Corpus, Cognition and Causative Constructions (Studies in Corpus Linguistics)

Corpus, Cognition and Causative Constructions Studies in Corpus Linguistics (SCL) SCL focuses on the use of corpora t...

Author: Gaëtanelle Gilquin

229 downloads 1568 Views 7MB Size Report

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

DOWNLOAD PDF

Corpus, Cognition and Causative Constructions

Studies in Corpus Linguistics (SCL) SCL focuses on the use of corpora throughout language study, the development of a quantitative approach to linguistics, the design and use of new tools for processing language texts, and the theoretical implications of a data-rich discipline.

General Editor

Consulting Editor

Elena Tognini-Bonelli

Wolfgang Teubert

The Tuscan Word Center/ The University of Siena

University of Birmingham

Advisory Board Michael Barlow

Graeme Kennedy

Douglas Biber

Geoffrey N. Leech

Marina Bondi

Anna Mauranen

Christopher S. Butler

Ute Römer

Sylviane Granger

Michaela Mahlberg

M.A.K. Halliday

Jan Svartvik

Susan Hunston

John M. Swales

Stig Johansson

Yang Huizhong

University of Auckland Northern Arizona University University of Modena and Reggio Emilia University of Wales, Swansea University of Louvain University of Sydney University of Birmingham University of Oslo

Victoria University of Wellington University of Lancaster University of Helsinki University of Michigan University of Nottingham University of Lund University of Michigan Jiao Tong University, Shanghai

Volume 39 Corpus, Cognition and Causative Constructions by Gaëtanelle Gilquin

Corpus, Cognition and Causative Constructions Gaëtanelle Gilquin Université catholique de Louvain

John Benjamins Publishing Company Amsterdam / Philadelphia

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.

Cover design: Françoise Berserik Cover illustration from original painting Random Order by Lorenzo Pezzatini, Florence, 1996.

Library of Congress Cataloging-in-Publication Data Gilquin, Gaëtanelle Corpus, cognition and causative constructions / Gaëtanelle Gilquin. p. cm. (Studies in Corpus Linguistics, issn 1388-0373 ; v. 39) Includes bibliographical references and index. 1. Corpora (Linguistics). 2. Causative (Linguistics) 3. Cognitive grammar. I. Title. P128.C68G55 2010 410.1--dc22 2009045792 isbn 978 90 272 2313 5 (Hb ; alk. paper) isbn 978 90 272 8849 3 (Eb)

© 2010 – 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 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

In memory of Joséphine Gilquin (1911–2005), my guardian angel, and Louis Brasseur (1920–2005), who taught me the importance of a job well done and the value of hard work

Nobody trips over mountains. It is the small pebble that causes you to stumble. Pass all the pebbles in your path and you will find you have crossed the mountain. (Author unknown)

Table of contents

List of tables List of figures Acknowledgements chapter 1 Introduction chapter 2 Corpus linguistics and theory 2.1 Corpus linguistics: Theory or methodology? 5 2.2 The place of theory: Bottom-up vs. top-down 7 2.2.1 From data to theory 7 2.2.2 From theory to data 8 2.2.3 To-ing and fro-ing between data and theory 10 2.3 Combining corpus linguistics and cognitive linguistics 11 2.3.1 “Computer-aided armchair linguistics” 11 2.3.2 The theoretical framework of cognitive linguistics 12 2.3.3 Compatibility of cognitive linguistics with corpus linguistics 14 2.3.4 Corpus linguistics and cognitive linguistics: Added value 16 2.3.5 Illustrations 17 2.3.6 The case of English periphrastic causative constructions 19 2.4 Summary 23 chapter 3 English causative constructions: Collecting the data 3.1 Corpus data: A semi-automatic method 25 3.1.1 Two problems with the extraction of the data 25 3.1.1.1 The retrieval of syntactic structures 25 3.1.1.2 Periphrastic causative construction: An ambiguous structure 29 3.1.2 Semi-automatic method 30 3.1.2.1 The corpus 30 3.1.2.2 The text retrieval software 34 3.1.2.3 Pilot study 36

xiii xvi xvii 1 5

25

viii Corpus, Cognition and Causative Constructions

3.1.2.4 Automatic stage: Using BNCweb to query BNC-10 43 3.1.2.5 Manual stage: Post-editing the concordances 44 3.1.2.6 Semi-automatic search of BNC-10 45 3.2 Experimental data: Elicitation test 48 3.2.1 Designing the questionnaire 48 3.2.2 Data collection 54 3.3 Combining corpus and experimental data 55 3.4 Summary 59 chapter 4 Causative constructions in action: The realisation of the action chain 4.1 The notion of action chain 61 4.2 The action chain of periphrastic causative constructions 63 4.3 Action chains in the corpus data 68 4.3.1 Different types of action chains 69 4.3.1.1 Literal vs. symbolic energy flow 69 4.3.1.2 Caused motion vs. caused rest 69 4.3.2 Length of the action chain 70 4.3.3 Linguistic realisation of the action chain 72 4.3.4 The causing event 77 4.3.5 Identity between participants 84 4.4 Action chains in the elicitation data 91 4.5 Summary 95 chapter 5 The syntax and semantics of causative constructions: The Principle of No Synonymy 5.1 No (more) synonymy 97 5.2 From data to database 99 5.3 Individual approach: Chi-square analysis 102 5.3.1 Methodology: Chi-square test with adjusted residuals 102 5.3.2 The causative verb 104 5.3.3 The causer 108 5.3.4 The causee 115 5.3.5 The effect 123 5.3.6 The patient 131 5.3.7 Interim summary: Distinctiveness and frequency 134 5.4 Global approach: Hierarchical feature selection 137 5.5 Syntactic and semantic features in the elicitation data 139 5.6 Summary 142

61

97

Table of contents

chapter 6 Defining the causative prototype 6.1 Prototypicality in cognitive linguistics 145 6.2 Prototypical causation 149 6.2.1 Ordering of the participants 149 6.2.2 Nature of the participants 151 6.2.2.1 Billiard-ball causation 151 6.2.2.2 Direct manipulation 151 6.2.3 The three models 153 6.3 Linguistic validation of the models of prototypical causation 153 6.3.1 Strict validation 155 6.3.2 Loose validation 157 6.4 Cognitive validation of the models of prototypical causation 160 6.5 Explaining the discrepancies 162 6.5.1 Theoretical constructs and language data 162 6.5.2 Corpus and elicitation data 164 6.6 Summary 166 chapter 7 Lexical co-occurrence in causative constructions 7.1 The co-textual approach 169 7.2 Methodology 171 7.3 Collocational profiles in the corpus data 172 7.3.1 [X cause Y Vto-inf ] 172 7.3.2 Get and have constructions 175 7.3.3 Make constructions 183 7.4 Word clusters in the corpus data 185 7.5 The co-textual approach in the elicitation data 189 7.6 Summary 190 chapter 8 Collexemes in the effect slot 8.1 Periphrastic causative constructions: Always safe? 193 8.2 Collostructional analysis and the technique of multiple distinctive collexeme analysis 195 8.3 Lemma-based vs. sense-based approach 196 8.4 Methodology 198 8.5 Results of the sense-based multiple distinctive collexeme analysis and discussion 200 8.5.1 Deviation from expected frequency 200

145

169

193

ix

x


8.6 8.7

8.5.2 [X make Y Vpp] 202 8.5.3 [X make Y Vinf ] and [X be made Vto-inf ] 204 8.5.4 [X get Y Vpp] and [X have Y Vpp] 208 8.5.5 [X get Y Vprp] and [X have Y Vprp] 211 8.5.6 [X get Y Vto-inf ] and [X have Y Vinf ] 213 8.5.7 [X cause Y Vto-inf ] 215 The effect slot in the elicitation data 216 Summary 219

chapter 9 The influence of register on causative constructions 9.1 Studying variation 223 9.2 Frequency according to register 225 9.2.1 Medium 225 9.2.2 Genres 227 9.3 Semantic and syntactic features in speech vs. writing 231 9.3.1 Shared differences 232 9.3.2 Specific differences 233 9.4 Lexical features in speech vs. writing 239 9.4.1 Collocation and semantic prosody 239 9.4.2 Word clusters 242 9.4.3 Collexemes in the effect slot 243 9.5 Summary 249 chapter 10 Causative constructions and foreign language teaching 10.1 Foreign learners: Lost in causative constructions 251 10.1.1 Data and methodology 252 10.1.2 Under- and overuse of causative constructions 254 10.1.3 Syntactic misuse of causative constructions 256 10.1.4 Idiomaticity of causative constructions: Collexemes in the effect slot 259 10.2 Lost in pedagogical tools 268 10.3 Grammatical unit 272 10.4 Summary 276

223

251

chapter 11 Conclusion

277

References

285

Table of contents

appendix 1 Elicitation test

305

appendix 2 Distinctive syntactic and semantic features of causative constructions

311

appendix 3 Most frequent syntactic and semantic features of causative constructions 315 Author index Subject index

317 321

xi

List of tables

Table 1. Table 2. Table 3. Table 4. Table 5. Table 6. Table 7. Table 8. Table 9. Table 10. Table 11. Table 12. Table 13. Table 14. Table 15. Table 16. Table 17. Table 18. Table 19. Table 20. Table 21. Table 22. Table 23. Table 24. Table 25. Table 26. Table 27. Table 28. Table 29. Table 30. Table 31. Table 32.

English periphrastic causative constructions 20 Composition of BNC-10 34 Proportion of periphrastic causative constructions in SUB_ARTS 37 Examples of mistagged non-finite complements with causative make in SUB_ARTS 40 Queries used and their recall/precision rates 42 Comparison of recall and precision rates with and without repair mechanisms 43 Number of hits, number of causative constructions and precision rate 46 Absolute frequency and relative frequency per million words of periphrastic causative constructions 48 Hypotheses underlying exercise A of the elicitation test 50 Types of problems involved in exercise D of the elicitation test 53 Types of interpretations possible in exercise F of the elicitation test 53 Length of the action chain (percentages) 72 Number of participants profiled (percentages) 73 Profiling and gapping of the causing event (percentages) 80 Relation between causer and causee (percentages) 87 Relation between patient and causer/causee (percentages) 91 Parameters of causative constructions investigated 100 Contingency table and adjusted residuals of the effect type 103 Status of the causative verb (distinctiveness and percentages) 105 Tense of the causative verb (distinctiveness and percentages) 107 Form of the causer (distinctiveness and percentages) 109 Grammatical person of the causer (distinctiveness and percentages) 110 Definiteness of the causer (distinctiveness and percentages) 110 Specificity of the causer (distinctiveness and percentages) 110 Animacy of the causer (distinctiveness and percentages) 112 Semantic nature of the causer (distinctiveness and percentages) 114 Form of the causee (distinctiveness and percentages) 116 Grammatical person of the causee (distinctiveness and percentages) 116 Definiteness of the causee (distinctiveness and percentages) 116 Specificity of the causee (distinctiveness and percentages) 118 Animacy of the causee (distinctiveness and percentages) 118 Types of causation (distinctiveness and percentages) 120

xiv Corpus, Cognition and Causative Constructions

Table 33. Table 34. Table 35. Table 36. Table 37. Table 38. Table 39. Table 40. Table 41. Table 42. Table 43. Table 44. Table 45. Table 46. Table 47. Table 48. Table 49. Table 50. Table 51. Table 52. Table 53. Table 54. Table 55. Table 56. Table 57. Table 58. Table 59. Table 60. Table 61. Table 62. Table 63. Table 64. Table 65. Table 66. Table 67. Table 68. Table 69.

