguistic speculations beyond different frameworks. Such a continuity, which has often been emphasised and is among the goals of the CELT series, of which this vol­ ume is a part, is even more important in clinical applications. The study of the relationship between language and the brain, or rather between language disor­ ders and brain lesions goes back to the beginning of the 19th century. Begin­ ning with the pioneering, albeit contentious, work of Franz Joseph Gall (17581828), the observations, arguments and opinions advanced in the research of Paul Broca (1824-1880), Carl Wernicke (1848-1905) or Pierre Marie (18531940) are still today crucially important, considering the frequency with which they are cited in current research papers.1 Since then, linguistics has included data and results from speech and language pathology, broadening its empirical basis and reframing data and research in speech and language pathology. The structuralist research, by reconsidering the traditions of this branch of science, has paved the way for the inclusion of language disorders research within linguistic theory. Considerations of the 19th-century discoveries of the French surgeon and anthropologist Paul Broca can be found in Ferdinand de Saussure's (1857-1913) work. We may also refer to some of the key concepts identified and discussed by Saussure himself, such as the methodological and ontological notions of langue and parole, which support and justify a large part of clinical linguistic research. They may be considered part of the implicit or explicit distinctions systematically adopted in any clinical work. Thirty years later, Roman Jakobson's systematic search for what was later described as the 1 See Paul Eling (ed.), Reader in the History of Aphasia: From [Franz] Gall to [Norman] Geschwind (Amsterdam & Philadelphia: John Benjamins, 1994).

INTRODUCTION

XI

elementary quanta of language offered tentative generalisations of acquisition and loss of grammatical systems. His hypotheses on child language acquisition and language breakdown, together with his inquiry into language change, are still part of the research into the structuration, restructuration and loss of lan­ guage. They deal with general tendencies, considered 'lawful' by Jakobson, whose status is still debated in the most recent literature.2 Empirical evidence in favour or against some of his hypotheses is offered and explicitly reconsidered in this volume. With Noam Chomsky's work, the impact of linguistic theory on the study of language in wider context as part of the scientific investigation of the archi­ tecture of the mind became significant. Many analyses proposed in this volume refer to rules and principles elaborated by Chomsky, although in slightly dif­ ferent frameworks, including various 'minimalist 'versions. One of the main phenomena that has motivated Chomsky since Syntactic Structures is the un­ limited possibility of expression in human language, what he now calls the dis­ crete infinity of language.3 In order for speakers of a language to create and understand sentences that they have never heard before, there must be a way of combining some finite number of memorised units into phrases and sentences of arbitrary length. The speaker's knowledge of the language must include a set of principles of combinations that determine which combinations are well formed (and which are ill-formed) and what they mean. The construction of minimal pair paradigms is one of the major features of experimental design concerning different abstract levels of representations. In their search for rigorous stipulations, linguists, like physicists, biologists or chemists, manipulate the environment experimentally with a central concern on issues of acceptability, as a pretheoretical term, and of grammaticality. The manner in which the linguist discusses pairs of grammatical sentences, juxta­ posed with their non-grammatical counterparts, has offered a way to under­ stand sentences, not to reject them. This approach has had a dramatic impact on clinical linguistic applications and in this volume too, this major point, inex­ tricably connected with a cluster of other issues, is reconsidered. Some fre­ quently observed features in the clinical literature, which are registered as 'er­ rors' concerning inflectional properties, determiners, or Wh-features, are not considered merely impressionistically: in this volume, the nature of the various kinds of assertions involved in identifying them as errors is discussed con2

Cf. Roman Jakobson, Studies on Child Language and Aphasia (The Hague: Mouton, 1971 [1941]), p. 51. 3 Cf. Noam Chomsky, New Horizons in the Study of Language and Mind (Cambridge: Cam­ bridge University Press, 2000), pp. 3-4, 184.

Xll

ELISABETTAFAVA

sistently. The way Chomsky sets up problems is an inducement to reflection and research not only when there is a convergence of methodological tools and theoretical apparatus, but also with differing and contrasting frameworks. Continuity in linguistic research has been even stronger in the last few decades, as issues on cognitivism and functionalism have been reconsidered, sometimes radically. It is claimed that cognitive abilities and experientially de­ rived cognitive models have direct and pervasive linguistic manifestations, and, conversely, that language structure furnishes important clues concerning basic language phenomena.4 By stressing that the conceptual setting for the work the linguist is doing is a scientific one, the research focuses on the role of larger contextual frames in describing language disorders: properties of different lan­ guages, principles of conversation, and various language functions. An under­ lying common feature of this approach, which is considered in this volume, concerns the role of semantic and pragmatic factors in assessing clinical data and evaluating therapeutic interventions. Many contributions to this volume show how delicate and difficult the treatment of these data is, involving deci­ sions which do not always find a consensus among researchers, as the debate on the empirical adequacy of many analyses suggests. Elsewhere I have de­ fended the relevance of semantics and pragmatics in controlling linguistic stipu­ lations in non-pathological frameworks: in dealing with descriptive and, more importantly, explanatory adequacy for some well-known phenomena such as interrogative or relative clause distinctions, crucial evidence is offered by re­ considering semantic and pragmatic aspects systematically.5 Attention to se­ mantics and pragmatics becomes even more crucial when dealing at the same time with both sentences and utterances, and when trying to apply distinctions at the boundaries between the grammatical and the pragmatic properties of what is said in contexts such as the linguistic production of a child with SLI or an aphasic patient. In this volume issues of appropriateness are discussed and considered as instruments to clarify the nature of such deficits. Moreover, it is an important merit of this volume that it discusses the role of semantic and pragmatic factors not only in the evaluation of deficits, but also in the perspec4 See Ronald W. Langacker, "Reference-point Constructions", Cognitive Linguistics 4.1-38 (1993). 5 Cf. Elisabetta Fava, "Questioning Interrogative Interpretation in Some Indo-European Lan­ guages", Contrastive Semantics and Pragmatics, vol. I: Meanings and representations ed. by Katarzyna Jaszczolt & Ken Turner (Oxford: Elsevier, 1996), 87-110; "Langue and Parole in Speech Act Theories: Some considerations and a proposal", Langue and Parole in Synchronic and Diachronic Perspective: Selected proceedings of the XXXIst Annual Meeting of the Societas Linguistica Europea (26-30 August 1998, St Andrews) ed. by Christopher Beedham (Oxford: Pergamon, 1999), 263-283.

INTRODUCTION

Xlll

tive of the therapy intervention with an attempt to take a larger perspective, opening up a discussion about the consequences and fruitfulness of different approaches. Another general remark should be made here. In volume contains some currently used terms which may cover empirically and conceptually distinct notions. A notion for which there may be no congruence of criteria is that of 'word': the same term may cover empirically and conceptually distinct concepts such as phonological word, content word, syntactic atom, morphological ob­ ject, lexical item, etc. Here phonological, syntactic and morphological criteria may be in conflict. Other crucial notions are 'subject' and 'object': although such terms appear frequently in linguistic argumentation (including in this vol­ ume), the notions underlying them may be radically different. They may be re­ garded as relating to semantic roles or considered pre-theoretically. The Rela­ tional Grammar model, instead, treats grammatical relations like Subject and Direct Object as primitive and central to grammar, while in the Government and Binding framework, X-bar type constituent structure representations are con­ sidered as basic.6 As a result, the interactions of the properties of Subject and Object, considered in rather different frameworks, may be analysed in rather different ways in this volume. After this lengthy introductory statement, let me now offer comments on the individual contributions to this volume. As the table of contents indicates, the volume is organised into five sections. The first section considers clinical applications in phonology. The data analysed are taken from different populations: child acquisition, deafness, flu­ ent and non-fluent aphasia, and other pathologies. The first essay, by Yishai Tobin (Ben Gurion University), is on Phonology as human behavior: Theoret­ ical implications and cognitive and clinical applications. By defending the con­ tribution of Structural Phonology to different clinical data applied to a large va­ riety of languages, he discusses the conflict between the communication and the human factors in language users' search for maximum communication with minimal effort both in the diachronic development and the synchronic state of a language. This conflict is even more keenly felt both in language acquisition, 6

Cf. Elisabetta Fava, "Contextualising Corpora in Testing Grammatical Hypotheses: Searching for preposed and postposed syntactic subjects in North Italian and Veneto childadult conversations", Lingua, Discourso, Texto: I Simposio internacional de analisis del discourso, vol. I, ed. by José Jesús de Bustos Tovar, Patrick Charaudeau, José Luis Giron Alconchel, Silvia Iglesias Recuero & Alonso Covadonga Lopez (Madrid: Visor Libros, 2000), 263-280; cf. David Perlmutter & Carol Rosen (eds.), Studies in Relational Grammar, vol. II (Chicago: University of Chicago Press, 1984).

XiV

ELISABETTAFAVA

where functional errors and processes may be observed, and in the clinical set­ ting, where developmental and pathological errors and processes become ap­ parent. The theory of phonology as human behaviour can explain, in a princi­ pled way, the connection and interrelationship between the phylogeny, the on­ togeny, and the pathology of the development of sound systems in human lan­ guages. Phonotactic skewing in language (diachronic, synchronic, and in de­ velopmental and pathological data) reflect the learning process of speakers. Tobin includes data of a hearing child of deaf parents and other series of func­ tional processes from different pathologies. The second contribution, by Dirk-Bart Den Ouden (Groningen University), is on Segmental vs. syllable markedness: Deletion errors in the paraphasias of fluent and non-fluent aphasics. It applies Optimality Theory to the analysis of deletion errors in the paraphasias of fluent and non-fluent aphasics, by compar­ ing the literal paraphasias of fluent and non-fluent aphasie speakers on a repeti­ tion task, aimed at determining the influence of syllable structure on error pat­ terns. Markedness is one of the key concepts of this chapter. Where phonolog­ ical theory has been applied to clinical data, the concept of markedness has of­ ten played a significant role. Not often taken into account, however, is the fact that the markedness value of linguistic structures may not be the same at all psycholinguistic levels of processing. What is marked at some linguistic level of representation may well be unmarked at another. The influence of different types of markedness on literal paraphasias may be related to specific psycholinguistic levels of processing and a difference between the pre-phonetic and the phonetic level of processing is hypothesised. The second section discusses problems of words in deafness and stutter­ ing. Although deafness is a pathology, which is widely discussed in the litera­ ture, the phenomena involved are not so often considered with sophisticated linguistic tools, especially with regard to writing systems where some linguis­ tic generalisations seem to be missed. The two papers on stuttering, instead, provide insight into a major area of speech phenomena that have traditionally been treated outside of linguistics proper. Stuttering is typically thought to re­ flect aspects of an individual's speech output that suggest problems with what are vaguely referred to as the performance systems. As well as other speech disorders demanding an explanation outside of the realm of formal grammar is not too much studied in clinical linguistics. Roberto Ajello, Giovanna Marotta, Laura Mazzoni & Florida Nicolai (University of Pisa) deal with The Morphosyntactic fragility in the spoken and written words of the deaf They analyse the linguistic production of profoundly deaf Italian people who received an oralist instruction in specialised institutions

INTRODUCTION

XV

and who never wore a prothesis. Their results reveal a discrepancy between a fairly good lexical competence and a poor morphological competence heavily dependent on the input, and a similarly poor syntactic competence, which relies fundamentally on pragmatic communication principles. The not sufficiently mastered morphology, and free rather than bound morphology, is explained on the basis of the process of learning, based mainly on general, not specifically linguistic, cognitive mechanisms. Peter Howell and his research group at Uni­ versity College London have elaborated a new perspective on stuttering, the EXPLAN theory, based on some linguistic insights, and offer an account of how differences between fluent and stuttered speech arise. Although the main focus is on fluent speech control, it is also relevant to the diagnosis and treat­ ment of stuttering. In the first paper, by Peter Howell & James Au-Young, The EXPLAN the­ ory of fluency control applied to the diagnosis of stuttering, the use of 'phono­ logical word' to predict the development of stuttering from childhood to adult­ hood is discussed. According to EXPLAN, the distinguishing characteristics of stuttering emerge in late childhood; a contrasting view is that people who stutter have incipiently different problems in controlling speech from the onset of language. Diagnosis of the disorder involves measuring and specifying the types of fluency failures that occur in late adolescence in order to see if they have changed in a way commensurate with adult stuttering. Early intervention may not be advisable in such suspected cases, not as long as the child is ex­ hibiting the types of fluency failure common to all children. The second paper, The EXPLAN theory of fluency control applied to the treatment of stuttering, Peter Howell poses the question of how speech can be manipulated, that is, how fluent speech can be made to contain fluency failures and how stuttered speech can be changed to make it more fluent. He proposes a linguistically motivated operant procedure for treating stuttering. A treatment for stuttering is considered successful if it decreases the incidence of fluency failures. EXPLAN theory is more specific as it requires a decrease in the inci­ dence of fluency failures that involve production of parts of words. Howell's results show the reduction in stalling fluency failures may be a result of the op­ erant procedures reducing overall speech rate, thereby cutting down all types of fluency failure. In the course of the treatment sessions, content word (advanc­ ing) fluency failures decreased significantly. The third section concerns morphology and syntax in child language disor­ ders. Four contributions discuss data from Dutch, Greek, Bantu, and Japanese in relation to a major issue, Specific Language Impairment (SLI). This term, applied to a child whose language development is substantially below age level

XVi

ELISABETTAFAVA

for no apparent cause, describes delays and/or disorders in the procedure of the acquisition of grammar in the absence of any hearing loss, mental retardation (performance IQ is within normal range), articulator motor impairment or psy­ cho-emotional disorders. The language development of SLI children is char­ acterised by severe problems in the acquisition of morphology and syntax, while their cognitive, motor and social development is considered to fall within the normal range. In attempts to identify the nature of the linguistic deficit in SLI, diverse linguistic models have been proposed. Although most authorities agree that SLI is heterogeneous, there is little consensus about how it should be classified. Based on cross-linguistic research findings indicating that inflec­ tional and derivational morphology is severely impaired in SLI children, several explanatory models of the linguistic deficit in SLI have been developed. They advance different hypotheses on the locus of the deficit in SLI grammar as well as on the nature of developmental patterns followed in SLI, i.e., whether lan­ guage development in SLI is a normal but delayed procedure or an abnormal process. The chapter by Roelien Bastiaanse, Gerard Boll, Sofia van Mol & Shalom Zuckerman (Groningen University) on Verb Movement and finiteness in lan­ guage impairment and language development poses two central questions. First, what is the origin of the problems with finite verbs in three populations, normally developing children, SLI children, and agrammatic aphasics? Second, whether is it true that SLI children deal with the problems of the production of finite verbs in the same way as normal children do? Although there are clear similarities between normally developing children and SLI children with respect to the production of finite verbs, the SLI children resemble agrammatic aphasics when it comes to strategies in circumventing their problems. According to the authors, the three populations have the same underlying problem, i.e., verb movement, but different ways to solve it. The normally developing children have problems with Verb Second position and circumvent these problems by inserting dummy auxiliaries, either from their dialect, or from a construction that in adult grammar has a different meaning. There is a clear dichotomy be­ tween the errors produced by the normally developing children and the aro­ matic speakers. The results of the Bastiaanse research group do not confirm the hypothesis that language decay is simply the opposite of language learning, as suggested by Jakobson sixty years ago. Agrammatic aphasies have knowledge of their language that young children have not yet acquired. Although there are certain similarities, given that both populations have problems with verb move­ ment, the solutions are different.

INTRODUCTION

XVii

In her contribution A-bar movement constructions in Greek children with SLI: Evidence for deficits in the syntactic component of language, Stavroula Stavrakaki (University of Thessaloniki) presents experimental data concerning the production of structures involving A-bar movement, that is, relative clauses and wh-questions, by Greek SLI children and their normally developing peers. Her results show that the performance of SLI children is qualitatively different from that of the control group. She discusses the status of the underlying lin­ guistic mechanism in SLI and normal grammar, making an attempt to evaluate the implications of her research findings for the core of the theoretical issues surrounding SLI, that is, the locus of the deficit and the way language devel­ opment takes place in SLI children. Assuming that linguistic development in SLI is an explicit rather than an implicit procedure, she argues that SLI children have problems with purely syntactic operations, such as A-bar movement. Within the minimalist framework, this is seen as due to a severe deficit in the in-interpretable features of grammar, that is, features with no semantic interpre­ tation. Susan M. Suzman (University of the Witwatersrand) analyses Morpho­ logical accessibility in Zulu. She explores a well-known problem, the vulner­ ability of morphology in language impairment. Such vulnerability is well at­ tested and varies cross-linguistically according to language type. Such a differ­ ential access to morphology, observed by several researchers, is discussed in two case studies of language impairment in Zulu. She investigates the morpho­ logical development by considering the multiple and diverse rules of morphol­ ogy: the range of noun classes, the agreement, and the agglutinative mor­ phemes used by normal and language-impaired Zulu-speaking children. The data considered reflects differential access to morphology depending in part on the obligatory or optional status of morphemes in the language. Considerations on the access to core grammar mediated by language-specific organisation and representation of basic concepts are offered. The study on Language production in Japanese preschoolers with SLI: Testing theories is a joint research project by Yumiko Tanaka Welty (Univer­ sity of Tochigi), Jun Watanabe (Arts Junior College, Osaka) and Lise Menn (University of Colorado). In their investigation they reconsider two claims: the etiological claim, according to which SLI is a unitary disorder with a single cause and the cross-linguistic claim, independent of the etiological claim, ac­ cording to which SLI (or different SLI's, if there are several) will be underlyingly comparable across languages. That is, if both of these claims are true, then a theory, which fails to explain SLI in even one language, cannot be an ad­ equate theory of SLI in general. If a general explanation offered for SLI cannot

XViii

ELISABETTAFAVA

apply to a particular language, but children speaking that language in fact have SLI, then either they have a different type of SLI, violating the etiological claim, or the explanation is inadequate. From this theoretical standpoint, Japanese SLI is discussed. The authors hypothesise that children with JSLI, and by implica­ tion, children with SLI in general, have a general language-processing deficit, which means that they are under 'overloaded' conditions. In the fourth section, issues of grammar and cognition are discussed. The linguistic and non-linguistic accounts, the interface of grammar and pragmatics in the characterisation of some deficits are reconsidered. Among the assump­ tions going back to the early 1950s is that there is a component of the human brain dedicated to the language faculty, which interacts with other systems. After the publication of Jerry Fodor's Modularity of the Mind in 1983, issues of modularity have made a profound impact on linguistic research. One major issue is whether pragmatics should be considered a module of the grammar or not.7 The evidence for the nature of genetic deficits is offered mainly by anal­ ysing SLI and Williams Syndrome (WS). WS is a rare genetic disorder, which presents a variety of cardiovascular difficulties, failure to thrive in infancy, etc. The neurolinguistic profile is characterised by relative strengths in language, facial processing and social cognition, and profound impairment in spatial cog­ nition, planning, and problem solving. Testing linguistic concepts: Are we test­ ing semantics, syntax or pragmatics? is the question raised by Leah PaltielGedalyovich (Ben-Gurion University). She investigates what kind of knowl­ edge is being assessed in the variety of standardised tests of language acqui­ sition available to speech/language clinicians, which are purported to assess children's linguistic, i.e., syntactic or semantic, knowledge. In particular, she considers the linguistic knowledge necessary to comprehend and produce coordinated sentences, and the interaction of this knowledge with pragmatic knowledge. She argues that in some cases it is pragmatic, rather than syntactic or semantic knowledge, which is required to complete test items successfully. Test items may reflect the interaction between the semantics of co-ordinators and the influence of the Grice's pragmatic principle of quantity. It appears that the failure of children to complete the task successfully reflects impaired prag­ matic and not impaired syntactic ability. An analysis of a failure as a pragmatic failure as opposed to a linguistic failure will affect the direction of a treatment program. In order to assess children's communicative abilities accurately, and 7

On the two opposite claims and their different underlying assumptions, cf. Asa Kasher. "Pragmatics and the Modularity of Mind". Pragmatics: A reader ed. by Steven Davis (Oxford: Oxford University Press, 1991), 567-582; and Deirdre Wilson & Dan Sperber. "Pragmatics and Modularity", ibid., 583-595.

INTRODUCTION

XiX

to design appropriate remediation programs for them, it is crucial to analyse ac­ curately the tools of assessment. The need for pragmatic knowledge to com­ plete test items, which purport to test purely linguistic knowledge, challenges standard interpretation of these tests. However, the usefulness of these tests is not denied. Rather, it is necessary to recognise the various skills needed in or­ der to use test results appropriately in diagnosis and in the planning of the treatment of language delayed and disordered children. The contribution by Dusana Rybárová (Slovak Academy of Sciences) on SLI and modularity: Linguistic and non-linguistic explanations argues that the current status of the debate between the proponents of linguistic and non-lin­ guistic approaches to explanation of SLI is characterised by a severe divide between these two approaches. Both are outlined from the point of view of one of their proponents, so as to show their premises and weak points in relation to the modularity issue. She argues that, although claiming to be completely dif­ ferent, both the linguistic and non-linguistic accounts are based on the same idea of 'strong modularity', characterised as nativist, static and non-interactionist. Such understanding of modularity causes them to appear as mutually ex­ clusive. To her, the concept of 'strong modularity' seems to be an unproduc­ tive and inappropriate view in accounting for the broad array of SLI symptoms, symptoms that probably result from different primary deficits. Rybárová ar­ gues in favour of a possible way to bridge the gap between those two posi­ tions, by suggesting the abandonment of the 'strong modularity' thesis in fa­ vour of a finer-grained view on modularity, 'weak modularity'. Vesna Stojanovik, Mick Perkins & Sara Howard (University of Sheffield) debate The language/cognition interface: Lessons from SLI and Williams Syn­ drome. Despite years of research in this area, the issue of language/cognition dissociation as evidenced in cases of WS and SLI is far from being resolved. According to Stojanovik, Perkins & Howard, there have been hardly any stud­ ies which consider in detail the linguistic and non-linguistic functioning in the same subjects with WS or SLI, respectively, or which provide linguistic and non-linguistic data from both populations with WS and with SLI. They investi­ gate the complex patterns of impairment in WS and SLI, highlighting aspects of the language/cognition interface. The question addressed is what aspects of language might develop independently, and whether pragmatics belongs to the linguistic or the cognitive system or whether it is a domain where both linguis­ tic and cognitive modules interact. When detailed information about linguistic, cognitive and communicative functioning is obtained for the same subjects, the two profiles may not be as opposed to each other as has been claimed in the lit­ erature. Their results suggest that the language/cognition interface is still terra

XX

ELISABETTAFAVA

incognita and may have been approached too simplistic ally up until now. This has important implications both for linguistic theory and for speech and lan­ guage therapy. It is extremely important for clinicians to be aware of the degree of individual variation possible in conditions such as WS and SLI. Given the current state of our knowledge about these language deficits, intervention pro­ grammes should be sensitive to the unique interplay of linguistic, cognitive, and pragmatic abilities in each individual, and should not assume that all cases of WS and SLI conform to an even relatively homogeneous profile. The final section is devoted to acquired language problems subsequent to cerebral damage, to be precise non-fluent aphasia, associated with lesions in the frontal part of the left cerebral cortex, and fluent aphasia, associated with lesions in the posterior areas of the left cerebral cortex grammatical structure. The data discussed derive mainly from Catalan, Dutch, Spanish, English, Ger­ man, and Greek. In Grammar and fluent aphasia Susan Edwards (University of Reading) offers some evidence regarding the many frequently spurious generalisations on fluent aphasia. These generalisations are due to the fact that the much of the research is not sufficiently linguistically oriented. Fluent apha­ sia is usually characterised as a disorder of lexical-semantic access. Any prob­ lem in sentence structure is seen as arising from difficulties in accessing lexical items which in turn arise from either semantically based problems or from problems in phonological representation. Although these lexical problems are characteristic of all fluent aphasic speakers, lexical deficits cannot account for all the errors found in fluent aphasia. The view that grammar is preserved in fluent aphasia and that any errors observed result from faulty lexical retrieval has gone largely unchallenged, perhaps because the unequivocal lexical errors in fluent aphasia are more obvious, often more 'flamboyant' and more frequent than the subtle, less frequent, grammatical deficits. On the contrary, the evi­ dence offered by the fluent aphasic subjects described by Susan Edwards is not so neat. They were better at sentence construction than typical agrammatic pa­ tients and they produced utterances that were considerably longer than agram­ matic speakers can produce. However, the errors made by the fluent aphasie speakers were of the same kind if not the same magnitude as those made by agrammatic speakers. The data discussed support the conclusions that, in the same way as non-fluent aphasie patients, fluent aphasies have faulty access to their grammar. The assumption that syntactic abilities are intact in this particu­ lar aphasie population can no longer be assumed. Anna Gavarro (Universitat Autònoma de Barcelona), in Failure to agree in agrammatism, focuses on the linguistic evidence provided by agrammatism, which has a bearing on the evaluation of competing linguistic theories. She

INTRODUCTION

xxi

explores how linguistic theory informs the study of agrammatism, especially the empirical domain of inflection, within competing Minimalist theoretical constructs. While the deficit of inflectional markers, including case markers, and function words associated with aphasia was recorded long ago in the sci­ entific literature, the first approach to inflectional disorders in agrammatism within generative grammar occurred in the 1970s. Results considering phono­ logical clitics, phonological words, morphological objects and syntactic atoms are debated: the question is whether errors derive from syntactic impairment rather than a morphological or phonological impairment. The categories af­ fected by this syntactic impairment are tense, aspect and person on verbs, gen­ der or number on Determiner/Noun categories, or are related to displacement of constituents within a structure. Gavarro points out that asymmetries may arise between production and comprehension. Grammaticality judgments in agrammatic subjects are not central to the thesis she defends. She attempts to charac­ terise the disruption that affects grammatical derivations; the reasons why this disruption is not apparent in all cases remain a topic for future research. According to her analysis, inflectional and word-order deviations constitute a natural class, and agrammatism results in part from inability to apply the op­ eration Agree. This hypothesis lends support to the latest version of Chom­ sky's minimalist position over previous ones. The Verb and Sentence Test: Assessing verb and sentence comprehension and production in aphasia by Judith Rispens & Roelien Bastiaanse (University of Groningen) and by Susan Edwards (University of Reading) offers some crucial reasons for constructing a new test for the diagnosis and treatment of aphasics, the Verb and Sentence Test (VAST), assessing verb and sentence com­ prehension and production in aphasia. From clinical experience and theoretical investigations, it has become clear that disorders at the level of verbs and sen­ tences are frequent in aphasic patients. Verbs play an important role in sentence comprehension and production; a disorder in processing verbs therefore has a great effect on linguistic processing and, very importantly, on communicative ability. There are currently few standardised assessment materials, which sys­ tematically investigate disorders at the level of verbs and sentences. It seems that a gap exists between neurolinguistic findings and clinical application of this knowledge. The test battery for verb and sentence processing developed uses linguistic insights and is theoretically motivated; it is regarded as suitable for different types of aphasie patients and it is clinically relevant. The only tasks which have been included are those which reveal impairments, which have been described adequtely in the literature and for which therapy pro­ grams or methods exist. These principles have led to the inclusion of ten sub-

XXii

ELISABETTAFAVA

tests which assess the processes involved in comprehending and producing verbs and sentences: verb retrieval, processing information regarding the gram­ matical and thematic roles, 'mapping' of thematic roles onto grammatical roles and, for verb and sentence production, retrieving morpho-syntactic features such as subject-verb agreement inflection. This evaluation of sentence compre­ hension and sentence production deficits is constructed to apply to many different languages. The second part of the study analyses the first two case studies of the application of the VAST test to Dutch and English patients. Direc­ tions for therapy are also proposed. Esther Ruigendijk (Utrecht University) focuses on Case assignment as an explanation f or determiner omission in German agrammatic speech. She offers a linguistic, psycholinguistic, and aphasiological background to agrammatic aphasics. Her thesis is that the production of determiners and pronouns is not impaired because they are grammatical morphemes, but because their realisa­ tion is dependent on the production of finite verbs. The basic problem is the production of finite verbs; poor determiner and pronoun production can then be considered a side effect. The distinction between functional categories is at the heart of present-day grammatical theory, but plays an equally central role in, among other topics, theories on language acquisition and aphasia. In the course of time, various diagnostic criteria have been identified which distinguish one class from another: e.g., productivity, distribution, and absence versus pres­ ence of semantic content. Her results on German determiner production and case assignment demonstrate that once the case-assigning verb is realised, the production of determiners is possible in agrammatic aphasics. This means that determiner production as such is not impaired, but rather that the problems with determiners are closely related to those of the production of verbs. When no case-assigning verb is realised, no determiners can be produced. Incorrect or incomplete retrieval of the lemma information of the verb can account for case substitution errors that are made with object noun phrases. Ruigendijk's results demonstrate that once the verb is realised, the production of determiners is less impaired in agrammatic aphasies who omit determiners in their sponta­ neous speech production. This has interesting implications for the treatment of agrammatic patients. Training in the production of isolated determiner phrases cannot be successful, since these are related to verbs. It is better to train apha­ sies in verb production, as the results of the present study suggest that this will increase the production of complete noun phrases. Kyrana Tsapkini, Gonia Jarema & Eva Kehayia (University of Montreal) discuss The role of verbal morphology in aphasia during lexical access: Evi­ dence from Greek. They explore the role of regularity in language breakdown

INTRODUCTION

XXiii

by investigating how these phenomena are manifested in the performance of Greek-speaking aphasics. The dissociation between regular and irregular mor­ phology in aphasia has been documented in different languages. Broca's apha­ sics were found to be impaired in their application of the past-tense inflection while having no difficulty with irregular stored forms. Conflicting results have been obtained in different languages on the question of whether regular and ir­ regular inflection can be described as two qualitatively different operations, or a single mechanism based on statistical probability or associative memory. The issue addressed is whether morphological regularity in languages allows the dissociation of the effects of morphological regularity from those of form and semantic similarity. This would allow the maintenance of the claim of univer­ sality of the distinction between rule-based and storage mechanisms in lan­ guage processing and breakdown. Greek is a language that offers the oppor­ tunity to investigate morphological regularity while keeping form and meaning overlap between base and past tense forms constant across regular and irregu­ lar forms. First, the study of morphologically regular and irregular verbs in Greek allows for the addressing of the issue of morphological regularity per se, independently of form and semantic considerations. Second, it clarifies whether the difficulties that Greek-speaking aphasies have with the past tense are due to the different morphological operations postulated, or whether they reflect task-specific deficits related to the lexical access procedures involved in each task. It is the very presence of a stem-allomorph that marks the most 'ir­ regular', in the sense of idiosyncratic and unpredictable, past tense formation, the authors note. In conclusion, this volume investigates relevant intersections among differ­ ent linguistic frameworks, languages, clinical approaches, and pathologies. There is unity, however, in the great effort made to provide some key aspects of the framework adopted and to discuss general problems posed by it. The theoretical arguments advanced and the empirical evidence proffered are bound to offer deeper insights into the factors that shape the nature of language. In this perspective, theoretical linguistics has been and will continue to be of par­ ticular significance for the study of speech pathology and speech therapy, giv­ ing it new life and scope. At the same time, it can be said that clinical research has offered new perspectives on investigative techniques and suggestions as to how to revise theoretical and methodological tools.

I. PHONOLOGY IN CLINICAL APPLICATIONS

PHONOLOGY AS HUMAN BEHAVIOR THEORETICAL IMPLICATIONS AND COGNITIVE AND CLINICAL APPLICATIONS

YISHAI TOBIN Ben-Gurion University of the Negev Phonology, then, is for us an attempt to under­ stand the non-random distribution within the signal, and grammar is the attempt to under­ stand the non-random distribution of the signals themselves. (William Diver 1975:13)

1. Introduction The theory of Phonology as Human Behavior (PHB) was developed by William Diver (1979) in an analysis of the non-random distribution of certain classes of initial consonant clusters in English, which he later expanded to explain the non-random combinations of vowels and consonants in English (Diver 1993) and in language in general (Diver 1995). PHB has been extended further to explain the combinatory phonology of consonant and vowel pho­ nemes in other languages such as Italian (Davis 1987 [1984]), Hebrew (Tobin 1990b,c), Urdu (Azim 1989, 1993, 1995, 1997; Hameed 1999; Jabeen 1993), Mewati (Fatihi 1987) and Spanish (Flores 1997). PHB has also been employed to explain the non-random distribution of initial consonant clusters in fortytwo different languages representing diverse language families as well in arti­ ficial languages such as Esperanto and Klingon (Tobin 2000a); it has been ap­ plied to the areas of developmental and clinical phonology and prosody in languages such as Hebrew, Finnish and Japanese (Moore 1991a,b, 1993; Moore & Korpijaakko-Huuhka 1996; Moore & Rosenberg-Wolf 1998; Tobin 1995, 1997a,b, 1999, 2000b, 2001a; Tobin & Miyakoda 2001b); and it has been compared to and contrasted with other functional and quantitativeoriented phonological theories (Tobin 1988c, 2000b). 1.1 Historical background PHB may be viewed as part of the historical development of a structural, functional and cognitive tradition in 20 century linguistics. This tradition begins with Ferdinand de Saussure's (1959 [1916]:34-68) concept of system, and the

4

YISHAI TOBIN

dichotomies of langue and parole and phonetics and phonology. The latter, fundamental dichotomy between the abstract code and its realization, based on a classification of sounds according to their articulatory and acoustic features, was further developed by the functional, communication-oriented Prague school phonology developed by Nikolai Trubetzkoy (1969 [1939]) and Roman Jakobson (1968 [1941], 1971). The communication factor adhered to by the Prague school was then supplemented by André Martinet's (1955) introduction of the human factor through the concepts of "assymetry" and "economy of ef­ fort in phonological change." Martinet maintained that phonological systems are arranged asymetrically and change in such a way that the non-random diachronic distribution of sounds reflects the search for equilibrium and harmony within the system as it is affected by the principle of least effort in human be­ havior. The principle of minimal effort postulated by Martinet implies that speakers strive for a minimal number of phonemes which require the least amount of effort to be produced and combined together in what Sampson (1980:112, reviewed in Tobin 1986), referred to as a "therapeutic view of sound change." It is Diver (1974, 1979), however, who has shown that a more complete theory of phonology has to take both the communication factor and the human factor into account together. Diver maintains that there is a constant struggle between our need for maximum communication and our desire for minimum effort. The communication factor (requiring a large number of pho­ nemes demanding a great deal of effort) is in conflict with the human factor (striving for minimal effort), which results in a trade-off between the two. This synergetic compromise between the communication factor and the human factor is reflected in the fact that there is a similar number (20-30) of phonemes of varied proportional degrees of difficulty acquired in a similar order in the languages of the world: less than 20 phonemes would reduce the com­ munication potential and more than 30 would be too difficult to learn, remember and produce. That is, Diver extends and enhances Martinet's more strictly diachronic view of the human factor so that it can become a means of explaining the non-random distribution of phonemes in language in coordination with the needs of communication as originally established by Saussure and the Prague school (discussed in Liberman 1991; Tobin 1988a-c, 1990a, 1996). 1.2 Theoretical background Diver (1995:61) justifies in the following way this synergetic combination of the roles played by the human factor and the communication factor in pho­ nology as orientations which motivate the hypotheses used to discover and analyze the potential observational data of sounds:

PHONOLOGY AS HUMAN BEHAVIOR

5

The observations, the sound waves, are produced by people. The problem is to discover what motivates people to produce particular sound waves on a particular occasion. The purpose of the hypotheses is to reveal motivation by demonstrating the relation of the observations to the orientations, for the orientations represent the con­ trolling circumstances of the entire procedure. The way this works out, in analytic practice, is that the essential motivation is communication, and that the details of how communication is effected are controlled by considerations of the available human physiology, by principles of acoustics, and by normal characteristics of human behavior; that is, by the other orientations.

Although PHB has maintained the basic Saussurian and Praguian notions of phonemes, distinctive features, and morphemes, etc., Diver (1995:61) views these familiar concepts in an innovative way. His particular view of language and linguistic analysis develops from the orientations and the hypotheses out­ lined above: The first task of the hypotheses is thus to establish the identity of the inventory of the units of language (beginning with phonemes and morphemes) and the nature of those units, their nature being an inference from how they are made use of in the process of communication. The other orientations constantly feed in to the process of reaching the primary goal, particularly in regard to why the units are made use of in the way they are. As analysis proceeded (in our work), the hypotheses sorted themselves into three distinct groups, distinct in the way in which they are related to the process of communication: hypotheses about phonological units, about grammatical units, and about lexical units.

Diver (1979, 1995) thus has postulated different kinds of distinctive features in phonology, which are more directly related to human physiology, perception, cognition, and behavior. Based on these new distinctive features, it is possible to classify different sets of phonemes according to the degree of difficulty required to learn, perceive, and produce them. Diver has also introduced the notion of the quantitative analysis of favored and disfavored distributions of phonemes, which can be directly related to their degree of difficulty. These distributional preferences or 'preferred combinations' are further paralleled by similar ones observed in human behavior requiring the control of fine motor movements in extralinguistic contexts (e.g., common daily tasks or sports), that is in other instances of learned behavior, as language is. In practice, Diver's (1995:62) theoretical perspective presents the phonological units of language in a unique way based on the interaction between the communication factor, which enables him to tackle the following questions: why are there certain kinds of sounds in human language and why do these sounds combine in a non-random way to form larger units in a similar way across languages?

6

YISHAI TOBIN The sounds are produced through the control of the musculature of the vocal tract. The first requirement is that audible sound be produced, if the communication is to be transferred from one person to another; clearly audible sound if the communication is to take place over any distance. The second, that a fairly extensive inventory of distinct sounds be built up, if the resources of human communication are to exceed those of animal cries. There are two important considerations here: first, given the variety of the ways in which sounds could be produced by the vocal apparatus, why is there so much agreement in general across languages, in the midst of a great deal of differences in matters of detail? Second, given that an inventory of grammatical and lexical units is to be built up through the combination of phonological units, just how is an expanding inventory of phonemes related to the expanding structure of the mor­ pheme?

Diver (1995:62-65) establishes the criteria for the basic inventory of audi­ ble sounds on the grounds of the physiological and acoustic properties of the human vocal tract, which serve to produce maximum acoustic differentiation with minimum precision of control (e.g., the vowels of the /i, a, u/ triangle can be used independently for the formation of morphemes; cf. the English indefi­ nite article a). He then shows how the phonemic inventory is expanded by ad­ ditional techniques of shaping and exciting the vocal tract to include interme­ diate vowels and consonants (involving the use of the dorsum, lips, apex and other active articulators). These consonants then 'flank' the 'keystone' vowels to form morphemes of the cvc variety. The possible order of phonological development in language is presented and explained by Diver (1995:67) based on the interaction of the communication factor and the human factor as follows: For the 'primary' units we thus get a picture of an imagined sequence of development ordered in terms of the need for precision of control. The units are here symbolized in ways that foreshadow their later phonological status: (1) The single-cavity /a/, furnishing undifferentiated resonance for the excitation of the vocal folds. (2) The development of a two-cavity system, using the dorsum and lips as articu­ lators, introducing /i/ and /u/ and thereby converting /a/ to a third member of a system, maximally differentiated from the other two, rather than a unique unit. (3) The use of the apex in a fairly undemanding way, giving another shape to the cavity, /1/, without recourse to dorsum and lips, still with excitation by the vocal folds. (4) The development of fine motor control over the apex, as it is brought into use as a means of excitation, as well as shaping, of the cavity, in the maximally differ­ entiated positions of/t/ and /s/.

PHONOLOGY AS HUMAN BEHAVIOR

7

Diver (1995:68-69) then shows how the communication and the human factors interact in the relation between phonemes and morphemes in an iso­ morphic way: In general, the principles that underlie the ordering of acquisition, as presented above, are the same as those reflected in the construction of morphemes in current speech; that is, in the utilization of the various members of the inventory of phonemes. Those that can be characterized as being more difficult to control (e.g., in terms of intermediate position), are the same ones that are, in general, made use of less frequently (or perhaps not at all) in the structuring of the morpheme. The most audible members of the inventory - ill, /a/, /u/, together with intermediate units regularly form what may be called the keystone in the arch of the structure of the morpheme. It is rare indeed for a 'word' to be constructed without one of the clearly audible members as keystone.

The flanking members are drawn from the less audible members of the in­ ventory, and the preference among these is very much in terms of how much demand is made on precision of control. Further, the extent to which a number of flanking members are combined in the formation of a single morpheme is very sharply restricted. Diver (1995:69) then attributes the restrictions on the phonotactic distribu­ tion of the consonants flanking the vowels in a syllable, and more generally the phonotactic favorings and disfavorings of sounds in larger units of communication such as morphemes and words, to his particular view of the human factor as a form of precision of control: In terms of precision of control, it is evident that the more phonological units are combined in the formation of a single morpheme, the more control is required, in terms of the number of individual members to be produced and in terms of the coor­ dination of the sequencing. An increase in the number of phonological units in a morpheme type leads to an in­ crease in the potential number of morphemes, but to a decrease in the actual number, in terms of actual numbers as well as in comparison with the potential. To the extent, then, that the selection of a morpheme in the text leads to the characteristics of the sound waves, we see that the "human factor" motivation, preci­ sion of control, is favoring some kinds of combinations of sound waves over others.

Apparent skewings in the form of favorings of certain phonological units on the morpheme level resulting from the interaction between the communica­ tion factor and the human factor play a major role in Diver's (1995:71) theo­ retical approach and even determine the kind of data he uses in his analyses 'words' or canonical morphemes:

8

YISHAI TOBIN A number of hints have been given that there is a quantitative basis for the comments about the favoring of certain uses of the phonological units over others. The following procedures have been used. A collection is made of all 'words' constructed with a single keystone [vowel], within which there is no morphemic boundary. Thus cat would be counted, but not cats. The rationale is that the combination t-s here is determined by something other than purely phonological considerations.