Semantic nature of the causee (distinctiveness and percentages) 122 Transitivity of the effect (distinctiveness and percentages) 124 Dynamicity of the effect (distinctiveness and percentages) 125 Functional category of the effect (distinctiveness and percentages) 127 Volitionality of the effect (distinctiveness and percentages) 128 Volitionality of the effect and animacy of the causer and causee (distinctiveness and percentages) 130 Form of the patient (distinctiveness and percentages) 132 Definiteness of the patient (distinctiveness and percentages) 132 Specificity of the patient (distinctiveness and percentages) 132 Semantic nature of the patient (distinctiveness and percentages) 133 Number of distinctive features in common and in opposition (grey shading) 135 Number of most frequent features in common 136 Results of hierarchical feature selection 139 Parameters of transitivity (Hopper & Thompson 1980: 252) 147 Parameters of the models of prototypical causation 154 Frequency of the models of prototypical causation 154 Frequency of the parameters of the billiard-ball model 159 Frequency of the parameters of the model of direct manipulation 159 Salience of the models of prototypical causation 161 [X cause Y Vto-inf ] and the semantic prosody of technical and scientific terminology 174 [X get Y Vpp] and the semantic prosody of difficulty 176 [X have Y Vpp] and the semantic prosody of service 180 [X have Y Vprp] and the semantic prosody of machinery 182 [X get Y Vprp] and the semantic prosody of machinery 183 [X make Y Vinf ] and the semantic prosody of bodily and inner sensations 184 Frequent noun phrases and verbs in [X get Y Vpp] and [X have Y Vpp] (Francis et al. 1996: 306) 194 Observed frequency, expected frequency and distinctiveness value for look_seem 200 Distinctiveness value of verb senses with SumAbsDev larger than fifty 201 Distinctive collexemes for [X make Y Vpp] 203 Distinctive collexemes for [X make Y Vinf ] 205 Distinctive collexemes for [X be made Vto-inf ] 207 Distinctive collexemes for [X get Y Vpp] 209 Distinctive collexemes for [X have Y Vpp] 211 Distinctive collexemes for [X get Y Vprp] 212 Distinctive collexemes for [X have Y Vprp] 212 Distinctive collexemes for [X get Y Vto-inf ] 214 Distinctive collexemes for [X have Y Vinf ] 214

List of tables xv

Table 70. Distinctive collexemes for [X cause Y Vto-inf ] 215 Table 71. Meanings distinctively associated with periphrastic causative constructions 220 Table 72. Main semantic and syntactic differences between speech and writing 234 Table 73. More semantic and syntactic differences between speech and writing 237 Table 74. Semantic prosody of technical and scientific terminology with [X cause Y Vto-inf ] in speech and writing 240 Table 75. Semantic prosody of difficulty with [X get Y Vto-inf ] in speech and writing 240 Table 76. Semantic prosody of service with [X have Y Vpp] in speech and writing 241 Table 77. Semantic prosody of bodily and inner sensations with [X make Y Vinf ] in speech and writing 241 Table 78. Some idioms in speech and writing 243 Table 79. Type/token ratio of the effect in speech and writing 244 Table 80. Absolute and cumulative frequency of the most frequent effects in speech and writing 245 Table 81. Proportion of shared collexemes in speech and writing 247 Table 82. Distinctive collexemes in speech and writing (with collostruction strength) 248 Table 83. Composition of ICLEv2 253 Table 84. Relative frequency per million words of causative constructions in native and learner writing 255 Table 85. Frequency of non-standard patterns in learner writing 257 Table 86. Proportion of syntactic errors in learner writing 259 Table 87. Proportion of shared collexemes in the effect slot of native and learner writing 261 Table 88. Some distinctive collostructions in native and learner writing 261 Table 89. Distinctive collexemes of [X make Y Vinf ] in native and learner writing (significant values only) 262 Table 90. Distinctive collexemes of [X make Y Vpp] in native and learner writing (all values) 265 Table 91. Distinctiveness of [X make Y feel] in learner writing vs. native writing and in learner writing vs. native speech 267 Table 92. Proportion of dynamic and stative verbs in the effect slot of the make/faire causative constructions in native American English (US), learner English (ICLE) and native French (FR) (Gilquin 2000/2001: 110) 267

List of figures

Figure 1. Figure 2. Figure 3. Figure 4. Figure 5. Figure 6. Figure 7. Figure 8. Figure 9. Figure 10. Figure 11. Figure 12. Figure 13. Figure 14.

Domains in the BNC (Lee 2001a: 50) 32 The Corpus-Cognition Integrated model 58 Action chain (Langacker 1991: 283) 61 Linguistic realisation of different action chains 62 Effect of causativisation on the action chain 64 Action chain of Jealousy caused the Queen to kill Snow White 65 Schematic action chain of a periphrastic causative construction with a patient 66 Schematic action chain of a periphrastic causative construction with no patient 66 Action chains with kill and cause to die 67 Relative frequency per million words of causative constructions according to medium 226 Relative frequency per million words of causative constructions according to genre (speech) 227 Relative frequency per million words of causative constructions according to genre (writing) 231 Proportion (%) of [X get Y done] in native writing, native speech and learner writing 266 Grammatical unit on causative constructions 272

Acknowledgements

This book is the result of over ten years of interest in English causative constructions. I wish to thank the many colleagues, friends and family who helped me throughout these years. First and foremost, I would like to thank Sylviane Granger for her unfailing support and guidance. Despite her busy schedule, she always took the time to discuss my work and offer insightful comments. Thanks are also due to Liesbeth Degand, Jean Heiderscheidt, Maarten Lemmens and Fanny Meunier for the intellectual stimulation they provided. I am extremely grateful to Bengt Altenberg, who read the whole manuscript and gave detailed feedback on it. For their help with statistical analysis, I owe special thanks to Yves Bestgen and Marie-Catherine de Marneffe. I also acknowledge the financial support of the Belgian National Fund for Scientific Research (FNRS). In addition, my thanks go to those who crossed my path (if only fleetingly) and helped me in one way or another, or simply showed interest in my work. Every single word of encouragement was greatly appreciated and took me one step further towards my goal. My debt to three Professors, André Hantson, Sylviane Granger and Geoffrey Leech, is considerable, as they have each acquainted me with the fascinating domain of linguistics and have been a constant source of inspiration since then. Last but not least, I would like to express my affectionate gratitude to my family, who were always there when I needed them. And to my friends, whom I have neglected far too much over the last few years. They are true friends. I know they will understand and forgive me.

chapter 1

Introduction

The notion of causation is a fundamental one. Lakoff & Johnson (1980: 69) see it as a “basic human concept”, one which is used by people to “organize their physical and cultural realities”, and Baron (1974: 340) notes “the importance of causation to the underlying structure of human language”. This book looks into causation as it is expressed in English, and more precisely, causation as encoded by so-called periphrastic causative constructions (also referred to as analytic causative constructions), i.e. two-part configurations such as He makes me laugh or I had my hair cut, where a causative verb controls a non-finite complement clause and which express a causal relation in which the occurrence of the effect is entailed (see Wolff & Song 2003). In total, ten different periphrastic causative constructions will be examined, centring around the verbs cause, get, have and make.1 Kemmer & Verhagen (1994: 115) note that “[t]he grammar of causative constructions has inspired what is probably one of the most extensive literatures in modern Linguistics”, and one may wonder what yet another study such as this one has to offer. Its contribution is threefold – descriptive, methodological and theoretical. The first aim is to provide an exhaustive and reliable description of the behavioural profile of causative constructions in British English. It is a functional and cognitive assumption that the availability of alternative expressions to describe one and the same situation implies a difference in meaning and conceptualisation (cf. the “one form, one meaning” principle in functional grammar and the “Principle of No Synonymy” in cognitive linguistics). The existence of several causative constructions therefore raises the question of what distinguishes them. The literature does not provide any satisfactory answer to this question. The constructions are often presented as interchangeable beyond the obvious differences in complementation. In addition, the existing descriptions of English periphrastic causative constructions tend to suffer from a lack of comprehensiveness, with aspects such as style or lexis being largely ignored, and present contradictions which underline their unreliable character. As a starting point for a more adequate description, use will be made of corpus data. Since they represent

1. See Wolff & Song (2003: 286) for an exhaustive list of periphrastic causative verbs.

2


naturally occurring language, corpora give access to authentic instances of causative constructions, which form a sound basis for a reliable description of the phenomenon. Moreover, the corpus data are presented in their co(n)text, so that all the facets of causative constructions may be studied, including the stylistic and lexical ones. In order to bring to light the distinctive features characteristic of causative constructions, and thus uncover the factors conditioning the choice of a given construction, a number of syntactic, semantic, lexical and stylistic parameters will be investigated and compared in the ten constructions under study. The analysis will be abundantly illustrated with corpus examples, mainly from the British National Corpus (BNC), which constitutes the basis of the whole analysis, but also from other corpora when relevant. The second goal of this study is methodological and consists in demonstrating the usefulness of combining naturally-occurring data with experimental data. While corpus data will represent the main resource to investigate the behaviour of causative constructions, they will be supplemented with data coming from elicitation tests. By not exclusively relying on one source of evidence, the analysis is less liable to bias and more likely to lay claim to scientific validity. When the corpus data and the experimental data present converging evidence, their combination gives more weight to the conclusions. When they present diverging evidence, they help qualify one’s findings or reinterpret them. This combined methodology also helps avoid some of the problems linked with either of the two types of data taken separately, namely the necessarily finite state of a corpus and its lack of grammatical/ungrammatical annotation on the one hand, and the subjective and mainly qualitative (as opposed to quantitative) orientation of experimental data on the other. At the same time, it makes it possible to capitalise on the strengths of each of them, most notably the natural character of corpus data and what they reveal about probable facts, stylistic usage and preferential lexical company (domains which are largely out of reach without the aid of a corpus), and the proximity of experimental data to conceptualisation, as well as the indication they may give of what is possible (rather than probable) in language. Consequently, the combination of corpus and experimental data is expected to contribute to a better understanding of the phenomenon of causative constructions. Finally, this study pursues a theoretical goal in that it shows how corpus linguistics and cognitive linguistics can be fruitfully combined into a single, integrated theoretical model, which will be referred to as the “Corpus-Cognition Integrated model”. Throughout the investigation, the use of corpora and the philosophy it presupposes will be associated with a cognitively-based view of language, in which “the knowledge underlying grammar is not qualitatively different from other aspects of human understanding and reasoning” (Kemmer & Verhagen 1994: 115). Although such a symbiosis is by no means obvious, if only because

Chapter 1. Introduction

corpus linguistics and cognitive linguistics have different objects of inquiry – language use and language cognition respectively (Mukherjee 2004: 86) – the data and facts revealed by a corpus can generally be accommodated within a cognitive framework (see Schönefeld 1999). This double framework makes it possible to respond to the criticism levelled against purely theoretical approaches, especially their lack of empirical basis, and against approaches relying solely on corpus data, namely their atheoretical status and tendency towards number crunching. What is more, it will be shown that to-ing and fro-ing between the data and the theory (i.e. a combination of bottom-up and top-down approaches) allows for an exhaustive description of a particular phenomenon, that is a description which not only states the facts (and all the facts), but also seeks to explain them. In order to achieve these three goals, we will proceed as follows. Chapter 2 presents the framework used in this study, consisting in an integration of corpus linguistics and cognitive linguistics, and a combination of bottom-up and topdown approaches. It shows what the added value of such a framework is, both in general and in the case of periphrastic causative constructions in particular. Chapter 3 deals with the methodology used to collect the (corpus and elicitation) data. The corpus, a 10,000,000-word selection from the spoken and written components of the BNC, is described and a pilot study is carried out with the BNCweb software in order to determine the best possible search string to automatically retrieve causative constructions with cause, get, have and make. As regards the elicitation data, light is shed on the design of the questionnaire and its complementarity with the corpus data. The results of the analysis proper are presented in Chapters 4 to 9. While Chapters 4 to 6 adopt a mainly top-down approach, starting from a theoretical construct to observe the data from this perspective, Chapters 7 to 9 adopt a more bottom-up approach, where the data are carefully examined in order to arrive at useful generalisations. In Chapter 4, the starting point is the concept of action chain, which symbolises a transfer of energy between several entities and is shown to underlie all causative constructions. The realisation of the action chain at the linguistic level explains the existence of different structures and reveals some differences between the ten causative constructions. Following Goldberg’s (1995) “Principle of No Synonymy”, Chapter 5 seeks to identify the syntactic and semantic features that distinguish between the different constructions. The approach is both individual, considering each parameter one by one and testing its significance by means of a chi-square analysis, and global, in the form of hierarchical feature selection (backward elimination and forward selection), which examines all the parameters together with the aim of determining the best predictors for the choice of one construction over the other. In Chapter 6, the definitions of prototypical causation found in the literature are compared with linguistic frequency as attested in the corpus data and cognitive

3

4


salience as evidenced by the elicitation data. The discrepancies observed between the theoretical constructs and the two types of data are discussed and explained. From Chapter 7 onwards, the approach is corpus-driven, starting as it does from the data themselves. Chapters 7 and 8 deal with the lexical regularities displayed by causative constructions. In Chapter 7, the collocational profile of the different constructions is established by examining the words occurring in the immediate environment of the causative verb, how these words group into semantic classes and how they form recurrent chunks of language, whereas in Chapter 8, it is the non-finite verb slot of causative constructions that is focused on, and what its lexical preferences can tell us about the meaning of each construction. Chapter 9 consists in a stylistic analysis of the use of causative constructions across media and genres, with special emphasis on the phraseological dimension. Following this analysis, the focus in Chapter 10 shifts to practical applications, as we look into the possibilities of applying the findings of the previous chapters to foreign language teaching. This results in a grammatical unit intended for advanced learners, relying on the analysis of (native and learner) corpus data in an attempt to avoid the shortcomings exhibited by most descriptions of causative constructions. The book ends with a conclusion (Chapter 11) summarising the most important insights gained from the study and suggesting possible avenues for research.

chapter 2

Corpus linguistics and theory

This chapter sets the framework for the analysis carried out in the next chapters. It begins with the question of the status of corpus linguistics and shows that corpora may be used either as a starting point (bottom-up, or corpus-driven approach) or as evidence for an otherwise well-established theory (top-down, or corpus-tested approach).2 It is argued that the ideal is a mixture of the two approaches, with to-ing and fro-ing between corpus and theory. The theory chosen for this study, cognitive linguistics, is then presented, and its compatibility and complementarity with corpus linguistics are demonstrated.