Diver's (1979, 1993, 1995) focus on phonotactics, or combinatory phonology, is reflected in his analysis of the non-random distribution of pho­ nemes in general and in alternative canonical morpheme and word positions in particular, which are directly related to the interaction between the human factor and the communication factor. Despite the fact that consonants are more difficult to produce than vowels, there are more consonant than vowel phonemes in languages because consonants provide clearer communicative distinctions. In morpheme/word-initial position - where the burden of com­ munication is the highest - there is an almost random distribution of conso­ nant phonemes, which, however, reflects their different degrees of difficulty, that is, with a slight favoring of labial, or labio-dental, or apical-dental pho­ nemes, which are perceived both aurally and visually. In syllable/word-final position - where the burden of communication is the lowest - there is a highly significant statistical preference for phonemes requiring the least amount of effort to produce, namely apical (the easiest to produce) and voiceless (which require the excitation of only one set of oral articulators) rather than voiced (which require the excitation of two sets of articulators: the oral articulators and the vocal folds) or nasal (which require the excitation of three sets of articulators: the oral articulators, the vocal folds, and the uvula). 1.3 Methodological background Diver's approach differs from other phonological frameworks not only in its theoretical orientations and hypotheses, but in its methodology as well. Three different procedures for collecting data have been outlined by Diver (1995:71-72) as evidence in support of the hypotheses of his theory: The collection is intended to be complete, but there is a problem as to just what con­ stitutes 'completeness'. Three different kinds of collection will be noted here. In my own research, the collection consisted of all words in my active vocabulary: that is, in the active vocabulary of a literate, well-educated speaker. It is recognized that the resulting list might not be exactly the same as that of another person of a similar education, and that there might well be quite a number of words "in the dictionary" that would not be included.

PHONOLOGY AS HUMAN BEHAVIOR

9

At the other extreme is a study by Shazi Shah Jabeen (Jabeen 1993) of the phonology of Bihar Urdu in the neighborhood of Gaya, India. Here, informants were used, essentially illiterate persons without access to radio or television. The author is herself a native of Gaya, and checked for the occurrence of every possible phonological combination. A third alternative was a study by Joseph Davis (Davis 1987 [1984]). This study based the count on essentially all the entries in a moderately sized dictionary of Standard Italian. In each of these, counts were made of the frequency of occurrence of the phonological units in various positions and in various combinations within the mor­ pheme. In spite of the differences in the inventory of phonological units (between English, Urdu and Italian) and the differences in methods of collection, the general principles that have been sketched above were found to be quite uniform in their manifestation. It was thus possible to check the interrelations among orientations, hypotheses, and observations, and to demonstrate that there is a relationship of motivation flowing from orientation to observation.

In order to uncover all the data necessary for our analyses of initial conso­ nant clusters across languages, Tobin (2000a) examined all such clusters ap­ pearing in monosyllabic words in standard dictionaries. The developmental and clinical studies discussed here were based on recordings of spontaneous speech as well clinical exercises. For similar methodological reasons lexical analyses from standard dictionaries were performed by Tobin (1990b,c) for close to 3,000 triconsonantal (ccc) roots in Hebrew. Specific text or discourse analyses testing the theory on various texts, including a poetic text containing many neologisms, appear in Tobin (1997a: ch. 6). In all cases, regardless of the methodology employed in the collection of the data, the principles established by the theory were supported. 1.4 Theoretical and methodological conclusions supported by the theory The following phonological and phonotactic parameters have been ex­ plicitly derived from the theory (adapted from Diver 1979; Davis 1987 [1984]; Tobin 1990b,c, 1995, 1997a,b): (1) the identification of active articulators (versus the traditional category of place of articulation, which is often a label for passive receptors) and the relative difficulty of learning how to control them; (2) the identification of relative degrees of constriction and turbulent and non-turbulent airflow (versus the traditional category of manner of articulation), which require different articulatory control (mobile and stable) and produce different acoustic patterns for individual sounds and phonation processes (labialization, apicalization, velarization, nasalization, and glottalization);

10

YISHAI TOBIN

(3) the identification of the number of sets of articulators to be controlled (versus the traditional categories of voicing, the fortis-lenis distinc­ tion, and nasality) that require different levels of articulatory control and produce different acoustic patterns; (4) the identification of 'phonemes of constriction' and 'phonemes of aperture' (versus the traditional concepts of consonants and vowels). The following principles have been derived from the quantitative results obtained from the theory (adapted from Diver 1979; Davis 1987 [1984]; Tobin 1990b,c, 1995, 1997a,b, 1999): (1) additional articulators are disfavored; (2) coarticulation by near articulators is disfavored; (3) coarticulation by the same articulators or of the same phonemes is even more highly disfavored, particularly in the roots of Semitic lan­ guages (Tobin 1990b,c); (4) different word or root positions have different communicative force, and thus affect the favoring and disfavoring of different articulatory and acoustic features and phonemes; (5) visual articulators are favored, particularly in word/root initial position; (6) explosive (mobile) phonemes are favored in initial position; (7) turbulent (stable) phonemes are favored in final position; (8) transitions from one distinct constriction to another within a single phoneme are disfavored; (9) consonant clusters concerning different articulatory and acoustic fea­ tures are restricted (e.g., mobility/stability; Diver 1979); (10) among constrictions, maximal constriction is favored, and among apertures, maximal aperture is favored; (11) sequences of phonemes with the same articulators are disfavored unless their juxtaposition is, by virtue of some other factor, mutually beneficial; (12) apical consonants are favored. The following principles have been added to the theory based on the appli­ cation of the theory itself to clinical studies (adapted from the works found and cited in Tobin 1995, 1997a, 1999): (1) the preservation of as many distinctive features as possible (usually 2 out of 3) in substitution processes which require more effort than de­ letion processes;

PHONOLOGY AS HUMAN BEHAVIOR

11

(2) the preservation of as many communicative oppositions as possible in the original word (e.g., the number of phonemes per word) in substitution processes which require more effort than deletion processes; (3) the use of a phoneme already available in the speaker's repertoire in accordance with the immediate phonetic environment in substitution processes which require more effort than deletion processes; (4) the preservation of the original phonetic structure of the word in dele­ tion processes not involving syllable reduction and in reduplication processes; (5) if the original structure of the word is reduced by the deletion of syl­ lables, the stressed syllable bearing the most communicative infor­ mation is maintained; (6) if the original structure of the word is enlarged by epenthesis, the epenthesis makes the transition to or between more difficult sounds easier. PHB supports the idea that there is no pure synchronic phonology, and that instead the dynamic interplay between the communication factor and the hu­ man factor both motivates and serves as a constant control over language change. Therefore the following conclusions may be drawn for synchronic and diachronic phonological analyses: Conclusion 1: Language in general - and phonology in particular - can be seen as a synergetic mini-max struggle: the desire to create maximum communication with minimal effort (Tobin 1990a: ch. 3). As it has been applied to the articulatory errors and processes found in de­ velopmental phonology and in functional and pathological clinical phonology, PHB basically represents a more extreme version of Conclusion 1 where the human factor often overrides the communication factor, and clinical interven­ tion serves as an attempt to balance the two: Conclusion 2: Developmental and clinical speech processes and errors may be viewed as an extreme version of this synergetic mini-max struggle: there is less than maximum communication because of either extreme minimal effort or a lack of control over the articulatory tract or mechanisms. Greater effort will be exerted in order to achieve more efficient or better communication through clinical intervention (Tobin 1995, 1997a,b, 1999).

12

YISHAI TOBIN

2. Phonological processes and phonology as human behavior Natural phonology, one of the leading theories in developmental and clini­ cal phonology, introduced the concept of natural phonological processes (Stampe 1979 [1972], Dressier et al. 1987). The following include most of the major and minor natural phonological processes found in functional language acquisition (adapted from Grunwell 1987; Ingram 1990), accompanied by an explanation derived from the principles obtained from PHB. 2.1 Functional processes influencing syllable structure (1) Final consonant deletion: cvc → cv: out [au], hike [baI] Explanation: word-final position has less communicative force; consonants require more articulatory control (i.e., are harder to make) than vowels. (2) Deletion of unstressed syllable (usually in word-initial position): banana [næena] Explanation: stressed syllables give more communicative, perceptual, and cognitive information than unstressed syllables; the more syllables in the word, the more effort it takes to pronounce it; therefore, the word-initial position, which usually has the greatest communicative force, carries less information in non-initially stressed words. (3) Consonant cluster reduction: CC → C:floor [for], step [tep] Explanation: a consonant cluster requires greater effort than a consonantvowel sequence and may be reduced or replaced at the expense of maximum communication; in addition, coarticulation by near articulators is disfavored; phonemes of constriction give clearer communicative distinctions than pho­ nemes of aperture - that is why there are more consonants than vowels in lan­ guage - but they require more articulatory control (hence the ideal cv syllable). (4) Reduplication: repetition of a syllable or part of a syllable: rabbit [w→wæwæ], noodle [nunu] Explanation: reduplication is often a means to avoid more difficult sound combinations and/or to maintain the number of syllables in the word: se­ quences of phonemes with the same articulators are disfavored unless their juxtaposition is, by virtue of some other factor, mutually beneficial. We also found that newly acquired sounds were often reduplicated as a means of practice or of hypercorrection in the clinical situation.

PHONOLOGY AS HUMAN BEHAVIOR

13

(5) Coalescence: characteristics of 2 consecutive sounds merging into 1 sound: swim [. im], slide [ aid] Explanation: fewer articulatory gestures resulting in fewer distinctive units at the expense of maximum communication. (6) Epenthesis: addition of segments, usually an unstressed vowel: snow [səno , drum [dər m] Explanation: the additional unstressed vowel often eases the transition to more difficult consonants or clusters. The clusters then may be reduced at the ex­ pense of communication. Examples (5) and (6) above, as well as the following assimilation processes (1) and (2) below, may also be explained by the factor: sequences of phonemes with the same or near articulators are disfavored unless their juxta­ position is, by virtue of some other factor, mutually beneficial. 2.2 Assimilation processes (Consonant/Consonant-Vowel harmony) (1) Velar/Nasal/Labial, etc. assimilation: duck [];friend [fire]; top [bap] Explanation: a non-velar/nasal/labial sound changes to a velar/nasal/labial be­ cause of the influence of, or the dominance of, a velar/nasal/labial sound which entails fewer articulatory gestures at the expense of maximum communication. (2) Prevocalic voicing of consonants: pen [ben], tea [di] Explanation: an unvoiced consonant generally becomes voiced before a vowel: the speaker anticipates the control of two sets of articulators in what is usually a longer acoustic phonological segment. (3) Devoicing of final consonants: bed [bet], big [blk] Explanation: additional articulators are disfavored; voiced consonants become unvoiced in word-final position: where the communicative force is least im­ portant or crucial, the speaker opts to activate one set of articulators rather than two. This may also be related to the fact that vowels are shorter before voiceless rather than voiced consonants. 2.3 Substitution processes (1) Processes reflecting the substitution of active articulators: a. Fronting: back (non-apical) consonants are replaced by apical con­ sonants usually preserving the same manner and voicing values:  → t book [but], g → d dug [dd], → s shoe [ u ] , Z → z beige [], r→ n sing [sin]

14

YISHAI TOBIN

Explanation: the apex is the most flexible and easy to control of all the active articulators: the earliest and most frequent examples of the substitution of active articulators are fronting or apicalization, which sharply reduces the number of communicative distinctions made by the speaker. b. Backing: the back pronunciation of front sounds (usually conso­ nants): t/d → k/g: dog [], tiger [kaIgə] Explanation: a later, less frequent (possibly idiosyncratic) process where the dorsum (or another back articulator) replaces the apex (or other front articula­ tors); often found in children who have difficulty controlling the musculature of the apex (for organic or other reasons) and/or try to reduce the number of communicative distinctions made by the apex (or other front articulators), especially as a result of earlier fronting or other processes. (2)

Processes reflecting the substitution of turbulence and airflow: a. Stopping: fricatives/affricates are replaced by stops: s/ts —> t: seat [tit], soup [dup] Explanation: maximum constriction is favored particularly when mobile (stop) phonemes of constriction are easier to control than stable phonemes of less constriction which require greater control of the musculature to create and maintain a small aperture for a stronger turbulent airflow (the most frequent manner of substitution for children). b. Affrication: stops/fricatives are replaced by affricates: t → t : tree [t i] Explanation: maximum constriction precedes and leads into partial con­ striction and turbulent airflow since mobile (stop) phonemes of constriction require less articulatory control than stable phonemes of constriction which produce greater turbulence (a less frequent manner substitution for children, possibly to avoid more difficult consonant clusters or combinations).  Gliding of liquids: l/r → j/w: rock [wak], lap [jæp] Explanation: substitution of a higher for a lower degree of aperture (from con­ sonants to semi-vowels) which may also require less articulatory control. d. Vocalization: nasals and liquids (syllabic consonants) are replaced by vowels: apple [æp], flower [fawo] Explanation: the favoring of maximal aperture particularly when phonemes of aperture (vowels) require less articulatory control than phonemes of constric­ tion (consonants).

PHONOLOGY AS HUMAN BEHAVIOR

15

e. Denasalization: m → b, n → d, → g: no [do], home [hob], sung [s g] Explanation: additional articulators are disfavored; nasal stops become their equivalent oral stops: two sets of articulators are exploited rather than three. f. Deaffrication11 → ', d3 → 31 chips [∫vps], juice [3US] Explanation: transitions from one distinct constriction to another within a sin­ gle phoneme are disfavored; a more complex sound requiring greater effort or control is reduced to a less complex sound after the speaker has acquired the ability to produce the more difficult stable sounds. g. Glottal replacement: butter [Λ?] Explanation: additional articulators are disfavored; a glottal stop replaces an intervocalic consonant or a consonant in syllable-final position: articulatory control of one set of articulators rather than two in an appropriate phonetic environment. h. Gliding of fricatives: /f/ → [w], /s/→ [l,j] Explanation: a more extreme substitution of a higher for a lower degree of aperture (cf. the gliding of liquids in (c) above) requiring less articulatory control: an idiosyncratic process. Most of the processes affecting the turbulence and airflow of phonemes of constriction (manner of articulation) listed above confirm the following factors obtained from PHB: among constrictions, maximal constriction is favored and among apertures, maximal aperture is favored. It should be clear from the above examples that more than one process can appear in the same word and that most, if not all, of these functional processes can be directly related to the principles obtained from PHB. It should also be noted that there is a chronology of natural processes which determines and separates normal processes from deviant functional or organic ones. 2.4 Functional processes in Hebrew-speaking children Shaked (1990) examined the frequency of functional errors of 20 Israeli children (from 1:7 to 2:7 years of age), obtaining the following results: (1) The most frequent functional processes found in the speech of the 20 Israeli children included: a. Fronting: particularly of fricatives: ∫/x → s ∫alom → [salom]L b. Consonant cluster reduction: p a x i m —» [paxim] ("flowers") (fricatives deleted, stops maintained);

16

YISHAI TOBIN

c. Syllable-final consonant deletion / medial consonant cluster reduction: taxtonim → [tatonim] ("underwear"); d. Devoicing of final consonants1 be ez → [be es] ("faucet") (it has also been noted in the literature that children aged 2 to 4 have particular difficulty in the production of voiced stops, which requires activating two sets of articulators for mobile consonants); e. Deletion of unstressed syllables: ambatja → [batja] ("bath"); f. Gliding: 1→j : delet → [jedet] ("door") (+ metathesis); g. Deletion of syllable/word-initial consonant (in unstressed syllable): mita → ita ("bed"); h. Deaffrication: tsav → [tav] ("turtle"). (2)

The least frequent functional errors and processes included: a. Backing (considered to be idiosyncratic by Grunwell 1987 and others): mixtav → [mitax] ("letter") (+ consonant cluster reduction, + metathesis); b. Stopping: sus → [tut] ("horse").

In this and other developmental and clinical studies, the general order of acquisition of phonemes and functional processes conform to the principles of the theory, including: i the systematic order of the exploitation of active articulators (labial apical - posterodorsal); ii the systematic favoring of consonants requiring extreme degrees of constriction and minimal airflow (stops preceding fricatives or mo­ bile preceding stable phonemes of constriction); iii the disfavoring of phonemes of constriction requiring the excitation of more than one set of articulators (when there are phonemic oppositions based on this factor): voiceless (0), most frequent, followed by voiced (+1), and the absence or extremely limited use of nasals (+2). 2.5 Functional processes in a hearing child of deaf parents Samet (1993) investigated the functional processes of a hearing child of deaf parents and compared the frequency of the functional processes found in her study with those found for Hebrew speakers. She analyzed the speech of Elita (2:8), the daughter of deaf, signing parents who are Russian immigrants with a limited knowledge of Hebrew, whose speech is extremely difficult to understand. Elita was brought for therapy because of late language develop­ ment and a medial level of comprehensibility (possibly due to lack of input).

PHONOLOGY AS HUMAN BEHAVIOR

(1) a. b. c. d. e. f. g.

17

Elita showed the following functional processes in descending order of frequency (according to Shaked 1990): consonant cluster reduction: bgadim → [gadim] ("clothes"); deletion of unstressed syllable: tsalaxat → [laxat] ("plate"); syllable-final consonant deletion / medial consonant cluster reduction: taxtonim —» [tatonim] ("underwear"); final consonant deletion: sipuy → [sipu] ("story"); initial consonant deletion 1 yam —» [(?) Jam] ("to-there"); stopping: xatula → [katula] ("kitty"); deaffrication: otsa → [ ota] ("want").

Samet then applied PHB to explain the processes. All of these processes, except for (l)(e) initial consonant deletion, clearly conform to the principles of phonology as human behavior. The possible initial consonant deletion in example (l)(e) appears in an unstressed function word le- 'to- in the phrase [∫am -> [(?) Jam] ("to-there") which, from the point of view of communi­ cation, is less crucial than the stressed lexical item∫am- "there" to which it is attached, which makes this example similar to the previously mentioned dele­ tion of unstressed initial syllables. (2) Elita also has functional processes not found in Shaked (1990): a. assimilation: beged^ [geged] ("garment") /jeled—» [leled] ("boy"); b. epenthesis (very limited and included addition of both consonants and vowels). In the case of Elita, we have a situation where the number and the degree of functional processes require clinical intervention. As we have previously stated, natural functional processes have been as­ signed chronological values and may serve as norms for child language acqui­ sition. If deviations from these chronological values and norms are discovered in a child's speech, that child may be sent for speech therapy. Children who are labeled as having functional disorders in their speech will usually reach the clinic when: i ii iii

early processes continue past their normal period; early processes coexist with later errors and processes; normal functional processes are accompanied by idiosyncratic processes (such as backing, gliding of fricatives, affrication).

18

YISHAI TOBIN

Functional processes which reach the speech clinic clearly reflect an even more extreme case of the struggle for maximum communication with minimal effort, usually under the following circumstances: i

ii

iii

The simultaneous coexistence of several functional and/or idiosyn­ cratic processes of simplification (the human factor) reduce the num­ ber of communicative distinctions (the communication factor) ex­ ploited by the child. If and when the child becomes, or is made, aware that her communi­ cation is impaired, only then will further effort be exerted to produce more communicative distinctions in the quest for maximum commu­ nication with minimal effort. In other words, the child will usually only exert effort in order to enhance communication.

3. Summary and conclusions We can now summarize the major principles of PHB for phonological analysis and its applications for developmental and clinical phonology: (1) We begin with the phonetic observations, both articulatory and acoustic, within phonemes which are abstract and therefore 'unobservable' units (child language inventory and clinical intake). (2) By means of the communication orientation we can establish the number of distinctive units of a language as found in a child or patient. (3) Consideration of the acoustic and physiological characteristics of phonetic units (e.g., the ones elicited from a child or patient) suggests a variety of characterizations. (4) In choosing among these characterizations, it is apparent that the characteristics of the units must be of such a kind that the human user can learn them both in normal developmental and pathological lan­ guage acquisition. (5) We do not know in advance, deductively, in exactly what way the human factor will interact with the communication factor and others. (6) Phonotactic skewings in language (diachronic, synchronic, and in de­ velopmental and pathological data) reflect the learning process of speakers, including children and clinical patients. (7) This skewing, viewed consistently with the human factor against the background of the other factors of communication, acoustics and

PHONOLOGY AS HUMAN BEHAVIOR

(8)

(9) (10)

(11)

(12)

19

physiology, provides us with information concerning the characteris­ tics of the phonological units. Phonology is not random, but motivated; the frequencies of the phonological units and the ways they combine are determined both by their phonetic make-up and by the speaker's (child's or patient's) exploitation of - or coping with - that make-up in the act of commu­ nication. Gestures enhancing communicative distinctiveness are favored, while articulatorily more difficult gestures are disfavored. There is a conflict between the communication and the human factors in language users' search for maximum communication with minimal effort both in the diachronic development and the synchronic state of a language. This conflict is even more keenly felt in language acquistion, where functional errors and processes may be observed, and even more so in the clinic, where developmental and pathological errors and processes are apparent. The theory of phonology as human behavior can explain, in a princi­ pled way, the connection and interrelationship between the phylogeny, the ontogeny, and the pathology of the development of sound systems in human languages. REFERENCES

Azim, Abdul. 1989. "Some problems in the phonology of Modern Standard Urdu". Paper presented at the First International Conference of the Columbia School of Linguistics, Columbia University, 24 August 1989. Azim, Abdul. 1993. "Problems of aspiration in Modern Standard Urdu". Paper pre­ sented at the Third International Columbia School Conference on Linguistics, Rutgers University, 11 October 1993. Azim, Abdul. 1995. "The phonology of the vocalic systems of Modern Standard Urdu". Paper presented at the Fourth International Columbia School Conference on Linguistics, Rutgers University, 20 February 1995. Azim, Abdul. 1997. "Revisiting the phonology of the vocalic systems of Modern Standard Urdu". Paper presented at the Fifth International Columbia School Con­ ference on Linguistics, Rutgers University, 16 February 1997. Davis, Joseph. 1987 [1984]. "A combinatory phonology of Italian". Columbia Uni­ versity Working Papers in Linguistics 8.1-99. Diver, William. 1974. "Substance and value in linguistic analysis". Sémiotext(e) 1:2.11-30.

20

YISHAI TOBIN

Diver, William. 1975. "Introduction". Columbia University Working Papers in Lin­ guistics 2.1 -20. Diver, William. 1979. "Phonology as human behavior". Psycholinguistic Research: Implications and applications, ed. by Dorothy Aaronson & Robert Rieber, 161186. Hillsdale, N.J.: Lawrence Erlbaum Associates. Diver, William. 1993. "The phonology of extremes: The correlation of initials and finals". Paper presented at the Third International Columbia School Conference on Linguistics, Rutgers University, 11 October 1993. Diver, William. 1995. "The theory". Meaning as Explanation: Advances in linguistic sign theory ed. by Ellen Contini-Morava & Barbara Sussman Goldberg, 43-114. Berlin & New York: Mouton de Gruyter. Dressler, Wolfgang U., Willi Mayertaler, Oswald Panagl & Wolfgang Wurzel, eds. 1987. Leitmotifs in Natural Phonology. Amsterdam & Philadelphia: John Benjamins. Fatihi, A. R. 1987. Economy of Articulation in Mewati Phonology. Ph.D. dissertation, Aligarh Muslim University, India. Flores, Nydia. 1997. "The distribution of post-vocalic phonological units in Spanish". Paper presented at the Fifth International Columbia School Conference on Linguistics, Rutgers University, 15 February, 1997. Grunwell, Pamela. 1987. Clinical Phonology. London: Croom-Helm. Hameed, Shahana. 1999. "Interaction of physiology and communication in the makeup and distribution of stops in Lucknow Urdu". Paper presented at the Sixth International Columbia School Conference on Linguistics, Rutgers University, 11 October 1999. Ingram, David. 1990. Phonological Disability in Children. London: Whurr. Jabeen, Shazi Shah. 1993. Economy of Articulation in the Phonology of Bihar Urdu (as spoken in and around Gaya). Ph.D. dissertation, Aligarh Muslim University, India. Jakobson, Roman. 1968 [1941]. Child Language, Aphasia, and Phonological Univer­ sals. The Hague: Mouton. Jakobson, Roman. 1971. Selected Writings I: Phonological studies. The Hague: Mouton. Liberman, Anatoly. 1991. "Postscript". The Legacy of Ghengis Khan by N. S. Trubetzkoy, ed. by Anatoly Liberman, 295-375. Ann Arbor: Michigan Slavic Publications. Martinet, André. 1955. Économie des changements phonétiques: Traité de phonologie diachronique. Berne: A. Franke. Moore, Kate. 1991a. "Speech rate, phonation rate, and pauses in cartoon and sports narrations". Studies in Logopedics and Phonetics ed. by R. Aulanko & M. Leiwo, vol. II, 135-143. Helsinki: Publications of the Department of Phonetics, Univer­ sity of Helsinki. Moore, Kate. 1991b. "A taxonomy of pauses in Finnish". Ibid., 145-150.

PHONOLOGY AS HUMAN BEHAVIOR

21

Moore, Kate. 1993. "Developmental disfluencies in preschool children". Nordic Prosody VI ed. by Björn Granström & Lennart Nord, 173-181. Stockhom: Almquist & Wiksell. Moore, Kate & Anna-Maija Korpijaakko-Huuhka. 1996. "The clinical assessment of fluency in Finnish". Advances in Clinical Phonetics ed. by Martin J. Ball & Martin Duckworth, 171-196. Amsterdam & Philadelphia: John Benjamins. Moore, Kate & Capita Rosenberg-Wolf. 1998. "Perceptions of hesitations in speech". Nordic Prosody VII ed. by Stephen Werner, 195-269. Frankfurt & New York: Peter Lang. Samet, Mirit. 1993. "An analysis of the speech of a hearing child (2:8) of deaf parents according to the theory of phonology as human behavior". Ms. Depart­ ment of Communication Disorders, Speech, Language and Hearing, The Sackler Faculty of Medicine, Tel-Aviv University. [In Hebrew.] Sampson, Geoffrey. 1980. Schools of Linguistics. Stanford: Stanford University Press. Saussure, Ferdinand de. 1959 [1916]. A Course in General Linguistics. Transi, by Wade Baskin. New York: Philosophical Library. Shaked, Galia. 1990. Early Phonological Development: Phonological processes in children 1:7-2:7 years old. M.A. Thesis. Department of Communication Disor­ ders, Speech, Language and Hearing, The Sackler Faculty of Medicine, Tel-Aviv University. [In Hebrew.] Stampe, David. 1979 [1972]. A Dissertation on Natural Phonology. New York: Garland. Tobin, Yishai. 1986. Review of Sampson (1980). Lingua 68.99-108. Tobin, Yishai ed. 1988a. The Prague School and its Legacy. Amsterdam & Phila­ delphia: John Benjamins. Tobin, Yishai. 1988b. "Phonetics versus phonology: The Prague School and be­ yond". Tobin 1988a. 49-70. Tobin, Yishai. 1988c. "Two quantitative approaches to phonology: A contrastive analysis". Beiträge zur quantitativen Linguistik ed. by Hermann Bluhme, 71-112. Tübingen: Gunter Narr. Tobin, Yishai. 1990a. Semiotics and Linguistics. London & New York: Longman. Tobin, Yishai. 1990b. "A combinatory phonology of the Hebrew triconsonantal (CCC) root system". La Linguistique 26:1.99-114. Tobin, Yishai. 1990c. "Principles for a contrastive phonotactics: The Hebrew tricon­ sonantal (CCC) root system a case in point". Papers and Studies in Contrastive Linguistics 26:137-153. Tobin, Yishai, ed. 1995. Phonology as Human Behavior: Theoretical implications and cognitive and clinical applications. Dibur u-shmiya (Speech and Hearing Disorders) 18 (Special Issue on Phonology). Tel-Aviv: The Israel Speech and Hearing Association. [In Hebrew.] Tobin, Yishai. 1996. "Will the real Professor de Saussure sign in, please? The three faces of Ferdinand". Semiotica 112:3/4.391-402.

22

YISHAI TOBIN

Tobin, Yishai. 1997a. Phonology as Human Behavior: Theoretical implications and clinical applications. Durham, N.C. & London: Duke University Press. Tobin, Yishai. 1997b. "Developmental and clinical phonology: Roman Jakobson and beyond". Acta Linguistica Hafniensia 29.127-166. Tobin, Yishai. 1999. "Developmental and clinical phonology: The Prague school and beyond". Travaux du Cercle Linguistique de Prague, Nouvelle série/Prague Linguistic Circle Papers 3.53-68. Tobin, Yishai. 2000a. "Phonology as human behavior: Initial consonant clusters across languages". Signal, Meaning and Message: Perspectives on sign-based linguistics ed. by Wallis Reid & Ricardo Otheguy. Amsterdam & Philadelphia: John Benjamins. Tobin, Yishai. 2000b. "Comparing and contrasting Optimality Theory with the Theory of Phonology as Human Behavior". The Linguistic Review 17:2-4.303-323. Tobin, Yishai. 2001a. "Trying to 'make sense' out of phonological reduplication in Hebrew". Proceedings of LP 2000: Item order and its variety and linguistic and phonetic consequences ed. by Bohumil Palek & Osamu Fujimura. Prague: Char­ les University Press. Tobin, Yishai & Haruko Miyakoda. 2001b. "An analysis of Japanese speech errors based on the theory of phonology as human behavior". Proceedings of the Second Malaysian International Conference on Languages, Literatures and Cultures. Kuala Lampur, 18 April 2001. Trubetzkoy, Nikolai S. 1969 [1939]. Principles of Phonology. Transl, by Christine A. M. Baltaxe. Berkeley & Los Angeles: University of California Press. Trubetzkoy, Nikolai S. 1991. The Legacy of Ghengis Khan. Ed., and with a post­ script, by Anatoly Liberman. Ann Arbor: Michigan Slavic Publications.

SEGMENTAL VS SYLLABLE MARKEDNESS DELETION ERRORS IN THE PARAPHASIAS OF FLUENT AND NON-FLUENT APHASICS

DIRK-BART DEN OUDEN University of Groningen 1. Introduction Where phonological theory has been applied to clinical data, the concept of markedness has often played a significant role (Blumstein 1991). Not often taken into account, however, is the fact that the markedness value of linguistic structures may not be the same at all psycholinguistic levels of processing. What is marked at some linguistic level of representation may well be unmarked at another. In combination with currently maintained assumptions about the generation of aphasic (phonological) errors at different psycholin­ guistic levels (e.g., Kohn 1988), this finding allows for interesting and useful comparisons, through which the influence of different types of markedness on literal paraphasias may be related to specific psycholinguistic levels of processing. For this study, we have compared the literal paraphasias of fluent and nonfluent aphasic speakers on a repetition task, aimed at determining the influence of syllable structure on error patterns. The results of both groups have been shown to be equal to a large extent (Den Ouden & Bastiaanse 1999 and in press), but the observed differences in complex coda cluster reductions lead us to conclude that different types of markedness apply at the different affected levels of processing in fluent and non-fluent aphasies. Our analysis is in terms of conflicting and violable constraints, which leads to the application of phonological Optimality Theory (Prince & Smolensky 1993) to our data. An important section of this paper is devoted to possible and, in our view, necessary, adaptations to mainstream Optimality Theory, for it to correspond more closely to a plausible psycholinguistic model of speech processing and deal with aphasie data such as presented here. The first section discusses phonological markedness, the background to this concept, and its application to the study of aphasia in general. The difference between fluent and non-fluent aphasia is explained and argued for in the second section. After this, we present the data obtained from a

24

DIRK-BART DEN OUDEN

repetition task and our analysis of these data, focusing on the difference between fluent and non-fluent patient groups in the reductions of coda clusters. This analysis is formalized in the fourth section, where we introduce Optimality Theory () and discuss how its tools can be used to give plausible representations of our findings. 2. Phonological markedness An important factor underlying the relative frequency of occurrence of lin­ guistic structures in language is the markedness of such structures.1 In different languages, and also during the course of child language acquisition, more marked structures will be less widely distributed than less marked structures sharing the same domain. According to Jakobson (1971), the sound system of a language starts off with the sharpest contrast between sounds, i.e., a wide vowel (usually /a/) and a stop with occlusion at the front of the mouth (usually /p/). In other words, vowels are as extremely vowel-like as possible, and consonants are as extremely consonant-like as possible. After this first acquired contrast, the next is that between oral consonants and nasal consonants. This order of acquisition is argued to be based on ease of articulation. The generally acknowledged least marked sequence of segment types, namely Consonant + Vowel (cv), however, is argued to develop because the phonemes "need to be correctly identified by the listener, and [...] the best graspable clue in discerning consonants is their transition to the following vowels" (Jakobson 1971:25).2 If markedness effects are the result of factors that are inherent to language, it is interesting to see what the role of markedness is in pathologies that are specific to language, viz. different forms of aphasia. Jakobson himself already stated that "[a]phasic regression has proved to be a mirror of the child's acqui­ sition of speech sounds" (Jakobson 1971:40), when it really had not, but com­ parable arguments did spark off many research projects into the relation be­ tween markedness and aphasia. Blumstein's (1973) starting hypothesis was very Jakobsonian, in that she appears to have expected to find, and indeed did find, similar effects of the relative phonological markedness of segments in patients with different types of aphasic syndromes, as, "regardless of the area of brain damage, the more 1

Here we use the term structure to imply all phonological particles of analysis, i.e., features or feature combinations, segments or segment combinations, syllables or syllable frames, feet or combinations of feet, etc. 2 Evidence of CV as the syllable structure that is first acquired by children is provided by a number of authors, such as Smith (1973) and Fikkert (1994).

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

25

complex phonological structures are impaired and the less complex phonological structures are relatively preserved" (Blumstein 1973:136). Blumstein studied the error patterns in the speech of patients with Broca's aphasia, conduction aphasia and Wernicke's aphasia. In a nutshell, the phonological errors of Broca's aphasics are claimed to result from the inability to translate a correct phonological speech plan into its correct phonetic counterpart and articulation. Conduction aphasics, in a very broad definition, mainly have (postlexical) problems with the sequencing of correctly retrieved sounds and sound patterns from the lexicon. Wernicke's aphasies' phonological problems are caused by deficient retrieval from the lexicon, or even by distorted lexical representations themselves. These three groups all made more errors on segments that had been classified as more marked beforehand. In segment substitutions, marked segments were generally replaced by less marked segments. Besides segment type, studies by Nespoulous et al. (1984, 1987) took into account syllable structure. They found that the error patterns of Broca's apha­ sies were generally constrained by the relative markedness of different segments and syllable structures, whereas such constraints were not found in the errors of conduction aphasies. Such results of studies into markedness and aphasia lead to hypotheses about the nature of markedness. If aphasie deficits exist at different levels of language processing and markedness effects can be related to specific aphasie syndromes and therefore to specific levels of processing, deepens our knowledge about the origin of markedness effects. One problem, of course, lies in the fact that aphasie syndromes are far from 'specific' (e.g., Poeck 1983; Ellis & Young 1988). Also, as noted by Nespoulous et al. (1984, 1987), if phonological markedness at underlying, ab­ stract levels of processing is heavily influenced, or indeed formed, by phonetic markedness, i.e., motor complexity, it may not be much use looking for the differences between markedness effects at both levels of processing, as they cannot be separated for analysis. 3. Fluent vs non-fluent aphasia and levels of processing A broader classification of patients displaying phonological impairment than that described above (into Broca's, conduction, and Wernicke's aphasies) is through a division between fluent and non-fluent aphasia. In the absence of extralinguistic factors, such as dysarthria, non-fluent patients are generally claimed to suffer from difficulty in the timing and coordination of articulatory movements in speech (Blumstein et al. 1980; Blumstein 1991; Hough et al.

26

DIRK-BART DEN OUDEN

1994). This is related to a deficit at a cognitive phonetic level of processing (Code & Ball 1988). This level is quite peripheral to the language processing system, but it may still be considered linguistic and it is not so peripheral that symptoms of a deficit at this level can be ascribed merely to inadequate bucco-facial muscle strength. Articulatory muscles themselves are intact and all movements necessary for speech production can be correctly executed, but the problem lies in the adequate coordination of and voluntary control over the articulators. An even more cognitive approach says that non-fluent patients are impaired in translating the phonological speech plan into a phonetic plan, which should be fully specified for correct articulation (Code & Ball 1988). Such marginally different interpretations are difficult to disentangle (Code 1998; Croot et al. 1998). Non-fluent aphasics are mostly patients that would be classically diagnosed as suffering from Broca's aphasia with apraxia of speech. The label of fluent aphasia seems to cover a wider range of traditional syn­ dromes. It includes lexical as well as postlexical disorders (Kohn 1988), for example the classical syndromes of Wernicke's aphasia and conduction apha­ sia. What these disorders have in common is that they yield incorrect phonological plans. This may be caused by incorrect lexical access or repre­ sentations, or by incorrect phonemic sequencing, the mapping of speech sounds and features onto metrical frames (i.e., phonological encoding). The difference, then, between fluent and non-fluent aphasics is that fluent aphasies create an erroneous phonological plan that may be correctly executed phoneti­ cally, whereas non-fluent aphasies incorrectly execute, or phonetically imple­ ment, a correct phonological speech plan. Den Ouden & Bastiaanse (1999) argued that this division provides the op­ portunity to investigate whether certain structural markedness effects, such as preferred syllable structure, are the result of phonological or of phonetic level constraints. We studied the effects of positional syllable-internal markedness on the deletion patterns of segments in the paraphasias produced by fluent and non-fluent aphasies on a repetition task. On the basis of a syllable template with relatively strong and weak segment positions (van Zonneveld 1988), we predicted that in a word such as sprints /sprints/, the segments printed in bold would be less susceptible to deletion than the others, if the factors responsible for positional syllable-internal markedness were active. The syllable template model itself was based on language typology (frequency of occurrence of structures) and data from child language acquisition (order of acquisition and tendencies within error patterns).

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

27

The graphs in Figure 1 show the results of the 9 fluent and the 6 nonfluent Dutch aphasic patients on the monosyllabic repetition task that was presented to them. The items in the task were all Dutch monosyllables, with different syllable structures, which, for analysis, were mapped onto the discussed syllable template. In this particular model, the onset and coda satellites (osat and csat) can only be filled with glides, liquids or nasals (i.e., sonorant consonants), the pre-margin (pre) can only be filled with the segment /s/ and the appendix (app) with coronal voiceless obstruents (/s/, /t/) or, in very rare (highly marked) cases, /k/ or /p/. The pre-margin and appendix positions can be considered extrasyllabic: they violat binary branching and their 'behavior' is exceptional in other ways as well (cf. Harris 1994). Positions dependent on other positions are only filled if the position they are dependent on is filled. For example, the pre-margin and the onset satellite are dependent on the onset core (ons). In this model, everything depends on the peak. The position for this peak (the vowel) is left out in these graphs, as the comparison was only between consonant positions. Vowels were hardly ever deleted in the monosyllabic repetition items. The graphs display the number of deletions within a certain segment posi­ tion, relative to the actual occurrence of that position in the items presented for repetition. Using the scores of individual patients, Wilcoxon tests were em­ ployed to calculate the significance of the difference in the mean proportions of deletions. The illustrative word used here is sprints, which did not actually occur in the item list itself, because it is morphologically complex in Dutch.

s

pre

p

r

ons

osat

n

t

csat

s

cod

s

app

pre

ons

p

osat

r

n

csat

Fig. 1 : Deletion errors in different syllable positions (Den Ouden & Bastiaanse

t

cod

s

app

1999)

Although the tendencies that are visible in these graphs show that the core onset and coda positions are least prone to deletion, significance was only reached for the difference between deletions in onset and onset satellite posi­ tions and between deletions in coda and appendix positions for both the fluent and the non-fluent patients. The fluent patients also deleted significantly more segments in appendix positions than in coda satellite positions. The overall

28

DIRK-BART DEN OUDEN

conclusion of this study was that the literal paraphasias of fluent and nonfluent aphasics show the influence of positional markedness on deletions. Nevertheless, what remains is the fact that the proportions of deletions in coda and coda satellite positions differ less for fluent aphasics than for nonfluent aphasies, whereas there is no particular difference in onsets.3 This coda observation, as we choose to name it, may have been obscured by the particu­ lar template that we used and by the rules for associating particular segments to particular syllable slots. The data analysed in Den Ouden & Bastiaanse (1999) containCVCsyllables, in which a sonorant coda would be associated to the coda core position, as well as CCCVC and CVCCC syllables, which may blur the view on what happens to (pure) CC-onsets and CC-codas. For this reason, we decided to take a closer look at the relevant data, taking into account only those items with CC-onsets and/or CC-codas, in order to abstract away from influences by other structural positions as much as possible. 4. The Coda observation In this section, the relevant data and the methods of collecting them are discussed. We analysed the responses to a subset of target items used in Den Ouden & Bastiaanse (1999). For this renewed visit to the repetition study, four more non-fluent patients were tested. The other patients and their data were taken from the previous, 1999 study. 4.1 The experimental investigation Subjects were 10 non-fluent aphasies, 4 male and 6 female, with a mean age of 61 (range 50-79), and 9 fluent aphasies, 7 male and 2 female, with a mean age of 58 (range 38-84). The non-fluent patients had been diagnosed by their speech therapists as suffering from apraxia of speech, without dysarthria. This diagnosis was confirmed by the examiner. The fluent aphasies did not suffer from apraxia of speech. All patients produced literal paraphasias on language tests and in sponta­ neous speech. All were native speakers of Dutch and more than 3 months aphasie due to a single left-hemispheric stroke. The repetition task consisted of 114 Dutch monosyllabic words, of which 41 were analysed for the present study. These were the items with complex 3

The relatively high proportion of deletions in pre-margin position by non-fluent aphasies is explained by the fact that apraxics generally have problems with initiating movement (Code 1998). Non-fluent patients with apraxia of speech will have difficulty with the beginning of words, independently of syllable position (cf. Dogil & Mayer 1998). All syllable onsets in the monosyllabic items in the test obviously coincided with word onsets.