2.1

Corpus linguistics: Theory or methodology?

It might seem surprising that, more than 40 years after the appearance of the first electronic corpus, the status of corpus linguistics may still be relatively unclear. Yet, that this is the case appears for example from the panel discussion that took place at the 26th ICAME Conference in Ann Arbor and whose title was “Corpus linguistics: methodology or sub-field?”. The fact that the title was formulated as a question, and that this question deserved a panel discussion at a meeting of corpus experts in the first place, proves that we are far from having reached an agreement about the exact nature of corpus linguistics. Some consider that corpus linguistics is a methodology, providing linguists with a set of tools and methods to analyse language (e.g. Kennedy 1998, Meyer 2002). In this view, no matter what theory linguists advocate (if any) or from what angle they approach language, as long as they use authentic data, they are corpus linguists. A sociolinguist using corpora in his/her research, for example, is a corpus linguist, just like the historical linguist who relies on corpus data to study the evolution of language (see Mittelberg et al. [2007: 21] for a list of subfields of the language sciences in which corpora may be used). Others argue that 2. The term “corpus-based”, normally used in opposition to “corpus-driven” to refer to topdown approaches, will be avoided here because of its ambiguity and will be used instead as an umbrella term, referring to any study based on the use of corpus data. Thanks are due to Bengt Altenberg for suggesting the term “corpus-tested”.

6


corpus linguistics is “much more than an assortment of some computational tools plus some small and large corpora” and that it is “an approach in its own right” (Teubert & Krishnamurthy 2007: 2). Tognini-Bonelli (2001: 1) goes one step further and claims that “corpus linguistics has (…) a theoretical status” (see also Teubert 2008: 82). In fact, even those scholars who view corpus linguistics as a methodology sometimes suggest that it might actually be more than this. Leech (1992: 106), for instance, describes corpus linguistics as a methodology rather than a domain of study, but still acknowledges that corpus linguistics is a “new research enterprise”, and the computer “an ‘open sesame’ to a new way of thinking about language”. Similarly, Partington (1998: 1) presents corpus analysis not only as a “new technological device”, but also as a “new philosophy for language description”. (See Taylor 2008 for a brief overview of the status granted to corpus linguistics in the literature.) At the ICAME panel discussion referred to above, a majority of the participants seemed to tip the scales in favour of corpus linguistics as a methodology – although some of them found this status unfortunate. Yet, even if one agrees that corpus linguistics refers essentially to the use of corpora (and this is the position that will be taken here), there is still a sense in which corpus linguists may feel that they belong to a community, and recognition among some people that certain works are more representative of corpus linguistics than others. For instance, probably no one would want to describe Poutsma’s (1926) Grammar of Late Modern English as a corpus-based grammar, even though it is illustrated with naturally-occurring sentences. Of course, the term “corpus-based” has now come to imply computer-assisted (Kennedy 1998, Mason 2000a) and Poutsma’s study certainly does not meet this criterion. But additionally, authentic data in Poutsma’s work occupy a relatively modest place and one would likely want them to be more central in order to consider Poutsma a corpus linguist. In fact, it is this reason, namely the relative marginality of the corpus data, that leads Mukherjee (2006: 342) to regard the Comprehensive Grammar of the English Language (Quirk et al. 1985) and the Cambridge Grammar of the English Language (Huddleston & Pullum 2002) as “corpus-aware”, rather than strictly “corpus-based”. As Teubert (2008: 62) puts it: “[b]y itself, (…) working with corpus data does not make one a corpus linguist. More and more linguists (…) may underpin their investigations with examples discovered in corpora. But corpus linguistics is more”. There seem to be two important features that characterise the “community” of corpus linguists, besides the use of machine-readable corpora. The first one is the importance attached to the integrity of the corpus data. Corpus linguists seek to do justice to the richness of the data, rather than select only those data that would be convenient for them because they confirm a particular hypothesis or fit in nicely with their theory. Second, many corpus linguists approach data not only

Chapter 2. Corpus linguistics and theory

qualitatively, but also quantitatively. This quantitative approach may take the form of simple frequencies and percentages, or more complex statistical calculations. These two concerns, although generally fruitful, tend to be badly thought of by linguists outside the “community” (and sometimes within the community itself). The integrity of the data is often taken to imply lack of theoretical sophistication, whereas a merely quantitative perspective tends to translate as “number crunching” (Mair & Hundt 2000: 3). While this criticism may sometimes be justified, it is important to bear in mind that it mainly pertains to one way of approaching corpus data, namely the bottom-up approach – and an extreme form of it, at that. The distinction between bottom-up and top-down approaches is the topic of the next section.

2.2 The place of theory: Bottom-up vs. top-down 2.2.1 From data to theory One way of approaching corpus data is through bottom-up (or corpus-driven) research. Bottom-up research is inductive and uses the corpus data to elaborate a theory. The linguist is committed to “the integrity of the data as a whole” (Tognini-Bonelli 2001: 84), so that new discoveries, which were not accessible through introspection, are now available. As Granger (2004: 133) puts it, the bottom-up approach is “the required passage to gain new insights into language” (emphasis original). Typically, one examines the whole set of data with no a priori hypothesis and makes generalisations on this basis. Grabe & Kaplan (1996: 46) note the post-hoc nature of such research, resulting from the absence of a theory before the analysis: The general dilemma facing most projects on corpus research is the lack of a theoretical foundation for the interpretation of the results prior to the analysis. Thus, most corpus research has been of a post-hoc nature, looking at the frequency counts and deciding what can be said about these results.

But deciding what can be said about the results is not necessarily an easy task. When faced with the jungle of data, the analyst may find it difficult to see the wood for the trees and thus risks losing sight of the wider perspective, hence falling victim to what Granger (1998) has called the “so what?” syndrome. The fifth and sixth most common words in a given spoken corpus are uh and um, and the corpus contains several hundred tokens of I don’t know (to take Swales’s [2002: 151] examples). So what? This attitude Pullum (2009) condemns as “corpus fetishism” (see also Pullum 2003).

7

8


Taken to extremes, the bottom-up approach does not go beyond the level of data observation and description. It fails to reach a higher degree of abstraction and may content itself with a purely quantitative analysis. It is with such an extreme form of bottom-up research that scholars like Flowerdew (1998) or Swales (2002) have expressed some dissatisfaction. Flowerdew (1998) advocates adopting a framework such as systemic-functional grammar or genre and discourse analysis to study the lexico-grammatical patterning of a text, an approach which she thinks is still too rare, especially within the applied field. In order to avoid atomistic conclusions, she says, one needs to go beyond the sentence level and observe the underlying discourse functions of collocational patternings at the text level (ibid. 547). In a similar vein, Swales (2002: 152, emphasis original) argues that “many corporists most of the time seem to have trouble – in the present state of the art – in coming up with useful discoveries about the target language” (although he recognises their capacity to test the validity of previously made generalisations, i.e. the corpus-tested, as opposed to corpus-driven, approach). He also claims that “small gleanings” about language, though seductive, can eventually lead to “a kind of anecdotal gee-whizzery” and prove to be intrinsically “pretty meaningless” (ibid. 151). The accumulation of such “incidental findings”, he adds, provides “little in the way of a platform from which to launch corpus-based pedagogical enterprises” (ibid., emphasis original). Frequencies and statistics are interesting, and certainly represent one of the greatest insights into language that corpus linguistics has made possible. However, they should not be seen as an answer, but rather as an incentive to ask questions. The quantitative perspective, in other words, is not an end in itself, but a starting point for qualitative research (see Aarts 2000, de Beaugrande 2002), and it is only through this instrumental function that its importance can be demonstrated. It is therefore crucial that bottom-up research does not confine itself to number crunching, but goes all the way from the data to theoretical considerations.

2.2.2 From theory to data Teubert & Krishnamurthy (2007: 7) claim that “[c]orpus linguistics is bottomup linguistics, is parole-linguistics. The starting point is always the corpus, real language data”. While this may be the most widespread conception of corpus linguistics, there is an alternative way of doing corpus linguistics, and this is the topdown (or corpus-tested) approach. In contrast to the bottom-up approach, the top-down approach is deductive, starting as it does from a theory and using the data to expound, test or exemplify this theory (Tognini-Bonelli 2001: 65). In its most extreme form, it consists in verifying, against a corpus, a hypothesis which


has been formulated on the basis of introspection alone. As a consequence, the results are limited by the scope of the hypothesis and the linguist finds just what s/he is looking for, no more. To take Tognini-Bonelli’s (2001: 15–17) example, a top-down study of the word any would prove the statement that this pronoun is used in negative and interrogative sentences (a rule that is found in pedagogical grammars and is indeed confirmed by corpus data), but would fail to show that it can also be used under different circumstances, a point revealed by a careful and unbiased examination of a concordance of any. As another example, consider the version of construction grammar (one of the approaches within cognitive linguistics, see Section 2.3.2) associated with Goldberg. In order to show that a sentence pattern is not necessarily determined by independent specifications of the main verb, Goldberg (1996) gives authentic examples such as (1) and (2). (1) My father frowned away the compliment. (S. McCauley, 1992, Easy Way Out) (2) The Miami quarterback was boo-ed to the bench. (NPR, October 1995)

She does not, however, examine all the instances of a given construction within a particular genre, nor does she explore the role played by their frequency in actual use. And like most construction grammarians, she supplements her analysis with invented examples. What appears from the above is that, in their most extreme forms, the bottom-up approach and the top-down approach, which should be seen as situated along a continuum, present both strengths and weaknesses. The former approach makes a commitment to the integrity of the data, but lacks theoretical sophistication and tends to satisfy itself with (sometimes “anecdotal”) quantitative observations. The latter, on the other hand, shows a wider theoretical perspective, but fails to do justice to the richness of the data, examining only part of them and usually neglecting their quantitative aspect. This contrast between the two approaches can be summarised by Partington’s (1998: 144–145) statement that topdown linguists think “in terms of wider rules but [are] forever in search of ways to refine the details of the information they [have]”, while bottom-up researchers “investigate the behaviour of individual words but are constantly on the lookout for similarities of collocational behaviour among items which would enable them to be grouped together”. Each of the two approaches is good at investigating different aspects – data for the bottom-up approach, theory for the top-down approach (the starting point of each approach) – but neither is good at both. One way to make sure that both data and theory are treated efficiently, therefore, is to combine a bottom-up approach and a top-down approach, thus capitalising on the strengths of each of them and making up for their weaknesses. This possibility is discussed in the next section.

9

10


2.2.3 To-ing and fro-ing between data and theory Ideally, theory and data should be given equal importance, both of them contributing what they can to the investigation of a particular phenomenon. The data show how the item or construction under investigation behaves, while the theory provides a framework for explaining why it behaves the way it does. Since, as we have seen, the bottom-up approach tends to favour the data (the “bottom” part) whereas the top-down approach tends to favour the theory (the “top” part), combining the two approaches makes it possible to put data and theory on an equal footing. Thanks to such a combination, the linguist can avoid the problems of purely corpus-driven approaches (especially the lack of theoretical sophistication) and those of purely corpus-tested approaches (particularly the poverty of the data), while taking advantage of their respective assets. Some authors, considering the difficulties involved, are not in favour of such integration of theory and data. Swales (2002: 163), who envisages this integration for corpus data and discourse analysis, recognises the advantages of this “working from both ends toward the middle”. Yet, he thinks that “the procedural differences between discoursal top-down and corporist bottom-up approaches remain to be resolved” (ibid., emphasis original) and describes his own attempts as “a discourse-analytic shot in the dark” (ibid. 162). Others, by contrast, are more hopeful of a symbiosis of the two approaches. This is the case of Aarts (2006: 405), who notes that the corpus-tested and corpus-driven approaches are “complementary rather than mutually exclusive”. Similarly, Partington (1998: 145) refers to top-down grammarians who “burrow down into language whilst [bottom-up researchers] burrow up until eventually they will meet in the middle”. He gives as a good example of such a symbiosis Altenberg’s (1991) study of the collocational behaviour of individual items, resulting in the refinement of the two sub-groups of amplifying intensifier adverbs proposed in Quirk et al. (1985). This book, rather than working from data and theory towards the middle, aims at to-ing and fro-ing between data and theory. The objective is not to provide “some data and a little theory”, but to offer a thorough analysis of a representative selection of authentic data and a fully-fledged linguistic theory. In addition, a reciprocal relation between the two is necessary. The theory should enlighten the data, but the data should also shed light on the theory. In the next section, we examine the compatibility between corpus linguistics and theory, and explain why cognitive linguistics is a viable option when looking for a theoretical framework that can be combined with the use of corpora.

2.3


Combining corpus linguistics and cognitive linguistics

2.3.1 “Computer-aided armchair linguistics” In a well-known article, Fillmore (1992) draws a caricature of the armchair linguist, who does the thinking, and the corpus linguist, who studies concrete facts. After showing their respective shortcomings, he concludes that “the two kinds of linguists need each other. Or better, that the two kinds of linguists, wherever possible, should exist in the same body” (ibid. 35). This co-existence, or “computer-aided armchair linguistics”, is all the easier to accept if one considers corpus linguistics primarily as a methodology (see Section 2.1), as appears from Aarts’s (2000: 7) quotation: Theoretical linguists couch their work in the terms of a particular theoretical model; corpus linguists may or may not use models, the important point is that they use data from corpora in their claims about language. It follows from this that there is in principle no conflict between being a corpus linguist and being a theoretical linguist.

In this view, corpus linguistics provides the data and the methodology, whereas the theory provides the background against which the data can be analysed and explained. If a linguistic theory involves examining a linguistic phenomenon from a particular perspective and if corpus linguistics implies the use of authentic data to study the linguistic phenomenon, there should be no constraints on the types of theories that can be combined with corpus linguistics. There are admittedly theories whose tenets seem a priori less compatible with the principles of corpus linguistics than others. Thus, corpus linguistics has often been said to contrast sharply with the Chomskyan approach, in that it focuses on performance rather than competence, and on description of specific languages rather than universals, it puts the quantitative approach on a par with the qualitative approach, and it has more empiricist, rather than rationalist views of scientific inquiry (see Leech 1992). Chomsky himself has denigrated the corpus linguistic approach more than once, claiming for instance that corpus linguistics “doesn’t mean anything” (Andor 2004: 97) or even “doesn’t exist” (Aarts 2000: 5). Yet, as pointed out by Granger (1998: 3), the two approaches do not necessarily exclude each other. Aarts (2000) himself admits that Chomsky is right in emphasising the need to “go beyond looking at data” (ibid. 6) and ask probing questions. In April 2001, the relation between generative grammar and corpus linguistics gave rise to a long and heated debate on the Corpora List. While it was pointed out that “Chomsky has very little personal interest in the application of his theories in any practical

11

12


pursuits” (James L. Fidelholtz, April 5, 2001), several linguists agreed that there was no inherent incompatibility between generative linguistics and corpus linguistics (see Aarts 1992 for a practical application) and, more generally, that the distance between theoretical linguistics and corpus linguistics “is being narrowed somewhat as each camp begins to realize that the other has useful methods to offer” (Mary D. Taffet, April 1, 2001). The issue, however, was taken up again (even more vehemently) in the summer of 2007, and almost led to the exclusion from the Corpora List of more theoretical debates considered as “having little to do with corpus linguistics” (Antoinette Renouf, September 14, 2007). Be that as it may, once we accept the “non-incompatibility” of corpus linguistics with such an a priori self-contained theory as generative grammar, it becomes easier to envisage its combination with other linguistic theories, which perhaps have closer affinities with corpus linguistics, such as discourse analysis (see Swales 2002 above) or functional linguistics (Butler 2004). Another theory whose combination with corpus linguistics seems promising is cognitive linguistics. In Section 2.3.3, we will explain the relevance of such a combination and show how this combination has been successfully operationalised in some recent studies. Before this, however, a brief outline of the basic principles of cognitive linguistics is in order.

2.3.2 The theoretical framework of cognitive linguistics Cognitive linguistics refers to a set of approaches to language that began to emerge around 1970.3 Initially, the model was elaborated as a reaction against generative grammar: in opposition to the “autonomy thesis” of generative theory, which states that the human brain contains a specific and autonomous organ that is exclusively devoted to language use and understanding, cognitive linguists claim that language is an integral part of human cognition. As Croft & Cruse (2004: 2) explain, the organization and retrieval of linguistic knowledge is not significantly different from the organization and retrieval of other knowledge in the mind, and the cognitive abilities that we apply to speaking and understanding language are not significantly different from those applied to other cognitive tasks, such as visual perception, reasoning or motor activity.

3. It should be noted that the term “cognitive” is sometimes used to cover the Chomskyan paradigm as well (cf. Schwarz 1992) – although, as Seuren (2001: 235) points out, the interest of the Chomskyan tradition in cognition “has amounted to little more than lip service” (see also Geeraerts 1989: 591). The term as used here will only refer to the post-1970s tradition.


They add that the particular configuration of cognitive abilities needed for language “is probably unique to language, but the component cognitive skills required are not” (ibid.). Besides this denial of an autonomous linguistic faculty, cognitive linguistics is, according to Croft & Cruse (2004), characterised by two other major hypotheses. One is the view of grammar as conceptualisation. It is argued that “conceptual structure cannot be reduced to a simple truth-conditional correspondence with the world” (ibid. 3). Reality may be construed in different ways, and language is made to reflect these construals, for example by choosing between several alternative expressions for one and the same situation (consider the classic example of the glass that is half empty or half full). This is what is referred to as “imagery” in cognitive grammar. Next, knowledge of language is said to emerge from language use (“usagebased approach”). Speakers’ internal grammar gradually builds up on the basis of the specific utterances they encounter – a claim which, it will be noticed, is in stark contrast to the nativism postulated by generative grammar, according to which children are born with an innate knowledge of the algebraic core of language. As rightly emphasised by Geeraerts (1995: 114), cognitive linguistics is “not a single theory of language, but rather a cluster of broadly compatible approaches”. Among these are: i. cognitive grammar (Langacker 1987, 1991), which maintains that grammar, like lexicon, is intrinsically symbolic in nature and hence meaningful, and that meaning should be equated with conceptualisation; ii. cognitive semantics (Talmy 2000a, b), which focuses on conceptual content and its organisation in language; iii. (radical) construction grammar (Goldberg 1995, 2005, Croft 2001), which views grammar as an inventory of constructions that are claimed to carry a meaning of their own, independently of the words in the sentence (e.g. the Caused Motion Construction illustrated by She sneezed the napkin off the table); iv. conceptual metaphor theory (Lakoff & Johnson 1980, Lakoff 1987, Sweetser 1990), which asserts the central role of metaphors (e.g. argument is war, love is a journey) in thought and everyday language; v. conceptual blending (or integration) theory (Fauconnier 1985, Fauconnier & Turner 1998), which allows for the blending of “mental spaces” (e.g. the speech event space and the decision-making space in If I were you, I would accept his invitation, where the speaker is mapped onto the hearer); vi. frame semantics (Fillmore 1982, Petruck 1996), according to which a word (e.g. the verb sell) can only be understood against the background of a frame (the frame of commercial transaction) made up of a number of frame elements (Buyer, Seller, Goods, etc).

13

14


However, precisely because of the compatibility between the different approaches (which is already apparent from the above list), it is possible to adopt a cognitively-inspired framework without necessarily applying all the tenets of any one of these approaches. In addition to the three major hypotheses described above, a number of concepts and interests are common to all (or most) cognitive approaches and can therefore easily be integrated into a broad cognitive framework. Among the topics of special interest to cognitive linguists, Geeraerts (1995: 112) lists the following: the structural characteristics of natural language categorization (such as prototypicality, systematic polysemy, cognitive models, mental imagery and metaphor); the functional principles of linguistic organization (such as iconicity and naturalness); the conceptual interface between syntax and semantics (as explored by cognitive grammar and construction grammar); the experiential and pragmatic background of language-in-use; and the relationship between language and thought, including questions about relativism and conceptual universals.

It is this broad cognitive framework that will lie at the basis of this study, and several of the topics mentioned above will have a role to play in the analysis.

2.3.3 Compatibility of cognitive linguistics with corpus linguistics In 1996 Teubert, talking about cognitive linguistics and corpus linguistics, noted the restricted communication between the two paradigms, adding that it was “not (or not yet) possible to unify both views in an integrated theory” (1996: vi). In 2005, at the ICAME panel discussion referred to in Section 2.1, he expressed the same view, underlining the contrast between cognitive linguistics, which posits categories beforehand, and corpus linguistics, which extracts categories from the data (see also Teubert 2008). And indeed, it must be recognised that until recently the cognitive theories have tended to ignore corpus evidence and instead rely on the linguistic judgements of their proponents. It is symptomatic, for example, that two seminal works within cognitive linguistics, Lakoff & Johnson’s (1980) Metaphors We Live By and Langacker’s (1987, 1991) Foundations of Cognitive Grammar, are based on invented examples and introspection. Yet, the description of cognitive linguistics as a usage-based model (see Barlow & Kemmer 2000, Croft & Cruse 2004: 291–327), that is a model in which “[s]ubstantial importance is given to the actual use of the linguistic system and a speaker’s knowledge of this use” (Langacker 1987: 494, emphasis added), almost automatically calls for the use of naturally-occurring data – and Bybee (2006), for instance, makes a convincing case for including usage events, and in particular frequency of use (as attested in corpora), in a model of cognition. In addition,


the claim that “the processing constitutive of language has to be studied and described with reference to the social and contextual interaction of actual language use” (Langacker 1997: 248) is close to the corpus linguistic principle that “a word’s meaning can only be determined from its (linguistic and extra-linguistic) context” (Schönefeld 1999: 144). There is at least another element that points to a possible combination of the cognitive and corpus linguistic approaches, and strangely enough, it relates to the objection raised by Teubert about categories. It is true that corpus linguistics extracts categories from the data and cognitive linguistics posits them beforehand, but this is no different from saying that corpus linguistics starts with the data, while cognitive linguistics starts with the theory. What is crucial is that both paradigms recognise the existence of categories. More fundamentally, they both see category membership as a matter of degree rather than a yes-or-no question. In cognitive linguistics, this principle is expressed by means of the concept of prototypicality, originally a concept from psychology, according to which categories are organised around a maximally representative example (the prototype). Depending on the similarity they exhibit with the prototype, the other members occupy a more or less central position within the category. As a consequence, categories have no clear boundaries and overlap with one another. It is the same fuzziness that corpus linguistics has revealed. Mair (1994: 128) rightly points out that “if there is one lesson to be learnt from studying and analysing corpus examples it is the ‘basic non-discreteness of categories’”. Čermák (2002: 273), too, evokes the hazy edges and scalar nature of corpus data and the end of “black-andwhite truths and clear-cut classification boxes”. One important illustration of this is the discovery, in corpus linguistics, that lexicon and grammar interact at every step and are ultimately inseparable (Francis 1993). This observation has led to the extensive study of a hitherto largely unexplored level of language, commonly referred to as “lexico-grammar” and implying the existence of a continuum between lexicon and grammar. The same continuum is acknowledged in cognitive linguistics by e.g. Langacker, for whom syntax does not constitute an independent level of language. Like lexicon, it is said to be intrinsically symbolic and therefore no clear borderline can be drawn between the two. They form, with morphology, “a continuum of symbolic structures, which differ along various parameters but can be divided into separate components only arbitrarily” (Langacker 1987: 3). More generally, Schönefeld (1999: 149) notes the “close affinity between what corpus linguistics reveals about language as it is used and the way the functioning of language is explained by (…) cognitive linguistic models” and goes on to demonstrate how the data and facts revealed by a corpus linguistic approach to the “idiom principle”, delexicalisation and the description of meaning can easily be accommodated within a cognitive framework. To take the example of the idiom

15

16


principle, according to which speakers rely on a large number of (semi-)preconstructed phrases that represent single choices (see Sinclair 1991: 110ff.), it is shown to be relevant to cognitive linguistics as well, as illustrated for instance by the following quotation, where Langacker (2002: 262) states that it is psychologically plausible and reasonable to assume that many structures are learned as established units even when they also follow from general principles – the computability of a structure does not in principle preclude its learnability and inclusion as a distinct element in the cognitive representation of the linguistic system.