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

29

onsets or complex codas, which did not violate the sonority slope, meaning that the sonority value of segments rose from the margins to the peak (Clements 1990). This restriction bars /st/-onsets, for example.4 The restric­ tions on usable items left us with 21 items with complex onsets and 20 items with complex codas. Deletions were scored per segment position within the analysed onsets and codas. In onsets the opposition was between the onset core and the onset satellite, or, in this case, the first and the second position, respectively. In co­ das, the opposition was between the coda satellite (the first position) and the coda core (the second position). As an example, the word print is (orthographically) given below, with the appropriate position labels: p r i onset core onset sat.

n

t coda sat.

coda core

We wanted to know whether the two positions within a complex onset or coda were equally affected by syllable simplification errors. For this reason, we only took into account deletions of these positions, as they are the only true quantitative simplifications of phonological structure, as opposed to segment substitutions. Note that not all patients produced 41 valid responses to the 41 items under scrutiny, as target items yielding neologisms or no-responses were not included in the analysis. All in all, the literal paraphasias of fluent aphasic patients and non-fluent aphasic patients were analysed for the proportions of deletions per syllable position, in onsets and codas. Results are summarized in Table 1 and in Figure 2. Table 1 shows the ab­ solute number of deletions in the different positions. N is the total number of occurrences of relevant onset or coda clusters for this group of patients. The pvalues printed in bold, based on χ2-tests, show significant differences (α= .05). The graphs in Figure 2 show the mean number of deletions, proportionate to the number of occurrences of the relevant position in the target list, for each group of patients. 4

In the more detailed sonority hierarchies that have been proposed, voiceless fricatives, such as /s/, are considered to be more sonorant than stops, such as lil. Detailed sonority hierarcy (Jespersen 1904, cited in Clements 1990:285): low vowels > mid vowels > high vowels > r-sounds > laterals = nasals > voiced fricatives > voiced stops > voiceless fricatives > = voiceless stops In less detailed hierarchies, with a stronger claim on universality, fricatives and stops are joined in the category of obstruents, with equal sonority values, the adjacency of which also does not make for a well-formed sonority slope.

30

DIRK-BART DEN OUDEN

Onsets

deletions in onset core

deletions in onset sat.

N

x2

Fluent

4

16

186

X =6.394 p = 0.0058

Non-fluent

16

30

206

X=4.064

deletions in coda sat.

deletions in coda core

Fluent

13

11

178

X = 0.045 p = 0.6725

Non-fluent

30

13

198

X = 6.679 p =0.0097

Codas

p = 0.0285

Table 1 : Deletions in onset and coda CC-clusters

■ C[-son] C[+son]

fluent nonfluent Fig. 2a: Deletions in onset clusters for fluent and non-fluent aphasic patients

C[+son] ■ C[-son]

fluent nonfluent Fig. 2b: Deletions in coda clusters for fluent and nonfluent aphasic patients

J

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

31

The graphs visualize the effects shown in Table 1, namely that the patterns of deletions are equal for both groups in onsets, but not,in codas, where only the non-fluent patients delete the sonorant coda position (the coda satellite) significantly more often than the non-sonorant coda position (the coda core). For example, the non-fluent patients' rendition of the target word print (/print/) will characteristically be [pIt], while the fluent patients will turn it into either [pIn] or [pIt], in a seemingly random fashion. The 'coda observation' of Den Ouden & Bastiaanse (1999) thus holds true upon closer inspection. 4.2 Discussion It is clearly not the case that only non-fluent aphasics show effects of markedness relations, while fluent aphasics reveal a random distribution of errors, as the effects are the same for both groups within onset clusters. For our account of these data, we return to the notion of markedness. Jakobson's markedness hierarchy was very much based on a contrast be­ tween segments and segment categories. In order to achieve the biggest con­ trast between vowels and consonants, consonants should be as consonantal as possible, and vowels should be as vowel-like as possible. According to this hierarchy of segmental markedness, consonants are less marked if they are less sonorant. Segmental markedness applies non-contextually; it does not take into account the position of a segment within a syllable. This markedness hierarchy can account for the error pattern of non-fluent patients, but not for the coda observation in fluent patients' errors. If we do look at segments in the context of prosodie structure, a different picture emerges. Clements (1990) argued that the preferred sonority slope of syllables has a steep rise in sonority pre-vocalically and a slow decline in sonority postvocalically. Syllables with sonorant codas are more frequent than syllables with non-sonorant codas. This Sonority Cycle (Clements 1990) allows us to formulate a second markedness hierarchy, which we will call syllable markedness, according to which onsets want to be non-sonorant and codas want to be sonorant. The Sonority Cycle was indeed the object of investigation of Christman (1992), who showed that neologisms (produced by fluent patients) conform to this principle of a steep rise and a minimal decline of sonority in syllables, as formulated by Clements (1990). In a case study, Romani & Calabrese (1998) also investigated the influence of syllable complexity and segmental markedness on literal paraphasias. The Italian mother tongue of their patient, however, has so many restrictions on possible codas, that it does not allow for complex coda analysis in the way that Dutch or English might.

32

DIRK-BART DEN OUDEN

Syllable markedness alone cannot account for the pattern of deletions ob­ served in fluent patients' paraphasias. Note, however, that segmental and syl­ lable markedness reinforce each other in onsets, whereas they are in opposition in codas. It is this combination of the two types of markedness, or rather the crucial conflict between them, that may account for the error pattern of fluent aphasics. Relating the two types of markedness to the results obtained in this study, our claim is that for non-fluent aphasic patients, who have a deficit at a (cognitive) phonetic level of speech production, segmental markedness is dominant. Irrespective of syllable position, these patients 'prefer' to delete sonorant consonants and to end up with non-sonorant consonants. Fluent aphasie patients, who have a deficit at or before the level of phonological encoding, show the influence of both segmental markedness and syllable markedness. For codas, this means that there is a conflict which results in a draw, hence the almost equal distribution of deletions of sonorant and nonsonorant consonants in this syllable constituent. The full analysis of the presented data is therefore as follows: non-fluent aphasies have a deficit at a phonetic level of processing. At this level, where articulatory planning takes place, the markedness of individual segments, or feature combinations, is still an influential factor. The impairment allows this type of markedness to become dominant and this means that when clusters of consonants are reduced, the non-sonorant, segmentally least marked consonant will come out as the winner, irrespective of its position within a syllable. Before this phonetic level of processing, constraints on sonority sequencing, i.e., on preferred syllable structure, are active beside constraints on segmental markedness. At the affected level(s) of processing in fluent aphasies, the conflict between segmental markedness constraints and syllable markedness constraints emerges, as structure-preserving constraints lose control over the output of the speech production process. This yields a pattern of errors in which onsets are relatively systematically reduced to non-sonorant segments, as both types of markedness reinforce each other in onsets, while codas are reduced on a seemingly random basis to either sonorant or nonsonorant segments, as the constraints are in direct conflict over what is a preferred coda. The notion of a conflict between different phonological constraints on well-formedness makes it quite appealing to investigate the representation of the interaction of these constraints within the framework of , as this framework typically allows for constraints with opposing ends and as it has become the most influential phonological framework of the previous and of the current

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

33

decade. The remainder of this paper will be dedicated to the discussion of how the foregoing analysis may or should be represented in , the starting point being to stay as much as possible within the representational and conceptual limits set out by the still developing formal theory itself. 5. The coda observation in OT Since its introduction in the early 1990s,  has quickly gained ground in phonology, as it is presently doing in the domains of syntax and semantics. Its main appeal lies in two characteristics: the focus on well-formedness of the output, as opposed to a focus on rules that seem to exist for their own sake, and the softness of constraints, where a constraint can be violated in order to satisfy something more important. Constraints on the well-formedness of the output, so-called markedness constraints, compete with each other and with structure-preserving faithful­ ness constraints. A grammar is formed by the language-specific ranking of these violable, universal constraints. An unrestricted number of possible can­ didates for the eventual output form of the utterance are generated by a component named GEN (Prince & Smolensky 1993). These output candidates are compared, on the basis of the input form and the ranked constraints. The output candidate that has the least important constraint violations wins and, consequently, is the optimal output. This optimal output may violate con­ straints, in order to satisfy higher ranked, i.e., more important, constraints. There are two types of faithfulness constraints. Parse (or Max) constraints say that material in the input should also be present in the output. Such con­ straints block deletion. Fill (or Dep) constraints say that material in the output should also be present in the input, thus blocking insertion. Another conspicuous characteristic of classic  is that all constraints on the output should compete with each other at all times and that their hierarchi­ cal ranking is stable for adult speakers. This basically means that the 'con­ struction' of the output occurs in one step,  is therefore minimally deriva­ tional, the only derivation being that from the input to the output. If we consider, then, the focus on derivations and cyclicity of processes in previous decades (e.g., Kiparsky 1979), it is only logical that the main criticism of  has been based on evidence that some phonological processes simply cannot be adequately described without making reference to some notion of cyclicity or multiple levels of processing. Such criticism has led to a number of adaptations to the original theory, all aimed at giving satisfactory descriptions of morphophonological processes in which the output form seems to be opaque, and certain constraints appear to

34

DIRK-BART DEN OUDEN

have been applied only to specific substrings of the eventual output form (a phonological word, mostly) (cf. the contributions to Hermans and van Oostendorp 1999). Examples of such tools that aim to maintain the one-step evaluation are Output-Output Correspondence (McCarthy & Prince 1995), in which the optimal output form wants to be as similar as possible to other out­ put forms it is related to, and Sympathy Theory (McCarthy 1998), in which the optimal output form wants to resemble a fairly arbitrarily chosen other output candidate. Other optimality theorists have chosen to abandon the onestep derivation and to incorporate some type of rule ordering in , allowing multiple levels of evaluation, with constraints that apply only to specific levels of processing, or stages in the derivation (Booij 1997; Rubach 2000). Crucially, the 'founding fathers' of , Prince & Smolensky (1993:79), did not put an absolute restriction on the theory as having only one level of evaluation, although the current practice in 'standard'  is such that multiple levels of evaluation are considered a weakness. Beside the discussion over single or multiple levels of evaluation in the formal theory, stands a large body of evidence for multiple levels of processing, from the fields of psycholinguistics and neurolinguistics, in which the issue is not under debate (see Levelt 1989, among many others). Lesion studies going back to the 19th century have shown that different parts of the brain perform different functions (for an overview, see Whitaker 1998). The only questions left to the field, with respect to this subject, are on the defini­ tion of 'function', whether there are direct, one-on-one relations between brain areas and specific functions and, if so, what areas perform which functions (Poeppel 1996a,b; Démonet et al. 1996). The domain of phonological processing itself has also been dissected (cf. Kohn 1988) and results of studies into temporally successive brain activity point towards a 'phonological loop' (Baddeley 1986) in which abstract and articulatory levels are distinct, though possibly mutually influential,  should aim at ways to incorporate these multiple levels of (phonological) processing, instead of focusing on retaining the one-step hypothesis.  has not yet been applied in a systematic way to the study of language breakdown. However, for our purposes, it is worthwhile to consider the progress it is making in the field of child language acquisition, and consider the theoretical and representational questions raised in that domain of investi­ gation (Sternberger & Bernhardt 1997; Barlow & Gierut 1999; Hayes 1999; Tesar & Smolensky 2000). The relation between aphasia and language acqui­ sition has been sketched above and elsewhere (cf. Den Ouden & Bastiaanse, in press) and it is only natural to take it into account here.

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

35

5.1  and child language acquisition Whether the constraints of language are considered to be innate and uni­ versal (Prince & Smolensky 1993) or not (Boersma 1998; Hayes 1999), or whether or not there are different rankings for production and perception, the general analysis of child language acquisition in  terms is that children start off with a high ranking of markedness constraints and a low ranking of faith­ fulness constraints. This simply means that they will not be able to produce (or parse) the adult output forms of their mother tongue and that they will mainly show effects of markedness during the first stages of acquisition. Gradually, then, the markedness constraints will lower, relative to the faithfulness constraints, as the child tunes its grammar to that of its mother tongue (Sternberger & Bernhardt 1997). Communication becomes more accu­ rate as the faithfulness constraints become more prominent. It is not the case that children will reset their constraint rankings on the basis of one single piece of positive evidence. Language acquisition, though fast, does not occur in one day, and children hear a lot of adult output forms on the basis of which they eventually set their grammar and optimize their lexicon. Also, during this process, there is much variation in their own output. Boersma (1998) therefore argues that this should be represented by variable distances between different constraints, which may at times even overlap. Thus, every piece of positive evidence (an adult output form) will cause a constraint to change position within the ranking, but this does not necessarily mean that it will immediately outrank another constraint; it may simply move closer. Constraints have a 'moving space' within the hierarchy and as soon as constraint A comes within the scope of the space of constraint B, this means that both ranking A >  and  > A may occur, which accounts for variation in the output, if constraints A and  are in conflict. It is a common strategy in  to capture optionality, also in adult grammars, by representing adjacent constraints as ranked freely with respect to each other (Clements 1997; Demuth 1997).  applications to child language acquisition thus provide us with the tools to represent the domination of the unmarked and to represent variation or optionality. 5.2  and aphasia: constraints at different levels of processing As noted in the above sections, aphasia is generally characterized by a prominence of unmarked structures. Compared to 'normal' speakers, the aphasic speaker is less faithful to the input, the input here being the lexicon or, for example, utterances to be repeated, whether real words or non-words. The most straightforward way of representing this in  is by a lowering of faith­ fulness constraints, relative to markedness constraints. Note that it is theoreti­ cally also possible that the input itself is disturbed, so that the correct (i.e.,

36

DIRK-BART DEN OUDEN

adult) constraint ranking works on an incorrect input, or that the number and/or type of output candidates that are generated is in some way restricted. These options, however, do not directly account for the prominence of the un­ marked, as observed in aphasic speech. Any systematic way of constricting the input or the output candidates would somehow have to be by incorporating extra markedness constrictions on these domains. This would come down to an extratheoretical add-on for which there is no evidence or argument in nonpathological natural language. We start from the hypothesis that language im­ pairment is focal breakdown of the normal language system, crucially within its own terms. The impairment is assumed not to add new features to the nor­ mal system (cf. Caramazza 1991). The aphasic patients in this study show markedness effects in their impaired output, and our  representation of this fact consists in the lowering of faithfulness constraints, which allows markedness constraints to have greater influence on the choice of the optimal output candidate. Aphasie data are never homogeneous. There is much noise and variabity, which is precisely why statistics are used to determine whether some structures are significantly used more often than others. Variation, as we have seen above, can be represented by 'switching' of adjacent constraints. In an algorithm such as Boersma's (1998), with unequal distances between dif­ ferent constraints in the hierarchy, there may be 100% overlap of constraint space, yielding a 50-50 distribution of two forms in a variable pattern, but it is also possible to conceive of a partial overlap, yielding a different distribution, with one of the two forms occurring more often than the other. In this section, we present the  tableau that represent our analysis of the data discussed here, viz. the coda observation. In reduction of consonant clusters, non-fluent patients will render the target word print (/print/) as [pIt], while fluent patients will turn it into either [pIn] or [pIt], in a seemingly ran­ dom fashion. After the presentation and the explanation of the tableaux, we will go deeper into some considerations underlying our particular representa­ tion of the data. The constraints needed for this analysis are in (1): (1)

Markedness *C[+SON] H O N S (Onset Harmony) HCOD (Coda Harmony) *COMPLEX

Do not allow sonorant consonants Onsets do not want to be sonorant Codas want to be sonorant Do not allow complex onsets or codas

Faithfulness PARSE

Preserve input material

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

37

GEN generates an infinite number of possible output candidates, most of which are rendered irrelevant because they are too deviant from the input form and therefore incur too many violations of faithfulness constraints. The candidates that we will consider in Table 2 are the most relevant to our data and to our example input form print, and we should be able to distinguish between them and choose the correct optimal output form with the constraints given above. Other possible candidates are dealt with by other constraints, but this is outside the scope of this paper. The data discussed all concern cluster reduction. This is represented by the ranking of *COMPLEX over PARSE. This ranking applies to both groups of patients, fluent and non-fluent. Ta­ ble 2 shows how these two constraints and their ranking distinguish between the candidate outputs. The top left cell shows the input. The output candidates are given below that, in the first column. Constraint names are given in the top row, ordered by prominence in the hierarchy from left to right. A candidate's violations of a constraint are marked with '*' in the relevant cell. A crucial violation of a constraint, meaning that the candidate in question is no longer relevant for lower-ranked constraints, as others will always be more optimal, is marked with an exclamation mark !, to the right of which the irrelevant cells are grey. The winning, optimal candidate is marked with a pointed finger

/print/ print pInt

*COMPLEX

PARSE

** !

*!

*'

pt

**

pm

**

nt

**

rm

**

rmt

*!

*

prit

*!

*

prin

*!

* Table 2: Ranking of *COMPLEX over PARSE

The ranking in Table 2 yields four candidates, boldly printed in the tableau, between which the other relevant constraints should differentiate. Non-fluent patients will delete the non-sonorant consonants, regardless of their position within the syllable and we have argued that the constraint

38

DIRK-BART DEN OUDEN

responsible for this is one on segmental markedness, namely *C[+S0N]. The  representation of the impairment of non-fluent patients, on the basis of our data, is therefore fairly straightforward (Table 3): /print/

*COMPLEX

print

** !

pint

* !

PARSE

*C[+SON]

** *

*

**

pit pin

**

* !

[rIt

**

nn

**

* ! ** !

rint

* !

*

**

prIt

* !

*

*

prIn

* !

*

**

Table 3: Non-fluent patients: /print/ → [pit]

The fluent patients may turn example word print into either [pIn] or [pIt]. We have argued that this is because of a competition between a constraint on the preferred sonority value of the syllable constituent coda and a segmental markedness constraint that disallows sonorant consonants. With respect to these two constraints, HCOD and *C[+SON], two rankings are possible, yielding different results, as shown in Table 4a and Table 4b. /print/

*COMPLEX

print

** I

pInt

*!

PARSE

*C[+SON]

HCOD

*

*

**

pit

HONS

** *

pIn

**

*!

rIt

**

*!

rIn

**

Irmt

*!

*

*,*

prit

*!

*

*

prin

*!

*

* *

*

•'*

'f

** i

Table 4a: Fluent patients: /print/ → [prt] ; [pIn]

* /

*

!

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

39

Note that the specific ranking of H O N S is irrelevant here, as long as it is ranked below *COMPLEX. H O N S , as we have argued, merely 'strengthens' the effect of *C[+SON] on onsets. Following  conventions, this non-crucial ranking with respect to other constraints is marked with a dotted line. /print/ print

* COMPLEX

HCOD

*C[+SON]

HONS

**

* !

pInt

PARSE

** i

pIt

* **

* *!

**

pIn

*

rIt

**

rln

**

** | **

rInt

* !

*

prIt

*!

*

prIn

* !

*

*!

*

*

■ . .

*

"

-

'

.

"

■

.

■

■

.

.

■

'

*

* **

Table 4b: Fluent patients: /print/ → [pIt] ; [pIn]

We have now given an  representation of our interpretation and analysis of the data discussed in this paper, viz. the coda observation. For fluent apha­ sics, the tableaux show competition between and switching of the constraint on segmental markedness and the constraint on preferred sonority value of co­ das. The tableau for non-fluent aphasics shows dominance of segmental markedness and either the absence, or the non-competitively low ranking of the constraint(s) representing the Sonority Cycle. Thus, the data of two groups of patients, with deficits at different levels of speech processing, are repre­ sented with two different tableaux. It will not come as a surprise to the reader that we argue that there is a relation between the different levels of processing and the different tableaux. This relation will be specified further below. 5.3 Considerations In the analysis above, aphasia was represented as a lowering of faithfulness constraints and an increased instability of (markedness) constraints, causing switching of adjacent constraints. A major argument against an analysis in which different types of aphasia are represented only through structural reranking of markedness constraints is the fact that aphasie speech errors hardly ever violate the phonotactics of the mother tongue of the speaker, or, indeed, universal restrictions on well-formedness (see Buckin­ gham 1992). This would be unexplained if markedness constraints changed

40

DIRK-BART DEN OUDEN

position in the hierarchy on a large scale. The adherence to (mother tongue) phonotactics points towards a lowering of faithfulness constraints only. However, the variation found in the patterns of paraphasias belonging to different types of aphasia, such as observed in this study and contra Blumstein's (1973) hypothesis, acts as an argument against the mere lowering of faithfulness constraints in the representation of aphasia. To represent different aphasic symptoms only through different degrees of faithfulness lowering comes down to saying that aphasic 'syndromes', or rather, clusters of symptoms, only differ with respect to the degree of seriousness of impairment. For these reasons, rather than claiming that the constraints H O N S and HCOD are ranked non-competitively low at the level of impairment of nonfluent aphasies, we argue that they are non-existent at this level. In this way, we minimize the structural reranking of markedness constraints, while still being able to represent impairments at different processing levels, with dif­ ferent characteristics. This means, then, that our analysis allows for different levels of evaluation of constraints, where not all constraints are active (i.e., exist) at all levels. From a psycholinguistic and neurolinguistic perspective, this is the only natural way to conceive of linguistic processing, and we also point to the more formally optimality theoretic attempts we have cited to allow for multiple-step derivation by incorporating multiple levels of evaluation in , with the approach taken here is by no means incompatible. In psycholinguistic modeling, it is common practice to minimize the number of levels, modules or stages of processing to those necessary for an accurate representation of empirical findings. A similar principle, Level Minimalism, is formulated by Rubach (2000:313), for his modification of , Derivational Optimality Theory, which allows multiple levels of evaluation. Another principle he formulates to restrict the power of his framework is that of Reranking Minimalism: "[the] number of rerankings is minimal [...] reranking of constraints comes at a cost and needs to be argued for" (Rubach 2000:313). This principle is in line with our approach of unstable rankings to account for variation, but with level-specific constraints instead of structural reranking of markedness constraints to account for the influence of different factors at dif­ ferent levels of speech production processing. The constraint * COMPLEX, which in our analysis is crucially present at the levels of impairment of both fluent and non-fluent patient groups, is formu­ lated here in terms of syllable markedness. It therefore prevents us from claiming that it is syllable markedness that is active besides segmental markedness at the pre-phonetic level of processing and that at the phonetic

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

41

level of processing, only segmental markedness constraints are active. This is because we have chosen to follow the intuitive notion that the underlying cause of cluster reduction in the first place is the same for fluent and nonfluent aphasics, namely that syllable constituents prefer to be simple (cf. Gilbers & Den Ouden 1994). Although the assumption that all constraints on syllable markedness act as one group and are active at exactly the same levels is not necessarily true, our analysis would gain by a more straightforward division between constraints that work on different domains, at different levels of evaluation/processing. This 'problem' would be solved if *COMPLEX were to be conceived of as a constraint on, for example, adjacent segments, without making reference to syllable constituents. If, however, the formulation of the constraint is indeed similar to 'consonants do not want to be adjacent to consonants', the prediction is that there should be no difference between proportions of reductions within syllable constituents and across syllable constituents. This hypothesis needs to be tested. 6.

Conclusion On the basis of fluent aphasics' and non-fluent aphasies' responses to a monosyllabic real word repetition test, we have argued that there is a differ­ ence between the pre-phonetic and the phonetic level of processing. In our  approach, the constraints responsible for the Sonority Cycle, as formulated by Clements (1990), are active only at pre-phonetic levels of evaluation, whereas a constraint on segmental markedness, saying that consonants should be as consonantal as possible (and therefore non-sonorant), is active at the prephonetic, as well as at the phonetic level. Aphasia, in an  approach, comprises the lowering of faithfulness con­ straints at the affected level of processing. This accounts for the different types of aphasia that are distinguished clinically. Also, aphasia is characterized by unstable ranking of close (adjacent) markedness constraints, which accounts for the high degree of variation found in aphasie error patterns. Along the lines of Boersma (1998), this could be conceived of as a widening of the moving space of constraints from their relatively fixed place in the hierarchy, which increases the opportunity for overlap of constraints. In our study, this overlap and thus switching of constraints was particularly visible in the paraphasias of fluent aphasies, at whose affected level of processing (pre-phonetic) a competition between segmental markedness (*C[+SON]) and syllable markedness ( H C O D ) leads to variation in the output. Where the process of cluster reduction applies, fluent patients will render the example item print /print/ as either [pIt] or [pIn], with an equal distribution of

42

DIRK-BART DEN OUDEN

deletion of the sonorant or the non-sonorant segment. It is our position that language breakdown in the form of aphasia provides a window on the workings of the language system. Linguistic theories should be able to deal with the view thus offered. REFERENCES Baddeley, Alan D. 1986. Working Memory. Oxford: Clarendon Press. Barlow, Jessica A. & Judith A. Gierut. 1999. "Optimality theory in phonological acquisition". Journal of Speech, Language and Hearing Research 42:6.14821498. Blumstein, Sheila E. 1973. "Some phonological implications of aphasic speech". Psycholinguistics and Aphasia ed. by Harold Goodglass & Sheila Blumstein, 123-137. Baltimore & London: Johns Hopkins University Press. Blumstein, Sheila E. 1991. "Phonological aspects of aphasia". Acquired Aphasia ed. by Martha T. Sarno, 157-186. San Diego: Academic Press. Blumstein, Sheila E., William E. Cooper, Harold Goodglass, Sheila Statlender & Jonathan Gottlieb. 1980. "Production deficits in aphasia: A voice-onset time analysis". Brain and Language 9.153-70. Boersma, Paul. 1998. Functional Phonology: Formalizing the interactions between articulatory and perceptual drives. (= LOT International Series, 11.) Ph.D. dissertation, University of Amsterdam. The Hague: Holland Academic Graphics. Booij, Geert E. 1997. "Non-derivational phonology meets Lexical Phonology". Deri­ vations and Constraints in Phonology ed. by Iggy Roca, 261-268. Oxford: Oxford University Press. Buckingham, Hugh W. 1992. "Phonological production deficits in conduction apha­ sia". Conduction Aphasia ed. by Susan E. Kohn, 77-116. Hillsdale, N.J.: Lawrence Erlbaum Associates. Caramazza, Alfonso. 1991. "Data, statistics and theory: A comment on Bates, McDonald, MacWhinney, and Applebaum's 'A maximum likelihood procedure for the analysis of group and individual data in aphasia research'". Brain and Language 41.43-51. Christman, Sarah S. 1992. "Uncovering phonological regularity in neologisms: Con­ tributions of sonority theory". Clinical Linguistics & Phonetics 6:3.219-247. Clements, George N. 1990. "The role of sonority in core syllabification". Papers in Laboratory Phonology I: Between the grammar and physics of speech ed. by John Kingston & Mary E. Beckman, 283-333. Cambridge: Cambridge University Press. Clements, George N. 1997. "Berber syllabification: derivations or constraints?". Derivations and Constraints in Phonology ed. by Iggy Roca, 289-330. New York: Oxford University Press. Code, Chris. 1998. "Models, theories and heuristics in apraxia of speech". Clinical Linguistics and Phonetics 12:1.47-65.

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

43

Code, Chris & Martin J. Ball. 1988. "Apraxia of speech: The case for a cognitive phonetics". Theoretical Linguistics and Disordered Language ed. by Martin J. Ball, 152-167. London & Sydney: Croom Helm. Croot, Karen, Karalyn Patterson & John R. Hodges. 1998. "Single word production in non-fluent progressive aphasia". Brain and Language 61.226-273. Démonet, Jean-François, Julie A. Fiez, Eraldo Paulesu, Steve E. Petersen & Robert J. Zatorre. 1996. "PET studies of phonological processing: A critical reply to Poeppel". Brain and Language 55.352-379. Demuth, Katherine. 1997. "Multiple optimal outputs in acquisition". University of Maryland Working Papers in Linguistics 5.53-71. Dogil, Grzegorz & Jörg Mayer. 1998 "Selective phonological impairment: A case of apraxia of speech". Phonology 15:2.143-188 Ellis, Andrew W. & Andrew W. Young. 1988. Human Cognitive Neuropsychology. Hillsdale, N.J.: Lawrence Erlbaum Associates. Fikkert, Paula. 1994. On the Acquisition of Prosodic Structure. Ph.D. dissertation, University of Leiden. Gilbers, Dicky G. & Dirk-Bart den Ouden. 1994. "Compensatory lengthening and cluster reduction in first language acquisition: A comparison of different analy­ ses". Language and Cognition 4: Yearbook 1994 of the research group for Theo­ retical and Experimental Linguistics of the University of Groningen ed. by Ale de Boer, Helen de Hoop & Henriëtte de Swart, 69-82. Groningen: University of Groningen. Harris, John. 1994. English Sound Structure. Oxford: Blackwell. Hayes, Bruce. 1999. "Phonetically Driven Phonology: The role of Optimality Theory and Inductive Grounding". Functionalism and Formalism in Linguistics, vol. I: General Papers ed. by Michael Darnell, Edith Moravcsik, Frederick J. Newmeyer, Michael Noonan & Kathleen Wheatley, 243-285. Amsterdam & Philadelphia: John Benjamins. Hermans, Ben & Marc van Oostendorp, eds. 1999. The Derivational Residue in Phonological Optimality Theory. Amsterdam & Philadelphia: John Benjamins. Hough, Monica Strauss, Salvatore Demarco & Donna Farler. 1994. "Phonemic re­ trieval in conduction aphasia and Broca's aphasia with apraxia of speech: Un­ derlying processes". Journal of Neurolinguistics 8: 4.235-246. Jakobson, Roman. 1971 [1941]. Studies on Child Language and Aphasia. The Hague: Mouton. Jespersen, Otto. 1904. Lehrbuch der Phonetik. Leipzig & Berlin: . G. Teubner. Kiparsky, Paul. 1979. "Metrical Structure Assignment Is Cyclic". Linguistic Inquiry 10:3.421-441. Kohn, Susan E. 1988. "Phonological production deficits in aphasia". Phonological Processes and Brain Mechanisms ed. by Harry A. Whitaker, 93-117. New York: Springer. Levelt, Willem J. M. 1989. Speaking: From intention to articulation. Cambridge, Mass.: MIT Press.

44

DIRK-BART DEN OUDEN

McCarthy, John. 1998. "Sympathy and Phonological Opacity". Ms., University of Massachusetts, Amherst. McCarthy, John & Alan Prince. 1995. "Faithfulness and reduplicative identity". Pa­ pers in Optimality Theory. (= University of Massachusetts Occasional Papers in Linguistics, 18), ed. by Jill Beckman, Laura Walsh Dickey & Suzanne Urbanczyk, 249-384. Amherst, Mass.: Graduate Linguistics Student Association. Nespoulous, Jean-Luc, Yves Joanette, Renée Béland, David Caplan & André Roch Lecours. 1984. "Phonological disturbances in aphasia: Is there a 'markedness ef­ fect' in aphasic phonemic errors?". Advances in Aphasiology, vol. 42: Progress in Aphasiology ed. by F. Clifford Rose, 203-214. London: Raven Press. Nespoulous, Jean-Luc, Yves Joanette, Bernadette Ska, David Caplan & André Roch Lecours. 1987. "Production deficits in Broca's and conduction aphasia: Repeti­ tion vs. reading". Motor and Sensory Processes in Language ed. by Eric Keller & Myrna Gopnik, 53-81. Hillsdale, N.J.: Lawrence Erlbaum Associates. Ouden, Dirk-Bart den & Roelien Bastiaanse. 1999. "Syllable structure at different levels in the speech production process: Evidence from aphasia". Paper presented at HILP 4, University of Leiden, January 1999. Ouden, Dirk-Bart den & Roelien Bastiaanse. In press. "Syllable structure at different levels in the speech production process: Evidence from aphasia". Proceedings of HILP 4 ed. by Jeroen van der Weijer, Harry van der Hulst & Vincent van Heuven. Amsterdam & Philadelphia: John Benjamins. Poeck, Klaus. 1983. "What do we mean by 'aphasie syndromes'? A neurologist's view". Brain and Language 20.79-89. Poeppel, David. 1996a. "A critial review of PET studies of phonological processing". Brain and Language 55.317-351. Poeppel, David. 1996b. "Some remaining questions about studying phonological processing with PET: Response to Demonet, Fiez, Paulesu, Petersen & Zatorre (1996)". Brain and Language 55.380-385. Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. New Brunswick, N.J.: Rutgers University Center for Cognitive Science Technical Report 2. Romani, Cristina & Andrea Calabrese. 1998. "Syllabic constraints in the phonological errors of an aphasie patient". Brain and Language 64.83-121. Rubach, Jerzy. 2000. "Glide and glottal stop insertion in Slavic languages: A DOT analysis". Linguistic Inquiry 31:221\-\1. Smith, Neilson V. 1973. The Acquisition of Phonology: A case study. Cambridge: Cambridge University Press. Sternberger, Joseph P. & Barbara H. Bernhardt. 1997. "Optimality Theory". The New Phonologies: Developments in clinical linguistics ed. by Martin J. Ball & Raymond D. Kent, 211-245. San Diego & London: Singular Publishing Group. Tesar, Bruce & Paul Smolensky. 2000. Learnability in Optimality Theory. Cambridge, Mass.: MIT Press.

SEGMENTAL VS SYLLABLE MARKEDNESS IN APHASIA

45

Whitaker, Harry A. 1998. "Neurolinguistics from the Middle Ages to the Pre-Modern Era: Historical vignettes". Handbook of Neurolinguistics ed. by Brigitte Stemmer & Harry A. Whitaker, 27-54. San Diego: Academic Press. Zonneveld, Ron van. 1988. "Two level phonology: structural stability and segmental variation in Dutch child language". First Language Acquisition ed. by Fred van Besien (= Association Belge de Linguistique Appliquée Papers, 12), 129-162. Antwerpen: University of Antwerpen.

. WORDS IN DEAFNESS AND STUTTERING

MORPHOSYNTACTIC FRAGILITY IN THE SPOKEN AND WRITTEN ITALIAN OF THE DEAF ROBERTO AJELLO, GIOVANNA MAROTTA, LAURA MAZZONI & FLORIDA NICOLAI University of Pisa 1. Introduction Morphology has often been characterized as particularly 'fragile', since it is subject to reduction processes of various kinds. Grammatical morphemes are, in fact, highly arbitrary, as well as not particularly salient from the articu­ latory, perceptive or semantic point of view. Such fragility is emphasized in those contexts naturally tending towards linguistic simplification, especially in the emergence of pidgins and the development of interlanguages origi­ nating in the spontaneous acquisition of L2, in formally guided learning of L2, and in the acquisition of L1 (see, e.g., Simone 1988, Berretta 1987; 1990 and Banfi 1993). Both the speech and the writing of subjects with acoustic deficits, like the deaf, show great alterations in morphological structure, although with marked interindividual variability. The present work deals with the linguistic production of some profoundly deaf Italian people who never wore prothesis. We analyze first their speech and then a written, 'not elicited' production, consisting of fax messages sent by a deaf person to other deaf people. The choice of this kind of production is due to the fact that it has many features in common with speech, such as 'ellipsis', brachylogy, topicalization processes, nominal sentences, and in general micro-planning involving very short sentences or even single words. The results of such analysis are then matched with those of a test aimed at checking the morphological ability of some of the subjects who wrote the fax messages and produced the analyzed speech. The coding of the data in the CHAT system is still in progress. All the deaf subjects examined (aged 40-65 years) belong to a generation who received an oralist instruction in specialized Institutes, where neither communication in Italian sign language (LIS = Lingua Italiana dei Segni) nor the use of mimetic gestures was encouraged, since this kind of interactional behavior was considered an obstacle to the correct learning of vocal Italian.

50

AJELLO, MAROTTA, MAZZONI & NICOLAI

We present the state of the art in the field of research dealing with the lin­ guistic production by the deaf, we provide a morphological analysis of the subjects' speech, examine their written production, focus on orthography; we discuss a morphosyntactic test and, finally, offer a global interpretation of the data. 2. State of the art The writing ability of deaf people has not been thoroughly researched. Still, the results of the scanty literature in this field uniformly emphasize the point that most difficulties reside in morphology, and, more specifically, in so-called 'free' morphology. These data, resulting initially from studies on deaf subjects and their mastering of the English language (Kelly 1987, 1993; Maxwell & Falick 1992; Paul & Quigley 1994), have been confirmed by re­ search made on deaf Italian people. The latter results have higher relevance, in consequence of the greater complexity of bound morphology in Italian in comparison to English. Research in this area, carried out with different ex­ perimental modalities, and concerning subjects of different ages, is due mainly to the team of scholars from the Istituto di Psicología, C.N.R. (Rome), who sometimes work in collaboration with outside scholars (see, in particular, Taeschner et al. 1988; Volterra & Bates 1989; Fabbretti 1997, 2000; Fabbretti et al. 1998). In order to evaluate the linguistic competence of the deaf through the me­ dium of writing, researchers have mostly utilised a form of production which they consider as the most spontaneous, namely the epistle or the diary, without excluding other methodologies, such as tales or structured tests (see Quigley, Power & Steinkamp 1977). These written productions are chosen for analysis, on the basis of the commonly shared opinion that spontaneous writing may provide a reliable index of the level of linguistic competence of the deaf, in consideration also of the relevant role played by this modality in various educational programs. Historically, literacy has always been the basis for the education of the deaf, since it was believed that also vocal language learning should start from competence in writing.1 All the research works pertaining to this subject, carried out both on Italian and on English, agree on the following points: -

deaf subjects produce shorter sentences and simpler syntactic structures in comparison to hearing subjects;

1 For a recent development within this methodology based on the central role of writing, see Radelli (1998).

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

-

51

they also reveal a poor vocabulary and lexical rigidity; they have difficulties with relative, subordinate and pronominal clauses; although they appear to have difficulties in many aspects of the written language (on lexical, morphological, syntactic and pragmatic levels), the major problems concern morphology (omission, substitution, and, to a lesser extent, also addition of morphemes).

Let's consider the most relevant research works carried out on Italian deaf subjects. In Taeschner et al. (1988), 25 deaf students (aged 11-15 years), com­ pared with a control group of 125 hearing students (aged 6-16), were tested in the written modality by way of structured tasks aiming at eliciting answers relative to some morphosyntactic aspects, such as plural, clitics, and definite articles. As for the first two morphological categories, a delayed response is evident in the experimental group in comparison to the control group, but there is no deviance. Errors pertaining to the formation of plural consist mainly in the generalization of the most frequent morpheme -i and in treating feminine singular nouns in -e as plurals; moreover, the majority of errors take place in non-words (as happens with hearing subjects), and this suggests that plural nouns might be learnt by rote. The experimental group's responses to clitic items present a remarkable delay (more noticeable than in the case of plurals) and a prevailing pre-verbal collocation, even in cases when their position should be post-verbal. The general framework concerning the articles appears to be more complex in the experimental group and qualitatively different from the one relative to the control group: deaf subjects make more errors, generally due to an idiosyncratic strategy, since they make the article agree with the final vowel of the noun; thus le (f.pl.) fucile (m.sg.), instead of il (m.sg.) fucile (m.sg.) "the rifle"; i (m.pl.) notti (f.pl.), instead of le (f.pl.) notti (f.pl.) "the nights"; Ie (f.pl.) notte (f.sg.), instead of la (f.sg.) notte (f.sg.) "the night". The authors of this study take into consideration several factors that might influence the linguistic performance of the deaf, such as their delayed exposi­ tion to Italian, the reduced linguistic input they receive, due to their acoustic deficit, and the formal situation of teaching they experience. However, they realize that these general factors are not sufficient to explain why only a few specific morphological aspects are problematic. So, for instance, deaf subjects' difficulty in using articles might be due to the fact that these are homonymic with clitic pronouns, but this does not explain why difficulties concern mainly m.sg. lo and m.pl. gli. One might ascribe this latter type of misuse to the deaf subjects' lack of sensitivity to the phonological conditions requiring the selection of these articles. But also this explanation does not

52

AJELLO, MAROTTA, MAZZONI & NICOLAI

account for the hypergeneralization that makes the article agree with the final vowel of the noun. The errors are ascribed to the fact that free morphology is more difficult and more problematic than bound morphology, since free morphemes are typically short, not stressed, and cany relatively scarce semantic information.2 Volterra & Bates (1989) analyzed the written production (10 letters to a friend) of a congenitally deaf woman who has reached a good competence level in Italian. Her written production is compared with the one of 2 control subjects: one who has learnt Italian as L2 at the age of 16 and a native speaker. The written production of the deaf woman shows complex syntactic constructions and a lexical usage that are comparable with those of an Italian adult with a high literacy level; still, she makes frequent, systematic morphological errors, mostly in the domain of free morphology. The few errors in bound morphology concern long distance agreement. When the same deaf subject takes grammatical tests, she shows good metalinguistic knowledge of many rules governing free and bound morphemes of Italian, and makes very few errors. These results reveal that probably her problem does not lie at the level of morphological knowledge, but at the level of mor­ phological access and use. Also, her errors are different from the bilingual subject's. The deaf subject makes errors equally in substitutions, omissions, and insertions, while the bilingual subject makes error equally in bound and free morphology, but makes many errors of substitutions; additionally, the latter subject fails in making gender agreement, while this error is rare in the deaf subject; and above all, the bilingual subject makes lexical errors. All these data hint that the source of errors is not the same for the two subjects and is not referrable to their shared situation as L2 learners. In the hearing subject's responses, some errors may be plausibly interpreted as interference phenomena from L1, as, e.g., when he invents Italian words as a result of transliteration from English. As for the deaf subject, interference from LIS seems to be excluded, because her errors are similar to those of deaf adolescents who do not know LIs, and because they do not include either errors in the ordering of words, which is an element of great difference between the two languages, or errors in gender agreement, which is a category totally absent in LIS. They think that the morphological errors are not referrable to the fact that L2 was learnt after the critical period, first because, as a matter of fact, the subject in question was exposed to Italian very early, and secondly because the specific scheme of morphological enfeeblement is

This is the reason why deaf Italians very often are not able to recognize them in lip-reading.

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

53

clearly selective. The explanation offered relates the morphological errors to a degraded input, which to the deaf often appears to consist of 'islands' of content words inserted in syntactic sequences (free morphemes in fluent production are short, uttered rapidly and without stress). Such degraded input may mean that the deaf subject has fewer opportunities to observe the application of morphological processes: he/she knows the rules at a conscious level, but never reaches that level of automaticity which is necessary for a fluent and correct performance. It thus seems that it is not possible to imagine a unique model of linguistic production which can explain the morphological weakness in learners of L2, in deaf subjects, and in agrammatic aphasics, who never show morphological enfeeblement in the presence of integral lexical and syntactic capacity. The results of the latest work by Fabbretti et al. (1998) confirm on the whole those of previous studies (Fabbretti 1997), although with slight differences, namely the probable influence of Lis in the origin of a few forms. In this work, 10 native signing deaf subjects, born of deaf parents, perform 4 writing tasks: a •summary of a short comic movie sequence, a summary of a video shoot in which deaf people play cards and talk about the game in LIs, a summary of another video shoot with a tale in LIS, and the writing down of a letter containing a refusal to an invitation made by a friend. One control group consists of 10 hearing subjects born of deaf parents who are native signers, and a second one consists of 10 hearing subjects who have never had contacts with deaf people. The results are in line with those from previous research, in particular Fabbretti et al. (1998): deaf subjects present a pattern of selective difficulties with morphology, especially with functional words. Interestingly enough, although Italian offers many occasions for errors in the bound morphology of nouns and verbs, such errors are infrequent. Deaf people seem to have access to some aspects of bound morphology, such as the plural, by rote. Although the data in Fabbretti (1997) reveal more frequent non-conventional orthographic forms in the writing of deaf people than in the writing of hearing subjects, the results of Fabbretti et al. (1998), according to which deaf subjects make fewer orthographic errors than the hearing control group, seem to confirm the observation made by Mayer & Welles (1996), who state that conventional spelling is an area of written language in which the deaf show relatively few difficulties. Interference from LIs is excluded because there are no errors in the ordering of words and because the hearing signers do not make the same kind of errors as the deaf subjects. The explanation for the errors is to be sought in deafness: the results are compatible with the statement that acoustic perception plays a special role in the acquisition and in the use of grammatical morphology.

54

AJELLO, MAROTTA, MAZZONI & NICOLAI

Similar observations have been made, among others, by Hanson & Wilkenfeld (1985) in relation to deaf subjects acquiring English: the lack of acoustic/articulatory mediation would have heavy consequences for the acqui­ sition of grammatical morphology, including the production and (perhaps) the understanding of prepositions, articles, pronouns and other free morphemes. Moreover, the specific difficulties the deaf have with the written language might derive from the way in which they learn the vocal language as children, which most of the time involves an intense one-year-long period of formal training; their written language is based on a vocal language they learn late, when the critical period is over. Many recent research works emphasize that the learning of morphology, even the morphology of a sign language, is par­ ticularly difficult after the critical period. It is possible that the explanation based on the critical period and the one based on phonetic/phonological mechanisms are not mutually exclusive: both in vocal and in sign languages, the aspects of morphology more closely connected to the transmission code (the acoustic one in the case of vocal languages, and the visual one in the case of sign languages) must be learnt early. The recent work presented by Chesi (2000) is particularly stimulating. Following the theoretical framework of generative grammar, he analyzed a corpus of verbal and written production of a group of 13 deaf Italian subjects, ranging in age from 6 to 17 years. All the subjects were prothesized, although with different auditory gain. Both verbal and written productions show comparable results. The number of errors is relatively high, especially for the subjects with less auditory gain from the prothesis. For the sake of clarity, Chesi (2000) speaks of non-standard forms instead of errors, since he believes that the grammatical competence of the deaf is completely compatible with the Universal Grammar (UG) hypothesis. Omission of functional elements, like articles, clitics, prepositions, auxiliaries, is well attested. Interestingly, some peculiar asymmetries emerge from the data. For instance, the article appears to be omitted in the Object Noun Phrase more often than in the Subject Noun Phrase. Enclitic forms are omitted more than proclitic ones, which confirms a trend already emerging in the work by Taeschner et al. (1998) and Fabbretti (2000). In verbal inflection, the 3rd sg.ps. is the most frequently used by the deaf children. Finite forms of the verb are often substituted by the infinitive ones. As for the auxiliaries, avere "to have" is often employed instead of ess ere "to be". Therefore, in Chesi's data too, functional morphology does not appear to be perfectly acquired by the deaf children, because of the poor stimulus they get from the input. However, in the interpretation of the data he collected, Chesi (2000) tries to demonstrate that the structure of the spoken and written language

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

55

of the deaf does not conflict with the basic principles of Universal Grammar. In particular, Chesi (2000:193) believes that structural positions and the features they check are not absent, since they are activated and licensed by special lexical heads, even if some functional elements are omitted in the surface representation of the sentence. 3. Speech production We examined the speech production of 7 subjects, 4 males and 3 females. All of them but one (Al) normally use LIs as a communication language inside the deaf community, although they received an oral education during their childhood in the specialized Institutes they attended. The speech production corpus consists of spontaneous dialogues between two deaf people in the presence of a hearing participant or between a single deaf and a hearing person. The speech was recorded in a sound-proof room with professional instruments and then phonetically transcribed and acoustically analyzed. The total amount of speech recorded and analyzed is about 30 minutes. Let's first consider the bound morphemes in the recorded speech: the un­ stressed final vowels performing morphological functions are very often cen­ tralized or omitted. As a consequence, the marks of nominal as well as of ver­ bal inflection are often no more recoverable, and the meaning of the whole sentence may be affected by that. Table 1 shows the distribution of the possible outputs for the final vowels (correctly produced, omitted or centralized) in our speech corpus, with the in­ dication of the subjects analyzed. As we can see, the rate of final vowel re­ duction (26.0% omission plus 46.2% centralization) reaches the 72.2% of the total. + Total @ Al A2 G R1 R2 R3 S All

24.7 29.0 40.9 32.7 18.5 31.1 21.9 27.9

32.6 22.6 18.2 16.4 30.8 19.7 40.6 26.0

42.7 48.4 40.9 50.9 50.8 49.2 37.5 46.2

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

Table1: Percentage values of final vowels correctly produced {+), omitted (-) or centralized (@)

According to the limited literature on this topic (cf. Savy 2000), the same trend occurs also in the informal speech of hearing subjects, especially in the final position of intonational phrases or utterances, where vocalic reduction

56

AJELLO, MAROTTA, MAZZONI & NICOLAI

appears to be physiologically conditioned. As is well known, at the end of the breath group, glottal pressure goes down, global energy gets reduced, and ar­ ticulatory inertia increases. These same reasons may account for the word-final vowel reduction which can be observed in the speech produced by the deaf, with a crucial difference in the amplitude of the motor program window, which is no doubt wider in the hearing subjects than in the deaf. As a matter of fact, the deaf program their vocal production mainly in terms of phonological words; therefore the final position of a word normally coincides with the final position of an intonational phrase or utterance (see Ajello et al. 1998/1999). On the other hand, the occurrence of phonetic reduction in the speech of hearing people makes the visual input reaching the deaf blurred and ambiguous, which reinforces the natural trend towards segmental weakening in unstressed syllable. The processes of reduction and centralization involving final segments of words cannot be considered a pure performance effect on behalf of the deaf, but are supposedly related to incomplete morphological competence. That is why, in order to check the morphological competence of the deaf, we decided to eliminate the variable of the performance difficulties, and to analyze the written production where mechanical drawbacks are not present. 3.1 Free morphology in speech A second result of the analysis of speech was an imperfect use of free mor­ phemes. The trend towards the omission of determinants, prepositions, auxilia­ ries, and clitics is clear evidence of the difficulty of the deaf in mastering mor­ phological aspects. By limiting the exposition of the data only to the occurrence of determinants (in particular, definite articles) and prepositions, the definite articles are often omitted, although the relative percentage is not the same for all the morphemes. In Table 2, in comparison with a generalized omission of the definite m.sg. article in all its allomorphs (il, lo, l), a rather stable occurrence of the corresponding feminine form (i.e. la) is observed. il lo l la i All

+ 10.0 0.0 33.3 65.2 0.0 35.7

90.0 0.0 66.7 34.8 100.0 64.3

Total 100.0 0.0 100.0 100.0 100.0 100.0

Table 2: Percentage values of production (+) and omission (-)for the definite articles

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

57

Our data agree with those collected by Chesi (2000:181), who found the highest percentage of omission for i and il (m.plur. and m.sing., respectively). The difference between the gender (masculine = weak versus feminine = strong) in the treatment of the articles is confirmed by the data relative to the acquisition of Italian as L1 (cf. Antelmi 1997), as well as L2. The feminine form appears to be more salient both for hearing children and for the deaf. In particular, la is more salient, since it is an open syllable, produced with the most open, and thus most visible, vowel. Moreover, in the feminine paradigm, the allomorphy is more restricted than in the masculine one. As far as the use of prepositions is concerned, we observe a general trend towards omission. This finding too is not unexpected, since it is often reported in the literature. However, there are some interesting differences among the prepositions occurring in our corpus. As is shown in Table 3, omission is par­ ticularly strong in the case of a ("to"), di ("of") and da ("from"), but rather weak in the case of per ("for "). a di da in con per All

+

-

Total

6.5 46.2 16.7 54.5 58.3 75.0 37.8

93.5 53.8 83.3 45.5 41.7 25.0 62.2

100.0 100.0 100.0 100.0 100.0 100.0 100.0

Table 3: Percentage values of production (+) and omission (-) relative to prepositions

As a matter of fact, the first three of those prepositions not only are very frequently used in Italian, but also play more than one thematic role. These data seem to indicate that the deaf are rather able to use 'unambigous' prepo­ sitions, while they find it more difficult to select a preposition with different syntactic functions. As far as the production and the right use of the preposi­ tion per are concerned, we will see that the same priority emerges as from the analysis of the written production we examined. On the other hand, we would like to recall that in LIS there is a specific sign corresponding to Italian per, while all other Italian prepositions have no corresponding sign. 4. Written production The written production, consisting of 200 fax messages, involved 14 male and 26 female subjects, all of them expert signers, but, as in the previous case, lacking a mother tongue acquired naturally through exposition from birth. All the subjects we have examined learnt the vocal language through a hard ten-

58

AJELLO, MAROTTA, MAZZONI & NICOLAI

year training period in the Institutes, which started at the age of 6 to 9, while they spontaneously acquired a rudimentary sign language, also in the Institutes, simply by communicating with older pupils in limited situational contexts. The interpretation and normalization of fax texts has turned out to be a very complex, not always satisfactory, operation, though one of the authors of the present work, Laura Mazzoni, not only has had life-long experience of the deaf world, but also knows personally and very well the authors of the fax messages and meets them almost daily. The morphological analysis of the written corpus reveals heavy reduction of the bound morphemes, omission or misuse of functional elements, lexicalization of grammatical categories and weak consciousness of some derivational mechanisms. 4.1 Heavy reduction of the bound morphemes The reduction of the bound morphemes and the consequent overextension of a few forms concern mainly the verb forms, and much less the forms of the other grammatical categories. The 3rd ps.sg. of the present indicative and the infinitive tend to function as the basic verb forms in independent clauses: (1) a. io scrive I writes-IND "I writes" b. io rimprovera I scolds-lND "I scolds"  scrive writes-IND "writes" but see also:

target: scrivi write-2nd.PERS-SG-IMP "do write!"

(2) a. io mai pulire I never clean-lNF "I never to clean" b. tu dire tuo marito you-SG tell-INF your-M-SG husband "you do tell your husband" After a coindexed modal verb, an inflected form of the verb is sometimes pre­ sent where an infinitive is expected: (3) a. voglio want-1St.PERS-SG-IND-PRES

"I want comes"

viene COmes-IND

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

59

b. voleva andato wanted-3rd.PERS-SG-IND-IMPF gone-M-SG. "she/he wanted gone"  puoi tu viene can-2nd.PERS-SG-IND-PRES you-SG comes-lND "you can comes" The present tense is often overextended to past and future contexts: (4) a ieri sera scrivo subito yesterday evening write-1 st.PERS-SG-IND-PRES immediately "last night I immediately write" b. stamattina arriviamo La Spezia this morning arrive-lst.PERS-PL-IND-PRES La Spezia "this morning we arrive La Spezia" The past participle without auxiliary is used to indicate an action achieved in the past, and the aspectual value is sometimes reinforced by the use of the ad­ verb già "already": (5) a. io andato Lucca circolo I gone-M-SG Lucca club "I gone Lucca club" b. già morto in incidente already dead-M-SG in accident "he died in accident" The 2nd ps.sg. of the indicative tends to be extended to cover the corresponding forms of the imperative and the former is also used in the negative imperative: (6)

pensi target: pensa think-2nd.PERS-SG-PRES-lND think-2nd.PERS-SG-lMP "(you) think" "do think! " (7) non fa target: non fare not do-2nd.PERS-sING-IMP do-INF "not do ! " "do not do ! " As for nouns and adjectives, a slight tendency to overextension of the mascu­ line singular form is detectable: (8) a. la gonna azzurro the-F-SG skirt-F-SG blue-M-SG "the blue skirt" b. domenico scorso target: Sunday-M-SG last-M-SG "last Sunday"

domenica scorsa Sunday-M-SG last-M-SG

60

AJELLO, MAROTTA, MAZZONI & NICOLAI

c. nuovo luce target: nuova luce new-M-SG light-F-SG new-F-SG light-F-SG "new light" d. più bello festa target: bella festa more nice-M-SG party-F-SG nice-F-SG party-F-SG "nicer party" at the expenses of the feminine singular, and masculine and feminine plurals: (9)

siamo

molto

contentis simo

are-1 st.PERS-SG-IND-PRES

very

happy-SUPERL-M-SG

"we are very happiest" A slight tendency to isomorphism is detactable in the case of the -e ending of nouns-adjectives-past participles, which tends to be used as a mark of plural: (10) a. baci forte kiss-M-PL strong-M-SG "strong kisses" b. altre amici other-F-PL friend-M-PL "other friends" (11) tutti sono stati gentile all-M-PL are-3rd-PERS-PL-PRES-IND be-PAST-PART-M-PL kind-SG "all of them have been kind" Morphological reduction is related to a phenomenon of substitution of bound forms with free forms, as in the case of verbal morphology, where the use of personal pronouns compensates for the lack of personal endings (see the al­ ready quoted examples 2a, 2b, etc). In the case of nominal morphology, the free morphemes indicating definiteness, such as the articles, are very often replaced by demonstratives: (12) a. quello Beautuful that-M-SG (soap-opera) Beautiful "that Beautiful" b. quello registrator e that-M-SG tape.recorder-M-SG "the tape recorder" c. quella Giovanna I. that-F-SG Joan I. "that Giovanna I."

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

61

4.2 Omission or misuse of functional elements Functional elements (determinants, prepositions, auxiliaries, copula, clitics) tend to be omitted or misused. The correct usage of prepositions covers approximately 50% of the total amount of contexts, their incorrect usage summing up cases of omission and substitution of the correct preposition needed, and cases of incorrect insertion of a preposition. A more detailed examination of their occurrences reveals that the average percentage just given is illusive, since, if linguistic routines involving prepositions are neglected, the percentage of correct usage of all prepositions greatly decreases. Only one preposition turns out to be extensively used in a correct manner: the preposition per ("for"), which also undergoes a process of overextension at the expenses of a ("to") and di ("of "). This result agrees with what we already found in the analysis of speech production. The reason for this high percentage of correct usage cannot lie in the frequency of that preposition in the input language, because, for instance, the preposition per does not rank as the most frequent one, being preceded by di, a, and in. More probably, its correct usage, which also emerges from the analysis of the speech data, might be due partly to its visual salience, since it begins with a bilabial articulation, and partly to the interference from LIS, where there is a specific sign corresponding to Italian per, while all other Italian prepositions have no corresponding sign. The analysis of the definite article gives approximately the same results as the analysis of the speech data: more than 50% of omissions, but a higher percentage for the masculine forms. As for auxiliaries, they are frequently omitted, as we have seen, but, when used, they often show substitution of essere with avere. (13) a. hanno sconvolto délia morte have-3rd.PERS-PL-PRES-IND upset-M-SG-PAST.PART of.the-F-SG death-F-SG "(they) have upset of death" b. ci ha [target: è] piaciuto to.us-OBL. has-IND [ is] like-M-SG-PAST.PART "we liked it" There are no cases of auxiliary avere "to have" replaced by essere "to be". The only case of a substitution of this kind concerns essere as a main verb, not as an auxiliary: (14)

io non sono [target: ho] problema I not am [ have] problem "I am no problem"

62

AJELLO, MAROTTA, MAZZONI & NICOLAI

There are cases of existential generalized 3rd ps.sg. c'è with the function of possessive avere "to have": (15)

lui c'è un he/him there is a-M-SG "he there is a friend"

amico friend-M-SG

This probably represents interference from LIS. The form c'è is overextended at the expense of other forms such as c'era ("there was"), ci sono ("there are") and so on. Clitics are also often omitted; the use of the form si "oneself, him/herself, one/impersonal") is overextended at the expense of other forms of clitics: (16) a. si ricordo himself-3rd-PERS-SG-CL. remind/remember-1 st-PERS-PRES-IND "I remind himself' b. non candidarsi not candidate-lNF-REFL/3rd-PERS-SG-CL. "not to candidate one-/him-self ' Our data match the results by Taeschner et al. (1988), Chesi (2000) and Fabbretti (2000): enclitic forms are omitted more frequently than proclitic ones. The existential particle ci and the relative pronoun che are also fre­ quently omitted, while che as a complementizer is overextended: (17) a. è moho contento che venire is very happy-M-SG that-COMPL come-INF "(she/he) is very happy that to come" b. voglio sapere che want-1 st-PERS-SG-PRES-IND know-INF that-COMPL. tuo marito è contento your-M-SG husband-M-SG is happy-M-SG "want to know that your husband is happy" Sometimes it is also inserted without justification at the very beginning of a sentence: (18)

Che ieri sono andata a Comune that-COMPL yesterday am-IND gone-F-SG to Town hall "That yesterday I went to Town hall"

Two different determinants are sometimes piled up: (19) a. questo il fax this-M-SG the-M-SG fax "this the fax"

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF b. fra qualche settimana between some week "in a few weeks next"

63

prossima next

The verb fare "to do, to make" used to form causatives, is often omitted: saparti [target: farti sapere "to let you know"]; visitare [farmi visitare "to have myself examined"]; vedere [far vedere "to show"]. The most striking omission of morphemes is that of the main verb, which must be recovered from the co-text or context: (20) a. mattino casa tua, venerdi mercato morning house-F-SG your-F-SG Friday marke-M-SG "in the morning at your home, on Friday market" b. amici Mara dopo pranzo friend-M-SG Mara after lunch "Mara('s) friends after lunch" as happens in the case of the omission of the subject. (21)

è tutto bene il nostro viaggio; is-IND all-M-SG fme-ADV the-M-SG our-M-SG trip-M-SG anche è arrivato esatto also is-IND arrive-PAST.PART-M-SG exact-M-SG "our trip was all fine, even (the train) was on time"

What prevails in these cases is a pragmatic communicative strategy. As a consequence of the defective morphological competence, agreement be­ tween article and noun, possessive/demonstrative adjective and noun, adjective and noun, subject and verb is sometimes incorrect. (22) a. la tua notizie the-F-SG your-F-SG. news-F-PL "your news" b. parole bianco word-F-PL white-M-SG "white words" Sometimes the agreement seems to be oriented to the analogy of endings. (23)

i mani the-M-PL hands-F-PL "the hands"

64

AJELLO, MAROTTA, MAZZONI & NICOLAI

4.3 Lexicalization of grammatical categories This is apparent in at least two cases: the temporal adverb is sometimes the only mark of temporality: (24) a. dopo sto cucendo [future tense] after/afterwards stay-1st-PERS-SG-PRES-IND sew-GER "afterwards I am sewing" b. ieri sera scrivo subito [past tense] yesterday evening write- lst-PERS-SG-PRES-IND immediately "last night (I) immediately write" and the numeral is sometimes the only indication of plurality: (25)

due idraulico two plumber-M-SG "two plumber"

4.4 Weak consciousness of some derivational mechanisms A partial sense of derivational morphology is present in the corpus, as is shown by neoformations such as rabbio [mi arrabbio "I get angry"], accompagnia [compagnia "company"], non abbanzano [non sono abbastanza/non bastano "they are not enough"], intermometro [termometro "thermometer"], which all reveal some sort of consciousness of some derivational mechanisms. 5. Lexicon and interference from LIS Besides the four phenomena illustrated above, other characteristics are worth noticing, especially a fairly good level of lexical competence. Strikingly enough, the lexical competence of the deaf appears to be superior to their mor­ phological competence. The lexicon is not severely limited and is also properly used, with very few exceptions: le tue confidenza [i tuoi pensieri "your thoughts"]; vi auguro come bravi docenti [mi congratulo con voi perché siete dei bravi docenti "I congratulate you, since you are good teachers"]; ti spiego il successo [ti spiego l'accaduto "I explain to you what happened"]; rimango a mio marito [penso sempre a mio marito "I always think of my husband"]. The lexicon is not severely limited even if we include here the phenomenon of grammatical category change: [ho fotocopie = ho fotocopiato "I have photo­ copied"]; devo coraggio la mia vita [devo farmi coraggio "I must take courage"]; io sempre pensiero a te [io pens sempre a te "I always think of you"]; affari [sbrigare faccende "to attend to business"]; avanti [andare avanti "to go ahead"]; accordo [accordarsi "to reach an agreement"]. The lexicon is sometimes affected by interference from LIS, and, very interestingly, this hap­ pens in two types of circumstances. It occurs in cases in which LIS has just one

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

65

sign corresponding to several Italian words; thus, for example, the expression il treno è esatto [in orario "the train is on time"] derives from the partial cor­ respondence of a LIS sign to several Italian adjectives (puntúale, preciso, etc.). Alternatively, it occurs in cases where Italian is affected by the lexical choice of LIS; for instance, since LIS can choose between QUESTA MATTINA and ADESSO MATTINA, the expression adesso mattina in the written corpus derives from LIS (By convention, expressions written with capital letters represent LIS signs). In­ correct lexical elements or omissions of expected lexical items are sometimes due to interference in cases in which a certain specification in LIS is expressed not by a manual sign, but by a qualitative modification of the movement of the sign, a modification of the facial expression visto tempo [visto da tanto tempo "seen since a long time"] or a modification of the direction of the movement, which may explain the frequent confusion between ricevere/spedire (fax) "to receive/to send (fax)"; arrivare/venire "to arrive/to come". The inevitable loss or the change of information is then due to a linear rendering, on behalf of the deaf, of the multi-layered structure of signs, which does not take into account the holistic multi-componential nature of the sign itself. The interference from LIS at the morphological level seems to be very limited: apart from the case of the preposition per, our data show: 1.

2.

only one case of plural expressed through reduplication of the noun, as is common in LIS (cuori cuori "hearts"), although both occurrences of the noun are marked with the plural ending. The strategy of redu­ plication in the case of plurals may not be due to interference from LIS, but to its iconic expressiveness, which makes it frequently used in all cases of linguistic simplification (acquisition, pidgins, and interlanguages); a few cases of confusion between possessives (tuo/suo "your/his" or "her"); tua/vostra "your", sg. and pl.), which in LIS are expressed with deictic gestures pointing at different but close sectors of the space in front of the signer. The incorrect renderings in Italian may be due to the difficulty in identifying the portion of space relevant to the distinction or to the difficulty in rendering the directionality.

Interference from LIS seems to be on the whole very heavy in the domain of syntax, in particular in the order of linguistic elements, which very frequently follows the LIS ordering, as in interrogative and negative sentences; for instance: (26) a. porta scuola chi takes-IND school who "who takes (him /herto) school?"

66

AJELLO, MAROTTA, MAZZONI & NICOLAI

b. rispos ta non answer-F-SG not "(there is) not answer" The same happens in the case of focalization processes concerning the specifying items: (27) a. panchine dove vicino casa tua bench-F-PL where near house-F-SG your-F-SG "benches where near your place" b. ci vediamo che giorno us/ourserlves-REC see-1 st-PERS-PL-PRES-IND which/what day "What day are we meeting?" 9 gennaio 9 January "(On) January 9th" To speak of LIS interference does not imply that the status of L1 is con­ ferred on LIS: the visuo-gestural communication system is simply the more natural and the more frequently used.3 6.

Orthography As evidenced in the literature, orthography does not represent remarkable difficulties for the deaf, probably because literacy has been traditionally a do­ main of very intensive training (see Ajello et al 2000). The most frequent er­ rors made by our subjects concern: 1.

2.

3.

3

confusion between letters corresponding to voiced and voiceless con­ sonants: rimanco [rimango "I stay"]; statio [stadio "stadium"]; pangia [pancia "belly"]. frequent degemination or misplaced gemination: abracci [abbracci "hugs"]; Ricardo [Riccardo "Richard"]; Coseta [Cosetta]; maremana [maremmana "native or inhabitant of Maremma-F-SG"]; avvisso [avviso "warning, notice"]; occassione [occasione "occasion"]; costanno [costano "they cost"]. substitution of graphemes: golfio [gonfio "swallen"]; pasiensa (pazienza "patience; (it is a) pity"].

We intend to check in detail, in a future work, the hypothesis of syntactic interference from LIS upon the written production of the Italian deaf, by analyzing similar spontaneous texts, very close to speech, produced by a non-signer deaf and addressed to non-signers, either deaf or hearing.

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

4. 5.

67

iteration of contiguous vowel grapheme: volontieri [volentieri "gladly"]; Monaca [Monica]; sapare [sapere "to know"]. lack of graphic indication of stress: gia [già "already"]; e [è "is"]; lunedi [lunedi "Monday"];perche [perché "why, because"].

On the other hand graphemic clusters are, as a rule, correctly written with only one exception: accompagniare [accompagnare "to accompany"]. Another point of interest is a non-standard use of punctuation. In particular, the interrogation mark is generally overextended (stiamo bene, spero pure di voi? "We are fine, I hope so are you?"; volevo sapere come è andata a Roma? "I wanted to know how did it go in Rome?"), but is sometimes missing when it should be appropriate {porta scuola chi "takes (to) school who?"; panchine dove vicino casa "benches where near house?"). Full stop is often omitted and semicolon is totally absent, while very few commas are used. 7. A test of morphosyntax In order to check the grammatical competence of the deaf, a test was given to a group of the subjects who wrote the faxes examined so far and who pro­ duced the speech analyzed in §§ 3 and 4. We wrote 79 sentences, 12 correct, and 67 incorrect, i.e. containing a grammatical error.4 The typology of the er­ rors includes: agreement within the NP, agreement within the VP, omission of a preposition, selection of a wrong preposition, omission of the article, selection of a wrong allomorph for the article, omission of the auxiliary. Eleven subjects participated in the test. Ten of them are profoundly deaf and one is severely deaf. None of them has ever been prothesized. Their age ranges from 50 to 65 years. Six are females and five are males. All of them have good LIS competence. Their education in special institutes covers the elementary cycle of the school. The test was applied in an informal atmosphere, and it was presented by one of authors, who is a signer (Laura Mazzoni). Each sentence was written on a single piece of paper. Each subject received an envelope containing all the sentences given in a random order. The signer gave the following instructions to the subject: "This envelope contains some pieces of paper, where an Italian sentence is written. The sentences may have errors. Read every sentence and decide whether it is right or wrong. If the sentence is wrong, i.e., it has an error, write the right form on the same paper. Read all the sentences and when you finish, put all the pieces of paper in the envelope. There are no time limits to do the task." An every-day lexicon and a very simple 4

Here and henceforth, the term 'error' is used in a rather general meaning, basically as synonymous of non-standard form.

68

AJELLO, MAROTTA, MAZZONI & NICOLAI

syntax were used in the construction of the sentences, in order to guarantee a relatively easy task. The subjects' comprehension was checked by asking them to reproduce each sentence in LIS after reading it. The relevant parameters referrable to the subjects are presented in Table 4. The LIS competence and the speech intelligibility have been evaluated by the signer with reference to the linguistic abilities shown by the subjects in communicative interactions. Age of education refers to the age when the subject began the education in a special institute for deaf. Time refers to the number of minutes needed to do the task. Subject GI GL GM PB RS SP AS LM BN RM

1 VP

Sex F F F F F F M M M M M

Age of education 6 6 6 7 9 7 7 11 7 6 6

LIS competence 7 6 7 7 7 7 7 7 6 8 8

Intelligibility 4 5 4 4 4 4 2 5 3 3 4

Time 43' 20' 36' 35' 38' 31' 36' 30' 28' 22' 15'

Table 4: List of the subjects who participated in the test

Many subjects observed that the test was too long. Some of them did not understand the meaning of 'error' immediately, and asked for some examples. Almost all the subjects had great difficulties in maintaining their concentration for all the time needed. The attention decreased after roughly ten minutes. A pause was then introduced during the test. No subject knew the goal of the ex­ periment. Three subjects (GL-F, AS-M, RM-M) said that there were no errors in the sentences. These subjects were excluded from the analysis, since they were not reliable. The results are summarized in Tables 5, 6 and 7.5

The abbreviations used in the following tables are:

M: male, -Prep: Sub st Prep Prep: AgrNP:

F: female omission of a preposition substitution of a preposition total of errors referring to prepositions agreement within the Noun Phrase

AgrVP: -Aux: -Det: AlloDet: Det:

agreement within the Verb Phrase omission of an auxiliary form omission of a determinant wrong selection of the allomorph of the determinant total of errors referring to determinants

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

69

Tables 5 show the percentage values concerning the sentences containing an error, with reference to the total number of errors for each grammatical category we considered. It is divided into three main sections, the first one giving the number of uncorrected errors, the second one giving the wrong corrections, and the third one showing the right corrections made by the subjects. UNCORRECTED ERRORS

Subj

Prep

-Prep

Subst Prep Agr NP Agr VP

-Aux

-Det

Alio Dei

Det 19.6

BN-M

32.6

13.0

19.6

37.0

8.7

2.2

10.9

8.7

LM-M

33.9

16.1

17.9

32.1

10.7

3.6

14.3

5.4

19.6

VP-M

36.0

4.0

32.0

28.0

4.0

0.0

24.0

8.0

32.0

GI-F

100.0

100.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

GM-F

29.4

5.9

23.5

29.4

5.9

0.0

17.6

17.6

35.3

PB-F

35.6

18.6

16.9

35.6

8.5

1.7

11.9

6.8

18.6

RS-F

48.1

18.5

29.6

29.6

3.7

0.0

14.8

3.7

18.5

SP-F

41.7

25.0

16.7

29.2

4.2

0.0

16.7

8.3

25.0

Subj

Prep

-Prep

Allo Det

Det

WRONG CORRECTIONS

Subst Prep Agr NP

Agr VP

-Aux

-Det

BN-M

57.1

42.9

14.3

42.9

0.0

0.0

0.0

0.0

0.0

LM-M

66.7

50.0

16.7

16.7

0.0

0.0

0.0

16.7

16.7

VP-M

25.0

25.0

0.0

75.0

0.0

0.0

0.0

0.0

0.0

GI-F

33.3

0.0

33.3

66.7

0.0

0.0

0.0

0.0

0.0

GM-F

30.0

20.0

10.0

60.0

0.0

10.0

0.0

0.0

0.0

PB-F

0.0

0.0

0.0

100.0

0.0

0.0

0.0

0.0

0.0

RS-F

50.0

33.3

16.7

50.0

0.0

0.0

0.0

0.0

0.0

SP-F

27.8

11.1

16.7

44.4

11.1

0.0

11.1

5.6

16.7

Subj

Prep

-Prep

Subst Prep

BN-M

30.8

23.1

1.1

23.1

RIGHT CORRECTIONS

Agr NP Agr VP

-Aux

-Det

Alio Det

Det

15.4

1.1

23.1

0.0

23.1

LM-M

0.0

0.0

0.0

100.0

0.0

0.0

0.0

0.0

0.0

VP-M

35.1

27.0

8.1

35.1

13.5

5.4

5.4

5.4

10.8 19.4

GI-F

33.9

17.7

16.1

33.9

9.7

3.2

12.9

6.5

GM-F

38.5

23.1

15.4

30.8

12.8

2.6

12.8

2.6

15.4

PB-F

40.0

20.0

20.0

0.0

20.0

20.0

20.0

0.0

20.0

RS-F

21.2

15.2

6.1

36.4

15.2

6.1

12.1

9.1

21.2

SP-F

33.3

16.7

16.7

33.3

12.5

8.3

8.3

4.2

12.5

Table 5: Percentage values of evaluation of sentences containing an input error

70

AJELLO, MAROTTA, MAZZONI & NICOLAI

In Table 6 we show the results concerning the correct sentences in absolute terms, since here percentage values would be meaningless. The number of corrections wrongly made by the subjects are in the first section, the number of correct sentences judged as correct are in the second section. WRONG CORRECTIONS - Prep Subst Prep

Prep

NO CORRECTIONS

Agr SN Agr SV - D e t Allo Det

- Aux

BN-M

0

0

0

0

0

0

0

0

BN-M

LM-M

0

1

1

2

1

0

0

0

LM-M

12

VP-M

0

1

1

0

0

0

0

0

|VP-M

GI-F

0

0

0

0

0

0

0

0

|GI-F

GM-F

0

0

0

0

0

1

0

0

GM-F

10

PB-F

0

1

1

1

0

0

0

0

PB-F

10

RS-F

0

1

1

0

0

1

0

0

RS-F

10

SP-F

0

0

0

2

0

2

0



|SP-F

8

9 12 8

Table 6: Numerical results for correct sentences (no errors in input)

The subjects showed variable results. With reference to their grammatical competence, as judged on the grounds of the results obtained, our subjects can be grouped into two classes: a) b)

deaf with good grammatical competence (VP, GI, GM, RS); deaf with poor grammatical competence (BN, LM, PB, SP).

The subjects belonging to the first class show the highest percentages of right corrections (on average > 50%), and the lowest percentages of errors (wrong corrections and no correction). The results we collected seem to idicate that there is no strong correlation between LIS competence and grammatical competence: although all our subjects have good knowledge of LIS, with a small difference among them (cf. Table 4), their performance in doing the test was different. If we consider the wrong sentences in input (cf. Table 5), the number of wrong corrections is higher than the number of missed corrections. This is true for both the identified groups of deaf subjects, although the difference is obviously much more evident in the case of group b), that is, in deaf subjects with poor grammatical competence, since the amount of errors they make is greater. Apart from interindividual differences, in all our subjects there is a general trend towards the recognition that a sentence is wrong, but, at the same time, without the ability to select the right form. On the other hand, the number of wrong corrections on right sentences given in input is very low for both groups of deaf subjects (cf. Table 6), which suggests that they are able to judge a right sentence as grammatical. In Table

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

71

7, the percentage values of the distribution of errors in relation to the three different possible behaviours of the subjects are given. Our aim is to check whether there is a concentration of errors (wrong corrections as well as no corrections) on some categories more than on others. UNCORRECTED ERRORS

Prep

-Prep Subst Prep Agr SN Agr SV

- Aux

-Det

BN-M 32.6

13.0

19.6

37.0

8.7

2.2

10.9

8.7

LM-M 33.9

16.1

17.9

32.1

10.7

3.6

14.3

5.4

19.6

VP-M 36.0

4.0

32.0

28.0

4.0

0.0

24.0

8.0

32.0

100.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

GI-F

100.0

Allo Det Det 19.6

GM-F 29.4

5.9

23.5

29.4

5.9

0.0

17.6

17.6

35.3

PB-F

35.6

18.6

16.9

35.6

8.5

1.7

11.9

6.8

18.6

RS-F

48.1

18.5

29.6

29.6

3.7

0.0

14.8

3.7

18.5

SP-F

41.7

25.0

16.7

29.2

4.2

0.0

16.7

8.3

25.0

Prep

-Prep Subst Prep Agr SN Agr SV

WRONG CORRECTIONS

- Aux

-Det

Allo Det - Aux

BN-M 57.1

42.9

14.3

42.9

0.0

0.0

0.0

0.0

0.0

LM-M 66.7

50.0

16.7

16.7

0.0

0.0

0.0

16.7

16.7

VP-M 25.0

25.0

0.0

75.0

0.0

0.0

0.0

0.0

0.0

0.0

33.3

66.7

0.0

0.0

0.0

0.0

0.0

20.0

10.0

60.0

0.0

10.0

0.0

0.0

0.0

0.0

0.0

0.0

100.0

0.0

0.0

0.0

0.0

0.0

RS-F

50.0

33.3

16.7

50.0

0.0

0.0

0.0

0.0

0.0

SP-F

27.8

11.1

16.7

44.4

11.1

0.0

11.1

5.6

16.7

Prep

-Prep Subst Prep Agr SN Agr SV

GI-F

33.3

GM-F 30.0 PB-F

RIGHT CORRECTIONS

- Aux

-Det

Allo Det - Aux |

BN-M 30.8

23.1

7.7

23.1

15.4

7.7

23.1

0.0

23.1

LM-M 0.0

0.0

0.0

100.0

0.0

0.0

0.0

0.0

0.0

VP-M 35.1

27.0

8.1

35.1

13.5

5.4

5.4

5.4

10.8

GI-F

33.9

17.7

16.1

33.9

9.7

3.2

12.9

6.5

19.4

GM-F 38.5

23.1

15.4

30.8

12.8

2.6

12.8

2.6

15.4

PB-F

40.0

20.0

20.0

0.0

20.0

20.0

20.0

0.0

20.0

RS-F

21.2

15.2

6.1

36.4

15.2

6.1

12.1

9.1

21.2

SP-F

33.3

16.7

16.7

33.3

12.5

8.3

8.3

4.2

12.5

Table 7: Percentage values of distribution of errors in relation to the three different possible types of behavior of the subjects

72

AJELLO, MAROTTA, MAZZONI & NICOLAI

As a matter of fact, our subjects correct the errors pertaining agreement of VP, as well as omission of Auxiliary or Determinant more than the ones con­ cerning prepositions and agreement within the Noun Phrase. In particular, the subjects with better morphological competence show very clearly the dif­ ference between the two classes of errors. At the same time, the wrong corrections again concentrated on the selection of the prepositions and of the agreement marker within the NP. What is surprising is that even in the case of right corrections, the percentage was still higher in the case of prepositions and agreement of NP. The datum could be constrained, at least in part, by the higher number of sentences containing an error relative to prepositions and agreement within NP. However, we still believe that the selection of the preposition represents the most difficult task for a deaf person. 8.

Conclusions As is apparent from the above-mentioned phenomena, the deaf we exam­ ined made exactly the same errors in Italian as all other learners of Italian as L2 at the initial stages of their interlanguages. Unlike the latter, the deaf sub­ jects showed a discrepancy between a poor competence at the morphological level and a fairly good competence at the lexical level, thus revealing that morphology represents their weak point. Additionally, the deaf subjects made more errors in free rather than in bound morphology, which nonetheless is not sufficiently mastered. The condition of this homogeneous group of deaf people is a very peculiar one, very different from the one that characterizes hearing, spontaneous learners of Italian as L2 for these two reasons: (a) they have no mother tongue acquired from birth, through which they may have access to the principles of UG, since LIS has been acquired naturally but in a late period; (b) the input is always problematic for the deaf, because, even after hard formal training, speech-reading cannot give all the necessary infor­ mation about the target phonological elements. Moreover, the lin­ guistic input for the deaf depends on an interpretation hearers give about their presumed ability to decode the message. The linguistic in­ put is then 'tailored' and in fact reduced, simplified (along lines that have never been investigated so far), so as to fit' their competence, as it is conjectured by hearers: it is an input basically reduced to its lexical content and structured so as to follow semantic and pragmatic strategies.

MORPHOSYNCTATIC FRAGILITY IN ITALIAN DEAF

73

These data could match an interpretation according to which, in the case of deaf people with neurolinguistic parameters of the kind illustrated above, the vocal language does not have a complete access to UG, not even an indirect one, mediated by the characteristics already implied in LI, but is learnt mainly through an explicit and formal approach, which gives as a result a system of rules which is never completely internalized. Our data seem to fit the maturational hypothesis rather than the continuity one, since the grammar of the deaf would be based on semantic and thematic relationships, rather than on merely syntactic ones. The deaf s very peculiar way of learning the language explains also the interindividual variability, which includes cases of very good linguistic performance. The process of learning is based mainly on general, not specifically lin­ guistic, cognitive mechanisms, as is apparent from the discrepancy between a fairly good lexical competence and a poor morphological competence heavily dependent on the input, and a similarly poor syntactic competence which relies fundamentally on pragmatic communication principles. The data we reported suggest the necessity of early didactic approaches with specific targets, so that the new deaf generations may better cope with the difficulties of integration.

REFERENCES Ajello, Roberto, Giovanna Marotta & Florida Nicolai. 1998-1999. "Uno studio sperimentale della produzione vocale di sordi italiani". Quaderni della Sezione di Glottologia e Linguistica,Universitàdegli Studi di Chieti 10/11.17-40. Ajello, Roberto, Antonella Convalle & Florida Nicolai. 2000. "Strategie di apprendimento della scrittura in bambini sordi". Atti del II Convegno Nazionale sulla Lingua Italiana dei Segni, Genova 25-27 setiembre 1998: Viaggio nella città invisibile ed. by Caterina Bagnara, Giampaolo Chiappini, Maria Pia Conte & Michela Ott, 158-180. Tirrenia: Edizioni Del Cerro. Antelmi, Donatella. 1997. La prima grammatica dell'italiano: Indagine longitudi­ nale sull'acquisizione della morfosintassi italiana. Bologna:I1Mulino. Banfi, Emanuele. 1993. L 'altra Europa linguistica.Firenze: La Nuova Italia. Berretta, Monica. 1987. "Per uno studio suH'apprendimento dell'italiano in contesto naturale: il caso dei pronomi personali atoni". L 'apprendimento spontaneo di una seconda lingua ed. by Anna Giacalone Ramat, 329-352. Bologna:I1Mulino. Berretta, Monica. 1990. "Morfologia in italiano lingua seconda". Storie dell'italiano e forme dell'italianizzazione ed. by Emanuele Banfi & Patrizia Cordin, 181-201. Roma: Bulzoni. Caselli, Maria Cristina, Simonetta Maragna, Laura Pagliari Rampelli & Virginia Volterra. 1994. Linguaggio e sordità. Firenze: La Nuova Italia.