What precedes, in fact, suggests that even if one considers corpus linguistics as more than a mere methodology (see the discussion in Section 2.1), its combination with cognitive linguistics still remains possible. The existence of a continuum between lexicon and grammar, for example, or the idiom principle can both be seen as theoretical concepts that have come out of corpus linguistics, and, as we have just seen, they find a clear echo in cognitive linguistics. In the next section, it is argued that the combination of corpus linguistics and cognitive linguistics is not only possible, but also desirable, since it adds value to both paradigms.

2.3.4 Corpus linguistics and cognitive linguistics: Added value Compatibility between two frameworks is not enough to justify their combination. Such a combination is only really useful if it can be demonstrated to add value to each of the frameworks. That this is the case of a model integrating corpus linguistics and cognitive linguistics is clear if we consider what each approach can gain from the integration. For corpus linguistics, the combination with cognitive linguistics provides the theoretical sophistication that is often said to be lacking in corpus linguistics. As Schönefeld (1999: 165) notes, corpus linguistics still presents “massive gaps”. The interpretation it gives of the findings is still too sketchy and it fails to cover a number of important aspects or merely mentions them in passing (e.g. semantics, pragmatics, motivations for linguistic facts, or assumptions about the psycholinguistic procedures of language use). Because such aspects are important in cognitive linguistics, they become part of the integrated model, and corpus linguistics opens up to new issues, while developing its explanatory power and psychological plausibility. From the point of view of cognitive linguistics, the combination with corpus linguistics confirms (and in fact actualises) its status as a “usage-based model”,


truly reflecting usage as attested in corpora. The model goes beyond introspective judgements, and by using the tools and methods of corpus linguistics, cognitive linguistics gains in descriptive adequacy and linguistic plausibility, since it bases its claims on empirical evidence. This mutual enrichment can be summarised by Schönefeld’s (1999: 153) statement that the cognitive assumptions represent a valuable complement to the facts revealed by the analyses of massive amounts of language data in that they go beyond the stage of “merely” stating facts. From the opposite perspective, the facts drawn from data analyses provide additional evidence underpinning the hypotheses at which cognitive linguists have arrived.

The model resulting from the combination of corpus linguistics and cognitive linguistics (which one could call “cognitive corpus linguistics”, see Grondelaers et al. 2007 or Arppe et al. forthcoming) can therefore lay claim to descriptive and explanatory adequacy, while exhibiting empirical foundation and theoretical sophistication. In addition, it views language both as a textual and conceptual phenomenon, and seeks to investigate the relationship between the two. This model also addresses a wider range of issues than corpus linguistics or cognitive linguistics taken separately. While the corpus linguistic dimension results in a focus on “formal” aspects, such as frequency or co-occurrence, the cognitive dimension helps generate interest in more semantic aspects or in mental phenomena underlying the encoding and decoding of language (for example, the fact that frequency and co-occurrence, precisely, are not merely formal matters, but do have an effect on cognitive representations, and hence on language structure itself, see Bybee 2006). Corpus linguistics and cognitive linguistics, therefore, are not only compatible, but also complementary. They have a different status and approach language from different perspectives, but these differences are not irreconcilable. On the contrary, they enrich one another and create a more comprehensive and solid model. In the next section, we examine some of the attempts that have been made to combine the two paradigms.

2.3.5 Illustrations Over the last few years, several linguists have successfully combined cognitive linguistics and corpus linguistics, taking advantage of the added value offered by their complementarity (see, for instance, the studies brought together in Gries & Stefanowitsch 2006).

17

18


Probably one of the largest projects carried out within this twofold framework is the FrameNet Project, developed at the International Computer Science Institute, Berkeley, and aiming to define word senses and specify the semantic and syntactic combinatorial properties of the words in the lexicon (see Baker et al. 1998, as well as the special issue of the International Journal of Lexicography edited by Thierry Fontenelle in 2003). While the project lies within the (cognitive) theory of frame semantics, it relies on the methodology of corpus linguistics. The cognitive inspiration is obvious from the organisation of word senses around frames such as the perception frame, the imitation frame or the frame of coming-to-believe, each of which is made up of a number of frame elements (e.g. the perceiver and the phenomenon for the perception frame), which are realised by certain syntactic functions in actual language. In accordance with the corpus linguistic approach, all generalisations are based on authentic data and each word sense is accompanied by a few concordance lines illustrating the contexts in which it may occur and specifying the different uses it may have. The phenomenon of collocation, dear to corpus linguists, is not neglected either in the FrameNet database (Ruppenhofer et al. 2002). The same approach has been adopted by some individual authors, albeit on a smaller scale, e.g. Boas’s (2001) investigation of motion verbs in English and German. The corpus linguistic approach has also been applied to Langacker’s cognitive grammar. Mukherjee (2004: 85), for instance, shows how “the consideration of corpus data in cognitive grammar leads to an innovative and realistic model of speakers’ linguistic knowledge”. More precisely, he argues that frequency figures should be included in a cognitive model of language, following Schmid’s (2000: 39) “From-Corpus-to-Cognition Principle”, which establishes a link between frequency in text and entrenchment in the cognitive system. Lemmens (2002) uses the same combination of Langackerian principles and corpus evidence to investigate Dutch posture verbs. Using corpus-extracted occurrences of posture verbs, he primarily explains the choice of one particular verb, not on the basis of real properties of the entities to be located, but on the basis of “image schemata”, that is abstract cognitive structures derived from our physical and cultural experience of the world. Likewise, the collostructional approach (see Stefanowitsch & Gries 2003 and Chapter 8) explicitly builds on the combination of the frameworks of cognitive linguistics (and more precisely construction grammar) and corpus linguistics. The term “collostruction”, in fact, is a blend of “collocation” and “construction”. This approach, relying on the cognitive assumption that grammatical structures are meaningful units, uses the methodology of (quantitative) corpus linguistics to identify the words that are statistically associated with a given construction. On the basis of this list of words (the “collexemes” of the construction), it determines


the meaning(s) most typical of the construction. (See also Lemmens 1998 for a similar approach.) As a final example, consider conceptual metaphor theory. It was mentioned earlier that Lakoff & Johnson (1980), who contributed to the popularity of metaphors, did not base their findings on a systematic analysis of naturally-occurring data. Some of their followers, however, have chosen to use authentic material. This is the case, among others, of Wikberg (2003), Deignan (2005) and the collection of papers in Stefanowitsch & Gries (2006). Semino (2006), for example, conducts a study of metaphors for speech activity in a corpus of written narratives. She finds that the conceptual metaphor argument is war, initially proposed by Lakoff & Johnson (1980), is too restrictive and should be replaced by a more general metaphor, which Semino refers to as antagonistic communication is physical conflict. In all these works, supplementing the cognitive approach with corpus data has enabled the authors to go further in their findings and thoughts than would have been the case if they had relied on one of the two paradigms only. As will become clear as we go along, the same gain may be achieved when investigating periphrastic causative constructions. The next section briefly shows how cognitive linguistics and corpus linguistics will be combined here to tackle this particular topic and outlines some of the reasons why such a combination is expected to provide good results.

2.3.6 The case of English periphrastic causative constructions The framework within which the present discussion will take place is a combination of cognitive principles and corpus evidence. The influence of the cognitive paradigm will be noticeable from the very start, in the selection of the topic of investigation. It is quite common in the literature to deal with periphrastic causative verbs (their complements, their meaning, their stylistic features, etc) rather than periphrastic causative constructions, on the assumption that all constructions associated with one verb share the same characteristics. Thus, get is thought to have the same meaning whether it is used in [X get Y Vto-inf ], [X get Y Vpp] or [X get Y Vprp]. Construction grammar, however, claims that if two or more constructions are formally different, they must be semantically or pragmatically different (Goldberg’s [1995] “Principle of No Synonymy”, see Section 5.1) and therefore deserve to be considered individually. Following this principle, a distinction will be made between ten causative constructions, rather than four causative verbs. These constructions are listed and exemplified in Table 1.

19

20 Corpus, Cognition and Causative Constructions

Table 1. English periphrastic causative constructions4 [X cause Y Vto-inf ]

The recession caused the price of aluminium to fall.

[X get Y Vto-inf ]

At one time we couldn’t get Jessy to talk.

[X get Y Vpp]

We’ll get everything sorted out this week.

[X get Y Vprp]

Couldn’t get these earphones working.

[X have Y Vinf ]

I had Elsie go on a Wednesday night.

[X have Y Vpp]

Did you have the blades sharpened?

[X have Y Vprp]

You better not have that tape working, is it on?

[X make Y Vinf ]

But I made him put his coat on.

[X be made Vto-inf ]

They’re being taken to court and made to pay.

[X make Y Vpp]

They made their voices heard at the conference.

The cognitive theory of frame semantics will also be relevant to the analysis of periphrastic causative constructions. Not only does frame semantics bring the different causative constructions together under one and the same frame, the frame of causation, thus offering a unified account of these constructions. It also establishes a relation between semantics and syntax, showing how frame elements are realised at the syntactic level. Four different frame elements will be recognised in periphrastic causative constructions, namely causer, causee, effect and patient.5 In addition, following Shibatani (1976a), the first predicate will be called the “causing event”, and the second one as the “caused event”. This terminological apparatus is illustrated in the following sentence: (3) The teacher made the student read the book causer causee effect patient [causing event] [caused event]

4. All corpus examples are followed by a code in angle brackets which gives the reference of the sentence(s). The first part of the code indicates the corpus from which the sentence is extracted (BNC = British National Corpus, ICE-GB = British component of the International Corpus of English, ICLE = International Corpus of Learner English) and the second part locates the sentence within the corpus according to the reference system specific to the corpus (in the case of the BNC, a letter has been added to signal whether the sentence comes from the written [W] or spoken [S] part of the corpus). Elicitation data (see Section 3.2) will be given a code starting with EL, followed by the number of the test item and the identification number of the respondent. 5. When referring to the frame elements making up the causative construction, these terms will be written in small caps. Small caps are also used for lemmas, standing for all the inflected forms of a word in the formal representation of causative constructions (cf. Table 1).


The causer is the entity bringing about the caused event, whereas the causee is the entity that is changed or influenced by the causer and carries out the effect of the caused event. The effect is the event or state brought about by the causee, while the patient, when present, is the object of the effect. The causing event is that which causes, and the caused event is that which is caused. A frame element may vary in its syntactic realisation. Consider the following sentences: (4) I had the boy leave. (5) The technician had the video working. (6) The emperor had the slave imprisoned.

Whereas in the first and second sentences the causee is realised as a direct object (as would appear from its pronominalisation: I had him leave), in the third it is the patient that takes on the function of direct object, while the causee is left unexpressed (cf. The emperor had [his guards] imprison the slave). In addition, a number of cognitive concepts will be used in the course of the investigation (and explained as we go along). They will serve both as a starting point for the analysis of the data (Chapters 4 to 6) and as an explanation for the results of the corpus study (Chapters 7 to 9). Among these concepts are “action chain”, “imagery”, “scenario” and “psychodynamics”. The main contribution of corpus linguistics will be the emphasis on authenticity of usage. All the claims about the features displayed by periphrastic causative constructions will rely on the careful observation and systematic analysis of large amounts of corpus data. The quantitative analysis will be based on the constructions extracted from a ten-million-word subset of the British National Corpus, which amounts to some 3,500 instances. In the qualitative analysis, examples coming from other authentic sources (e.g. ICE-GB, the British component of the International Corpus of English) will occasionally be given. The presence of corpus linguistics will also be visible in the methods used to analyse the data. Frequency and probability will have an important role to play in the analysis, as well as statistical calculation, in the form of chi-square tests and hierarchical feature selection. Furthermore, great attention will be paid to the textual environment of causative constructions: what words surround the construction, what verbs recur in the effect slot, etc. Finally, the influence of medium and genre on the behaviour of causative constructions will also be studied, thus reflecting corpus linguists’ interest in stylistic variation. A corpus analysis of English periphrastic causative constructions appears to be particularly welcome when one examines the literature devoted to this linguistic phenomenon. While it has given rise to “a vast amount of literature” (Song 1996: 2), its treatment in grammars (and in linguistic works in general) is far from satisfactory. As Altenberg & Granger (2001: 184) point out, “[i]t is very difficult to

21

22


find a good description of the usage differences between these verbs” – let alone between the constructions themselves. The main weaknesses of the descriptions of causative constructions, in fact, can all be shown to arise from their lack of empirical foundation. Studies all too often rely on their authors’ introspection, rather than authentic data, which results in contradictions (most strikingly at the semantic level), errors in grammaticality judgements, neglect of lexical and stylistic aspects (more difficult to evaluate on the basis of introspection alone), or atypical examples. A number of linguists have investigated causative constructions on the basis of corpus data, but their goals were essentially different from those pursued here. Stocker (1990) and Hollmann (2000, 2003) approach causative constructions from a basically diachronic perspective. Stefanowitsch’s (2001) study is based on American spoken data. Cottier (1991) gives a very detailed description of the four causative verbs investigated here, but her approach is purely qualitative and no attempt is made to provide a quantitative analysis. In a series of articles on have and get, Ikegami (1989, 1990a, 1990b) deals with syntactic patterns, not semantic constructions, and therefore makes no distinction between causative constructions and, say, experiential or existential constructions (see Section 3.1.1.2 on these types of constructions). Some studies have been devoted to one or several periphrastic causative verbs/constructions in present-day British English (e.g. Kemmer 2001 on causative make), but they are small-scale studies, not aiming at the degree of exhaustiveness which is set as a goal here. Causative constructions also lend themselves well to a cognitive approach, as witnessed by studies such as Kemmer & Verhagen (1994) or Talmy (1986). Cognitive linguistics is an essentially mentalistic theory, with mind and conceptualisation at its core. As such, it offers a suitable framework for the study of causation, which has often been described as one of the most fundamental categories of the mind (see Kant 1934, who regards causation as a “transcendental notion”, imposed by the mind upon experience). Moreover, for reasons that will become clear in the next chapters, the cognitive approach makes it possible to cast new light on causation and causative constructions. Given that corpus linguistics and cognitive linguistics are both suitable frameworks for the study of periphrastic causative constructions, likely to improve our knowledge of this (not so well-known) phenomenon, the combination of the two is expected to provide even more valuable insights, as already demonstrated by linguists like Stefanowitsch (2001) or Hollmann (2003). Before showing how this particular combination is operationalised, however, we have to discuss another combination which will lie at the basis of this study, namely that between corpus and elicitation data. This is the subject of Chapter 3.


2.4 Summary While the status of corpus linguistics is still to a certain extent unclear, many linguists see it more as a kit of tools and techniques than as an independent subdiscipline of linguistics. As a methodology, however, corpus linguistics tends to be unsatisfactory when used on its own. It often limits itself to a quantitative analysis and a descriptive perspective, and fails to reach a degree of generalisation high enough to add an explanatory dimension. This drawback may be overcome by combining corpus linguistics with a solid theoretical framework. The framework that has been chosen for this study of periphrastic causative constructions is cognitive linguistics. Not only is it compatible with corpus linguistics, but it is in fact complementary, resulting in mutual enrichment between the two approaches. This integration is therefore expected to throw new light on a phenomenon, that of causative constructions, which has been the topic of many studies but is still surprisingly obscure. The integrated framework proposed in this chapter finds a parallel in the data used to analyse causative constructions. As will be described in the next chapter, the data consist in a combination of corpus data and more experimental data (in the form of elicitation tests). While the former give access to authentic language performance in natural settings, the latter make it possible to address cognitive issues such as competence, conceptualisation or salience. The combination of the two types of data, like the combination of corpus linguistics and cognitive linguistics, will allow for a more comprehensive approach to causation.

23

chapter 3

English causative constructions Collecting the data

As suggested earlier, many of the weaknesses of the descriptions of causative constructions found in the literature may be related to their lack of empirical foundation. This study, by contrast, relies on solid empirical evidence. The evidence takes two forms: corpus data, which represent naturally-occurring language, and elicitation data, which correspond to directed language production. In this chapter, the methods to collect the two types of data are described. It is explained why the automatic extraction of the causative constructions from the corpus (a subset of the British National Corpus) has to be followed by a stage of manual post-editing, and how the results of the elicitation test may supplement the findings made on the basis of the corpus data. The chapter ends with a presentation of the “CorpusCognition Integrated model”, which summarises the frame of the present study as proposed in this and the preceding chapter.

3.1

Corpus data: A semi-automatic method

The choice of a methodology to extract periphrastic causative constructions from a corpus has to take account of two main problems, namely the difficulty of finding the right resources to extract syntactic structures automatically, and the ambiguity of sentences having the formal structure of causative constructions. As a result, the best option appears to be a semi-automatic method, combining a fully automatic search with manual post-editing. In this section, we first review the problems involved in retrieving causative constructions, and next we describe the tools and methodology used within the framework of this study.

3.1.1 Two problems with the extraction of the data 3.1.1.1 The retrieval of syntactic structures As suggested in Chapter 2, the notion of corpus has come to imply machine-readable. This technological support, together with the wider availability of personal computers, has made it possible for corpus linguists to automate a number of tasks


that used to be carried out by hand. Instead of spending months or even years perusing written texts and noting down relevant examples by means of a complex system of stencilled sheets, punched cards and paper stickers (see Olofsson 1981), today’s linguists can use their own computers to store huge bodies of text, representative of several varieties of language, and search through them automatically in a matter of seconds. Not all aspects of language have equally benefited from this automation, however. Lexical items can be retrieved easily from raw corpora with any text retrieval software. Not only does it represent a considerable saving of time and energy over manual data analysis, but the results are also likely to be more accurate, since the human mind is prone to errors when it comes to dull repetitive tasks such as counting linguistic events, which is precisely what computers are very good at (Mason 2000a: 4). More complex grammatical phenomena such as syntactic structures, on the other hand, are still difficult to extract automatically, mainly for lack of appropriate tools. These tools are of two kinds. First, one might need corpora that contain more linguistic information than plain orthographic text. Tagged corpora, i.e. corpora annotated with part of speech (POS), and parsed corpora, where the syntactic constituents of sentences are identified, are particularly useful to retrieve syntactic structures (although they impose a linguistic analysis on the data which may not always suit the user). Second, in order to exploit such annotated corpora, one needs adequate software, since basic text retrieval programs are not always capable of handling the annotation schemes of tagged or parsed corpora. These resources may be unavailable (query systems for parsed corpora, for instance, are not very common) or, if available, may leave something to be desired from a qualitative point of view (e.g. corpus annotation inaccurate or inadequate for one’s purposes, limited power of the query system) or a quantitative one (e.g. small size of certain corpora, especially when they have been annotated with a fine-grained annotation scheme and/or have been checked manually). As a consequence, we note an imbalance between lexical and syntactic studies in the corpus-based literature (Kennedy 1998: 8) and, more generally, a tendency among corpus linguists to address research questions that are easy to investigate and neglect those whose investigation would require more effort in terms of the retrieval of the data (Gilquin 2002). Oostdijk & de Haan (1994: 41), in a plea for more syntactic research in corpus linguistics, claim that [l]arge-scale quantitative studies of syntactic structures and phenomena are long overdue. While word frequency counts and concordances have been a common good to the linguistic community for quite some time now, corpora that have undergone a detailed syntactic analysis are few, and so are the quantitative studies that are based on these.

Chapter 3. Collecting the data

While in the above quotation Oostdijk & de Haan explicitly refer to the use of parsed corpora (“corpora that have undergone a detailed syntactic analysis”) for the study of syntactic structures, it should be emphasised that alternative methods do exist – although parsed corpora, of course, remain the “royal road” to the retrieval of such phenomena. Biber (1988), for instance, used a tagged corpus to study a number of syntactic features, some of which were extracted by means of complex algorithms combining lexical items, POS-tags and unspecified strings of words. It should however be noted that, sophisticated though it is, Biber’s method also reveals the limitations of automatic extraction. For one thing, manual intervention is necessary in some cases to discard irrelevant hits. For another, some of Biber’s algorithms miss part of their target, as convincingly demonstrated by Ball (1994). In other words, the search is problematic both in terms of precision, i.e. “the proportion of retrieved materials that are relevant” (Salton 1989: 248), and recall, i.e. “the proportion of relevant materials retrieved” (ibid.). In the absence of better suited tools, however, one has to make the best of the available resources. The same sort of compromise is involved in the extraction of periphrastic causative constructions. The fastest track is the use of a parsed corpus, combined with an appropriate query system which could search for, say, all the occurrences of the verb cause followed by a noun phrase and a to-infinitive. The more finegrained the parsing, the better the results. ICE-GB, the British component of the International Corpus of English (ICE), has been encoded with an extremely delicate level of annotation and manually corrected (see Wallis 2002). It includes a tag for “transitive” verbs, that is verbs followed by a noun phrase and a nonfinite clause, where the noun phrase may be described both as the object of the main verb and the subject of the non-finite clause (Fang 1996: 145–146). Thanks to this tag, ICECUP, the Corpus Utility Program specifically designed to process and query ICE-GB, is able to distinguish between verbs used with a direct object, tagged as “monotransitive” verbs (e.g. to cause an accident), and periphrastic causative verbs used with a clause, tagged as “transitive” verbs (e.g. to cause the show to be cancelled). While the annotation scheme of ICE-GB does not help disambiguate semantically ambiguous structures (see Section 3.1.1.2), it enables the user to retrieve causative constructions with a very long causee, as in (7), or with a non-canonical order, as in (8), where the causative verb and the effect directly follow each other and the patient (the tunnel [that]) precedes the causer (Olivia Vereno). The recall rate of queries performed with ICECUP is, as a consequence, very high.6 6. Missed causative constructions seem to be the result of errors in parsing. While get in the first sentence below is, as expected, labelled as a “transitive” verb, it is not in the second sentence, despite its similarity to the first one. The second sentence, therefore, cannot be retrieved

27

28


(7) Pig’s friends’ ambition was to make a nervous, forty-year-old, thin violinist from Hungary with long, black hair laugh or cry by making faces at him. (8) She’d apparently been lying unconscious in the tunnel Olivia Vereno had had constructed in the rock as one of several follies that gave the estate its air of careful grandeur.