74

AJELLO, MAROTTA, MAZZONI & NICOLAI

Chesi, Cristiano. 2000. Inferenze strutturali. Analisi sull'uso degli elementi funzionali neue produzioni verbali dei bambini sordi. Dissertation, University of Siena. De Mauro, Tullio, Federico Mancini, Massimo Vedovelli & Miriam Voghera. 1993. Lessico difrequenza dell'italianoparlato. Milano: Etas. Fabbretti, Daniela. 1997. Scrivere e segnare: la costruzione del discorso nell 'italiano scritto e nella Lingua Italiana dei Segni delle persone sorde. Ph.D. dissertation, University of Rome, Faculty of Psychology. Fabbretti, Daniela. 2000. "L'italiano scritto dai sordi: un'indagine sulle abilità di scrittura dei sordi adulti segnanti nativi". Rassegna di Psicología 17.1-21. Fabbretti, Daniela, Virginia Volterra & Clotilde Pontecorvo. 1998. "Written Language Abilities in Deaf Italians". Journal of Deaf Studies and Deaf Education 3.231- 244. Hanson, Vicki & David Wilkenfeld. 1985. "Morphophonology and lexical organization in deaf readers". Language and speech 28.269-280. Kelly, Leonard P. 1993. "Recall of English function words and inflections by skilled and average deaf readers". American Annals of the Deaf 138.288-296. Maxwell, Madeline M. & Tracey Gordon Falick. 1992. "Cohesion & quality in deaf & hearing children's written English". Sign Language Studies 77.345-372. Paul, Peter V. & Stephen P. Quigley. 1994. Language and Deafness. San Diego: Sin­ gular Publishing Group. Quigley, Stephen P., D. Power & M. Steinkamp. 1977. "The language structure of deaf children". Volta Review 79.73-83. Radelli, Bruna. 1998. Nicola vuole le virgole. Dialoghi con i sordi. Padova: DecibelZanichelli. Savy, Renata. 2000. "Riduzioni foniche nella morfologia del sintagma nominale nel parlato spontaneo. Indagine quantitativa e aspetti strutturali". Fonología e Morfo­ logia dell'italiano e dei dialetti d'ltalia ed. by Paola Benincà, Alberto Mioni & Laura Vanelli, 201-221. Roma: Bulzoni. Simone, Raffaele. 1988. "Fragilità delia morfologia e contesti turbati". L'italiano tra le altre lingue: strategie di acquisizione ed. by Anna Giacalone Ramat. Bologna: I1 Mulino. Spears, Arthur K. & Donald Winford. 1997. The Structure and Status of Pidgins and Creoles. Amsterdam & Philadelphia: John Benjamins. Taeschner, Traute, Antonella Devescovi & Virginia Volterra. 1988. "Affixes and function words in the written language of deaf children". Applied Psycholinguistics 9.385-401. Volterra, Virginia & Elisabeth Bates. 1989. "Selective impairment of Italian gram­ matical morphology in the congenitally deaf: A case study". Cognitive Neuropsy­ chology 6. 273-308.

THE EXPLAN THEORY OF FLUENCY CONTROL APPLIED TO THE DIAGNOSIS OF STUTTERING* PETER HOWELL & JAMES AU-YEUNG University College London 1. Introduction In the following two chapters, we present an overview of the EXPLAN theory. The main focus of EXPLAN is on fluent speech control, but it is also relevant to the diagnosis and treatment of stuttering.1 Fluency failures can af­ fect normally fluent as well as stuttered speech: it occurs where speech control falters even though the speaker does not produce an overt error. An extract in­ cluding several types of fluency failures (detailed later) is "I got on, on the seven ... fffifty ... three train t.to Mac.clesfield". According to EXPLAN, all fluency failures arise from the same generic problem. This problem leads to specific types of fluency failures that are shown by all children. The difference between fluency failures in child and adult stuttering is due to a change in the way the underlying problem is dealt with by people who stutter. The EXPLAN theory offers an account of how differences between fluent and struttered speech arise. This change leads to fluency failures with different characteristic surface (speech output) forms. As all speakers experience the same generic problem, they all could, in principle, tackle the problem in the same way as adults who stutter. According to EXPLAN, the distinguishing characteristics of stuttering emerge in late childhood; a contrasting view is that people who stutter have incipiently different problems in controlling speech from the onset of language. Viewed from the EXPLAN perspective, then, diagnosis of the disorder involves specifying and measuring the types of fluency failures that occur in late adolescence to see if they have changed in a way commensurate with adult stuttering, EXPLAN supports the view that young This work was supported by the Wellcome Trust. Thanks to Scott Yaruss for reading a preliminary version of this chapter. 1 On fluent speech control, see Howell, Rosen, Hannigan & Rustin 2000; Howell & Sackin, submitted. On diagnosis of stuttering, see Au-Yeung, Howell & Pilgrim 1998; Howell, AuYeung, Davis, Charles, Sackin, Williams, Cook, Rustin & Reed, in press; Howell, Au-Yeung & Sackin 1999. On treatment of stuttering, see Howell, Au-Yeung & Vallejo-Gomez, in press; Howell & Sackin, in press.

76

PETER HOWELL & JAMES AU-YEUNG

speakers whom clinicians suspect might develop the disorder need to be monitored. However, the theory suggests that early intervention may not be advisable in such suspected cases (not as long as the child is exhibiting the types of fluency failure common to all children). We will begin by examining how fluency control has been explained in a representative psycholinguistic model (Levelt 1989) that contrasts in many respects with EXPLAN. Levelt's model was selected for this comparison because it attempts to explain all events between intention and action in language. The points made about Levelt's (1989) model also apply to more recent versions by the same author (e.g., Levelt, Roelofs & Meyer 1999). A final reason for choosing Levelt's (1989) model is that it has also been the basis of a model of adult stuttering (the Covert Repair Hypothesis or CRH [Kolk & Postma 1997]). Several implications consistent with EXPLAN are discussed. The problem of stuttering is not expected to be associated with problems of lexical retrieval but, rather, to be dependent on the complexity of the phonological form of words. 2. A psycholinguistic model of language production A summarizing diagram of Levelt's model is given in Figure 1. The lefthand side is the route for language production that goes from the conceptualizer through the formulator to the articulators. The speech compre­ hension system is on the right.

overt speech

Fig. 1: A sketch of Levelt's (1989) model

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

77

The conceptualizer is responsible for generating a message and monitoring that it is delivered appropriately. Monitoring devices like the one included in the conceptualizer take the output of a process and compare it with what the process intended to produce (the initial input to the conceptualizer). If there is a discrepancy (an error), an adjustment is made to the output to reduce the er­ ror. Feedback is a route by which output is returned to the monitor to be used for comparison with the original intention. Thus the presence of a feedback route implies that a monitoring process is involved. After the message is output from the conceptualizer, it goes through the formulation stage. At the output of the formulation stage, the message is repre­ sented as a phonemic string. In Levelt's model as in many others (Dell 1986; Dell & O'Seaghdha 1992), two sub-stages are identified where the message is represented as a lemma and in phonological forms, respectively. Two different patterns of speech errors occur that have been associated with these two for­ mulation levels: word exchange errors that could arise at the lemma level, and sub-word level speech errors (Dell 1986) that involve phonemic or syllabic transmutations and may arise at the phonemic level. Recently, Caramazza & Miozzo (1997) and Miozzo & Caramazza (1997) have provided evidence on the tip-of-the tongue phenomenon (Brown & McNeill 1966) that they interpret as indicating there is only one stage in formulation. For EXPLAN, it does not matter whether one or two sub-levels are involved in formulation. What is important is that the output of formulation is a phonemic plan generated from left to right. This allows the plan to be available for the first part of the word before the rest of it. Word onset is the point at which most sub-word speech errors occur and is also the point at which speakers sometimes find it difficult to span from word onset to the subsequent part of the word during fluency failures. Articulation is the final step in translating the abstract phonemic represen­ tations to overt speech. In Levelt's model, speech production and perception systems are coupled together on the assumption that audition can recover overt speech from articulatory output and process it through the speech comprehension system (the external loop that will be considered in the following chapter). Information from the speech comprehension system is then transmitted back to the monitor in the conceptualizer. In this chapter, we will only consider internal monitoring as we concentrate exclusively on the production route (the left-hand side of Figure 1, above).

78

PETER HOWELL & JAMES AU-YEUNG

3. Speech repairs The monitor in the conceptualizer is an essential component in Levelt's model. (A similar monitor also appears in Levelt et al.'s more recent 1999 model). Monitors, in general, are rejected by EXPLAN because of the interactions between high and low levels of processing they engender. Such interactions place unrealistic processing demands on language output in terms of the representations that need to be available and the amount of time needed for processing these representations during on-going speech control. We will evaluate Levelt's (1983) work on speech repairs to assess one line of support for the internal monitoring process. Levelt regards structures like "Turn left at the, no, turn right at the crossroads" as instances of self cor­ rection of errors after they have been detected by internal monitoring (termed repairs). When errors are detected by the monitor, speech can be interrupted at any point between conceptualization and articulation and the message can be corrected and restarted. To illustrate with the example used earlier, the speaker gives the wrong direction, realizes this, and replaces the reparandum ("left") with the alteration ("right"). The substitution is not the only thing that happens: The speaker overshoots the reparandum (goes on with the words "at the") and retraces to "turn" when the message recommences (the word said before "left" that is not incorrect). The repair contains an interruption (which can range from a pause to a short phrase; here the word "no"). Most of these parsed components are optional. Levelt considers that speakers can detect and repair errors before they are overtly spoken. These are called covert repairs and an example is "Turn right at the, at the crossroads". Such repairs are problematic to interpret as there is no outward sign of what the error was (i.e., why the speaker hesitated after "turn"), nor even whether an error occurred at all. Though Levelt does not rely on covert repairs in his account of speech control, Kolk & Postma (1997) have described stuttering as a result of covert repairs. Other types of repairs are Different (D) repairs, in which the topic is changed (like non sequirurs), and repairs in which the speech is not pitched at an appropriate level for the addressee (appropriateness repairs). 4. Modelling errors versus fluency failure Basing a model on fluency failures (episodes where speech control falters, but there is no overt error) stands in marked contrast with basing one on error patterns using monitoring processes. It is not denied that errors provide im­ portant information about points of difficulty in a message, but their explana­ tion by a monitoring process is questionable. First of all, errors are infrequent events in language (Shallice & Butterworth 1977). Fluency failures, on the

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

79

other hand, are common. For instance, estimates from the data of Howell, AuYeung & Sackin (1999) show that fluent speakers produce around 2.57% fluency failures in function words and 0.97% in content words. While a case can be made that it is important to model error patterns, it also seems advisable, at the very least, to consider how the most frequent breakdowns in fluency control could arise. Next, we consider whether errors need explanation by an internal moni­ toring process. Monitoring all speech for errors would be inefficient, given that errors are infrequent. However, this does not necessarily mean that monitoring should be dismissed. We next examine more critical points about the operation of a monitor once an error has been detected. The monitor takes corrective action when a difference is detected between intended and actual versions (the actual version assumed to be in error). To establish whether a difference has occurred or not, like needs to be compared with like, so it must be assumed that the monitor has multiple intended representations equivalent to the actual forms produced at all stages during the output of a message. One problem is that it is not apparent how an errorless version of the intended form at lower levels can be available in the conceptualizer for comparison when an erroneous output form is generated as actual output at these lower levels. If a correct version is available for monitoring, why not use it for controlling the respective level? Another look at the repairs, described earlier from Levelt's perspective, reveals that these actually provide little support for an internal errormonitoring process. Evidence that an error has occurred is only obtained when speech output reveals this. However, this selfsame evidence for an error would suggest either that no monitoring takes place or that the monitor has not worked on this occasion as the speech goes right through to output. The only cases Levelt describes where there may be evidence for the operation of an internal monitor is where errors are intercepted and the result is a covert repair. However, in these repairs another interpretation is that no error occurred in the first place (see the later discussion of stalling fluency failures for an alternative explanation of some surface form features considered to represent components of repairs). In the case of D repairs, the message is abandoned rather than repaired, and appropriateness repairs are a matter of style rather than anything else. Abandonment and restart of a message (as in D repairs) may be a general process applying to all repairs rather than arising from internal monitoring for errors and on-line repair to remove any detected discrepancy. In the abandon­ ment conception, the new message would not be an on-going alteration to the

80

PETER HOWELL & JAMES AU-YEUNG

current message, but would be regarded as a new attempt altogether. This view would not preclude there being a large measure of overlap (overshoot and retrace) between the initial and subsequent messages, as is observed to occur. The abandonment view would also explain why no error is evident in several types of repair and no definitive evidence of errors is available at all until they are translated into concrete articulatory output form. 5. Stutterings and repairs Stuttered speech is often described as fluent stretches that are interspersed with dysfluent events that we term fluency failures. Johnsonet al. (1959) listed eight types of events he considered to be often associated with stuttering, and these are given in Table 1. Stalling

Advancing

1. Word repetitions (e.g., "Katy, Katy"). 2. Phrase repetitions (e.g., "if he, if he").

4. Part-word repetitions (e.g., "k..k..Katy"). 5. Prolongations (e.g., "mmother").

3. Interjections (e.g., editing terms like "uh"). This can include filled and abnormally long silent pauses too.

6. "Other" fluency failures such as broken words (e.g., "di.nosaur"). 7. Idea abandonment (like non sequiturs) 8. Phrase revisions as in "my uunc, my mother's brother".

Table 1 : Stuttering events used by Johnston et al. (1959)

It is apparent that the form of these events is closely related to that of repairs and, as with repairs, none of the stuttering events involve errors at the phonemic level or in articulation (for example, none involve using the wrong phone to start a word). We term events 1-3 stalling fluency failures. These involve whole words (the ones said immediately prior to the point of interruption) being reiterated (types 1 and 2), or interruption by silence or filler word(s) (type 3). We call events 4-6 advancing fluency failures which are, respectively, where the first part of a word is repeated (type 4), is elongated (type 5) or has a gap before the remainder of the word is produced (type 6). The overt speech appears to show that only the first part of the utterance was available (types 4-6). These events could be a consequence of the formulated plan not being complete but, nevertheless, the speaker attempts what is available (the first part), EXPLAN, as described later, develops this idea. The remaining two categories in Table 1 are similar to some classes of re­ pairs Levelt describes. Category 7 events are like D repairs, but, as said earlier, the speaker has not repaired the message, but abandoned it and started another one. The phrase revision, category 8, could be described as an error

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

81

repair involving lexical selection, as in the "turn left, no turn right" example discussed above. In cases like that and their stuttering equivalents (type 8), we would agree that a high level error has occurred, but, even so, we would not interpret them as showing that the error has been repaired by an internal monitoring process, as the error has not been intercepted. Type 7 and 8 stuttering events can be removed by the procedure, described in Howell, AuYeung, Sackin, Glenn & Rustin (1997), for selecting only fluency failures that do not involve a phonemic or lower-level error. To summarize, EXPLAN regards the surface form of stuttering events (rather than putative errors that may lie behind them, according to Kolk and Postma's CRH) as reflecting two alternative forms of fluency failure (stalling versus advancing) in the vicinity of a word. The EXPLAN view that all fluency failures arise at points where planning cannot keep up with output speed is now considered. 6. The EXPLAN theory of stuttering According to EXPLAN, the fluency failures just described arise because segments take different amounts of time to plan and execute. The linguistic formulator process generates a plan (PLAN), and motor processes execute it (EX). The model assumes that PLAN and EX take place in parallel and that PLAN is independent of EX. This independence precludes feedback loops from EX output to a central monitor that could be used to correct and reinitiate a PLAN. The independence also allows the current word to be executed, while the plan of one to be produced subsequently is being generated. Planning can be put under time pressure when a segment that is difficult and, therefore, time-consuming to generate has to be prepared, and this plan is required quickly as when the planned segment follows a word that is executed rapidly. Though ordinarily the planning process is rapid so that most of the time a plan is delivered ahead of execution (Sternberg, Monsell, Knoll & Wright 1978), in the circumstances just described, the process can fail to deliver the complete plan on time. If this happens, only the left-most part of the plan is ready at the point where execution of the segment should start. Speech can be delayed until the remainder of the plan arrives (stalling) or the speaker can continue and attempt the speech with the partial plan (advancing). Dividing words into function and content types provides a rough basis for characterizing planning time differences of English words. Function words are simpler than content words and require less planning time. Howell, Au-Yeung & Sackin's (1999) transcriptions of samples of spontaneous English speech were analysed to show this (these data include speech from several different age

82

PETER HOWELL & JAMES AU-YEUNG

groups of speakers who stutter). 94% of the function words used were monosyl­ labic and 84% of them carried no primary stress. Among the content words, 62% were monosyllabic and 23% of them carried no primary stress. 88% of the content words started with consonants, whereas only 54% of the function words did. 14%) of the content words started with consonant strings, while only 1% of the function words did. 35% of the content words had a developmentally lateemerging consonant at the start, as defined by Sander (1972), whereas this ap­ plied to only 21% of the function words. The consonant string and later emerging consonant properties co-occurred on the onset of 1% of the function words, but on 12% of the content words. These properties remain roughly con­ stant across normally fluent people and those who stutter, and across ages 2 to 40 years. In sum, then, function words are phonologically simpler than content words, with the onset of content words posing particular difficulties. Parts of wrords that are phonologically difficult do not necessarily present problems in themselves. The difficult word parts are a particular problem when they occur in a message context that itself adds extra pressure on the time at which plans for words must be formulated ready for execution. To demonstrate this, the contexts that cause difficulty when leading up to a word with a phonologically complex onset are described. The segments that can be executed rapidly are those that are simple to plan. There may not be sufficient time during the execution of such simple segments to plan a subsequent word particularly when the later word has a complex focus on it. To be concrete, such demanding contexts are likely to occur at the juncture between a simple function word (executed rapidly) and a content word that starts with a complex phonological structure (planned slowly). In effect, the short execution time for the prior function words puts an extra strain on the planning time of the content word that cannot be met in the short time allowed during the execution of the simple function word. The preceding discussion shows that the function/content word distinction and the context in which these words are combined offer an approximate esti­ mate of regions where fluency may fail in English. Two words of caution about using function and content words are necessary. First, though lexical word type has been discussed, the important property leading to fluency failure is that a segment requiring a long planning time occurs after a segment or segments requiring a short execution time. This could equally well be de­ scribed at the phonological or motorlevels (although the superordinate relation between length of execution and planning times would still need to apply). Second, the situation described for English would not operate in other lan­ guages. In Romance languages, for instance, function words can be long and

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

83

nearly as complex as content words. Comparison across languages like English and, for example, Spanish potentially allows linguistic and motor involvement in this process to be distinguished. Distinguishing planning and execution has implications for control both where speech is fluent and where fluency fails. Figure 2 shows speech that is controlled fluently. Time runs horizontally with planning shown at the top and execution underneath. The message starts when planning the first word com­ mences (word n). At this time there is nothing to execute. Once PLAN(n) is ready, execution of word n can begin. EX(n) and PLAN(n+l) commences. When EX(n) has been completed, PLAN(n+l) is ready and its execution can start, and the whole process repeats cyclically. PLAN(n) has been completed earlier and is just retrieved for execution at this point (planning of subsequent segments is proceeding independently). Note that speech control does not involve an ex­ ternal timekeeper (it involves what Fowler 1980 would call an intrinsic timing process). The situation that can arise in the contexts described at the beginning of this section is shown at the point in time that EX(n+l) has been completed in Figure 2, of timing patterns (horizontal axis) for PLAN (top line) and EX (bottom line) of serially arrayed words (n, n+1 ...) during fluent speech. If comparison is made vertically, the plan for all words (e.g., word n+1) is always complete prior to when the execution of a preceding word is finished (e.g., word n) so that speech can proceed fluently. The PLAN(n+2) is not complete. The speaker cannot embark on the execution of word (n+2) with a complete plan. PLAN(n)

PLAN (n+1)

PLAN (n+2)

EX(n) 1

\\

EX(n+l)

i

Fig. 2: Timing patterns (horizontal axis) for PLAN (top line) and EX (bottom line)

When the plan is not complete, the message cannot progress fluently. In this situation, the speaker can retrieve the plan of a word recently used and execute it again (Blackmer & Mitton 1991). The word usually selected is the previous (function) word, which leads to the often observed function word repetition prior to a content word (the first type of stalling fluency failure given in Table 1). As planning takes place in parallel with execution, the planning of the content word can continue unaffected during this repetition. After the function word (word n) has been repeated, the plan for the content word n+1 may be complete and, if so, execution can proceed.

84

PETER HOWELL & JAMES AU-YEUNG

In Figure 3 the two lines indicate what words are planned and executed (respectively). At the start, planning and execution are operating appropriately on the first word (word n). After word n is first completed, the plan for n+1 is not ready (its planning continues). In this case, the speaker retrieves plan n and executes that again (leading to a word repetition), at which point the plan for word n+1 is ready and can be completed fluently. A pause introduced at the function-content word juncture or a phrase repe­ tition (which are the other two types of stalling fluency failure in Table 1) would serve the same role of gaining time so the plan can be completed. The consequences of repeating a function word that precedes a content word is to slow message rate down (where 'message' corresponds to Levelt et al.'s (1999) usage as the technical term for the conceptual, rather than the actual, output structure that is ultimately going to be formulated). More specifically, function word repetition leads to a message, but not word, rate adjustment at this point. Speech timing has been controlled within the hand-shaking process and so speech continues to be intrinsically timed (Fowler 1980). Stalling types of fluency failures are not problematic, as the speaker successfully avoided having to commence a word on the basis of a plan that was only partly pre­ pared. The next type of fluency failure is problematic insofar as the speaker does not avoid this situation. PLAN(n)

PLAN(n+l)

EX(n)

EX(n)

EX(n+l)

fig. 3: Scnemanc representation oj a word repetition fluency failure

In this second type of fluency failure, the speaker commences the execu­ tion of a problem (usually content) word when only the first part of the plan is available. Since the generation of the remainder of the plan continues while the first part is being executed, the plan may be completed by the time this part is required for execution, so speech will appear fluent. However, the plan may not be completed and will run out of input as it is being executed. The speaker then only has the beginning of the content word available to work with, and fluency characterised by elongation, repetition or interruption of the first part of the word occurs. EXPLAN regards the advancing fluency failure types 4-6 as cases where this happens. A schematic representation of this situation is given in Figure 4, where, at the start, planning and execution are operating appropriately on the first word in the sequence shown (word n).

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

85

After word n is completed on the first occasion after planning, the plan for word n+1 is not ready. Even so, the speaker executes the part plan of word n+1. Fluency failure involving parts of words (due to advancing) is problematic, as it is a characteristic of persistent stuttering (Conture 1982). On a priori grounds, it also seems risky to attempt to produce words in this manner. The next chapter describes a simple, biologically plausible check that suffices to identify points where failure has occurred so that this problem can be avoided. This has the result that it changes timing control from an intrinsic to an extrinsic mode. PLAN(n)

PLAN(n+l)

EX(n)

EX(part(n+l))

Fig. 4: A stretch of fluent speech

The EXPLAN model stresses that motor levels are as important as the lin­ guistic planning levels in leading to fluency failure. Fluency failures are not errors and do not involve a monitoring process. Stalling fluency failures and advancing fluency failures both arise as a consequence of the plan not being available in time. No fluency group is precluded from producing any type of fluency failure. However, adult stuttering is characterised by frequent use of advancing fluency failures rather than stalling ones, so this type becomes ha­ bitual and, once acquired, is very hard to lose. Note that though adult stuttering is not easy to treat, it should be reversible, as people who stutter do not have defective brain or articulatory structures. 7. Some lines of evidence consistent with EXPLAN The contexts in which fluency is likely to fail are when EX(n) is short and PLAN(n+l) is long, particularly when PLAN(n+l) starts with a phonologically complex structure. (Problems in these contexts happen to all speakers, not just people who stutter.) Several implications of this view are addressed in this section. First, the problem of stuttering is not expected to be associated with problems of lexical retrieval but, rather, to be dependent on the complexity of the phonological form of words. There is evidence that children (whether or not they stutter) do not have differential problems in lexical retrieval. The re­ maining lines of support are derived from an examination of the contexts

86

PETER HOWELL & JAMES AU-YEUNG

mentioned above. To locate such contexts, extensive use is made of phonological words (Selkirk 1984; see below). The important property of phonological words is that they group a single content word with its adjacent function words. For English, this serves the role of differentiating (approximately) short simple (function) words from long complex (content) words with difficult initial phonemes. In sub-section two, phonological words allow predictions about fluency failure on function words in different positions relative to their content word. Sub-section three uses phonological words to test the prediction that people who stutter change from stalling to advancing fluency failures in late adolescence. The fourth sub-section looks at properties of content words to establish whether those with complex onsets are more likely to lead to fluency failure. Planning problems are going to be exaggerated when speaking at a rapid rate. Analyses of spontaneous speech are reported in sub-section five that establish that those stretches spoken rapidly are more likely to lead to fluency failure. According to EXPLAN, repetition of function word segments is only one way of stalling. In the sixth and final sub-section, evidence that suggests function word repetition and pausing have equivalent roles is reviewed. 7.1 The role of lexical retrieval in fluency failure In the tip-of-the-tongue () phenomenon (Brown & McNeill 1966), speakers often report being in a feeling of knowing state, but cannot retrieve the actual word for articulation. In laboratory studies, it has been shown that speakers in these TOT states, can recall the first sound of the word and the number of syllables in the word reliably and at greater than chance levels. The consensus is that, in TOT, the concept for the word(s) is available, but the com­ plete phonological form of the word is not (see Figure 1). TOT states experienced during spontaneous speech may reflect the same problem as we have described that lies behind fluency failures (i.e., all the speech plan is not available for execution, only its onset is ready). Thus, the onset information shown to be available in TOT states corresponds with the information people appear to have available in advancing fluency failures. Recent work in our laboratory has shown that TOT states occur with the same frequency in age-matched groups of fluent speakers and speakers who stutter. This suggests that, if the TOT state reflects lexical retrieval problems allied to fluency failures, then lexical retrieval failure is no more frequent in people who stutter than in fluent speakers. The fluency failure problem in peo­ ple who stutter would appear to arise, as EXPLAN predicts, because speakers who stutter go on to translate this partial information into the problematic out­ put surface forms (advancing fluency failures).

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

87

7.2 Phonological word contexts, and stalling and advancing fluency failures Function word position relative to a content word needs to be established objectively to assess the stalling role of function word repetitions. The reason this is necessary is apparent in an utterance like "I look after my mother". Here there are two function words between the content words "look" and "mother". So does one or both of the function words occur before "mother" (and, consequently could it be used to stall) or do they appear after "looK" (when they cannot be used to stall production of "look")? Our research group has developed an analysis procedure based on Selkirk's (1984) phonological words to establish function word position (Au-Yeung & Howell 1998; AuYeung et al. 1998 and Howell, Au-Yeung & Sackin 1999). Phonological words consist of a single content word as its nucleus and any number of function words that serve as prefixes or suffixes to the content word (F n CF m , where n and m can be zero or a positive integer in value). Function words are then associated with content words by sense rules and their position can be determined. The application of this procedure to the preceding example reveals that "after" is part of the phonological word that includes "look" and cannot be used to delay production of the content word, as it occurs in final position in its phonological word, while "my" is part of the phonological word that includes "mother". In this case, as "my" appears prior to "mother", it can be used to delay production of the content word. Au-Yeung et al. (1998) used speech segmented into phonological words to show that the majority of stuttering on function words occurs when the function word appears prior to a content word (e.g., "my" in "my mother") with little stuttering on function words that appear after a content word ("after" in "I look after"). The fact that the likelihood of fluency failure on a function word depends to a marked extent on whether these words appear in initial position in a phonological word (Au-Yeung et al. 1998) is consistent with the view that these words are used to delay production of the content word. 7.3 Function word and content wordfluency failures It was argued above that people who stutter change from producing stalling fluency failures to advancing ones in their teenage years, whereas fluent speakers do not. Phonological words can be used to trace the change from stalling to advancing fluency failures across age groups. Fluency failures on initial function words within phonological words are mainly stalling, whereas advancing fluency failures occur predominantly on content words. Consequently, the relationship between fluency failure rate on function and content words over age groups can be used to establish changes from stalling

88

PETER HOWELL & JAMES AU-YEUNG

(young) to advancing (older) fluency failures in people who stutter (control data were also obtained for fluent speakers in the same age groups). To examine the change from fluency failure on function to content words, phonological words of the form F n CF m , (where the integers n > 0 and m >= 0) were selected so that the relationship between stalling on phonological wordinitial function words and advancing on content words could be investigated. For both fluent speakers and speakers who stutter, very few fluency failures occurred on function words that occupied a position after a content word. For both fluency groups, fluency failure within each phonological word occurred predominantly on either the function word preceding the content word or the content word itself, but not both. Fluent speakers showed, statistically speaking, constant rates of fluency failure on function and content words across age groups, with higher rates on function words. The data for speakers who stutter are shown in Figure 5. 25 T

>17 Age Graup (in year) - Content Word

 Function Word

Fig. 5: Fluency failure rates in groups of stuttering speakers who differ in age

Fluency failure rates are shown separately for function and content words. Function word fluency failures were significantly higher in the two youngest age groups (2-6 and 7-9 years), but were not significantly different in older age groups. This shows the change from function to content word fluency failures (stalling to advancing) as age increases. These data support the view that as speakers who stutter get older, they stop stalling when the plan of a

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

89

word is not complete, and instead advance and attempt a content word on the basis of an incomplete plan. Similar exchange patterns have been found in German (Dworzynski 1999; Rommel 2000). 7.4 The role of phonological difficulty in fluency failure The issue of word onset difficulty was addressed by Howell, Au-Yeung & Sackin (2000) in a study based on earlier work by Throneburg, Yairi & Padden (1994). The latter authors examined how stuttering on words was affected by phonological properties of their constituent consonants. The consonantal factors examined were: a) whether the word contained a developmentally Late Emerging Consonant (LEC, as defined by Sander 1972); b) whether the word contained a Consonant String (cs) and c) whether the word contained more than one syllable (MS). CS and LEC can occur anywhere in a word. So, for instance, a word could have a simple CS in initial position and an LEC in word final position (e.g., quiz). Throneburg et al. (1994) analysed CS and LEC factors irrespective of the word-position they occupied. EXPLAN suggests that CS and LEC should be more of a problem when they cooccur at word onset than when either occurs alone or when they occur at different word positions because when they co-occur they then require a long planning time. If the word containing CS and LEC is preceded by a simple function word, there may not be sufficient time to complete the plan before it is needed for execution. Howell, Au-Yeung & Sackin (2000) found that when both CS and LEC occurred at the beginning of a content word, stuttering rate on the content word was particularly high compared with that on words that did not contain these factors or where the factors were spread throughout the word. In addition, if phonological words are divided into classes based on whether their content word contains a CS and an LEC in initial position, the difficult stretches of speech can be identified to some extent, as these phonological words are the ones that cause most fluency failure. 7.5 Rate change as a general result of stalling and advancing fluency failures Howell, Au-Yeung & Pilgrim (1999) reported a study designed to examine whether a rapid speech rate in short stretches of speech increases the chance of fluency failure. (It is assumed that the onsets of difficult content words, discussed in the fourth sub-section are randomly distributed across all the stretches so that rate alone can be studied.) Howell et al. (1999) located all tone units in speech and measured speech rate on the fluently produced and stuttered words in each tone unit. They then divided the tone units in which the initial words were fluent, as to whether they were spoken at slow,

90

PETER HOWELL & JAMES AU-YEUNG

medium, and fast rates. They found that speech rate of the fluent words correlated with fluency failure rate, with the most rapid rate of speech being associated with the highest rate of fluency failure. This suggested a local effect of speech rate in spontaneous production that was linked to fluency failure. This finding also carries an implication for treatment: although a global change can lead to improvement in fluency, a local change in these regions should be sufficient to bring about such improvements (see the next chapter). A second prediction about the relationship between speech rate and fluency is that fluent speakers will produce stuttering-like fluency failures if speech rate goes up and speech has to continue, as this will increase the chance of execution getting ahead of planning. An interpretation of a study by Blackmer & Mitton (1991) offers some support for this hypothesis. These authors analysed speech errors and repairs of fluent speakers recorded from late night radio chat shows. They found that fluency breakdown occurred when the speakers were talking rapidly. They reported that this often involved word repetition, which they explained as re-issuing a plan already available (this would be similar to the role we attribute to stalling fluency failures, assuming that the repeated words were function words). Given that Blackmer and Mitton were not in control of the speaking situa­ tion, obviously they had no way of getting the speakers to advance the message at the points where fluency failures occurred. Howell & Sackin (2000) examined this question using a task in which fluent subjects had to provide a running commentary on the actions on a cartoon video they were watching (without the soundtrack). As the speakers were under pressure to continue to keep up with the video action, they tended not to repeat function words, it was found that these speakers produced advancing fluency failures on content words. 7.6 Other ways of stalling The findings of Beattie and Bradbury's study of the temporal structure of speech, support the view that word repetition and pausing act equivalently to slow down speech rate. Beattie and Bradbury reduced pauses in fluent speakers by using an operant procedure while requiring the speakers to maintain a similar articulation rate to that without the procedure. The speakers then produced more word repetitions. This suggests that word repetition has an equivalent role to pausing. Unfortunately, the authors did not report what word type (function or content) was repeated, nor whether the repetitions were of parts of words, words or groups of words. Howell and Sackin (in press) partially replicated Beattie and Bradbury's study and extended the analysis to

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

91

show that when pauses were conditioned out, it was in fact function words that were repeated. Slowing speech by introducing pauses or encouraging function word repe­ tition (see the next chapter) would be expected to enhance the fluency of peo­ ple who stutter, It is noteworthy in this connection that the strategy of inserting filled pauses has been proposed as a way of reducing stuttering (Christenfield 1996). 8.

Conclusions We consider that our experimental investigations may have implications for the practical purpose of diagnosing stuttering. In our research database, we have long-term recordings of eleven children who stutter, who were first seen when they were in the 7-9 year age group and are now in the 10-12 years group. All these children had been admitted to an intensive speech therapy course and the first recording we have was a sample obtained immediately prior to that course. Besides this sample and several follow-up recordings, the children were assessed using therapists' own criteria and researchers' ratings, and on the basis of parents' and children's attitude to communication as re­ vealed in questionnaires. (These criteria were made independently of the speech samples.) In most cases, all the non-speech assessments agreed about whether a child had recovered or not (in the remaining cases there was always a majority in agreement). One classification obtained was based on these scores (recover/persist). After this categorisation had been made, the speech of each child was examined to see what proportion of fluency failure on function (stalling) and content (advancing) word was shown on initial attendance at clinic. The fluency failure rates, provided separately for function and content words, are shown in Figure 6. An 'R' above a child's scores indicates that the independent non-speech measures classified the child as recovered. The children who recovered showed a much higher proportion of function word than content word fluency failures. The recovered cases predicted by applying a criterion of a ratio of function to content word fluency failures of two or greater are indicated with a ' * '. Even this crude measure predicts two thirds (4/6) of the cases of recovery with only a single false alarm in the five recovered cases. This classification procedure appears to offer considerable promise bearing in mind that the analysis was based on a sample of speech obtained before the children received any treatment, and the analysis is limited, as it does not take account of differences in speech rate across these speakers (that is also operative in stuttering

92

PETER HOWELL & JAMES AU-YEUNG

according to EXPLAN). The question that remains to be addressed (see the next chapter) is whether therapies that lead to speech changes that EXPLAN predicts will improve fluency (rate slowing globally or locally, change in type of stuttering pattern) are an effective treatment for stuttering.

Function Word

Conte nt Wond

Fig. 6: Fluency failure rates on function and content words on first attendance at clinic

REFERENCES Au-Yeung, James & Peter Howell. 1998. "Lexical and syntactic context and stuttering". Clinical Linguistics and Phonetics 12.67-78. Au-Yeung, Peter Howell & Lesley Pilgrim. 1998. "Phonological words and stuttering on function words". Journal of Speech, Language, and Hearing Research 41.1019-1030. Beattie, Geoffrey W. & R.J. Bradbury. 1979. "An experimental investigation of the modifiability of the temporal structure of spontaneous speech". Journal of Psy­ cholinguistic Research 8.225-248. Blackmer, Elizabeth R. & Janet L. Mitton. 1991. "Theories of monitoring and the timing of repairs in spontaneous speech". Cognition 39.173-194. Brown, Roger & David McNeill. 1966. "The 'Tip of the Tongue' phenomenon". Journal of Verbal Learning and Verbal Behavior 5.325-337. Caramazza, Alfonso & Michele Miozzo 1997. "The relation between syntactic and phonological knowledge in lexical access: evidence from the 'Tip-of-the-tongue' phenomenon". Cognition 64.309-343.

EXPLAN APPLIED TO THE DIAGNOSIS OF STUTTERING

93

Christenfeld, Nicholas. 1996. "Effects of a metronome on the filled pauses of fluent speakers". Journal of Speech and Hearing Research 39.1232-1238. Conture, Edward G. 1982. Stuttering. Englewood Cliffs, N.J.: Prentice-Hall. Dell, Gary S. 1986. "A spreading-activation theory of retrieval in sentence pro­ duction". Psychological Review 93.283-321. Dell, Gary S. & Padraig G. O'Seaghdha. 1992. "Stages of lexical access in language production". Cognition 42.287-314. Dworzynski, Katharina. 1999. Locating where Fluency Breaks down using the Phonological Word: A cross-linguistic study using speech samples of English and German stutterers. Unpublished MSc dissertation, University College London. Fowler, Carol A. 1980. "Coarticulation and theories of extrinsic timing". Journal of Phonetics 8.113-133. Howell, Peter, James Au-Yeung, Steve Davis, Nicole Charles, Stevie Sackin, Roberta Williams, Francis Cook, Lena Rustin & Phil Reed. In press. "Factors implicated in the diagnosis and prognosis of children who stutter". Proceedings of the Oxford Dysfluency Conference 1999. Howell, Peter, James Au-Yeung & Isabel Vallejo-Gomez. In press. "Why does the position of stuttering change during language development?". Proceedings of the Child Language Seminar, City University, September 1999. Howell, Peter, James Au-Yeung & Lesley Pilgrim. 1999. "Utterance rate and lin­ guistic properties as determinants of speech dysfluency in children who stutter". Journal of the Acoustical Society of America 105.481-490. Howell, Peter, James Au-Yeung & Stevie Sackin. 1999. "Exchange of stuttering from function words to content words with age". Journal of Speech, Language and Hearing Research 42.345-354. Howell, Peter, James Au-Yeung & Stevie Sackin. 2000. "Internal structure of content words leading to lifespan differences in phonological difficulty in stuttering". Journal of Fluency Disorders 25.1-20. Howell, Peter, James Au-Yeung, Stevie Sackin, Kazan Glenn & Lena Rustin. 1997. "Detection of supralexical dysfluencies in a text read by child stutterers". Journal of Fluency Disorders 22.299-307. Howell, Peter, Stuart Rosen, Geraldine Hannigan & Lena Rustin. 2000. "Deficits in auditory temporal resolution in children who stutter and its relation to dysfluency rate". Perceptual and Motor Skills 90.355-363. Howell, Peter & Stevie Sackin. 2000. "Speech rate manipulation and its effects on fluency reversal in children who stutter". Journal of Developmental and Physical Disabilities 12.291-315. Howell, Peter & Stevie Sackin submitted. "Timing interference to speech in altered listening conditions". Perception & Psychophysics. Howell, Peter & Stevie Sackin. In press. "Function word repetitions emerge when speakers are operantly conditioned to reduce frequency of silent pauses". Journal of Psycholinguis tic Research.

94

PETER HOWELL & JAMES AU-YEUNG

Johnson, Wendell & associates. 1959. The Onset of Stuttering. Minneapolis: Univer­ sity of Minnesota Press. Kolk, Herman & Albert Postma. 1997. "Stuttering as a covert repair phenomenon". The Nature and Treatment of Stuttering: New directions (2nd edition) ed. by Richard F. Curlee & Gerald M. Siegel, 182-203. Boston, Mass.: Allyn & Bacon. Levelt, Willem J. M. 1983. "Monitoring and self-repair in speech". Cognition 14.41-104. Levelt, Willem J. M. 1989. Speaking: From intention to articulation. Cambridge, Mass.: MIT Press. Levelt, Willem J. M., Ardi Roelofs & Antje S. Meyer. 1999. "A theory of lexical ac­ cess in speech production". Behavioral and Brain Sciences 22.1-75. Miozzo, Michele & Alfonso Caramazza. 1997. "Retrieval of lexical-syntactic fea­ tures in Tip-of-the-Tongue states". Journal of Experimental Psychology: Learning, Memory, and Cognition 23.1410-1423. Rommel, Dieter. 2000. "The influence of Psycholinguistic Variables on stuttering in childhood". Fluency disorders: Theory, research, treatment and self-help ed. by H.-G. Bosshardt, J.S. Yaruss & H.F.M. Peters. Nijmegen University Press. Sander, Eric K. 1972. "When are speech sounds learned?". Journal of Speech and Hearing Disorders 37.55-63. Selkirk, Elisabeth. 1984. Phonology and Syntax: The relation between sound and structure. Cambridge, Mass.: MIT Press. Shallice, Tim & Brian Butterworth. 1977. "Short-term memory impairment and spontaneous speech". Neuropsychologia 15.729-735. Sternberg, Saul, Stephen Monsell, Ronald L. Knoll & Charles E. Wright. 1978. "The latency and duration of rapid movement sequences: comparison of speech and typewriting". Information processing in motor control and learning ed. by George E. Stelmach, 118-152. New York: Academic Press. Throneburg, N. Rebecca, Ehud Yairi & Elaine P. Paden. 1994. "The relation between phonological difficulty and the occurrence of disfluencies in the early stage of stuttering". Journal of Speech and Hearing Research 37.504-509.