The downside, however, is the size of ICE-GB. With about one million words, this corpus may be suitable for the study of frequent phenomena, but not for a relatively uncommon structure such as the periphrastic causative construction, whose relative frequency per million words varies from 0.9 for [X make Y Vpp] to 126.3 for [X make Y Vinf ] (relative frequencies in ICE-GB). In addition, given the current state of parsing (in 1993 Black referred to the “dismal state of the art in the parsing of English” [ibid. 5], and although progress has been made since then, parsers have not yet reached the level of accuracy of automatic taggers), a large parsed corpus, if available, is unlikely to be the ideal solution, as the odds are that the annotation of this corpus would be less delicate and/or less reliable than that of a small and manually corrected corpus like ICE-GB. One therefore has to turn to alternative methods of extraction, less automatic but more reliable. Theoretically, we could extract all the occurrences of cause, get, have and make from a raw corpus and manually select those occurrences that are causative. This method, however, seems unrealistic, considering the low proportion of causative uses of these four verbs, ranging from some 10% for cause to hardly 0.5% for have (proportion in a 10-million-word subset of the BNC).7 Alternatively, we can make use of a tagged corpus, applying a method similar to that proposed by Biber (1988). More precisely, we can look for all the occurrences of one of the four verbs followed, within a given span, by an infinitive, present participle or past participle. Since the intermediate noun phrase may take on a variety of forms (pronoun, adjective + noun, determiner + noun + relative clause, etc) and since adverbs or other elements may occur between the causative verb and the effect, as in (9), it is preferable to leave the string of words in-between unspecified. This method, like Biber’s, produces irrelevant material, as in (10), where make is followed by an infinitive but the

by the query. (See Duvieusart [2003] on the sometimes inconsistent annotation of –ing clauses in ICE-GB.)

i. Stewart’s looking for instructions lucky to get play moving ii. Let’s get America moving again

7. Only the uses of these verbs in periphrastic causative constructions with a non-finite complement are taken into account in these figures.


infinitive is not dependent on make, but as will be shown in Section 3.1.2, it has a relatively good recall rate – a sine qua non for a reliable analysis. (9) Honestly, I don’t think it matters knowing how to program, or knowing how to make the machines actually work. (10) It will tell us about the individuals’ ability to make and maintain new relationships.

3.1.1.2 Periphrastic causative construction: An ambiguous structure We saw in the preceding section that the main problem with the automatic retrieval of syntactic structures is the unavailability of finely-grained parsed corpora large enough to provide a fair number of examples. Tagged corpora, therefore, appear as the next best solution, despite their poorer precision rate. There is another problem, however, for which neither parsing nor tagging offers a satisfactory solution, namely the semantic ambiguity of certain patterns. Some sentences, while not expressing causation, share the formal structure of causative constructions, as abundantly discussed in the literature, especially for have (see e.g. Chomsky 1965, Poldauf 1967, Freed 1979, Palmer 1988, Ritter & Rosen 1993, Inoue 1995). I had a book stolen (Chomsky’s example), for instance, may be paraphrased as “I caused someone to steal a book for me” (causative construction), “it happened to me that someone stole a book of mine” (experiential construction), “I possessed a book that had been stolen” (lexical meaning) or “I succeeded in stealing a book” (conclusive perfect [Kirchner 1952: 401ff.], included here among causative constructions). Note also the similarity in structure between a causative construction with have such as I had the boiler going and an existential construction like I have two buttons missing on my jacket (Quirk et al. 1985: 1411). As for [X be made Vto-inf ], it may represent both a causative construction and a lexical verb with an infinitive of purpose, as appears from the sentence I wasn’t made to be a judge, taken from Andersson (1985: 149), which may be paraphrased as “No one forced me to become a judge” or “It was not my destiny to be a judge”. Although some sentences are intrinsically ambiguous, most of them can be disambiguated by using one of the criteria mentioned in the literature (see e.g. Inoue 1995 on the disambiguation between causative and experiential have) or, more simply, by referring to the context of the sentence. But while this disambiguation process is quite straightforward for a human subject, who can make a judgement on the basis of both form and content, it is much less so for a machine, which has essentially access to one side of the coin only, viz. form. In fact, so long as substantial advances in the field of semantic annotation are not made, it looks as if the disambiguation of such structures is out of reach of computers, all the more so since, as pointed out by Ritter & Rosen (1993: 526),

29

30


minimal changes may influence the availability of one interpretation or the other, as the following pairs of sentences (taken from Ritter & Rosen 1993) make clear: (11) a. John had his daughter accepted at Dawson College. [most likely interpretation: experiential] b. John paid fifty dollars to have his daughter accepted at Dawson College. [most likely interpretation: causative] (12) a. Sherry had George water her plants. [most likely interpretation: causative] b. Sherry had George overwater her plants. [most likely interpretation: experiential]

This ambiguity and the impossibility, so far, of resolving it by automatic means imply a decreased precision rate and, consequently, further manual post-editing. Although ICECUP is able to retrieve only those sentences that have the formal structure of causative constructions, by taking advantage of the “transitive” feature (see Section 3.1.1.1), it cannot make a distinction between the patterns that are causative and those that are not. An experiential construction, just like a causative construction, will contain a “transitive” verb (as defined above). Among the transitive uses of get in ICE-GB, 28.9% have to be discarded because they are not causative. With have, the proportion of irrelevant materials amounts to 57.5%. In a tagged corpus, this proportion comes on top of the proportion of materials that have to be discarded because they do not have the formal structure of a causative construction.

3.1.2 Semi-automatic method 3.1.2.1 The corpus POS-tagged corpora are less richly annotated than parsed corpora, since they only contain information on word classes, but say nothing about phrases and functions, as parsed corpora do. Yet, as emphasised in the preceding section, they have the advantage of giving access to large amounts of data, whose annotation is arguably more reliable than that of parsed corpora. For these reasons, and although their use involves more manual post-editing than the use of a parsed corpus, tagged corpus data constitute the basis of this analysis. More particularly, the data come from the British National Corpus (BNC), a 100-million-word tagged corpus of contemporary written and spoken British English, collected between 1991 and 1995 in different areas of the United Kingdom. Although the BNC is biased towards written English (90% of writing vs. 10% of speech), the spoken section still represents a good 10 million words. In


addition, it is a well-balanced corpus illustrating a wide range of genres, subject fields and registers. The written component of the BNC is divided into two main sections: informative prose (75%) and imaginative prose (25%). The former is further subdivided into eight categories, viz. natural and pure science, applied science, belief and thought, commerce and finance, arts, social science, world affairs and leisure. The proportions of the different sections and subsections loosely reflect the distribution of books published in the UK (Kennedy 1998: 50–51). The spoken component is also made up of two main sections, namely context-governed material and spontaneous conversations classified by socio-economic class (see Crowdy 1993). The context-governed material consists of recordings of leisure events (e.g. sports commentaries or broadcast phone-ins), business events (e.g. demonstrations, consultations, interviews), educational events (lectures, tutorials, etc) and public and institutional events such as political speeches or parliamentary debate. The spontaneous conversations (demographic part) involve male and female speakers of a broad spectrum of ages and geographical backgrounds. The structure of the whole corpus is shown in Figure 1. In addition, the texts are marked up with metatextual information such as estimated circulation size and perceived level of difficulty of written texts, age and gender of author/speaker, or region where a spoken text was captured. Three different versions of the BNC are available, BNC version 1 (1995), BNC World Edition (2001) and BNC XML Edition (2007). Here, use was made of the BNC World Edition, which presents two major improvements when compared to the first version. First, the BNC World Edition has been completely retagged (Leech & Smith 2000). The tags used are the same as in the first version (CLAWS5 tagset, see http://ucrel.lancs.ac.uk/claws5tags.html), but the tagging system has been improved. Errors remain (about 1.15% of all words according to Leech & Smith), as well as “ambiguity tags”, indicating that the tagger was unable to determine which of two tags was the correct one (approximately 3.75% of all words), but they are less numerous than in the first version of the BNC. The second improvement has to do with the classification of texts into categories. While BNC version 1 was categorised according to a number of domains such as pure science, educational speech or imaginative writing, Lee (2001a: 53) considers that such categories are too broad to be of great worth. For instance, Lee notes, no distinction is made within imaginative prose between novels, short stories and poems. In addition, texts are not always classified correctly, usually due to misleading titles. A text with “lecture” in its title, for example, may well be a classroom discussion or a tutorial seminar. It is this lack of clarity around text classification that has led Lee to propose a new classification based on “genres”, that is text categories associated with “typical configurations of power, ideology, and social purposes” (ibid. 47). These genres include print advertisements,

31

 

32




  

   





              







   



   

 



 

  

   

 

            













Figure 1. Domains in the BNC (Lee 2001a: 50)

 school essays, personal letters, tabloid newspapers, TV documentaries, parliamentary speeches, university-level tutorials and many more.8  Among the 100 million words of the BNC World Edition, a selection was made of some 10 million words, half spoken, half written. It is this corpus, henceforth BNC-10, which will be used as a basis for the investigation of the behaviour of periphrastic causative constructions in present-day British English.9 As one of the aims of this study is to compare the use of these constructions in writing and speech (see Chapter 9), special care was taken to choose typically written and spoken text categories. Categories that are “written-to-be-spoken” (e.g. plays) or “spoken-from-written” (e.g. broadcast news or speeches) were excluded, as they are situated somewhere in the middle of the continuum linking speech and writing. Although plays are usually found in written form, they are meant to be played and therefore seek to imitate speech as closely as possible. By contrast, broadcast news is spoken, but the script on which it is based results in a number of written characteristics: formal language, compact expression, lack of spontaneity, 8. See Lee (2002) for a complete list of the genres and “super-genres” of the BNC World Edition, together with the percentage each of them represents in the corpus. 9. See Hoffmann et al. (2008: 45–46) on why the data in the BNC may be said to represent “present-day English”, even though they contain texts produced at least 15 years ago.






etc. By the same token, the categories of fiction and journalese, albeit large, were discarded due to the elements of spoken material they contain. Koubali (2000) has demonstrated that newspaper articles and editorials contain a large proportion of quotations (i.e. transcription of spoken language) and Cowie (1992) suggests that the time constraints imposed by the production of dailies may result in certain spoken features. As for fiction, the many dialogues most novels contain also contribute to the spoken-like character of this genre (see Oostdijk 1990 on the language of dialogue in fiction). Following the “typical spoken versus typical written” cline (Lee 2001b: 268), with “spontaneous conversations at one end and highly information-heavy, specialist expository texts at the other” (to be related to Biber’s [1988] “first dimension” of involved vs. informational production), the spoken part of BNC-10 (BNC-10S) was made up of conversations, and the written part (BNC-10W) of academic prose. Because the super-genre of academic prose contains over 15 million words and only 5 million were needed, the first 863,200 words or so of each genre within academic prose were taken into account (“humanities”, “medicine”, “natural science”, “politics, law, education” and “social and behavioural sciences”), as well as the whole of the “technology, computing, engineering” genre (686,004 words). The total of BNC-10W amounts to exactly 5,003,007 words. As for BNC-10S, since the genre of conversations contains only 4,206,058 words, it was supplemented with what seemed to be the most spontaneous categories, namely “TV and radio discussions” and “live sports commentaries and discussions”. The latter was included as a whole, representing a total of 33,320 words. For the former, on the other hand, a number of files were eliminated because they contained long stretches of read text. This concerns the weather, traffic and general news, all characterised by a lack of interaction and the existence of a written script. These passages were retrieved by looking for keywords like “headline”, “news”, “weather”, “traffic” or “roadwatch”. Sometimes, the “spoken-fromwritten” passage was not included in the file (13), but when the keyword turned out to be part of a long scripted speech, as in (14), the whole file was discarded (it was not possible to delete only part of a file). When a headline was read in order to launch a debate, as in (15), it did not lead to the exclusion of the file in question. (13) Right, erm, the travel before the six o’clock news. Thought we’d play a little er different version this morning, just to er change things round a bit, but the message is the same. (14) Radio Nottingham News with Andy . Kent police say at least ten people have been killed and a further two are believed to have died in a coach crash on the M two near the Faversham turn-off. The coach carrying forty

33

34


four American tourists on a day trip to Canterbury collided with a van and plunged down an embankment. Dozens of other people were injured. <end of voice quality> From the scene, Martin reports. (15) I’m just <pause> you’re just making me think actually because I saw in one of the papers this morning erm a headline, if I can find it, it said something to the effect that -- yes here it is, in the Independent, it says “TV soaps are first to survive rescheduling” and it goes on to tell us that the British television schedules as we all know were the first casualties of war in the Gulf except for the nation’s daily diet of Australian soap. So erm yes, I’m just trying to scan down to see exactly what happened, but erm it was actually erm as we all know, a major reorganisation of television schedules. Are you surprised that it’s the soaps that survive?