THE EXPLAN THEORY OF FLUENCY CONTROL APPLIED TO THE TREATMENT OF STUTTERING * PETER HOWELL University College London 1. Introduction In this chapter, the question of how speech can be manipulated (i.e., how fluent speech can be made to contain fluency failures and how stuttered speech can be changed to make it more fluent) is addressed with particular emphasis paid to the treatment of stuttering. Generally speaking, a treatment for stuttering is considered successful if it decreases the incidence of fluency failures.1 EXPLAN theory is more specific as it requires a decrease in the inci­ dence of fluency failures that involve production of parts of words (termed advancing fluency failures in the preceding chapter). The current chapter commences by describing the effects Altered Auditory Feedback (AAF) procedures have on fluent speakers and speakers who stutter, AAF manipulations are widely considered as affecting fluency in these speakers in different ways.2 Original explanations of the effects of AAF are then evaluated with particular emphasis on whether an auditory feedback monitor is required to explain their effects. The EXPLAN account of how AAF changes fluency in fluent and stuttering speakers is presented. This explanation, in contrast to those considered earlier in the chapter, does not involve a feedback monitoring process. A contrasting way of manipulating stuttering is then described (verbal operant procedure) and a way this could be explained within the EXPLAN framework is also considered. Finally, lines of

This work was supported by the Wellcome Trust. Thanks to Scott Yaruss for reading a preliminary version of this chapter. * It is recognised that in practical circumstances, treatments bring about various other important changes in the speaker's outlook. However, such effects cannot be used as an outcome measures until there is agreement as to what these changes are, and objective ways of measuring these attributes are available. 2 Inclusion of the word 'feedback' begs the question of how the findings should be explained. Existence of an auditory loop that can be disrupted by altered listening does not necessarily require a monitoring process.

96

PETER HOWELL

evidence that support the EXPLAN account of how speech output control is affected by various procedures are reviewed. 2. Altered listening conditions and auditory feedback monitoring A monitor is defined as a device that takes the output of a process, com­ pares it with what was intended to be produced and makes a correction to re­ duce any discrepancy detected. The auditory exteroceptive loop that features in auditory monitoring accounts (Fairbanks 1955; Lee 1950; Levelt 1989), suggests the speaker recovers speech output through the auditory system and uses this to determine whether there is a discrepancy between intention and output. The main source of evidence for an exteroceptive auditory loop is that alterations interposed between articulation and audition have dramatic effects on speech. Moreover, their effects are somewhat different for fluent speakers and speakers who stutter. Three types of alteration to the auditory loop can be made: timing, spectral content and intensity. Speech timing is almost always studied by delaying time of arrival of the speaker's voice, when it is called Delayed Auditory Feedback (DAF). In fluent speakers, DAF slows speech mainly by elongating vowels. The speech also has a monotone pitch and high amplitude (both effects being easily discernible on the vowels of DAF speech). The effects of DAF on fluent speakers have been explained by proposing that a feedback monitor is used for speech control. Speakers continue to use the altered sound for voice control when the sound of the voice is experimentally altered, and speech problems occur because the monitor acts on this misleading feedback. Timing alterations to feedback lead to adjustment to output timing where none was needed and this causes odd-sounding speech that can contain errors. In speakers who stutter, DAF has the effect of reducing the incidence of stuttering events dramatically. It was originally considered that the response to DAF of adults who stutter was opposite to the response of fluent speakers. However, this position is not sustainable: DAF disruption occurs mainly on medial vowels, (Howell, Wingfield & Johnson 1988), whereas the stuttering events that are lost occur on initial consonants (Brown 1945). From the dif­ ferent distributions of the two types of fluency failure, it can be established that speakers who stutter, while losing stuttering events under DAF, experience the same disruption to speech as fluent speakers. These differences rule out proposals that people who stutter have a timing problem in the monitoring process that once rectified by an external DAF delay allows these speakers to regain fluent control (Webster & Dorman 1970).

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

97

Fluent speakers shift vocal pitch down when the speech spectrum is shifted up and vice versa. Frequency-shifted Feedback (FSF) has been the subject of intense study since Howell, El-Yaniv & Powell (1987) found that it has equally dramatic effects as DAF in controlling the fluency of people who stutter without any of the side-effects on vowels that occur with DAF. For instance, the intensity of speech output does not change under frequency shift for fluent speakers or speakers who stutter (Howell 1990). Also, speakers who stutter do not make compensatory shifts in voice fundamental frequency as fluent speakers do (Natke & Kalveram, in press). The final form of alteration, intensity, has been extensively studied in fluent speech control. One variant of this manipulation is to amplify or attenuate the speech directly. When speech is amplified, speakers reduce voice level (Lane & Tranel 1971). Another way of investigating voice intensity change is by changing ambient noise level. A high noise level operates like attenuating the voice (i.e., speakers increase loudness when speaking under noise). No work has been done to investigate the changes in voice intensity in people who stutter. However, masking the voice of people who stutter has been investigated and this is reported to improve fluency control in these speakers (Cherry & Sayers 1956). These authors proposed auditory feedback improved speech control in people who stutter by masking the voice. 3. Problematic relevance of the auditory loop It was not necessary to consider an articulation-audition (external) loop, like that in Levelt's (1989) model, in the previous chapter. This loop needs considering here, though, as the powerful effects of AAF may be due to inter­ ference with this loop. As arguments were only given for dismissing monitor­ ing of internal loops in the earlier chapter, nothing said so far requires that an auditory feedback loop should be ruled out. Borden (1979) pointed out two fundamental problems for auditory feedback monitoring loops. First, auditory processing takes time (about 100 ms even for simple reaction time). So, at least an additional 100 ms would be needed after every segment has finished, to process the auditory output from it, establish that it was as intended and then to carry on with the message. This would lead to slower speech rates than speakers achieve. Second, hearing-impaired adults can control their speech even though they are not able to use audition to any great extent (Borden 1979). Thus, it appears that auditory feedback is not necessary for controlling the voice, as has been assumed in auditory monitoring accounts.

98

PETER HOWELL

A third problem, not noted by Borden, is that from an auditory monitoring perspective, it is necessary to be able to retrieve a veridical record of what was produced; otherwise, establishing if and what error has occurred with the intention of correcting it would not be possible. However, it is not clear how veridical the representation of articulatory output provided by the auditory system is about the intended message. The auditory representation the speaker receives while speaking is affected by internal and external noise sources dur­ ing transmission and the external noises vary with speaking environment. Internal noise arises through bone-conduction and is heard along with di­ rect sound. These sound-sources are at approximately equal levels (von Bekesy 1960). Formant structure is heavily attenuated in bone-conducted sound and this sound is dominated by the voice fundamental (Howell & Powell 1984). Bone-conducted sound limits the information about articulation that can be provided by the auditory loop as it is very different from speech output. External sources of noise are mainly a result of extraneous environmental sounds and echo. Extraneous noise affects the intensity-time profile of speech (Rosen & Howell 1981), and rooms with hard walls are reverberant, which affects the acoustic properties of speech salient for perception (Watkins 1992). Thus speech is output into different environments that have unique effects on the sound of the voice. Again, the effect of these variations limits what infor­ mation the speaker can obtain through hearing about how the intended speech was articulated. In brief, internal and external noise added during output makes the distal source of speech a non-reflexive form of the originally conceptualized message that cannot be used for control by a monitor. The first requirement imposed on the architecture of EXPLAN is that plan­ ning is independent of execution. This independence does not allow an audi­ tory feedback monitoring loop so some other explanation why AAF modifies speech is needed. Dismissing a monitor (specifically for monitoring auditory events) was also suggested by the three problems discussed in this section. A second requirement is that a satisfactory theory needs to avoid the latter problems while, at the same time, incorporating (requirement three) an expla­ nation of control by fluent speakers and (requirement four) control by speakers who stutter under normal listening and AAF conditions. An explanation of why disparate forms of treatment (such as AAF and operant procedures) are effective with people who stutter and why the different forms of treatment work over different time periods is also needed. A development of EXPLAN is proposed in the following sections and assessed against these requirements.

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

99

4. Speech output control without feedback monitoring The basics of EXPLAN as they apply to spontaneous speech produced in normal listening conditions are presented, and developments of the theory to extend it in order to explain the effects of AAF are then given. In this section, EXPLAN's operation is described for word production as in the preceding chapter. Words were used to describe how EXPLAN operates in the previous chapter as most of the evidence considered used these segments. However, syllables are more useful when considering rhythm-control that lies behind the EXPLAN account of speech modification. Words will continue to be used for describing the extension of EXPLAN to speech modification, though monosyl­ labic words should be understood to establish connection with the rhythm of speech. EXPLAN proposes that planning takes place independently of, and in parallel with, execution. Provided that there is enough time during the execu­ tion of one word to generate the next, the following plan can be picked up when the first word is completed and speech will be fluent. The fluent situa­ tion is depicted schematically in Figure la. The diagrammatic representation of the temporal relationship between plan­ ning and execution for three words (n, n+1, n+2) when speech is proceeding fluently is given. Time is along the abscissa. The epoch during which planning (PLAN) and execution (EX) occur are shown as bars in the top and middle rows respectively. The bottom row indicates the computation made on the plan and its execution (the result is zero in these fluent cases).

Fig. la: Planning and execution for three words

Figure lb gives a stretch of fluent speech where, at the start, planning and execution are hand-shaking appropriately on the first word (word n). After word n is first completed, the plan for n+1 is not ready (its planning continues to the right of the bold arrow). In this case, the speaker retrieves plan n, executes that again (leading to a word repetition), at which point the

100

PETER HOWELL

plan for word n+1 is ready and can be completed fluently (the difference computation between the plan and its executed form is zero).

Fig. lb: Stretch of fluent speech on the first word

Figure 1c shows a stretch of fluent speech where, at the start, planning and execution are hand-shaking appropriately on the first word in the sequence shown (wordn). After word n is completed on the first occasion after planning, the plan for word n+1 is not ready. Even so, the speaker executes the part plan of word n+1. After the execution of this part plan has been completed, an efferent copy of the timing pattern is differenced with the complete plan and results in a non-zero difference. This alerts the speaker who then makes a rate change to prevent this situation continuing.

Fig. lc: Stretch ofluent speech on the first word in this frequence

Two types of fluency failure were described in the previous chapter. In both of these, the plan for a word was not available within the time necessary for executing the previous one. This occurs in Figures lb and lc, at the point where execution of word n finishes before the time when the plan of word n+1 is complete. Stalling fluency failures deal with this situation by producing an immediately previous word (or words) again, or by pausing. In the example in Figure lb, word n should be re-executed. Planning of word n+1 continues and

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

101

when execution of word n has finished the second time, if the plan of word n+1 is then ready it will be executed. Importantly in stalling types of fluency failure, no word is executed until its plan is complete. The situation where the plan is not complete gives rise to different type of fluency failure {advancing fluency failures). Instead of waiting for the whole plan to be ready, the speaker commences execution of the word with the part of its plan that is available. In advancing fluency failures, represented sche­ matically in Figure lc, fluency can fail in ways that involve production of the first part of words. This has been observed to be a feature of persistent stut­ tering (Conture 1982). 4.1 Alerts, stalling and advancing The speech control system ought to detect when speech is advancing pre­ maturely so that something can be done to avoid the inherent danger of fluency failure. The necessary control may be achieved by checking whether the whole plan is supplied for execution. Howell and Sackin (in press) proposed that the check on speech timing is made by taking the difference between the planned and executed versions where the executed version is an efferent copy of the timing plan (e.g., a marker for each vowel onset) delivered to the motor effectors. This aspect of the theory will be illustrated for fluent speech, and then the situation that occurs in stalling and advancing fluency failures will be considered. In the stretch of fluent speech shown in Figure la, the efferent copy (1) of the timing plan of a speech segment is taken at the point in time that its execu­ tion commences (signified for word n as c(EX(n))). A copy of the timing plan of the word is obtained when execution is completed (c(PLAN(n))). c(PLAN(n)) is then differenced with c(EX(n)). If the timing patterns correspond, the dif­ ference is zero, and the two timing signals cancel. This result indicates that execution rate is not too rapid for planning to keep up. The plans of preceding words remain available for a short time after they have been executed (Blackmer & Mitton 1991). So when the plan for the next word is not complete, prior words whose plan is still available can be reexecuted (word repetition). When the plan for word n+1 is not ready, the plan of word n is retrieved (Figure lb) and executed. The timing pattern of the exe­ cuted form of word n (c(EX(n))) will still correspond with the efferent copy of the timing plan of the word (c(PlAN(n))) so the difference will still be zero (as in speech produced fluently). Consequently, word repetition is essentially a fluent speech control mechanism. The point where the speaker has advanced to word n+1 while planning of this attempted word is still going on is depicted after word n has been

102

PETER HOWELL

executed in Figure 1c. If the speaker advances and produces the first part of the word, more, but not necessarily all, of the plan will be produced during the time that this part plan is being executed because the speech plan continues to be updated for execution in this period. When this updated plan, taken at the time execution is completed (c(PLAN(n+l))), is differenced with the efferent copy of the plan at the point in time that execution commences (c(EX(part P(n+1)))), the result is always non-zero. Non-cancellation of the signals provides an alerting response that the speaker is speaking at too high a rate. The intermittency of the alerting response and the relatively crude information it provides (that something has gone wrong, not whether a component of articulation has failed to be realized properly) is a major difference between the current proposal and auditory feedback monitoring accounts. In particular, in EXPLAN no measure of the error in articulation occurs that could be used to minimize any discrepancy between intended and actual speech. Consequently, a monitor has no error information to work with. A block diagram giving an overview of what happens after the EXPLAN process breaks down and the additional processing components that then come into play is given in Figure 2.

Fig. 2: Block diagram of the processes involved in executing speech

Fluent speech and stalling are ways of adjusting local speech rate that do not involve external timing changes (control occurs within the section top left

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

103

of Figure 2, the response loop). According to EXPLAN, the alerts generated during advancing fluency failures are a series of pulses (the non-zero pulses after differencing each segment). These are input to a timekeeper (bottom right of Figure 2). Because of the selective involvement of the timekeeper during advancing fluency failures, an equivalent way of describing the operation of sending input to the timekeeper is that timing control changes from an intrinsic to an extrinsic schedule (Fowler 1980). Figure 2 shows the inputs that would be received by the timekeeper when a fluent hearing individual is speaking easy material at a comfortable rate and dealing with any occasional incomplete plans by stalling fluency failures. Speech is controlled within the response loop where PLAN and EX handshake in a chaining process (as EX of one word is complete, the PLAN of the next is supplied). Serial activity associated with the execution of each segment arises during this process and, as with all such activity, it is an input registered by the timekeeper. At the same time, the timekeeper also receives a serial input via the auditory system (the speech output). The load that arises from these two synchronised inputs is within the timekeeper's capacity and no change is made. Effectively, the PLAN-EX cycles in the response loop and operates inde­ pendent of the timekeeper. Consideration will be given in the remainder of this chapter, to how changes in the inputs to the timekeeper can change operation of the response loop. In Figure 2 the response loop (top left) performs the operations detailed in Figure 1. The timekeeper is bottom centre. The bulbs represent inputs to the timekeeper that are connected automatically when activity arises in these input lines. When load to the timekeeper is exceeded, the parameters of the coupled oscillator (top centre) are adjusted and the response loop adjusts automatically. The coupled oscillator is under timekeeper control can undergo global (top input line, response to DAF) or local (bottom input line, response to FSF) slowing. 4.2 Timekeeper and load Situations where the inputs to the timekeeper change can now be con­ sidered, starting with the case of advancing fluency failures. In this case, an alert is added to the inputs to the timekeeper. According to EXPLAN, adding an alert sequence raises the number of inputs to the timekeeper. The relationship between the alert and the other concurrent inputs to the timekeeper determines load and can affect the timekeeper's operation. 4.2.1 Number, rate and precision requirements of inputs. In EXPLAN, the load on the timekeeper is determined by sequential internally-generated neural signals, signals arising from other external sensory events and any similar ac-

104

PETER HOWELL

tivity from ongoing motor activities. As alerts are intermittent, they add a neu­ ral input to the timekeeper when they arise. This raises load as the more signals there are, the higher the load. Changing rate of inputs also affects load. If the timekeeper is monitoring a fixed number of inputs, other things being equal, load will increase when the rate of these inputs increases. Load also in­ creases when concurrent tasks are performed that adds inputs and can make demands for more precision on the timekeeper. 4.2.2 Synchrony. Synchronous signals are easier for a timekeeper to deal with than asynchronous ones. Howell, Powell & Khan (1983) illustrate this with canon, a simple form of song in which a synchronised rhythm is heard while singing. Conversely, load increases when sensory or motor inputs are out of synchrony, as when dancing or clapping offbeat (Howell et al. 1983). 4.3 Response of the timekeeper to load variation The timekeeper mechanism is governed by the principle that it needs to keep load within its capabilities. In the spontaneous speech of a fluent speaker speaking at moderate or slow speech rates with sounds that are simple to plan and spoken in an ordinary listening environment (Figure 2), no alerts are gen­ erated. Change in any of these circumstances will cause load on the time­ keeper to increase and potentially to exceed its capacity. If capacity is ex­ ceeded, the mechanism can offset the increase that occurs, though the options for this are limited. Two such ways of reducing load are now considered. Both involve the timekeeper changing the parameters of an oscillator that is coupled with the response loop. This oscillator has its parameters (centre frequency and bandwidth) set initially so that they are in tune with the modulation rate of speech in the response loop (e.g., centre frequency set at word rate, bandwidth set to allow expressive variation around that frequency). The coupled oscillator does not then affect the response loop providing speech rate remains within these limits, preserving the independence of the response generation cycle from the timekeeper. 4.3.1 Global rate control DAF slows overall (global) speech rate because asynchronous inputs that occur with this type of manipulation, overload the timekeeper. The timekeeper responds to asynchronous inputs by shifting the centre frequency of the coupled oscillator to a lower mean value. The coupled oscillator is represented as a resonator in the frequency domain and the shift in centre frequency is shown in the top branch going to the response loop at top right of Figure 2. This alteration to the parameters of the coupled oscillator then perturbs the responses generated in the response loop. That is,

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

105

resonance principles suggest that the responses are entrained to a lower rate so that they are, once again, not affected by the coupled oscillator. The slower response rate that is induced, besides having an inherently lower load itself, would also remove or reduce the incidence of alert inputs to the timekeeper, as they occur less often at slower rates because planning has more time to catch up with execution rate. 4.3.2 Local rate control Episodes of speech that contain a word that starts with a difficult string of phonemes (Howell, Au-Yeung & Sackin 2000), and where the speech rate in the immediate context is rapid (Howell, Au-Yeung & Pilgrim 1999) are most prone to fluency failure in people who stutter. A change to speech rate focussed on these regions (local rate change) is all that is necessary to reduce advancing fluency failures and reduce load on the time­ keeper. The rate in more extensive stretches of speech, such as phonological words (Selkirk 1984), can be represented as a distribution of rates of segments within these stretches. The segments that lie at the fast rate end of this distri­ bution require slowing. The incidence of these fast segments can be reduced without affecting mean speech rate if an equivalent reduction in the incidence of slow segments is made. This would be revealed if variance in the rate distribution is reduced. FSF has been reported to lead to such a reduction in the variance of rate responses in repeated readings of a target sentence by speakers who stutter (Howell & Sackin 2000). The bottom branch at the top right of Figure 2 depicts the effects of a local rate change. When load is exceeded under FSF, the coupled oscillator is tuned more sharply. The narrower bandwidth entrains the speech rates made in the response loop into a narrow band around the centre frequency (bottom branch). One effect of this is to reduce the incidence of segments spoken fast that can be problematic in that the plan may not be ready for execution in time in these regions. Removing these sections would make speech fluent and remove associated alerts, reducing the load on the timekeeper. 5. EXPLAN addressed to the problems faced by a monitoring account The problems raised in connection with a monitoring account are now re­ considered. Borden's first concern was the amount of time involved in processing feedback for monitoring speech. It is quick to make an efferent copy and perform a differencing operation. The proposed operations in EXPLAN would not, therefore, slow speech down to the extent that obtaining and checking an auditory representation does.

106

PETER HOWELL

Borden's (1979) second point was that auditory feedback is not essential in adults who sustain hearing loss. Sound input goes to the timekeeper, not to the response loop in EXPLAN. So, loss of an input to the timekeeper would not affect operation of the response loop. The loss of auditory input to the timekeeper consequent on hearing loss would reduce load. The timing mechanism responsible for speech output can still operate efficiently when this input is lost. EXPLAN does not include a monitoring mechanism to explain the effects of AAF treatment. Specifically, execution output is not used to revise or tune the speech plan (Postma 2000). The requirement was also made that the speech control mechanism has to operate without veridical information about speech output. In EXPLAN, the auditory input channel only operates as serial input to the timekeeper, not as a channel that provides information about how speech is articulated. So, as the speaker is not retrieving information about placement of articulators to produce the sounds, it does not matter whether the auditory version of the speaker's own speech is veridical, only that its timing pattern does not interfere with speech activity. 6. Account of AAF and secondary tasks in fluent speakers Listening environment, the fluency of the speaker, and the number and complexity of tasks being performed simultaneously, can all affect the time­ keeper and its control structures. All these factors add extra signal/s (in or out of synchrony) or change the rate of signals so the load on the timekeeper alters. The effects of AAF that have been traditionally explained as due to feedback monitoring, really reflect interference to timing control mechanisms (Ivry 1997 locates the timekeeper in the cerebellum). In EXPLAN, FSF and DAF both add an extra input to the timekeeper, increasing its load. The DAF procedure produces an extra signal that is out of synchrony with that due to direct speech output responses from the response loop and unaltered auditory output of the voice. The extra asynchronous DAF input to the timekeeper increases its load. The timekeeper then needs to reduce load, which it does by reducing mean response rate in the oscillator coupled to the response loop (global rate change). FSF also creates a second signal from manipulations of vocal output that is input to the timekeeper. This sound is in synchrony with sensory and response inputs from speech that occur as with DAF. Speakers decrease the bandwidth of the oscillator coupled with the response loop that leads to a reduction in variance of subjects' responses when they are presented with this type of alteration (a local change in response rate that decreases load). The EXPLAN model also incorporates an explanation for secondary task disruption. All serial events associated with sensory, response or neural

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

107

activity from a secondary task are merely treated as more serial inputs. A con­ current task that involves monitoring and/or responding to a serial signal adds load to the timekeeper because of the additional sensory (and possibly response) input. The extra input/s add load that can lead to response rate or accuracy changes (Howell & Sackin submitted). 7. Effect of rate changes in people who stutter EXPLAN predicts that improvement in speech is always attendant on change in rate control (locally or globally). Both ways of reducing rate will affect how much time is allowed for EX in the EX-PLAN system. Voluntary rate changes should tax or relieve problems where speech planning cannot keep up with execution rate. There is much evidence that reducing the speech rate of speakers who stutter decreases frequency of fluency failure, as EXPLAN pre­ dicts (Johnson & Rosen 1937; Perkins, Kent & Curlee 1991; Starkweather 1985; Wingate 1976). It is also widely reported that rate increase has the opposite effect, again as EXPLAN predicts. Increasing global speech rate increases frequency of fluency failure (Johnson & Rosen 1937; Bloodstein 1987). Besides the evidence where speech rate is altered voluntarily, many clinical procedures known to alleviate stuttering have equivalent effects insofar as they slow speech execution rate. A speculative model about long-term stuttering, suggested by EXPLAN, is that the disorder persists because no adjustment to rate is made when alerts occur. Speakers who stutter experience many fluency failures that lead to alerts. However, although many alerts occur, the speakers do not respond to them. Speech responses that lead to alerts will continue if no action to avoid them is taken. The speaker then gets saturated with alerts that are ignored and loses sensitivity through adaptation (Howell, Rosen, Hannigan & Rustin 2000). The converse applies to fluent speakers: Most of their fluency failures are of the stalling type that do not involve alerts. Responding appropriately to alerts on the rare occasions that advancing fluency failures occur, obviates their occurrence and maintains sensitivity to the alerts by keeping their occurrence infrequent. According to this perspective, treatments should help speakers who stutter by ensuring the timekeeper keeps its sensitivity to alerts, and by treating the alerts as serial signals to a timekeeper that are controllable if speakers respond appropriately. A period where alerts are not experienced as frequently (as ex­ perienced by fluent speakers) may be necessary to regain sensitivity. While DAF and FSF operate on response output, they do this by adding load to the timekeeper that removes alerts. This transient exposure to added load may be too short to restore sensitivity to alerts. To restore sensitivity, it may be neces-

108

PETER HOWELL

sary to prevent alerts for long periods of time or target AAF specifically on stuttered sections and according to prescribed schedules (Reed & Howell 2001). EXPLAN suggests that the extended period of treatment time associated with operant procedures (such as the Australian Lidcombe program) may be reinstating sensitivity by preventing alerts. These procedures may prevent alerts in a different way to AAF (i.e. preventing too rapid execution, thereby influencing the response loop directly). Whichever way alerts are prevented, the timekeeper can regain its sensitivity and fluent speech can be reinstated. 8. Comparing architectures of monitoring and EXPLAN outputs Table 1 gives a summary comparison of error monitoring and the alert view to highlight the differences discussed so far. EXPLAN during advancing fluency failures works on neural alerts whereas the monitors discussed earlier on in this chapter use auditory information to detect equivalent problems (row 1). Neural alert signals are rapid to obtain and process and the alerting activity would be unaffected in people who sustain a hearing loss (Borden 1979). The monitoring and EXPLAN outlooks differ in terms of the signal information they provide (row 2). The error information provided to a monitor is much more detailed than that provided by an alert, as it needs to specify what alterations are needed to correct or retune a speech plan after an error. In the previous chapter, it was pointed out that speech errors are rare and a different strategy for dealing with them is simply to let them happen and then restart speech. Whereas monitoring needs to be continuous and produces a signal all the time, alerts are intermittent (row 3). A knock-on effect of this is that continuous signals have to be continuously monitored, thus affecting overall speech rate (Borden 1979). The two perspectives also differ with respect to the response that is initiated when fluency fails (row 4): in an error monitor, speech has to be replanned or tuned to remove the error, whereas in EXPLAN, rate slows as an emergent property of the timekeeper, not on the basis of a reformulation of the message.

1. 2. 3. 4.

Signal Information provided Existence Response

Error monitors Auditory Contains information what was wrong Signal continuously generated Message replanned when fluency fails

Alerts CNS-internal Indicate where failure occurred, not what the failures were Signal intermittently generated Slower rate is an emergent response

Table 1: Differences between error-monitoring and alert perspectives on speech control

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

9. Some experimental lines of support for the

EXPLAN

109

control structures

9.1 Is the delayed signal under DAF speech or noise? According to EXPLAN, the delayed sound does not have its effect because the speech information it contains is deciphered and used for control; rather it is a signal with particular rhythmic properties that make it disruptive. Consequently, any noise with the same spectral and intensity properties that appears at the same points as the delayed signal should have an effect equivalent to a delayed speech sound. Howell & Archer (1984) confirmed this with an experiment, in which a non-speech noise with the same intensity profile as the original speech was substituted for the delayed sound under DAF. There was no difference in the time taken to read a list in this stimulus condition compared with that in a delayed speech condition. This suggests that any sound that stands in this temporal relationship with the direct speech will cause equivalent disruption. The interpretation that the delayed signal disrupts the rhythm, not the speech content, would also explain why DAF sounds produce an increase in volume as with noise (Howell 1990). The act of delaying speech creates a sound that disrupts rhythm, which is responded to as noise. 9.2 Does DAF selectively affect the timekeeper process? EXPLAN maintains that the effects of alteration to auditory feedback are due to activity sent specifically to the timekeeper mechanism. Manipulations that influence the timekeeper can be separated from manipulations that affect motor responses using Wing & Kristofferson's (1973) model. As originally developed, the Wing-Kristofferson model decomposes the total variance in a set of tapping responses that have to be made at a specified rate into components associated with the motor and clock processes. The data for the delayed auditory feedback conditions are shown in Figure 3 a and Figure 3b.

Fig. 3a: Motor variance

110

PETER HOWELL

• ♦ - -66ms -

-

133ms

Fig. 3b: Timekeeper (clock) variance

Howell and Sackin (submitted) had speakers do a speech version of the Wing-Kristofferson task (their subjects repeated the syllable /bae/ at pre­ selected rates), and performed the Wing-Kristofferson analysis after syllable onsets had been marked. Subjects performed the experiment at two rates (600 and 800 ms repetition rates) and at three DAF delays (66, 133 and 200 ms). DAF delay of the points connected together was 66, 133 or 200 and the delay used for connected points can be identified from the symbol in the caption. Axes are variance on the ordinate and repetition rate on the abscissa on both sides of the graph. Clock variance increases as repetition rate increases as re­ ported by Wing (1980) in a tapping task. Importantly, clock, but not motor variance also shows dramatic increases as DAF delays are changed from 66 through to 200 ms delays. This demonstrates that most of the effect of DAF is specifically to timekeeper variance, as EXPLAN predicts. 9.3 Serial inputs and speech EXPLAN predicts that serial input signals, in general, are signals going to the timekeeper that should lead to improvement in speech performance in people who stutter because of the effects they have on speech rate. Howell & El-Yaniv (1987) reported that a click synchronous with syllable onset improved the speech of people who stutter as much as a regularly-timed metronome click. The pacing effect of a regularly-timed metronome could lead to improved fluency by slowing speech rate, if (as is usual) the metronome is set to a slow rate. However, the speech synchronous click does not impose a prescribed rate; it is synchronised to the rate the subject uses. The EXPLAN account would pre­ dict that the timekeeper receives an extra input from the metronome pulse what­ ever mode it is operating in, and this leads to the timekeeper regulating its load, affecting speech rate control. Both types of metronome should reduce rate-

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

111

variability as the clicks are speech synchronous. In turn, this should lead to local slowing that improves speech by reducing the incidence of fast, problematic stretches of speech. Howell et al. (1983) showed that interrupted vowels have some abstract similarities to DAF. Delaying a sound's onset leads the speaker to hear sound at a low level and for the level to increase when the sound onsets after the delay. Interrupting a vowel sound also leads to such steep increases in intensity during vowels. Based on the similarity in disruption to intensity patterns, it would be predicted that, as DAF improves the speech of speakers who stutter, interrupted vowels should also do so, as confirmed recently by Kalinowski, Dayalu et al. (2000). EXPLAN predicts that this would be associated with global slowing as the interruption creates an intensity increase that is asynchronous with direct speech activity. All serial inputs, whatever their modality, are fed to the timekeeper. All serial input signals to the timekeeper would have equivalent effects on load. In people who stutter, the effect of the timekeeper reducing response rate due to a load increase allows planning to catch up with execution and stuttering rate to drop. EXPLAN predicts that even serial inputs from different modalities would produce improvement in much the same way, Kuniszyk-Jozkowiak, Smolka & Adamczyk (1996) have confirmed this prediction. They demonstrated that speech control by speakers who stutter improves when a concomitant flashing light input occurs. 9 4 Affecting fluency failures by operant procedures Beattie & Bradbury (1979) reported changes in stalling fluency failures in fluent speakers using a verbal operant-conditioning procedure. Silent pauses were detected electronically and a light came on when they occurred. Partici­ pants were instructed to try to prevent the light illuminating while maintaining the same speech articulation rate. Findings revealed an increase in the number of word repetitions. Howell & Sackin (in press) replicated this study and established that this repetition occurred mainly on function words. According to EXPLAN, pauses and function word repetitions both have a stalling role. Stalling is essential for maintaining fluent speech, so when the experimenter prevents speakers from using one way of stalling, another automatically emerges. New­ man (1987) also suggested that rate control is important for fluent speech con­ trol. He found that punishing word repetition led to a reduced word output rate« Operant procedures are effective with young speakers who stutter though less so with older speakers (Onslow, Costa & Rue 1990) whereas the reverse has been reported for FSF (Howell, Sackin & Williams 1999). The ineffective­ ness of the operant work with adults may suggest that if the ability to alert the timekeeper is lost it is difficult to re-establish its usual function. Conversely. operant procedures appear to prevent this happening when applied in childhood.

112

PETER HOWELL

10. Conclusions If the EXPLAN account of why operant procedures work is correct, it is only necessary to stop advancing fluency failures in order to prevent alerts. EXPLAN considers that stalling fluency failures are ways of controlling fluency within the response loop and should be reinforced as they may be beneficial in offering a way of avoiding more serious fluency failures. Advancing fluency failures, on the other hand, lead to alerts and should be targeted for reduction to prevent stuttering (assuming that the speaker will then respond by using stalling fluency failures or speak at a slower rate commensurate with fluent speech). Currently my research group is employing a variety of procedures reported to increase stalling fluency failures while at the same time punishing advancing fluency failures. Three children who were diagnosed as non-recovered stutterers based on both the speech and non-speech criteria described in the previous chapter have been treated. The results for one child are given in Figure 4. Percentage of content (mainly advancing) and function (mainly stalling) word fluency failures at baseline, various points during intervention, and six weeks post-treatment are shown. Results show that relative to baseline speech data during diagnosis, stuttering rate of both advancing and (more surprisingly) stalling fluency failures decreased. The reduction in stalling fluency failures may be a result of the operant procedures reducing overall speech rate, thereby cutting down all types of fluency failure. Within the treatment sessions, content word (advancing) fluency failures reduced significantly over sessions. The follow-up six weeks after treatment had finished showed that most of the improvement obtained during therapy was maintained.

—

f

u

n

c

t

i

o

n

word

Fig. 4: Results for subject ED during intervention and post treatment.

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

113

REFERENCES Beattie, Geoffrey W. & R. J. Bradbury. 1979. "An experimental investigation of the modifiability of the temporal structure of spontaneous speech". Journal of Psycholinguistic Research 8.225- 248. Blackmer, Elizabeth R. & Janet L. Mitton. 1991. "Theories of monitoring and the timing of repairs in spontaneous speech". Cognition 39.173-194. Bloodstein, Oliver. 1987. A Handbook on Stuttering. 4th ed. Chicago: National Easter Seal Society. Borden, Gloria J. 1979. "An interpretation of research on feedback interruption in speech". Brain & Language 7.307-319. Brown, Spencer F. 1945. "The loci of stuttering in the speech sequence". Journal of Speech Disorders 10.181 -192. Cherry, Colin, & Bruce Sayers. 1956. "Experiments on the total inhibition of stam­ mering by external controls and some clinical results". Journal of Psychosomatic Research 1.233-246. Conture, Edward G. 1982. Stuttering. Englewood Cliffs, N.J.: Prentice-Hall. Fairbanks, Grant. 1955. "Selective vocal effects of delayed auditory feedback". Jour­ nal of Speech and Hearing Disorders 20.335-348. Fowler, Carol A. 1980. "Coarticulation and theories of extrinsic timing". Journal of Phonetics 8.113-133. Howell, Peter. 1990. "Changes in voice level caused by several forms of altered feedback in normal speakers and stutterers". Language and Speech 33.325-338. Howell, Peter & Alexander Archer. 1984. "Susceptibility to the effects of delayed auditory feedback". Perception & Psychophysics 36.296-302. Howell, Peter & James Au-Yeung. 2001. "The EXPLAN theory of fluency failure and the diagnosis of stuttering". This volume, 75-94. Howell, Peter, James Au-Yeung & Lesley Pilgrim. 1999. "Utterance rate and lin­ guistic properties as determinants of speech dysfluency in children who stutter". Journal of the Acoustical Society of America 105.481-490. Howell, Peter, James Au-Yeung & Stevie Sackin. 2000. "Internal structure of content words leading to lifespan differences in phonological difficulty in stuttering". Journal of Fluency Disorders 25.1-20. Howell, Peter & Nirit El-Yaniv. 1987. "The effects of presenting a click in sylla­ ble-initial position on the speech of stutterers: comparison with a metronome click". Journal of Fluency Disorders 12.249-256. Howell, Peter, Nirit El-Yaniv & David J. Powell. 1987. "Factors affecting fluency in stutterers when speaking under altered auditory feedback". Speech Motor Dy­ namics in Stuttering ed. by H. Peters & W. Hulstijn, 361-369. New York: Springer. Howell, Peter & David J. Powell. 1984. "Hearing your voice through bone and air: implications for explanations of stuttering behaviour from studies of normal speakers". Journal of Fluency Disorders 9.247-264.

114

PETER HOWELL

Howell, Peter, David J. Powell & Ian Khan. 1983. "Amplitude contour of the delayed signal and interference in delayed auditory feedback tasks". Journal of Experi­ mental Psychology: Human Perception and Performance 9..112-784. Howell, Peter, Stuart Rosen, Geraldine Hannigan & Lena Rustin. 2000. "Deficits in auditory temporal resolution in children who stutter and its relation to dysfluency rate5'. Perceptual and Motor Skills 90.355-363. Howell, Peter & Stevie Sackin. 2000. "Speech rate manipulation and its effects on fluency reversal in children who stutter". Journal of Developmental and Physical Disabilities 12.291-315. Howell, Peter & Stevie Sackin submitted. "Timing interference to speech in altered listening conditions". Perception & Psychophysics. Howell, Peter & Stevie Sackin. In press. "Function word repetitions emerge when speakers are operantly conditioned to reduce frequency of silent pauses". Journal of Psycholinguistic Research. Howell Peter, Stevie Sackin & Roberta Williams. 1999. "Differential effects of fre­ quency shifted feedback between child and adult stutterers". Journal of Fluency Disorders 24.127-136. Howell, Peter, Trudie Wingfield & Michael Johnson. 1938. "Characteristics of the speech of stutterers during normal and altered auditory feedback". Proceedings Speech 88, 7th Federation of Acoustical Societies of Europe conference, Edin­ burgh 1988 ed. by William A. Ainsworth & John N. Holmes, vol 3, 1069-1076. Edinburgh: Institute of Acoustics. Ivry, Richard. 1997. "Cerebellar timing systems". The Cerebellum and Cognition ed. by Jeremy Schmahmann, 555-573. San Diego: Academic Press. Johnson, Wendell & Leonard Rosen. 1937. "Effect of certain changes in speech pat­ tern upon frequency of stuttering". Journal of Speech Disorders 2.105-109. Kalinowski, Joseph, Vikram N. Dayalu, Andrew Stuart, Michael P. Rastatter & Manish K. Rami. 2000. "Stutter-free and stutter-filled speech signals and their role in stuttering amelioration for English speaking adults". Neuroscience Letters 293.115-118. Kuniszyk-Jozkowiak, Elzbieta Smolka Wieslawa & Bogdan Adamczyk. 1996. ''Effect of acoustical, visual and tactile reverberation on speech fluency of stut­ terers". Folia Phoniatrica & Logopedics 48.193-200. Lane, Harlan L. & Bernard Tranel. 1971. "The lombard sign and the role of hearing in speech". Journal of Speech and Hearing Research 14.677-709. Lee, Bernard S. 1950. "Effects of delayed speech feedback". Journal of the Acousti­ cal Society of America 22.824-826. Levelt, Willem J.M. 1989. Speaking: From intention to articulation. Cambridge, Mass.: MIT Press. Natke, Ulrich & Karl T. Kalveram. In press. "Fundamental frequency and vowel du­ ration under frequency shifted auditory feedback in stuttering and nonstuttering adults". Proceedings of the Third World Congress on Fluency Disorders, 7-11 August 2000, Nyborg, Denmark.

EXPLAN APPLIED TO THE TREATMENT OF STUTTERING

115

Newman, Linda L. 1987. "The Effects of punishment of repetitions and the acquisi­ tion of 'Stutter-Like' behaviors in normal speakers". Journal of Fluency Disor­ ders 12.51-62. Onslow, Mark, Leanne Costa & Stephen Rue. 1990. "Direct early intervention with stuttering". Journal of Speech and Hearing Disorders 55.405-416 Perkins, William H., Ray Kent & Richard Curie. 1991. "A theory of neuropsycholinguistic function in stuttering". Journal of Speech and Hearing Research 34.734-752. Postma, Albert. 2000. "Detection of errors during speech production: A review of speech monitoring models". Cognition 77.97-131. Reed, Phil & Peter Howell. 2001. "Presentation of frequency-shifted feedback to sustain long-term improvements in fluency in people who stutter". European Journal of Analysis of Behaviour 1.89-106. Rosen, Stuart & Peter Howell. 1981. "Plucks and bows are not categorically per­ ceived". Perception and Psychophysics 30.1256-1260. Selkirk, Elisabeth. 1984. Phonology and Syntax: The relation between sound and structure. Cambridge, Mass.: MIT Press. Starkweather,  Woodruff. 1985. "The development of fluency in normal children". Stuttering therapy: Prevention and intervention with children ed. by Hugo Gregory, 9-42. Memphis, TN: Speech Foundation of America. von Bekesy, Georg. 1960. Experiments in hearing. New York: McGraw Hill. Watkins, Anthony J. 1992. "Perceptual compensation for the effects of reverberation on amplitude Envelopes - Cues to the Slay-Splay distinction". Proceedings of the Institute of A coustics 14.125-132. Webster, Ronald L. & Michael F. Dorman. 1970. "Decreases in stuttering frequency as a function of continuous and contingent forms of auditory masking". Journal of Speech and Hearing Research 14.307-311. Wing, Alan M. 1980. "The long and the short of timing in response sequences". Tu­ torials in motor behavior ed. by George E. Stelmach & Jean Requin, 469-486. Amsterdam: North Holland. Wing, Alan M. & Alfred B. Kristofferson. 1973. "Response delays and the timing of discrete motor responses". Perception & Psychophysics 14.5-12. Wingate, Marcel E. 1976. Stuttering: Theory and treatment. New York: IrvingtonWiley.

III. MORPHOLOGY AND SYNTAX IN CHILD LANGUAGE DISORDERS

VERB MOVEMENT AND FINITENESS IN LANGUAGE IMPAIRMENT AND LANGUAGE DEVELOPMENT ROELIEN BASTIAANSE, GERARD BOL, SOFIE VAN MOL & SHALOM ZUCKERMAN University of Groningen 1.

Introduction Both normally developing children and children with Specific Language Impairment (SLl) have problems with the production of finite verbs (e.g., Leonard 1998; Wexler et al. in press). The same phenomenon has been de­ scribed for agrammatic aphasic patients (e.g., Saffran et al. 1989; Thompson et al. 1995; Bastiaanse et al. 1995). In the spontaneous speech of these popula­ tions many non-finite verbs are used, usually in combination with a restricted number of finite verbs. In normally developing children this is called the op­ tional infinitive stage. It has been argued that children with SLI show a delay in the development of language skills and therefore the optional infinitive stage takes longer to disappear, if it disappears at all. Rice and Wexler (1996) have named this period in specific language impairment the extended optional infinitive stage. Several authors in the past and present suggested that language loss due to brain damage is characterized by a pattern that is the reverse of language ac­ quisition (e.g., Jakobson 1941; Grodzinsky 1990; Avrutin 1999). At first sight, the characteristics of verbs produced by agrammatic aphasics in spontaneous speech show similarities with those in children in the optional infinitive stage. Agrammatic aphasies produce a significantly diminished number of finite verbs (Thompson et al. 1995; Bastiaanse & Jonkers 1998) compared to non-brain­ damaged speakers, but still produce a considerable number of finite verbs. The present study focuses on two central questions: what is the origin of the problems with finite verbs in these three populations and is it true that children with SLI deal the same way with the problems of the production of finite verbs as normal children do. We will first give a short description of relevant linguistic aspects of Dutch, followed by a summary of the empirical findings in the three populations. Then a test will be presented, meant to tear apart the aspects that are related to finite verbs. The results of testing the three populations will be given and discussed.

120

BASTIAANSE, BOL, VAN MOL & ZUCKERMAN

1.1 Linguistic background Dutch has been analyzed as an sov-language, meaning that the basegenerated position of the verb is after the object (Koster 1975). In the Dutch declarative matrix sentence, the finite verb has to be moved to second position. This movement is known as Verb Second. If the main verb clusters with a modal verb or auxiliary, the main verb remains in situ and the modal verb or auxiliary is moved to Verb Second position. In embedded clauses, the finite verb remains in its base-generated position. This is illustrated in (1-3), where t designates the canonical verb position, which is co-indexed with the Verb Second position. (1)

(2)

(3)

matrix sentence without modal verb/auxiliary de jongen koopti een fiets t¡ the boy buys a bike matrix sentence with modal verb de jongen wili¿ een fiets kopen

ti

the boy wants a bike (to) buy "The boy wants to buy a bike." embedded clause without modal verb/auxiliary (ik denk) dat de jongen een fiets koopt (I think) that the boy a bike buys "I think that the boy buys a bike."

Here it is assumed that a verb is inserted in the structure in fully inflected form (Chomsky 1995; for Dutch, see Zwart 1993). In the embedded clause, the finite verb remains in its base-generated position; in matrix clauses, only the finite verb moves to the left, the nonfinite verb stays behind. Hence, in the matrix clause the finite verb has to fulfil two requirements: it has to move to the second position and it has to check its features. 1.2 Psycho- and neurolinguistic background The speech of young normally developing Dutch children (around the age of 2) is characterized by the use of uninflected verbs (Bol & Kuiken 1990). In Dutch, these verbs are in clause final position, as in (4)-(5). This implies that these verbs are not only non-finite, but also non-moved. This phenomenon oc­ curs both with objects (4) and with subjects (5): (4) (5)

koekje eten cookie eat-lNF papa bouwen daddy build-lNF

VERB MOVEMENT AND FINITENESS

121

In the optional infinitive stage, the children also use a restricted number of finite verbs, usually used as dummy auxiliaries. In Dutch, the third person sin­ gular form of gaan (to go) is often used by children (see, for example, Jordens 1990), as shown in (6). (6)

auto gaat vallen car goes fall-lNF "the car falls" This sentence type is regular in adult Dutch, but it has an inchoative meaning ("is going to"). Children in the optional infinitive stage, however, use this dummy auxiliary for the present tense as well. Other finite verb forms that are used by children in the optional infinitive stage are the copula is and the empty element doet (does). It has been suggested by De Haan (1987) that these dummy auxiliaries which are produced in this stage of acquisition stage are base-generated in Verb-Second position and, hence, no movement is involved. Rice & Wexler (1996) proposed that the problems with the production of finite verbs encountered by children with SLI are caused by a similar mecha­ nism, as shown by their label extended optional infinitive stage. Bastiaanse & Bol (2001) also showed similarities with respect to the production in finite verbs in the two groups of children. The proportion of finite verbs produced by nor­ mally developing children is indeed the same as the proportion produced by children with SLI with the same 'mean length of utterance'. They also demon­ strated, however, that the variation in the proportion of finite verbs in children with SLI is dependent on the diversity of the lexical verbs they produce. The children with SLI can be divided into two subgroups. One group that produces a relatively normal proportion of finite verbs, combined with a relatively low di­ versity of the produced lexical verbs and one group that produces relatively few finite verbs but has an adult-like variation of lexical verbs. This reverse relation­ ship does not appear in the speech production of normally developing children, but there is a similarity with verb production in agrammatic aphasics. Brain­ damaged people suffering from this aphasia type, also known as Broca's apha­ sia, produce a relatively large number of non-finite verbs (both infinitives - to refer to the present - and participles - to refer to the past), as shown in (7)-(8). (7)

(8)

moeilijke zinnetjes vormen difficult sentences form-lNF "to form difficult sentences" na de oorlog gebouwd after the war built-PART "built after the war"

122

BASTIAANSE, BOL, VAN MOL & ZUCKERMAN

Just like in children with SLI, the proportion of finite verbs is reversibly related to the variety of lexical verbs. Bastiaanse & Bol (2001) interpret these data as an inability of languageimpaired people to integrate lexical-semantic and morpho-syntactic informa­ tion, due to a lack of resources needed for complex linguistic operations. In this respect, language impaired individuals differ from children who are acquiring language in a normal way: although the production of finite verbs is difficult to learn, it does not interfere with the retrieval of lexical elements (verbs) from the lexicon. Hence, the similarity between these three (Dutch) populations is the rela­ tive large number of non-finite clauses in their spontaneous speech. One may wonder what makes finite verbs difficult for them. As mentioned before, in order to produce a finite verb in a Dutch matrix verb, two operations must be performed: a morphological operation, that is verb inflection, and a syntactic operation, that is, verb movement. The central question is at makes the pro­ duction of finite verbs in Dutch difficult for these speakers. Is it the morpho­ logical operation (verb inflection) or the syntactic operation (movement)? Ac­ cording to Bastiaanse & Van Zonneveld (1998), for agrammatic aphasics it is movement that causes the problems. In the present paper it will be tested whether this hypothesis can be supported and whether the syntactic account also holds for the two other populations. In order to find an answer to these questions, a test was developed in which finite verbs were elicited in both base-generated position (in the embedded clause) and Verb Second position (in the matrix clause). If the speakers perform at the same level in the matrix and the embedded clause, a morphological account is most likely: they cannot produce finite verbs properly, regardless of movement. If, however, the production of finite verbs in the base-generated position is easier than in Verb-Second position, then support is found for the hypothesis that it is verb movement that is the issue at stake and hence a syn­ tactic operation is causing the difficulties with finite verbs. 2.

Methods

2.1 Subjects Three different groups were tested: 10 children with a normally developing language system (age 3;0-3;11), 10 children with a specific (grammatical) language impairment (mean age 6;2; range 4;10-6;11), as diagnosed by their speech therapists and 6 agrammatic Broca's aphasics (mean age 46.5, range 27-67), for which the aphasia type was established by the Aachen Aphasia test (Graetz et al. 1992) and confirmed by both the

VERB MOVEMENT AND FINITENESS

123

speech therapist and the examiner.1 All patients were right-handed and aphasic due to a single stroke in the left hemisphere, at least 3 months prior to onset (mean time post-onset 35.2 months; range 3-66 months). The choice for normally developing children of this age was motivated by the fact that this is about the youngest age band that can reliably be tested with a language task. The children with SLI were children who were available for testing. One more child participated, but was later excluded from the group, because he was much older (11 years) than the rest of the group. Fourteen older children (age 4;8-5;0) were tested as a control group. 2.2 Materials The participants were presented with two pictures in which the same per­ son was performing the same action with a different person or object (see Figure 1). They were asked to complete a sentence with a finite verb and an object. There were two conditions: 1) a matrix clause that should be comple­ ted with a finite verb - object, 2) an embedded clause with the intended answer object -finite verb. There were 17 sentences in each condition; two of these were used as training items. Two examples are: Condition 1: matrix clause (+movement) Tester: Dit is de man die de tomaat snijdt en dit is de man die het brood snijdt. Dus deze man snijdt de tomaat en deze man "This is the man that the tomato cuts and this is the man that the bread cuts. So, this man cuts the tomato and this man " Participant: "snijdt het brood" "cuts the bread" Condition 2: embedded clause (-movement) Tester: Deze man snijdt de tomaat en deze man snijdt het brood. Dus dit is de man die de tomaat snijdt en dit is de man die "This man cuts the bread and this man cuts the tomato. So this is the man that the bread cuts and this is the man that " Participant: "het brood snijdt" "the bread cuts"

1

Not all children with a normally developing language system finished the entire test. The minimum number of items required for inclusion in the analysis was 16 (50% of the items, equally divided over matrix and embedded clause).

124

ASTIAANSE, BOL, VAN MOL & ZUCKERMAN

The reason to prompt the participants with the verb and the noun several times before s/he was supposed to produce it, was to avoid interference with word-finding problems. Each set of pictures was used twice, once with a ma­ trix clause and once with an embedded clause to complete. The test started with two examples that were repeated until it was clear that the subject under­ stood the task.

Fig. 1: An example of the pictures of the test

2.3 Scoring A correct/incorrect scoring system was used, in which determiner omission was ignored. Self-corrections were allowed and the final answer was the one that was analysed. If requested, the examiner repeated the cueing sentences one time. Post-hoc, an error analysis system was developed on the basis of the most frequently occurring errors: auxiliary insertion, verb omission, word order errors, and a class 'others', which contained semantic paraphasias, nil reactions etc. 3. Quantitative analysis The older children who were used as control subjects performed perfectly in both conditions. Their results will further be ignored. The quantitative re­ sults are given in Table 1. Percentages are used, since not all normally developing children completed the test (see note 1) matrix clause

embedded clause

57

82

normally developing children 49

89

children with SLI

84

! agrammatic aphasics

58

Table 1: Percentages correct for the three populations.

125

VERB MOVEMENT AND FINITENESS

Each population performs significantly better on the embedded clauses than on the matrix clauses (normally developing children t(9) = 6.76, p cvc, CV. However, he used selected aspects of the noun class and agreement system productively. At 3;7 Sipho's speech showed considerable develop­ ment. He initiated some conversation and talked well on his own, although he tended to go off on his own tangent. He used several Tense Aspect Markers (TAMS), alternated OMS with full noun objects and used simple sentences occasionally with more complex semantic intention. His speech showed a number of non-standard complex verbals. His social interaction was more normal than it had been at 2;7, a development that, together with his speech, initially led to an interpretation that Sipho was at the lower end of normal language development, and that his social isolation contributed to his delayed speech. However, comparing his speech with that of an older language impaired child, revealed similarities between the two children's speech. The similarity strengthened the interpretation of his speech as language impaired (Suzman & Tshabalala submitted). Nompumelelo was a 5;6-year-old girl who had been adopted into her family (parents and children) at 8 months. She was a lively, interactive, outgoing child who enthusiastically entered into all activities. She had a history of minor medical problems, drooling, ear infections, and delayed perceptual and

160

SUSAN M. SUZMAN

motor development. She was brought to the Department of Speech Pathology and Audiology at the University of the Witwatersrand because of her parents' concern with her language development in light of her imminent school atten­ dance. She was diagnosed as having immature speech with some phonological and morphological errors, word-finding difficulties and poor topic mainte­ nance. No neurologically-triggered language problems were found and her hearing was normal. Nompumelelo's speech was voluble, at times incoherent. In her enthusiastic responses, she became quite excited and talked very fast, sometimes stuttering and repeating partial utterances. This was most noticeable in talking on her own and most controlled in ordinary conversa­ tional interaction. She told a structured story, but adhered less to the story line than normal children. Comparison between the two children was partial, given their personality and age differences. Nompumelelo was older and had greater language com­ petence than Sipho. An earlier comparison of Sipho with an age-matched nor­ mal child (Demuth & Suzman 1997) showed that his speech was restricted compared to normal development. Impressionistically, Sipho's speech was re­ stricted, whereas Nompumelelo's was disfluent. According to Suzman & Tshabalala (2000), however, shared properties of the children's speech may allow to identify a syndrome of language impairment in Zulu. Data collection procedures were similar, naturalistic, unstructured conversational interactions between the children and their caretakers and/or the researcher. In addition, Nompumulelo was given a story-telling task. Conversations were tape re­ corded and transcribed. Representative samples of the two children's speech are given in Appendix B. 4.

Analysis Word and morpheme Mean Length of Utterance (MLU) were calculated for the two children and a younger normal control, Nqoba, a child of 3;5 years of age with a loquacious speech style like Nompumelelo. Comparisons were made of 55 word speech samples. MLU

Word Morpheme

Nompumelelo Language-delayed girl, 5;6 years. Narrative 3.87 (213/55) 7.50 (412/55)

Sipho Language-delayed boy 3;7 years. Conversation 1.73 (95/55) 4.2 (231/55)

Table 3 : Word and morpheme ML U counts

Nqoba Normal girl, 3;5 years. Conversation 3.1 7.08

(176/56) (397/56)

161

MORPHOLOGICAL ACCESSIBILITY IN ZULU

Table 3 shows that Sipho had the lowest MLU of the children, slightly un­ der 2 words. His non-imitative speech was comparable to normal children between 2-2;6 years both in length and content. He asked simple questions, and his complex verbals typically contained 3 morphemes. As will be seen, the absence of certain types of morphology in his speech might be due to length limitations on what he said. This was not the case with Nompumelelo, whose word and morpheme MLUs were comparable to the normal younger child, Nqoba. Low MLU on its own provides a diagnostic of language disability (Leonard 1998), but these case studies suggest that low MLU may be optionally associated with language impairment but is not diagnostic in Zulu. More case studies are needed to test the importance of the MLU findings. 4.1 Lexicon The results of a comparison of the number and frequency of words in the major word categories, noun and verb, are in Table 4. Verbs Nouns

Nompumelelo .387 (24/62) .429 (26/61)

.44 .50

Sipho (20/45) (15/30)

.620 .607

Nqoba (36/58) (31/51)

Table 4: Type-token ratios in major lexical classes

The language-delayed children had less elaborate and more repetitious vo­ cabularies than the normal child Nqoba, as the lower type/token ratios of these children indicate. The type-token ratio for verbs is particularly interesting in this respect, indicating that whatever the topic, the normal child had access to greater resources for describing it. However, more data are also needed to estimate the importance of this finding. Nompumelelo was telling a story in­ volving a single actor engaged in somewhat similar activities, a factor that may account for her type/token ratio. Sipho had restricted speech that may be partly a consequence of being shy and living in an isolated environment. 4.2 Morphology The children varied considerably in the extent to which they used mor­ phology, due partly to their age differences. Sipho's spontaneous speech re­ vealed a restricted but productive NC and agreement system, NCs 1 and 9, and their respective SMs at age 2;7 with some evidence of development at 3;7, as shown below in Table 5. He over-generalized rather than omitted morphologi­ cal markers, a finding consistent with younger normal Zulu children (Suzman 1991) and typical of languages with obligatory morphology (Leonard 1998). Naturalistic data do not preclude a more extensive morphology for Sipho than that represented in his speech samples.

162

SUSAN M. SUZMAN

Marker Totals NCPrefix Poss ADJ REL

SM OM

DP

Nompumelelo, 5;6 144 (54) 38% 6% (9) 2% (3) (1) — (54) 38% 2% (3) 14% (20)

Sipho, 3;7 75 33% (25) .04% (3)

— —

— —

57%

(43)

—

(1) (3)

.04%

Nqoba, 3;5

|

150 (56)

37% 5% 3% 3% 39% 14% 13%

(8) (4) (4) (26) (21) (19)

Table 5: Noun class and agreement morphemes

Profiles of the older language-delayed child and of the younger normal control were similar with the exception of the use of OMs and RELs. Neither language-delayed child used either of these markers with any frequency, but otherwise their speech provided evidence of productive NC and agreement morphology. Table 6 summarizes verb morphology used by the three children. Tense Present yaPast-e, -ile Future yoRemote Conting remote Totals Agglutinative Verb part. SeVerb extensions -is -el -iw

Nompumelelo 12 1 23 13 49

Sipho 11 24 1 5 1 42

Nqoba 11 6 6 11

1

—

1

—

1 1

2 2 2

1

2

7

|

34

-ek | Totals

Table 6: Frequency counts of verbal morphology

The profiles of tenses and optional agglutinative morphology are quite dif­ ferent. All three children used a range of tenses and rather fewer agglutinative morphemes. The actual tenses they used reflected what they were talking about. In Nompumelelo's case, she was telling a story and using the Remote

MORPHOLOGICAL ACCESSIBILITY IN ZULU

163

Past tense appropriate for relating a story that had happened some time ago. Sipho's samples included an account of another child's bad experience that he told in several versions in the past tense. Nqoba used mainly the present tense to talk about ongoing activities in the immediate context. More interesting is the difference in use of verb extensions. Although numbers are low, the normal child produced several verb extensions lacking in the speech of the language-delayed children. A larger speech sample con­ firmed her productive use of these morphological devices for extending argu­ ment structure of sentences. Summarizing morphological differences, all children used nominal and verbal morphology. In some respects, children's morphological profiles were very similar, namely in their use of NC prefixes, SMs and DPS. However, results showed differences between normal and language impaired children in OMs, RELs, ADJs and Verb Extensions, morphemes that were both optional and contribute to conceptual complexity in children's speech. 4.3 Syntax With two exceptions of infinitives, Sipho's utterances at 3;7 were simple sentences. The majority were Verb-Subject constructions, e.g., u-ya-dlala um-fana SMl-PRES-play NC 1-boy "he plays, the boy does" and transitive sentences, e.g., um-fana u-dlala i-bhola NCl-boy SMl-play NC5-ball "the boy plays ball". Sipho formed new sentence types building upon early, learned forms that were reported on in Demuth & Suzman (1997). At 2;7 he used frequently iya- and a verb or predicate. Although iya- in Zulu is a SM9-Present tense construction, he used it as an unanalyzed form to create new sentences as in (2). (2)

Sipho a. iya-phuza ???-drink b. iya-m-sula ???-OMl-wipe  iya-phansi d. iya-baba

target ngi-ya-phuz-a 1 sg.-PRES-drink-FV ngi-ya-yi-sul-a lsg-PRES-OM9-V-FV ngi-phansi ngu-baba

Gloss "I drink" "I wipe it (floor)" "I am (sitting) down" "It is Father"

Sipho presents atypical complex verbals. Forms of iya persist in sentences at 3;7 years, as in (3).

164

SUSAN M. SUZMAN (3)

Sipho Hamb -a a. Ayiya-FV NEG- SM9- PRES- go "He doesn't go" b. Ayishek -a NEG- SM9- shake -FV "She (mother) doesn't shake"  U afun' u-ngen-a SMl- PRES- want INF-enter-FV "She wants to enter" d. Iy a- phuz' ama-nzi SM9- PRES- drink NC6-water "It drinks water" e. Iya- dl -e SM9- PRES eat -PAST "She ate"

Target Ayi-

hamb

NEG- SM- go

A-

y i-

s hek

NEG- S M 9 - shake

-i -NEG

-i -NEG

U- fun ' uku-ngen-a SM 1 - want' INF-enter-FV Iphuz' ama-nzi SM9- drink NC6-water /dl -He SM9- eat -PAST

These verbs share the elements SM-ya-V, e.g., i-ya-hamba "he goes", one of the first simple sentence types seen in young children's speech. Sipho used the present tense in his basic structure in the negative (3a), the infinitive (3c), the transitive (3d), and the past (3e), none of which have the present tense marker in Zulu. This type of error occurs occasionally in young normal children's speech between 18 months and 2;4 years but Sipho's speech provided evidence of prolonged reliance on early forms in formulating new sentences. As stated earlier, normal Zulu- and SeSotho-speaking children produce passives as early as 2;6 years (Demuth 1984, Suzman 1991). However, the passive is rarely used by the two language impaired children in this study. Sipho had the concept, word order differences, but lacked the appropriate passive verb morphology, as it appears in (4). (4)

Sipho talked about someone being bitten by the toy horse and fish Sipho Target a. U- lum -ile i-hhashi U-luny-w-e -i-hhashi SMl- bite -PAST NC5-horse SMl-bite-PASS-PAST COP-NC5-horse "He bit, the horse did" "He was bitten by the horse" b. U- lum -ile i-fish U-luny-w-e y-i-fish SMl- bite -PAST NC5-fish SMl-bite-P ASS-PAST COP-NC5-fish "He bit, the fish did" "He was bitten by the fish"  Iya- hamb -a i-hhashi SM9- PRES- go -FV NC5/9-horse "He goes, the horse does"

MORPHOLOGICAL ACCESSIBILITY IN ZULU

165

Sipho did not use the passive verb extension -w. However, the non-linguistic context and the morphological contrast between (4a) and (4b), and (4c) show that (4a) and (4b) have passive meaning for him. When he talked about the person (object) being bitten by the horse, the SM agreed with the object (SM1). When he talked about "the horse (subject) going" (4c), the SM agreed with the subject (SM5/9). Nompumelelo also used primarily simple sentences, but there were many instances of complex sentences in a larger data sample (5). The most frequent of these were compound sentences linked by "and then" (5a) and serial con­ structions (5b). Like Sipho, she relied on a few learned forms, producing not so much deviant forms, but simple svo sentences strung together with or without a fixed conjunction. Other complex sentence types occurred infrequently and are illustrated in (5c)-(5f). The absence of these sentences distinguished Nompumelelo's speech from younger normal children who use infinitives between 2 - 2;6 years and other complex sentences between 2;6 and 3 years. (5)

Compound sentences with aqeda lapho "and then" a. Aceda lapho wayopheka waceda lapho wayofun 'khaya "And then she went to cook and then she was looking for a home" Serial constructions used productively b. Abantwana baye bahamba no Mama "The children went and walked with Mother" Infinitives c. Waceda lapho aceda wayofun ' u-casha la (waqeda lapho aqeda wayofun' uku-casha la) "And then, and then, she was looking to hide there" Consecutive actions marked by present subjunctive suffix -e d. Bamshaya akhal -e they hit her and Subjunctla- cry -subjunct "They hit her and she cried" Concurrent actions marked by participial prefix (irregular) on Noun classes la, 2, and 6. e. a- buy -e bekhal -a SM2- Return -PAST PART2- cry -FV "They returned crying" Relative clause f. Lo-mama wa-shäy-a lo o-gqcok' im-pahla zakhe DP1-mother REM1-hit-FV DP1 REL1-dress' NC1lO-clothes POSSlO- 3PS "This mother hit this one who dressed in her clothes"

166

SUSAN M. SUZMAN

The absence of the relative marker was further probed in two structured tasks, both having picture recognition of relativized complex sentences, instructions and paraphrases. (6)

Instruction a. Ngi-bon-is-e in-tombazane e-khomb-a phezulu lsgOM-see-CAUS-subjunct NC9-girl REL9-point-FV above "Show me the girl who points upwards" Paraphrase b. Na-yi. Um-ntwana lo ukhomb-a eceleni. Here-SM9. NCl-child DPI SMl- point-FV on the side. "Here she is. This child points on the side" Nompumelelo rephrased the relative clause as two simple sentences. Results on 25 input relative clauses are given in Table 7. While the child did use the relative marker, the majority of the paraphrases are simply simple sentences. Total input sentences

Responses: 2 simple sentences

25

Responses: Relative clauses 8 (36%)

16(64%) Table 7: Paraphrases of relative clauses

The weakness in relativizing was further demonstrated in a repetition task (Table 8) where the child demonstrated less ability in repeating this type of sentence than other sentence types. Utterance type Verbal complexes Relatived S V + Caus/Bene N + Adjective

Accurate 13 100% 3 43% 5 100% 5 100%

Errors

— 4 SS (57%)

— —

Total 13 7 5 5

Table 8: Repetition task

Nompumelelo did not use relative clause markers in the majority of the rela­ tive sentences, although, interestingly, she had no difficulty with adjective modifiers that had a similar or identical form. Normal children tend to learn adjectives as part of rote-learned expressions but relatives are rule-governed. These children used early, learned forms in new sentences, producing non­ standard forms or simple sentences rather than complex ones. They did not appear to recognize the role played by certain morphological markers in altering verb arguments (Sipho's passive) and in sentence embedding (relatives). The use of familiar and stereotyped structures is common in language delay. The

167

MORPHOLOGICAL ACCESSIBILITY IN ZULU

language-specific consequence in Zulu was compromised development of morphologically expressed complex syntax. Table 9 summarizes differential access to Zulu morphology. Grammatical Morphology Selected NCs SMs Adjs Tense OMs Vextensions Relative clause markers [ Complex Sentence morphology

Normal X X X X X X X X

Zulu Impaired X X X X Rare Rare Rare Rare

|

Table 9: Normal and language impaired morphology

In Table 9, tenses and some NC morphology were used by all children. Less robust morphology included OMs, VExtensions and complex sentence markers (RELs, participials and subjunctives). The OM's absence may be explained by its lack of salience (SM-(Tns)-OM-V) preceding the verb, comparable to Italian object clitic pronouns (e.g., e poi lo hacia "and then kisses him"; Leonard 1998:92) that are also omitted. The other missing morphemes are associated with the expression of more complex propositions, being either verb extensions or morphemes that embed a subordinate sentence within a complex sentence. Summarizing morphology and syntax results, language impaired children have productive noun class and agreement morphology, but lack many grammatical morphemes signaling structural or conceptual complexity. Their absence results in prolonged reliance on early, structurally simple constructions and avoidance or lack of perception of morphological markers of syntactic complexity. 5. Discussion Grammatical morphology in English involves a handful of morphemes while in Zulu it consists of substantial numbers of grammatical and agglutina­ tive morphemes that combine in extremely varied ways. Pinker (1994) estimates the possible combinations of morphemes on an 8-morpheme verb in Kivunjo, another Bantu language, as approximately half a million. A similar projection can be made for the African languages of Southern Africa. Abundant mor­ phology is predicted to both focus children's attention on morphology and to present learning problems. Research bears out these predictions. Children have

168

SUSAN M. SUZMAN

access to some NCs and agreement markers and have problems with agglutina­ tive, anaphoric morphology. Normal children develop NCs and agreement before verb extensions and OMs (Suzman 1991). Language impaired children use NCs and agreement but generally omit agglutinative morphemes. Aphasics make more mistakes on agglutinative morphology than on NC prefixes and agreement (Herbert 1992). Some selective access to Zulu morphology cleaves along the lines of Herbert's distinction between obligatory and optional prefixes in Afri­ can languages. Based on traditional discussions of Bantu morphology, he notes that NCs and agreement prefixes are obligatory. They do not contribute meaning to nouns and agreeing constituents and their omission generally produces ungrammatical utterances. Verb anaphors (and verb extensions) are optional; they contribute meaning to the sentence and their omission produces incorrect but grammatical sentences. Herbert observes that the two sets of morphemes are not fully comparable. Their lack of comparability is reflected in different patterns of acquisition and breakdown. Acquisition sequences, persistent omissions in lan­ guage delay and difference in type and rate of error by aphasics lend credence to the obligatory/optional distinction. It predicts early acquisition of NC prefixes and SMs compared to OMs. While OMs belong to obligatory morphology, the NC and agreement system, they behave like optional morphemes. Besides being op­ tional, they contribute meaning to an utterance and their absence does not pro­ duce an ungrammatical utterance. The following examples illustrate. The OM is a pronominal object when used alone (u-ya-ba-bona SMl-PRES-OM2-see "he sees them") and a emphasis or definiteness marker when used with a full noun object (u-ya-ba-bona abantwana SM1-PRES-OM2 -see Nc2-child "he sees the children" vs. u-bona abantwana "he sees children". The OM over-generalizes like the SM and is acquired after the NC prefix OM and Poss by normal children (Suzman 1991). It is used infrequently by language impaired children and omitted in repetition tasks by aphasies. Suzman (1991) proposed that SMs and OMs over-generalize because they are learned first as general referential pro­ nouns. Subsequently, the SM becomes part of the obligatory prefix system whereas the OM remains an optional pronominal object. The OM is acquired later because it is less likely to occur. As mentioned earlier, its word position also makes it less perceptually salient than the SM. The relative marker is another exception to the predictions made by the obligatory/optional distinction. Al­ though this issue needs more investigation, its absence in language impairment appears to be associated with sentence embedding. Relative markers on relative stems that function like adjectives tend to be rote-learned (in frequent modifying expressions) in normal acquisition (Suzman 1991). RELs in relative clauses align with VExtensions that add semantic complexity to the sentence. While

MORPHOLOGICAL ACCESSIBILITY IN ZULU

169

VExtensions are not prefixes, they are meaningful and optional and profile like OMs and RELs in delayed acquisition. While the issues are complex, the functional status of morphemes within Zulu partly accounts for patterns of acquisition. Comparability of morphemes on a functional basis may be of relevance cross-linguistically. Superficially, Zulu and English findings are conflicting. From the study of plurals and agree­ ment in English SLI children, Rice & Oetting (1993) proposed that nominal morphology is easier to learn than verbal morphology for children. According to the Zulu data discussed in this paper, nominal and verbal morphology are equally accessible. There is no evidence that there is any difference in acquisi­ tion difficulty between nominal (NC) and verbal (SM) inflection. The fact that the NC prefix is more likely to be omitted than the SM in young children's speech might suggest that verbal morphology is easier than nominal morphology. How­ ever, the invariable inclusion of the NC prefix on nouns following the verb, e.g., bhek'umu-ntu look at'NCl-person "look at the person", indicates that processing factors account for omission of the NC prefix. Ric, Wexler & Cleaves's (1995) proposal that late acquisition of tense, but not agreement, is diagnostic for SLI children is not supported by the Zulu study. Normal and language-delayed chil­ dren have little difficulty with tense, a morpheme that profiles like obligatory morphemes. There is no period of uninflected verb usage by Zulu children com­ parable to the Optional Extended Infinitive (OEl) period found in English SLI children's speech. The lack of difficulty of number, tense, and agreement displayed by Zulu language-impaired children contrasts with the persistently problematic nature of these morphemes and categories in English. The obliga­ tory/optional distinction would predict late acquisition of English grammatical morphology. Concepts like English agreement profile more like optional mor­ phology than Zulu agreement does. It does not have the same morphological status as Zulu agreement. The comparison of Zulu and English suggests that there are language-specific differences in the organization and representation of functional categories. If they do not have the same function and status across languages, they may not be directly comparable. Other aspects of NCs and agreement to consider are its systematicity and its alliterative nature. Zulu NCs and agreement prefixes are also systemic; they are integrated in an exceptionally regular, pervasive, 'closed' system (Levy 1996). She attributes the surprisingly early acquisition of morphological systems like gender in Hebrew, English, German, Polish, French, and Russian, to their being closed, their rules requiring no input from other systems. Further, the system is variably alliterative. Evi­ dence for the significance of this factor is relatively minor to date. Herbert (1992) found that aphasics could repeat agreement morphemes more accurately

170

SUSAN M. SUZMAN

when utterances contained full noun subjects than when had pronominal subjects, e.g., isilwane si-ba-lum-ile NC7-animal SM7-OM2-bite-Past "The ani­ mal bit them" versus siba-lumile SM7-OM2-bite-Past "It bit them". Normal chil­ dren also provide more accurate SMs in conversations with nominal rather than pronominal discourse-established topics. Languages like Zulu hold research potential in the study of homonyms, morphophonological patterning, contrasts (Carstairs-McCarthy 1994), and other system-internal features in acquisition and breakdown. For example, Nompumelelo employs adjective agreement with adjective stems, but does not use the morphologically similar relative agreement markers with relative clauses (see Appendix A for a comparison of AD J and REL forms). Further re­ search will investigate the relative contribution of factors identified in the discussion above, semantics, frequency, systemic and alliterative properties, in explaining patterns of acquisition and breakdown. Beyond morphological issues lies the general concern with explaining dif­ ferent acquisition patterns in normal and delayed language acquisition. In Pinker's (1984) dual mechanism model of acquisition, rule and associative learning dependent on input frequency account for language acquisition. Oetting & Rice's (1993) observation of increased reliance on rote-learning in SLI is supported in the Zulu data. As discussed above, Sipho's and Nompumelelo's speech samples provide ample evidence of reliance on learned forms and lack of progressive restructuring of their grammars along the lines of the adult language. Predominantly simple sentences, often stereotyped, and/or non­ standard sentences, are characteristic. 6.

Conclusions This study of Zulu language impairment confirms findings about how lan­ guage impaired children learn language. It also supports findings of differential access to morphology. Its findings do not support more specific proposals con­ cerning ease of acquisition of particular functional categories and nominal ver­ sus verbal morphology. It identifies factors that partly predict differential access to morphology in Zulu and suggests that they might be of crosslinguistic rele­ vance in explaining developmental acquisition. The view expressed in this paper is that access to core grammar is mediated by language specific organization and representation of basic concepts. Specifically, the status of functional cate­ gories in particular languages influences their accessibility in acquisition. Or­ ganizational principles are anticipated to be general enough to provide general predictions for acquisition across language types and provide an intermediate level of analysis relating language specific and universal properties of language.

MORPHOLOGICAL ACCESSIBILITY IN ZULU

171

REFERENCES Bishop, Dorothy. 1994. "Grammatical errors in specific language impairment: Com­ petence or performance limitations?". Applied Psycholinguistics 15.507-550. Carstairs-McCarthy, Andrew. 1994. "Inflection classes, gender and the principle of contrast". Language 70.737-788. Connelly, Michael. 1984. Basotho Children's Acquisition of Noun Morphology. Ph.D. dissertation, University of Essex. Demuth, Katherine. 1992. "Acquisition of Sesotho". The Crosslinguistic Study of Language Acquisition ed. by Daniel I. Slobin, vol 3, 557-639. Hillsdale, N.J.: Lawrence Erlbaum Associates. Demuth, Katherine. 1992. "Accessing functional categories in Sesotho: Interactions at the morpho-syntax interface". The Acquisition of Verb Placement. Functional categories and V2 phenomenon in language development ed. by Jürgen Meisel, 83-105. Dordrecht: Kluwer Academic Publishers. Demuth, Katherine & Susan Suzman. 1997. "Language Impairment in Zulu". Boston University Conference on Language Development Conference Proceedings 1, ed. by Elizabeth Hughes, Mary Hughes & Annabel Greenhill, 124-135. Somerville, MA: Cascadilla Press. Doke, Clement L. 1954. Zulu Grammar. Cape Town: Longmans. Herbert, Robert K. 1992. "Patterns in language change, acquisition and dissolution: noun prefixes and concords in Bantu". Anthropological Linguistics 33:2.103-134. Kunene, Euphrasia  1979. The Acquisition of Siswati as a First Language: A mor­ phological study with special reference to noun prefixes, noun classes and some agreement markers Ph.D. dissertation, University of California, Los Angeles. Leonard, Lawrence B. 1992. "Specific language impairment in three languages: some crosslinguistic evidence". Specific Speech and Language Disorders in Children: Correlates, characteristics and outcomes ed. by Paul Fletcher & David Hall, 118-126. London: Whurr Publications. Leonard, Lawrence B. 1998. Specific Language Impairment in Childhood. Cam­ bridge: MIT Press. Levy, Yonata. 1996. "Why are Formal Systems Early to Emerge?". Children's Lan­ guage ed. by Carolyn Johnson & John H. Gilbert, 75-84. Hillsdale, N.J.: Lawrence Erlbaum Associates. Oetting, Janna . & Mabel L. Rice. 1993. "Plural acquisition in children with Specific Language Impairment". Journal of Speech and Hearing Research 36.1236-1248. Pinker, Steven. 1984. Language Learnability and Language Development. Cam­ bridge, Mass: Harvard University Press. Rice, Mabel L. & Janna . Oetting. 1993. "Morphological deficits of children with SLI: Evaluation of number marking and agreement". Journal of Speech and Hearing Research 36.1249-1257. Rice, Mabel L., Kenneth Wexler & Patricia Cleave. 1995. "Specific Language Im­ pairment as a period of Extended Optional Infinitive". Journal of Speech and Hearing Research 38.850-863.

172

SUSAN M. SUZMAN

Suzman, Susan M. 1991. Language Acquisition in Zulu. Ph.D. dissertation, Univer­ sity of the Witwatersrand. Suzman, Susan M. 1995. "The discourse origin of agreement in Zulu". The Complete Linguist. Essays in Honour of Patrick Dickens ed. by Anthony Traill, Rainer Vossen & Megan Biesele, 319-337. Cologne: Rüdiger Koppe. Suzman, Susan M. 1996. "Acquisition of noun class systems in related Bantu lan­ guages". Children's Language ed. by Carolyn Johnson & John H. Gilbert, 87-104. Hillsdale, N.J.: Lawrence Erlbaum Associates. Suzman, Susan M. 1998. "Morphology as a diagnostic of Language Impairment in +/- Pro-drop languages". Paper presented at the Annual Meeting of the Linguistic Society of America, New York, Jan. 1998. Suzman, Susan M. & Busi Tshabalala submitted. "An investigation of language im­ pairment in Zulu". The South African Journal of Communication Disorders. Tsonope, Joseph. 1987. The Acquisition of Setswana Noun Class and Agreement Morphology: With special reference to demonstratives and possessives. Ph.D. dissertation, Buffalo: State University of New York (SUNY). Weimers, William E. 1973. African Languages Structures. Berkeley, Calif: University of California Press. Appendix A: Morphological Structure of Zulu NC 1 la 2 2a 3 4 5 6 7 8 9 10 11 14 15

NCPrefix

Poss

Adj

Rel

SM

OM

DP

umuu-

wawababawayalaasazayazaIwabakwa-

omomabaabaomemieliamaesieziNeNeziNoluobuoku-

0-

uubabauiliasiziizilubuku-

mmbabawuyiliwasiziyizilubuku-

To

aba00-

umimi iamaisiizii(N)izi(N)u:ubuuku-

0-

abaabaoeeliaesiezieezioluobuoku-

lo laba laba lo le leli Iowa lesi lezi le lezi lolu lobu lokhu

MORPHOLOGICAL ACCESSIBILITY IN ZULU

173

Appendix B: Speech Samples of Sipho and Nompumelelo S[ipho], 2;7 years, picks up a small broom and sweeps the floor with G[randmother] and R[ose], 11 years. Target forms are given in parenthesis. S.

R.

R.

S.

I(ngiSM1sgI(ngilsgFaka put Fak' (faka put WSMlNahereNahereNa-y' Here-SM9'

yayaPresfun' fun' want phakathi inside phathi phakathi) inside enza do

myiOM9-

amaama NC6

sula sula) wipe nzi -nzi) water

"I am wiping it (floor)"

"I want water"

"Put (the water) inside" "Put (it) inside"

-n' -what

uNCla-

gogo? grandmother

"Here it (=cow) is"

yi SM9 ' SM9' inNC9-COW

"What is Granny doing?"

inNC9komo

komo cow

"Here is the cow" "Here is the cow"

Sipho, 3;7years Yi-ni COP-what INC9G.

S.

u2sgBaSM2-

S. S. S.

WaREM1WaREMlUSMl-

le? DP9?

mali money zoFutthath take shis ' burn yi-

Bheka lapha. Look here.

"What is this? Look here." "Money"

thenga

-ni

OM9-

buy -what mali -e i-PAST NC9- money imali NC9money shisa burn

shis burn

-il' -PAST

inNC9-

nge-mali? Instr-money le

"What will you buy with the money?" "They took this money"

DP9

le

"He burned this money"

DP9

"He burned it (money)" gubo blanket

"He burned the blanket"

174

SUSAN M. SUZMAN

Nompumelelo, 5;6 years. Excerpt from telling the Vava story. Glosses refer to the child's form, not the correct Zulu form, except when the child's form is incomplete. Ba(Ba-

yaya-

hamba hamba

nna-

abaaba-

ntanaa ntwana

ba ba-

SM2PRESgo "They go with their children"

Prep-

NC2-

child

POSS2- 3PS

Lo ?

Vava

lo

I-

ya-

phuma

DPI

Vava

DPI

SM9-

PRES-

go out

khe)

Iyaphuma la ifasitelo SM9PRESgo out there NC5window "This one? Vava this one. She goes o u t . . . she goes out there, the window" Za-

fika la izi-... (Afika la ama-... REM 10arrive there NC 10... "They arrived there, the mice did"

iiamaNC 10-

gundane gundwane) mouse

Nayi indiu bayahleka. ba... le inje Here- SM9 N C 9 - house S M 2 - PRES- Laugh S M 2 . . . D P 9 N C 9 - dog "Here it is, the house. They are laughing (children) They.. .this one, the dog" I-

hlak'

hlek -el -a (ISM9laugh BENE FV "She laughs with her children"

aba-

ntana

ba

abaNC2-

ntwana child

bakhe?) POSS2- 3PS

LANGUAGE PRODUCTION IN JAPANESE PRESCHOOLERS WITH SPECIFIC LANGUAGE IMPAIRMENT TESTING THEORIES

YUMIKO TANAKA WELTY, JUN WATANABE & LISE MENN* 1. Introduction: Japanese as a test of su theories Over the last two decades of the 20th century, theoretical debates about the innateness of grammar have increasingly focused on certain developmental language disorders in children: those which cannot be ascribed to brain damage or overall cognitive insufficiency. Such disorders are currently referred to as Specific Language Impairment (SLI). Researchers reason that if there is such a thing as innate universal grammar - assumed to consist of specific templates for rules and representations (including grammatical categories) - then Specific Language Impairment might be explained by deficits in such templates (Rice & Wexler 1996). Conversely, if such explanations are indeed adequate to describe the data, this may be taken as support for innate grammar. Some general claims are implicit in these arguments. First is the etiological claim: that SLI is a unitary disorder with a single cause. In order to reason across combined evidence from all cases, this claim is necessary; the alternative is to consider that there may be many subtypes of SLI, with different causes. If there are many subtypes, of course, each would have to be studied separately to see if is compatible with the various proposed explanations. Second, and independent of the etiological claim, is the cross-linguistic claim: that SLI (or the different SLI'S, if there are several) will be underlyingly comparable across languages. If both the etiological claim and the crosslinguistic claim are true, then for an explanation of SLI to be valid, it must be capable of applying to all languages. That is, if both of these claims are true, then a theory which fails to explain SLI in even one language cannot be an adequate theory of SLI. If there are several types of SLI, then of course a particular type might not be apparent in a given language because the structure of the language would not be vulnerable to a particular kind of Yumiko Tanaka Welty, International University of Health and Welfare, Tochigi; Jun Watanabe, Osaka University & Lise Menn, University of Colorado.

176

TANAKA WELTY, WATANABE & MENN

problem. For example, a language which does not have any reduced syllables is not vulnerable to problems with reduced syllables; a language that does not have agreement is not vulnerable to a deficit in agreement. But it is not plausible that a particular type of SLI is restricted to a particular language unless it is linked to a gene with limited population distribution (cf. Tay-Sachs disease), and that population only speaks that language. So it is unlikely to be the case that Japanese children with SLI have a different type of disorder from all other children with SLI. (This could be tested by looking at SLI in children of Japanese descent who speak other languages natively, on the model of medical studies. If there were a Japanese-population variant of a genetic SLI, it should be manifested in a certain percentage of children of Japanese ancestry who grow up speaking other languages, and SLI in those children should look different from SLI in other populations speaking those languages. The only reason to even discuss this unlikely possibility is to make clear what would be implied if someone tried to dismiss results from Japanese SLI as being irrelevant to explanations of SLI in other languages.) In summary, if a general explanation offered for SLI cannot apply to a par­ ticular language, but children speaking that language in fact have SLI, then either they have a different type of SLI, violating the etiological claim, or the explanation is not adequate. This is the theoretical standpoint from which we consider SLI in Japanese. 2. Theories of SLI There is no agreement on an exhaustive listing of the symptoms of SLI, but in Indo-European languages and in Hebrew, many children with specific lan­ guage impairment have difficulties using grammatical morphemes (e.g., in English, forms of the copula 'be', present third singular -s verb inflections, and regular past -ed verb inflections), and they tend to use them less frequently than younger, but linguistically normal children (see Leonard 1998 for a recent review). Hansson, Nettelbladt & Leonard (2000) divide current theories of SLI roughly into two groups. One group assumes that children with SLI have a grammatical deficit: they lack some grammatical knowledge that is essential for normal grammatical functioning. The other assumes that SLI children have a processing deficit: they are capable of grasping all grammatical notions, but have limited processing abilities, which slows their progress in identifying and interpreting morphological information from the input. Most theories of SLI are based on data from English-speaking children with SLI; now, however, other languages, specifically German, Italian, Hebrew, and Swedish, are also providing data to test their validity. In this

LANGUAGE PRODUCTION IN SLI JAPANESE PRESCHOOLERS

177

study, we present data from SLI children who are learning Japanese as their first language. Japanese is morphologically very different from Indo-European languages, for example, it has no infinitives, tense-number agreement, gender marking or articles; it almost has no irregular verbs, and its subject and object case markings, although commonly used, are optional, unless they serve to disambiguate the meaning of a clause. 2.1 Grammar deficit theories One account that assumes a deficit in grammatical knowledge in children with SLI is the Missing Agreement account of Clahsen research, based on evi­ dence from German and English (Clahsen 1989, 1991; Clahsen & Hansen 1993). They claim that SLI children have a selective deficit in establishing structural relationships of agreement (thus, e.g., they may fail to produce cer­ tain subject-verb agreements such as "He runs", but not others such as "They run"). This type of deficit cannot exist in Japanese. The inflectional, agglutinative suffixes in this language mark verbs for tense, aspect, voice, negation, and causation, among other things, but not for person, number or gender. Thus, there is no possibility that Japanese-speaking SLI children could have problems with subject-verb agreement. Another account within the grammatical deficit hypothesis is the Extended Optional Infinitives. This account assumes that children with SLI go through an extended period during which they fail to recognize that verbs must be marked for tense in main clauses (e.g., Rice et al. 1995; Rice & Wexler 1996). In its revised version (Wexler et al. 1998), this account allows for the possibility that these children treat agreement, too, as optional. For example, they may either select an appropriate (correctly marked) agreement and/or tense form or select an infinitive (comparable to the infinitive form in adult sentences such as I saw lohn play baseball). This account is also, apparently, irrelevant to Japanese. In Japanese there are no 'bare' or infinitive forms (although there is an extremely common non-finite form, the conjunctive -te). Every other verb form is neces­ sarily inflected for tense/aspect. Since the te- form is tenseless, it is possible, however, that the Extended Optional Infinitive account could be interpreted as a prediction that it would be optionally used as a main clause verb for an extended period of time by Japanese children with SLI. 2.2 Processing limitations One of the accounts that assumes processing limitations on the part of SLI children is the surface account (e.g., Leonard et al. 1997). According to this theory, SLI children have special difficulties with using the information that is encoded in relatively short grammatical morphemes. While they can perceive

178

TANAKA WELTY, WATANABE & MENN

non-syllabic consonantal inflections, weak syllable inflections, and function words, according to this account, they have a limited processing capacity, so their ability to obtain grammatical information from such short forms at nor­ mal conversational speeds is severely taxed. If grammatical morphemes are optional, these problems are made worse, since children would have fewer op­ portunities to observe their use. Japanese case particles, which mark syntactic relations, are short monosyllabic postpositions, such as  (direct object), ga (subject), and ni (indirect object/locative). However, they are not 'weak' like English or Italian inflections, because Japanese does not have a stress accent system, nor does it reduce the vowels of unaccented syllables (except in spe­ cific phonological conditions that case markers never satisfy). The case parti­ cles are, however, frequently omitted in casual conversations, especially if word order and semantics already permit the hearer to decode the basic gram­ matical relations. (When both subject and object are present but case particles are omitted, Subject-Object-Verb word order is assumed. For more informa­ tion on Japanese grammar, see Appendix). Since particle omission occurs even more in the speech addressed to young children (Rispoli 1996), Japanese children with SLI should lag behind their normally developing counterparts in the use of these particles. Another of the accounts that assumes processing limitations on the part of SLI children is the morphological richness account (e.g., Dromi et al. 1993; Leonard et al. 1987). According to this proposal, children with SLI devote their limited resources to the aspect of grammar that is dominant in a given language. Therefore, SLI children who speak English focus on word order, while those who speak Italian and Hebrew concentrate on inflectional morphology. In Japanese, neither word order nor case particles are 'dominant grammatical features', because they are not used consistently. Word order is relatively free (except that verbs appear in sentence-final position). Within the clause, word order is not contrastive or syntactically controlled, and both subject and object may be omitted when retrievable from the context. The most reliable grammatical feature is verb morphology, since verbs are always inflected for tense/aspect or carry the -te marker. It is also perceptually salient, since finite verbs are clause-final and can be followed only by sen­ tence-final discourse markers such as question particles. Japanese thus offers several advantages as a language for studying SLI. For example, if Japanese children show morphological impairment in SLI, this impairment cannot be due to difficulties with representing person-number agreement, or to using a grammar in which the verb of the main clause may optionally be infinitive. Japanese data could therefore require modifications of these particular

LANGUAGE PRODUCTION IN SLI JAPANESE PRESCHOOLERS

179

hypotheses as general explanations for SLI. The regularity of Japanese also permits the researcher to design test items using very high frequency verbs, without needing to use irregular forms. This makes the design of morphological experiments for very young children simpler than in European languages. However, Japanese also has a design disadvantage: the optionality of subject and object case markings in most contexts complicates error scoring. Designing contexts in which production of these case markings is obligatory continues to be a challenge to the researcher. 3. Typical development of verbs and particles Very young typically developing Japanese children appear to use the nonfinite -te form as their default. However, they soon show control of verbal inflections (Clancy 1985), appearing to do so earlier than children acquiring many other languages. Since either -te or tense/aspect markers are required on every form, this would be predicted by Mac Whinney and Bates' Competition Model (MacWhinney & Bates 1987; MacWhinney 1987), in which children most easily learn the cues to meaning that are the most available and reliable. In addition, because person-number agreement is absent, the finite verb system is less complex than in most other languages whose acquisition has been studied. The basic Japanese noun phrase consists of the unit Noun + Particle. Case particles are syntactic markers, and, as in Indo-European languages, they have varying degrees of correlation with semantic functions. Two of their properties, however, make them unlike familiar case suffixes. The Nominative and Accusative markers can be replaced by the Topic marker wa under appropriate discourse conditions, and, as mentioned above, they can be omitted entirely in colloquial discourse styles. Children attempt to produce the Noun + Particle unit early, even before they have learned the functions of the case particles. Children at this stage apparently realize that nouns are typically followed by particles, but do not know yet what kinds of particles exist, nor which particle to use. This parallels the way that children learning English, French, and many other languages learn that nouns are typically preceded by articles before they learn how to use them (Peters & Menn 1993; Veneziano & Sinclair 1997; Peters 2001). At a more advanced stage, Japanese children make systematic errors in using case particles because of the inconsistency of their semantic function. For example, the subject marking Case Particle ga has a wide range of semantic functions. As shown below, the ga-marked nominal indicates agentive/actor subjects of transitive verbs (la), certain types of objects of stative verbs (lb) or experiencers, others are in construction with intransitive verbs (lb) and (2b), and finally the patient of passive verbs (3).

180

TANAKA WELTY, WATANABE & MENN Agent of transitive verbs: (1) a. Kazuo-ga jusu-o nom-u Kazuo-NOM juice-ACC drink-Non past "Kazuo will drink juice" Agent of intransitive verbs where the subject controls the action: b. Yoshie-ga aruk-u Yoshie-NOM walk-Non past "Yoshie walks" Object (theme) of stative predicates: (2) a. Ringo-ga suk-i apple-NOM like-Non past "(I or some one) like apples" Object (theme) of intransitive verbs where the subject undergoes the action: b. Konoha-ga ochir-u Leaves-NOM fall-Non past "Leaves fall" Patient of passive verbs: (3) Okasan-ga tatak-are-ru Mother-NOM hit-Passive-Non past "The mother is hit"

These three major functions of ga appear at the same time in 2-year-old children (Clancy 1985). However, since ga does not consistently mark agents and is sometimes omitted or replaced by wa (and because the whole subject noun phrase is sometimes omitted), Japanese children have difficulty in acquiring the case marking for transitive events with agents and affected objects, as in (4a). (4) a. Kodomo-ga okasan-o tataku Child-NOM mother-ACC hit-Non past "The child hits the mother" b. Okasan-ga kodomo-ni tatak-are-ru Mother-NOM child-AGT hit-PASS-Non past "The mother is hit by the child" The oblique particle ni also has many semantic functions. It marks agent in passive sentences (AGT) ((4b) above), recipient in benefactor sentences, intermediate agent (causee) in causative sentences, and it also simply means "in'/'on' in locative constructions. Unlike ga and o, it is not optional in any of these uses.

LANGUAGE PRODUCTION IN SLI JAPANESE PRESCHOOLERS

181

The current study is based on the data from a large-scale ongoing study in Japan. Its major objectives are to identify preschool children with SLI, if possi­ ble, and to establish assessment criteria; to begin to characterize the language deficits in Japanese SLI (JSLI); and to contribute to the development of theoretical accounts of bases of SLI in general. 4. Initial study: does J(apanese)SLI exist? 4.1 Procedures: subject identification The identification of SLI children has not been well established in Japan for several reasons: 1) for preschool children (especially boys), not speaking as well as can be expected with age is not perceived as a developmental problem in Japan, as long as they behave appropriately. 2) There is a lack of comprehensive language assessment batteries. 3) The structural characteristics of Japanese - especially the optionality of both subjects and objects - hinder the identification of grammatical deficits. The JSLI identification process began by surveying teachers in regular kindergartens in Osaka, Japan, asking them to list children with whom they were concerned in terms of their language and communicative development in their classrooms. This is an ongoing project, involving the survey of 4 kindergartens with a total enrollment of about 600 children. So far, 34 children ranging in age from 3;10 to 6;4, have been listed as suspects for SLI. After the initial survey, each child was administered the tests for language and cognitive abilities by the first or second author. The tests included the Picture Vocabulary Test (PVT), two subtests of the ITPA (word association and grammatical completion), a subtest (digit span) of the K-ABC, and two of the WIPPSI subtests (picture completion and coding). These tests are standardized in Japanese. The session lasted for about 45 minutes per child. Typical inclusionary and exclusionary criteria were used for identifying SLI subjects: 1) language deficit, defined as -1.25 or more Standard Deviations (SD) below the means of at least two out of three language test scores (the PVT and two subtests of the ITPA); 2) performance IQ not lower than -1 SD on picture completion and/or coding of the WIPPSI; and 3) a confirmation of there being no hearing problem, no evidence of neurological dysfunction, and no anomalies in oral structure and motor function. Based on the scores of these standardized tests, 15 (10 boys and 5 girls) out of the 34 SLI suspects were identified as having SLI. They ranged in age from 4;4 to 6;9, with the mean of 5;6 (SD = 10 months). Fourteen Chronological Age matched children (8 boys and 6 girls) and nine Language Age-matched children (6 boys and 3 girls) served as the control subjects. As

182

TANAKA WELTY, WATANABE & MENN

shown in Table I, the CA-matched controls scored significantly higher than the JSLI children on expressive and receptive language (t=5.05, p canoeing") or he substitutes a verb for a noun (for example schilderen > schilder, "paint > painter"). He also sometimes changes a noun (related to the picture) into a verb (e.g., persen > sappen, "squeezing > juicing"). Filling in infinitives and finite verbs in a sentence: Filling in infinitives does not pose too much problems: 8/10 correctly. When he has to produce a fi­ nite verb, the score reduces to only 50% correctly (5/10 correct).

295

THE VERB AND SENTENCE TEST

Inflecting the verb is not problematic, but phonemic and a few semantic paraphasias occur. Sentence anagrams without pictures: The first four passive sentences are con­ structed in the following manner: agent-verb-theme. After four sentences he detects that this is wrong and does not make any more mistakes after this. Score: 16/20 correct. Sentence anagrams with pictures. Mr.  does not have any problems with this task: score 20/20 correct. Sentence construction. Mr.  tries to make sentences but often he does not succeed due to phonemic paraphasias in combination with great word finding difficulties. He experiences difficulties with verb production, which is surprising considering his score on the action naming task which shows that verb retrieval itself is relatively spared. The verb is sometimes omitted (target: the boy hits the girl > the boy and the girl), sometimes the infinitival form is used (target: de jongen duwt het meisje "the boy pushes the girl > de man ging omdraaien "the man went to turn"). Inflection of the verb is sometimes problematic (target: de baby kruipt "the baby crawls"> de baby kruip, kruip, kruipt "the baby crawl, crawl, crawls") and sometimes a noun is changed into a verb (target: de clown lacht "the clown smiles" > de man clownt "the man clowns"). Score: 7/20 correct. An overview of his results is displayed in Figure 8.

action naming

infinitives finite verbs

anagram -pict

anagram +pict

sentence const

Figure 8: Percentages correct of the Dutch patient

296

RISPENS, BASTIAANSE & EDWARDS

4.3 Conclusions and directions for therapy Figure 8 shows that retrieving verbs is only mildly impaired, but when the patient has to use them in a sentence, this becomes very difficult. The results on the task 'filling in verbs in sentences' show that producing a finite verb is particularly difficult in contrast with retrieving an infinitive. Therefore, it is assumed that verb movement, which underlies the production of finite verbs (as he has to in the sentence construction task), is impaired. The problems with fmiteness have a spin-off effect on lexical properties of the verb: semantic and phonemic paraphasias occur when he has to produce a finite verb. This can also be observed in his spontaneous speech; the diversity of the verbs is extremely low (in the sample printed above, only two different lexical verbs are used: think and work). In therapy, one can focus on expanding a linguistic strength of this patient, namely the ability to use verbs which are not finite. Treatment may consist of training sentences in which verbs do not need to be moved (e.g., hij is een boek aan het lezen (lit. he is a book to read; "he is reading a book") rather than hij leest een boek "he reads a book"). The Dutch therapy program of Bastiaanse, Jonkers, Quack & Varela Put (1996) can be used for this. Teaching this strategy is aimed at not only improving his ability to produce sentences, but also at increasing the diversity of verbs. 5. Conclusion Only few standardised assessment materials focusing on verbs and sen­ tences have been published. Considering the problems that aphasic patients have with verbs and sentences, it was decided to develop a test battery: the VAST. This test is based on theoretical insights and neurolinguistic findings. Deficits can be pinpointed for which treatment programs/methods exist or have been described in the literature. In the text above, two case studies have been described illustrating the clinical relevance of the VAST. We hope that the VAST contributes to the diagnosis and treatment of verb and sentence deficits in aphasia.

REFERENCES Balogh, Jennifer E. & Yosef Grodzinsky. 2000. "Levels of linguistic representation in Broca's aphasia: Implicitness and referentiality of arguments". Grammatical Disorders in Aphasia: A neurolinguistic perspective ed. by Roelien Bastiaanse & Yosef Grodzinsky, 88-104. London: Whurr.

THE VERB AND SENTENCE TEST

297

Basso, Anna, Carmelina Razzano, Pietro Faglioni & M. Ester Zanobio. 1990. "Confrontation naming, picture description and action naming in aphasic pa­ tients". Aphasiology 4.185-195. Bastiaanse, Roelien, Susan Edwards & Katalin Kiss. 1996. "Fluent aphasia in three languages: Aspects of spontaneous speech". Aphasiology 10.561-575. Bastiaanse, Roelien, Roel Jonkers, Christina Quack & Maria Varela Put. 1996. Werkwoordproductie op woord- en zinsniveau. Lisse: S wets & Zeitlinger. Bastiaanse, Roelien & Roel Jonkers. 1998. "Verb retrieval in action naming and spontaneous speech in agrammatic and anomic aphasia". Aphasiology 12.951-969. Bastiaanse, Roelien, Edwin Maas & Judith Rispens. 2000. De Werkwoorden- en Zinnentest (WEzT). Lisse: Swets & Zeitlinger. Bastiaanse, Roelien & Cynthia Thompson. 2000. "Verb finiteness in agrammatism: A cross-linguistic study". Brain and Language 74.503-514. Bastiaanse, Roelien & Ron van Zonneveld. 1998. "On the relation between verb in­ flection and verb position in Dutch agrammatic aphasies". Brain and Language 64.165-181. Berndt, Rita S., Anne N. Haendiges, Charlotte  Mitchum & Jennifer Sandson. 1997a. "Verb retrieval in aphasia: 1. Characterising single word impairments". Brain and Language 56.68-106. Berndt, Rita S., Charlotte  Mitchum, Anne N. Haendiges & Jennifer Sandson. 1997b. "Verb retrieval in aphasia: 2. Relationship to sentence processing". Brain and Language 56.107-137. Bird, Helen & Sue Franklin. 1996. "Cinderella revisited: A comparison of fluent and nonfluent aphasie speech". Journal of Neurolinguistics 9.187-206. Butterworth, Brian & David Howard. 1987. "Paragrammatisms". Cognition 26.1 -37. Butterworth, Brian, Marta Panzeri, Carlo Semenza & T. Ferreri. 1990. "Paragramma­ tisms: A longitudinal study of an Italian patient". Language and Cognitive Processes 5.115-140. Byng, Sally. 1988. "Sentence processing deficits: Theory and therapy". Cognitive Neuropsychology 5.629-676. Edwards, Susan. 2000. "A clinical assessment of verbs in an agrammatic patient". Grammatical Disorders in Aphasia: A Neurolinguistic Perspective ed. by Roelien Bastiaanse & Yosef Grodzinsky, 191-213. London: Whurr. Edwards, Susan & Roelien Bastiaanse. 1998. "Diversity in the lexical and syntactic abilities of fluent aphasie speakers". Aphasiology 12.99-117. Friedmann, Na'ama & Yosef Grodzinsky. 1997. "Tense and agreement in agram­ matic production: Pruning the syntactic tree". Brain and Language 56.397-425. Goodglass, Harold & Edith Kaplan. 1983. The Assessment of Aphasia and Related Disorders. Philadelphia: Lea & Febiger. Grodzinsky, Yosef. 1995. "A restrictive theory of agrammatic comprehension". Brain and Language 50.27-51. Grodzinsky, Yosef. 2000. "The neurology of syntax: Language use without Broca's area". Behavioral and Brain Sciences 23.1-73.

298

RISPENS, BASTIAANSE & EDWARDS

Grodzinsky, Yosef & Lisa Finkei. 1998. "The neurology of empty categories: Apha­ sics' failure to detect ungrammaticality". Journal of Cognitive Neuroscience 10.281-292. Hagiwara, Hiroko. 1995. "The breakdown of functional categories and the economy of derivation". Brain and Language 50.92-116. Huber, Walther, Klaus Poeck, Dorothea Weniger & Klaus Willmes. 1983. Der Aachener Aphasie Test. Göttingen: Hogrefe. Jonkers, Roel. 2000. "Verb finding problems in Broca's aphasics: The influence of transitivity". Grammatical Disorders in Aphasia: A Neurolinguistic Perspective ed. by Roelien Bastiaanse & Yosef Grodzinsky, 105-122. London: Whurr. Jonkers, Roel & Roelien Bastiaanse. 1996. "The influence of instrumentality and transitivity on action naming in Broca's and anomie aphasia". Brain and Lan­ guage 55.37-39. Kay, Janice, Ruth Lesser & Max Coltheart. 1992. Psycholinguistic Assessments of Language Processing in Aphasia (PALPA). London: LEA. Kohn, Susan E., Marjorie P. Lorch & Dawn M. Pearson. 1989. "Verb finding in aphasia". Cortex 25.57-69. Koster, Jan. 1975. "Dutch as an SOV language". Linguistic Analysis 1.111-136. Linebarger, Marcia, Myrna Schwartz & Eleanor Saffran. 1983. "Sensitivity to gram­ matical structure in so-called agrammatic aphasies". Cognition 3.361-392. Marshall, Jane. 1995. "The mapping hypothesis and aphasia therapy". Aphasiology 9.517-539. Miceli, Gabriela, M. Silveri, Giampiero Villa & Alfonso Caramazza. 1984. "On the basis of the 'agrammatics' difficulty in producing main verbs". Cortex 20.207-220. Schwartz, Myrna, Ruth Fink & Eleanor Saffran. 1995. "The modular treatment of agrammatism". Neuropsychological Rehabilitation 5.91 All. Schwartz, Myrna, Eleanor Saffran, Ruth Fink, J. Myers & Nadine Martin. 1994. "Mapping therapy: A treatment program for agrammatism". Aphasiology 8.19-54. Thompson, Cynthia, K. Lange, Sandra Schneider & Lewis Shapiro. 1997. '"Agram­ matic and non-brain damaged subjects' verb and verb argument structure pro­ duction". Aphasiology 11.473-490. Thompson, Cynthia, Lewis Shapiro, Mary Tait, Beverly Jacobs & Sandra Schneider. 1996. "Training wh-question production in agrammatic aphasia: Analysis of ar­ gument and adjunct movement". Brain and Language 52.175-228. Williams, Sarah E. & Gerald J. Canter. 1987. "Action naming performance in four syndromes of aphasia". Brain and Language 32:124-136. Zingeser, Louise B. & Rita S. Berndt. 1988. "Grammatical class and context effects in a case of pure anomia: Implications for models of language production". Cognitive Neuropsychology 5:473-516. Zurif, Edgar. 1995. "Brain regions of relevance to syntactic processing". Language ed. by Lila Gleitman & Mark Liberman, 381-398, Cambridge, Mass.: MIT Press.

CASE ASSIGNMENT AS AN EXPLANATION FOR DETERMINER OMISSION IN GERMAN AGRAMMATIC SPEECH* ESTHER RUIGENDIJK University of Utrecht 1. Introduction Agrammatic patients are known to have problems with the production of determiners and pronouns. Determiners are often omitted and a relatively low number of pronouns is used in (spontaneous) speech production (Saffran et al. 1989). Determiners and pronouns are function words, and these are vulnerable in agrammatic aphasia. Problems with function words, or grammatical mor­ phemes, is one of the main features of agrammatic aphasia (Saffran et al. 1989; Caramazza & Berndt 1985).1 Recently, it has been suggested that considering all grammatical morphemes as a homogeneous class is an overgeneralization (e.g., Friedmann & Grodzinsky 1997; Ruigendijk et al. 1999). According to Friedmann & Grodzinsky (1997) the production deficit in agrammatism is restricted to realisation of functional projections high in the tree (i.e., CP and TP). Ruigendijk et al. (1999) suggested that the production of determiners and pronouns is not impaired because they are grammatical morphemes, but because their realisation is dependent on the production of (finite) verbs. According to Ruigendijk et al., the basic problem is the production of (finite) verbs; poor determiner and pronoun production can then be considered a side effect (see also: Bastiaanse et al. in press). In contrast to function words (closed-class words), content words are said to be relatively intact in agrammatism. It has frequently been shown, however, that verbs are relatively difficult, and that finite verbs are even more difficult in agrammatism (Bastiaanse & Van Zonneveld 1998; Bastiaanse & Jonkers 1998; Kim & Thompson 2000). One can wonder whether these phenomena, that is, The authors thank Fedor Jalvingh, Martina Penke and Luise Springer for referring patients for this study. We are grateful to Dirk den Ouden for is comments on an earlier version of this paper. 1 'Function words' and 'closed-class words' refer to the same word classes: determiners, prepositions, pronouns, etc. As it is also mentioned that 'grammatical endings' are often missing, we prefer the term 'grammatical morphemes' as this refers to both fuinction/closedclass words and grammatical endings.

300

ESTHER RUIGENDIJK

reduced number of determiners and poor verb production, are related. Based on linguistic theory, which will be discussed in the next section, the expectation is that there is a causative effect, as suggested by a former study (Ruigendijk et al. 1999) in which the spontaneous speech of Dutch and German agrammatic aphasics was analysed. In the study presented here, we will test the hypothesis that there is a relationship between the production of de­ terminers and the production of (finite) verbs in agrammatic speech. This hypothesis is neurolinguistically interesting, because it theoretically relates two well-known characteristics of agrammatic speech, and it is also clinically relevant, since it may have implications for therapy. The first section of this paper will give the linguistic background to our topic. The next section describes the psycholinguistic background. In the third section, relevant aphasiological studies are presented. This section is followed by a description of the methods of the study. Subsequently, a quantitative and qualitative analysis of the results will be given and these will be discussed in the final section of this paper. 2. Linguistic background According to Chomsky, all languages have an abstract case system. Case may or may not be morphologically realised. In a language like English or Dutch, it can only be seen on the pronouns (7 vs. me; he vs. him). In the lan­ guage under investigation in this study, German, not only pronouns, but also determiners are marked for case (see (1) for the German case paradigm). (1) nominative genitive dative accusative

masculine der Mann (the man) des Mannes dem Mann den Mann

feminine die Frau . (the woman) der Frau der Frau die Frau

neutral das Kind (the child) des Kindes dem Kind das Kind

plural die Kinder (the children) der Kinder den Kindern die Kinder

Case, that is, abstract case, expresses the syntactic relation between a verb and its arguments. It specifies, for example, which noun phrase is the subject of the sentence and which one is the object. Noun phrases therefore require case. Chomsky (1986, 1995:111) formulates this in the Case Filter, "every phonetically realised NP must receive (abstract) Case". This means that a sen­ tence with a noun phrase that has not been assigned case is ungrammatical. Following Chomsky (1986), the head of a noun phrase is the noun, therefore it is analysed as an NP (noun phrase). Abney (1987) suggests another analysis, in which the determiner of the noun phrase is the head, therefore a noun phrase

CASE ASSIGNMENT IN GERMAN AGRAMMATIC SPEECH

301

should be analysed as a DP (determiner phrase).2 Ouhalla (1993) refines Chomsky's Case Filter, using the DP-analysis of noun phrases. He suggests that every DP, i.e., every complete noun phrase, must have abstract case. Bare noun phrases (NPs in his terminology) are not subject to the Case Filter. In other words: only the occurrence of DPS is restricted by the Case Filter, not the occurrence of NPs. The question is: how do DPS get case? According to Chomsky (1986, 1995), abstract case is assigned by a verb or a preposition in the following way: the finite verb, which has moved to AgrS, assigns nominative case to the subject of a sentence (in [Spec, AgrS], see figure 1). The subject DP is assumed to be base-generated vp-internally, following Koopman & Sportiche (1991), which means that the subject DP has to move to the specifier position of AgrS to receive its nominative case from the finite verb. AgrSP Spec

AgrS'

Fig. 1 : Case assignment to the subject and object noun phrase

Accusative and dative are both assigned by the verb to the direct and indi­ rect object of a sentence, respectively. Without a finite verb, nominative case cannot be assigned and thus no subject DP should be present. This means that the production of properly casemarked subject determiners depends on the presence of a case assigning finite verb that has moved to AgrS. In the same way, without a verb, no case can be assigned to the object of a sentence and therefore no object determiners can be present when there is no verb. Case-marked object determiners are thus de­ pendent on the presence of a verb.

See Abney (1987) for technical details.

302

ESTHER RUIGENDIJK

The German examples (2) and (3) show this. The finite verb wird (will) in (2) and (3) assigns nominative case to the subject of the sentence der Mann (the mannom). The object of sentence (2) den Jungen (the boyacc) is assigned accusative case by the verb tragen (to carry). German allows for dative case assignment to the object of a sentence as well (also depending on the verb), as can be seen in (3). This dative assignment is associated with the theta role 'benificiary'. In (3) the verb helfen (to help) assigns dative case to the object of the sentence dem Jungen (the boydat). (2) der Mann

wird

NOM

den Jungen ACC

the man-NOM will the boy- ACC "the man will carry the boy" (3) der Mann

tragen

wird

dem Jungen

carry helfen

DAT

the man-NOM will "the man will help the boy"

the boy-DAT

help

Whether a verb assigns dative case - or even genitive - to an object noun phrase has to be learnt for each individual verb. This information can be seen as part of the lexical entry of the verb (Haider 1984; Haegeman 1994). This kind of case assignment is called inherent case assignment, as opposed to structural case assignment, which is dependent on the syntactic structure, as described above, i.e., nominative case assignment to the subject of a sentence in [Spec, AgrS] and accusative case assignment to the object of a sentence, by V.3 3. Psycholinguistic background The information in the lexical entry of a verb and the retrieval of this in­ formation are important for case assignment and thus for the production of DPS. According to Levelt (1989) a lexical entry contains information about the meaning of a specific item, and about the syntactic properties of this item, for example, to which category it belongs (V for hit) and the syntactic arguments it can take {hit takes an external subject and an internal object). The semantic and syntactic information of an item constitute the lemma. Together with mor3

For a detailed discussion on the difference between 'structural' and 'inherent' or 'lexical' case in German see, for example, Haider (1984), Czepluch (1998).

CASE ASSIGNMENT IN GERMAN AGRAMMATIC SPEECH

303

phological and phonological information (i.e., the word form or lexeme) the lemma forms the lexical entry of a word. Systematic relations exist between these different kinds of information in the lexical entry. In his blueprint for the speaker, Levelt (1989) distinguishes several stages from intention to articulation. First a conceptualizer generates a preverbal message. This message is translated into a linguistic structure by a so-called formulator that can be divided into two components, a grammatical encoder and a phonological encoder. For grammatical encoding, the semantic and syntactic information, i.e., the lemma information, of a lexical entry is needed. Phonological encoding is triggered by word form information (the lexeme) and results in a phonetic plan. This phonetic plan, finally, is needed for articulation. For this study, specifically the retrieval of the syntactic information on verbs is important, in other words, the retrieval of verb lemmas. This is used for grammatical encoding. A lemma can only be activated when the conceptual conditions of the preverbal message can be matched with the semantic information of this lemma. When the lemma is retrieved, the syntactic information also becomes available. According to Haider (1984), the information on the verb not only contains how many and which arguments a verb can take, but also how these arguments are assigned case: Lexically (L) or Structurally (s). In (4) two examples are given of the case assignment information in a verb's lexical entry: note that in these examples A1 is normally realised as 'external argument' and A2 as 'internal argument' (examples from Haider 1984). In (4a) both arguments of the verb schlagen must be assigned structural case, which will result in nominative for the external argument (Al s ), the subject, and accusative for the internal argument (A2S), the object. In (4b) only the external argument (AI s ) of the verb helfen, the subject, is assigned structural nominative case, the internal argument (A2L), the object, is assigned lexical case, namely dative. (4) a. schlagen (A1s A2S) to hit b. helfen (A1s A 2 L ), L=Dative to help It can be concluded from psycholinguistic and linguistic theory that the verb is important for case assignment. First, the syntactic information of the lexical entry of the verb specifies how case is to be assigned to the arguments that a verb takes: lexically or structurally. Second, structurally, the subject DP of a sentence depends on the finite verb of a sentence for its case and the ob-

304

ESTHER RUIGENDIJK

ject DP is dependent on the verb for its case. In other words, the presence of a DP depends on the presence of a case assigning verb, and the correct case of this DP depends on the complete retrieval of the syntactic information of that verb. Verbs are known to be vulnerable in Broca's aphasia, the effects of this vulnerability on the ability of Broca's aphasics to produce determiners and as­ sign case will be clarified in the next section. 4. Aphasiological background Agrammatic aphasics have problems with the production of verbs on naming tasks. They also produce fewer verbs in their spontaneous speech than non-brain damaged speakers do, and when they produce verbs, these verbs are often non-finite (Bastiaanse & Jonkers 1998; Bastiaanse & Van Zonneveld 1998). Lexical information must be encoded grammatically for the production of sentences (Levelt 1989). Bastiaanse (1995) as well as Kim and Thompson (2000) ascribe part of the problems Broca aphasies have with verbs to an im­ pairment in grammatical encoding; the lemma information of the verb cannot be processed or accessed correctly. Structural case assignment, and consequently determiner production, de­ pends on the presence of a (finite) verb. Ruigendijk et al. (1999) analysed the spontaneous speech of German and Dutch agrammatic aphasies with regard to determiner production in relation to verb production and inflection. They showed that there is such a relationship. No object determiners (i.e., no accu­ sative or dative marked noun phrases) were produced in the absence of a case assigning verb. The patients did produce some nominative case marked noun phrases, but these were explained with a so-called default strategy: nominative case is assigned to a subject NP, when this cannot be done structurally by a finite verb in [Spec, AgrS]. This default strategy was originally proposed for normal speech (Van Zonneveld 1994), but can be used to describe agrammatic data as well. The most important finding of this study was that when no proper case assigner was present, the determiner was often omitted. German determiners are, as mentioned above, marked for case, but also for number and gender. De Bleser et al. (1996) demonstrated that a group of German agrammatic aphasies was able to inflect isolated noun phrases for gender. When these same patients had to produce the same noun phrases in the context of a sentence, where not only gender, but also case plays a role, their performance was far less accurate. This was interpreted as a problem with case marking, rather than with gender marking. Information about how case has to be assigned is assumed to be part of the lexical entry of a verb (Haider 1984).

CASE ASSIGNMENT IN GERMAN AGRAMMATIC SPEECH

305

Incompletely or incorrectly retrieved lemma information will negatively influence grammatical encoding. The syntactic information on the verb speci­ fies, among other things, how case is to be assigned to the arguments of a verb. If this information is not available for a patient, case errors are expected. Con­ sidering the issues described above, the expectation of this study is that there is a relationship between verb production and determiner production. Impaired verb retrieval will have a negative influence on determiner production. This leads to the following hypothesis for agrammatic speech production: (5) The production of complete noun phrases (DPS), that is, the pro­ duction of determiners, is related to the realisation of a case assigning verb. 5.

Methods

5.1 Subjects and materials Ten agrammatic Broca's aphasics participated in this study (8 male, 2 fe­ male; mean age 55,4). All patients were examined more than a year post onset. Nine patients were right-handed and one left-handed. All were aphasic due to a single stroke in the left hemisphere. The patients were diagnosed as Broca's aphasies on the basis of the Aachen Aphasia Battery (AAT, Huber et al. 1983) and this diagnosis was confirmed by the speech therapist and the examiner. The speech production of all aphasie patients was characterised as telegraphic. The experimental data of this group were compared to the data of a group of 16 German non-brain damaged speakers (7 male, 9 female; mean age 55,6). First, a spontaneous speech sample of each patient was elicited. This was done by a semi-structured interview, according to AAT standards. In this inter­ view, patients were asked to describe their history of illness, job, family and hobbies. These speech samples were then analysed to examine whether and how the patients produced determiners, and whether the production of deter­ miners depended on the production and inflection of verbs. Subsequently, the data of the spontaneous speech analysis were compared to the data of the two experimental tasks. The first experimental task was a sentence completion test that contained 30 items. For each item, subjects were presented with a picture and an incom­ plete svo-sentence (see figure 2) that was to be completed with a noun phrase. Either the subject or the object noun phrase was missing. The patients were asked to read the sentence (or repeat it in case of reading difficulties) and to complete it with the missing noun phrase. For figure 2 this would result in, e.g., der Junge ähnelt... [dem Mann] (the boynom looks like .... [the mandat]).

306

ESTHER RUIGENDIJK

This task meant to examine whether the patients were able to produce a deter­ miner when the case marking verb was given. After this task, the patients were presented with a sentence production task in which the same pictures were used, but now only the non-finite verb was given (see figure 2). The patients were asked to tell in one simple sentence, using the given verb, what was happening in the picture. The sentence for the picture in figure 2 aims to elicit something like die Frau überwacht den Mann (the womannom guards the manacc). The patients were told that they should try to change the infinite verb into a finite verb. This was done to avoid that the patients would think they had to produce a sentence with the infinite verb. Through this task it was possible to test the patients' ability to inflect a verb for tense and agreement in combination with the production of the correct de­ terminers for the noun phrases in the sentence.

Fig. 2: Two examples of the test for the production of determiners (first: sentence completion; second: sentence construction)

These two tasks aimed to evoke nominative subjects, accusative objects and dative objects. In each task, there were 10 items for each constituent, which makes 60 items in total. Note that in German only masculine nouns are unambiguously marked for case (see the German paradigm in 1), therefore the critical constituent was always represented by a masculine noun. 5.2 Scoring Since the question of this study was whether there is a relationship be­ tween the production of determiners and the production and inflection of verbs, the spontaneous speech data were analysed with regard to determiner production in relation to verb production. The noun phrases were counted, and then each noun phrase was examined to check whether a correctly case-marked determiner was produced, and whether a case assigning verb was present. The

CASE ASSIGNMENT IN GERMAN AGRAMMATIC SPEECH

307

difference between (1) noun phrases plus determiner, plus case assigning verb, and noun phrases without determiner without case assigning verb and (2) noun phrases without determiner plus case assigning verb and noun phrases plus determiner without a case assigning verb was statistically tested. Occurrences such as (1) are expected according to the linguistic theory described above, whereas occurrences such as (2) were not expected. In sentence (6a) and (6b), examples are shown of respectively a noun phrase plus determiner, plus case assigning verb, and a noun phrase without a determiner, without a case assigning verb. Examples (7a) and (7b) show the unexpected combinations, namely a noun phrase plus determiner, but without a case assigning verb (7a) and a noun phrase without a determiner, but plus a case assigning verb (7b). (6) a. dann habe ich ein Kind than have I one child "I have one child" b. Kohlrabi langsam kohlrabi slow (7) a. das Essen fertig the food-NOM ready b. Krankenwagen bestellt ambulance called-PART For the tests, the noun phrase was considered to be correct and complete when it was produced with a correctly case-marked determiner. For the sen­ tence production task it was also examined whether the given verb was in­ flected correctly by the patient. After this, the errors on the noun phrases were analysed for a qualitative comparison of the data. Four error types were identi­ fied: case substitution, gender substitution, determiner omission and other er­ rors (nil reactions or irrelevant reaction). See (8) for examples. (8) a. case substitution der Junge begegnet the boy meets Goal: der Junge begegnet the boy meets b. gender substitution Katze kratzen cat Goal: die Katze kratzt the cat scratches

den Vater the father-ACC dem Vater the father-DAT die Mann scratch-INF the man-FEM den Mann the man-ACC

308

ESTHER RUIGENDIJK

c. determiner omission das Mädchen beißt Mann man the girl bites Goal: das Mädchen beißt den Mann the girl bites the man-ACC d. other error (irrelevant completion) das Mädchen erwürgt mit immer die Klassenarbeit the girl strangles with always the examination Goal: das Mädchen erwürgt den Jungen the girl strangles the boy-ACC 6.

Results First, a quantitative analysis of spontaneous speech and the test results will be given. Table 1 shows the results of the spontaneous speech analysis of ten German agrammatic speakers (the expected combinations are bold). All pa­ tients omitted determiners in their spontaneous speech. As a group they pro­ duced 378 noun phrases. In 227 (60%) of these noun phrases the determiners were omitted, in 151 (40%) cases the determiners were produced. For 198 noun phrases no case assigning verb was present. There was a significant dif­ ference between (1) the expected combinations, namely the noun phrases plus determiner, plus case assigning verb, and the noun phrases without determiner, without a case assigning verb and (2) the unexpected combinations: the noun phrases without a determiner, but plus case assigning verb and the noun phrases plus determiner, but without a case assigning verb (t=5,586, df=9, p