As a result of this “cleaning-up” process, 43 files out of the 53 initial files of the TV and radio discussions were kept and included in the spoken subcorpus, for a total of 561,055 words. BNC-10S thus amounted to 4,800,433 words, and the whole BNC-10 to 9,803,440 words. Table 2 sums up the composition of the corpus, with the number of files and words for each component. Table 2. Composition of BNC-10 Medium

Super-genre

Writing Academic prose (BNC-10W)

Speech (BNC-10S)

TOTAL

Genre Humanities Medicine Natural science Politics, law, education Social & behavioural sciences Technology, computing, engineering

Total Conversations Broadcast TV and radio discussions Live sports commentaries and discussions Total

No. of files No. of words 23 19 29 26 27 23

863,377 863,163 863,071 864,280 863,112 686,004

147 153 43 4 200 347

5,003,007 4,206,058 561,055 33,320 4,800,433 9,803,440

3.1.2.2 The text retrieval software The choice of a text retrieval program is a function of the type of corpus used (e.g. annotated or not) and the sort of search that is envisaged (e.g. lexical or grammatical). Most programs nowadays can handle POS-tagged corpora, but not all of them are adequate to carry out complex syntactic searches. WordSmith Tools,


for instance, is an excellent tool as far as lexical searches are concerned, but offers limited possibilities for the retrieval of complex syntactic structures and discontinuous constructs (see Lee & Rayson 2000).10 By contrast, a program such as XKwic, developed at the University of Stuttgart and part of the IMS Corpus Workbench (see Christ 1994), allows for highly refined and specialised searches, in the shape of algorithms describing a succession of words and/or part-of-speech tags. The BNC World Edition is distributed with its own text retrieval software, SARA (SGML Aware Retrieval Application, see Aston & Burnard 1998), which makes it possible to carry out complex queries as well.11 Because there is no separate index layer for grammatical tags (Lehmann et al. 2000: 264), however, the program cannot extract all the occurrences of a given part of speech (or, for that matter, a sequence of parts of speech) without specifying a lexical item. Thus, while it can look for all the occurrences of cause as a verb, it cannot retrieve sentences where cause is followed by any infinitive, and cannot, therefore, be used for the particular purpose of retrieving periphrastic causative constructions automatically. SARA, however, lies at the heart of another query system, BNCweb Version 2.0 (2002) (see Lehmann et al. 2000).12 This program, developed at the University of Zurich, relies on SARA, but extends its functionality – although, as pointed out by Kreyer & Mukherjee (2002), it cannot overcome its system-internal limitations – and offers a number of additional features for corpus analysis. It is a web-based client program, whose interface is also more user-friendly than SARA. BNCweb makes it possible to query the BNC and retrieve lexical, grammatical and textual data. Both standard and lemma queries are feasible. One can also restrict the query to the written or spoken component of the corpus or to a subset of it (as determined by metatextual criteria such as publication date, domain of the text, age of the author/speaker, social class of the respondent, or a combination of several criteria). Once the search has been completed, the output of which can take the form of KWIC (keyword-in-context) or full-sentence concordances, a number of post-query options are available to the user, e.g. sorting the concordance lines alphabetically on the n-th word to the left or right of the search item 10. Lee & Rayson (2000) refer to the third version of WordSmith Tools (Scott 1999), but the same is still largely true of the latest version of the program (WordSmith Tools Version 5, Scott 2008). 11. The XML Edition of the BNC works with a new version of SARA, called Xaira (XML Aware Indexing and Retrieval Architecture). 12. Note that the current version of BNCweb no longer relies on SARA, but on the Corpus Query Processor (CQP) of the IMS Corpus Workbench (see Christ 1994 on the IMS Corpus Workbench). Some of the limitations of BNCweb Version 2.0 described in this book, therefore, do not apply to the current version. More information on the CQP-version of BNCweb can be found in Hoffmann & Evert (2006) and Hoffmann et al. (2008). See also http://www.bncweb.info.

35

36


(“node”), reducing the number of concordance lines, limiting the query to a previously defined subcorpus, displaying descriptive statistics about the distribution of the results over metatextual categories or carrying out a collocational analysis determining the words most significantly attracted to the node. The most innovative option according to Kreyer & Mukherjee (2002), and the most interesting one for the present purposes, is the tag sequence search option, which makes it possible to do what SARA does not permit, namely the search for a word followed, within a given distance, by a particular part of speech. The “simple mode” allows the user to specify the syntactic structure of the four words preceding and/or following the node. The “advanced mode” may involve up to ten words before and/ or after the node, including discontinuous constructs. It is also possible, in the advanced mode, to specify a particular word in one or some of the positions and decide whether the POS-tag (and word) should be necessary, optional or excluded. In addition to the search for words or patterns, BNCweb allows for a number of additional functions, such as browsing a file, searching the document headers for keyword, title or genre elements, compiling frequency lists of words or lemmas, consulting one’s query history or creating and editing subcorpora. Because of this wide range of features, including the tag sequence search option, and because of its great user-friendliness, BNCweb was chosen as the most appropriate tool for the automatic extraction of periphrastic causative constructions from the BNC.

3.1.2.3 Pilot study In order to define the most efficient query and determine its accuracy and reliability, a pilot study was carried out by means of BNCweb on a 300,000-word subcorpus from the “Arts” domain of the BNC, henceforth SUB_ARTS. This domain was chosen for its particularly long mean sentence-length (average of 22.17 words per sentence), for the longer the sentence is, the longer the distance between the causative verb and the effect (non-finite complement) may be. By allowing the longest span possible between these two elements, the other cases, with a shorter span, will automatically be taken into account. One way to extract periphrastic causative constructions is to look for all the occurrences of the verbs cause, get, have and make, and manually scan the matches in order to discard those sentences that are not causative. This method takes advantage of the tagging of the BNC, in that only verbs are retrieved (and not, say, the occurrences of cause as a noun), as well as of the lemmatisation facility of BNCweb, which brings together all the inflected forms of these verbs (e.g. make, makes, making, made). Because of the manual stage of post-editing, it also gives a guarantee that all causative constructions are retrieved (although mistakes can never be excluded), which results in a (normally) perfect recall rate. The precision rate, on the other hand, is poor, with many examples having to be discarded


manually. Table 3 compares the number of occurrences of cause, get, have and make in SUB_ARTS and, among these, the number of occurrences that are part of a periphrastic causative construction. It appears from this table that causative constructions represent a small proportion of the total uses of the four verbs. The highest proportion is found with cause, which is used causatively with a nonfinite clause in 44.44% of all its occurrences. This proportion falls to 9.71% with make, 3.64% with get and a mere 0.35% with have. In other words, for the verb have, it is necessary to scan as many as 3,437 examples in order to retrieve just 12 relevant instances. In consequence, this type of method, with an automatic search for the lemma and manual weeding out, is untenable on a large scale, especially with have, given the enormous amount of noise (i.e. irrelevant material) that would be included in the matches. The query therefore has to be automated to a greater extent than is the case here, in an attempt to improve the precision rate, while retaining a recall rate as close as possible to the perfect recall rate of the method just described. Table 3. Proportion of periphrastic causative constructions in SUB_ARTS

cause make get have

Verb

Causative verb

Proportion of causative uses

18 628 110 3,437

8 61 4 12

44.44% 9.71% 3.64% 0.35%

As already suggested, an alternative method to extract causative constructions from a tagged corpus is to specify the causative verb (cause, get, have or make) and the type of effect (infinitive, present participle or past participle), and allow for a number of indeterminate words between the two. Such a query further restricts the occurrences of cause, get, have and make that are selected (only those which are followed, within a given span, by an infinitive, present participle or past participle), while offering enough flexibility for the intermediate element, which could be a pronoun, a pre-modified noun, a noun followed by an adverb, etc. In order to determine the most efficient query, several tests were performed on SUB_ARTS, each with a different span. The results were evaluated against the actual frequency of causative constructions, as determined manually (see above and Table 3). In order to limit the number of queries, it was decided that all the constructions associated with a particular verb would be searched together. Thus, the search for [X get Y Vto-inf ], [X get Y Vprp] and [X get Y Vpp] did not give rise to three different queries, but just one query, including the lemma get and either of three non-finite complements, viz. infinitive, present participle or past participle.

37

38


[X make Y Vinf ] and [X be made Vto-inf ], although they present different structures (in the former construction make is followed by the causee, whereas in the latter it is directly followed by the effect), can also be retrieved by means of the same query. In one case the central element takes the form of a noun phrase, in the other it takes the form of the infinitive marker to. A critical step in the selection of the query is to establish the optimal distance between the causative verb and the effect. As a rule, a short span provides a good precision rate (most of the hits are causative constructions), but a poor recall rate (causative constructions with a long central portion are not retrieved). By contrast, a longer span provides a good recall rate (causative constructions are retrieved even if many words separate the causative verb from the effect), but a poor precision rate (the matches include many examples where the non-finite verb is not part of a causative construction). Consequently, a trade-off has to be found between precision and recall rates. One should bear in mind, however, that of these two measures, it is the latter that is more crucial, for while irrelevant hits can easily be discarded, missing material cannot be recovered (cf. Ball’s [1994] “You-Don’t-Know-WhatYou’re-Missing problem”). Priority was therefore given to queries favouring the recall rate, unless they brought about a large drop in precision for a very small gain in recall. With cause, increasing the span up to four words resulted in an improved recall rate (from 50% with one word to 100% with four words) and a constant precision rate of 100%. Since the maximum span of nine words technically possible for this query did not lead to a decrease in the precision rate, however, but could under different circumstances produce a better recall rate, it was established as the optimal distance to retrieve causative constructions with cause. With the other four verbs, the span was set to six words at most. With get and make, this corresponds to the longest object in SUB_ARTS. An example with make is: (16) Indeed, at a certain point the stories come to seem almost like pauses, moments of reflection perhaps, in a wider discourse that envelopes the written word, such as to make the readers of Celati’s narrative acknowledge something in common with Palomar listening to his blackbirds, uncertain which part of their communication is language, which silence.

With have, objects do occur which are longer than six words, but since these objects are also longer than the maximum span allowed by BNCweb (e.g. [17]), they have to be ignored anyway. (17) John Wain, who is said to reread Johnson’s “Rasselas” every year, has the heroine of his first novel, Hurry on Down (1953), call herself Moll Flanders because she has just been reading Defoe’s novel and scents a resemblance to herself; and Iris Murdoch, who seldom reads twentieth-century fiction at all,


is profoundly immersed in the great realistic fiction of earlier ages, whether English, French or Russian.

It should be added that with make, but not the other verbs, the minimum span was set to zero words, so as to retrieve sentences such as (18) and (19) which, though not present in SUB_ARTS, are possible causative constructions.13 (18) Slogans such as “art for the people” mask the cynicism of commercial and political manipulation, which would like to make believe that we all live in a homogeneous society of consumers. (19) First, the buyer has to make known to the seller, expressly or by implication, the purpose for which he intends to use the goods.

Albeit important, an optimal distance between the causative verb and the non-finite complement is not enough to assure the efficiency of a query. In addition, a number of “repair mechanisms” may have to be applied in order to get an “optimised” set of data (Granger 1997). The first repair mechanism applied here has to do with the tagging system, and its relevance is particularly clear in the case of make. Even if one were to take the longest span possible between make and the effect (up to nine words), over 30% of causative constructions with make would still not be retrieved from SUB_ARTS. This cannot be due to the length of the object, for, as we saw, the objects of make in the subcorpus contain no more than six words. This poor recall rate, instead, is the result of mistagged non-finite complements (see Table 4). In nine cases, the infinitive is tagged as a base form (VVB or VDB). Two non-finite complements (an infinitive and a past participle) are tagged as VVD (past tense form of a lexical verb). One infinitive is tagged as NN1 (singular noun). In six cases, it is tagged with a combination of VVB and NN1 (base form of a lexical verb or singular noun). Finally, one past participle has the double tag VVN-AJ0 (past participle form of a lexical verb or unmarked adjective). Because they represent such a large proportion, tagging errors were taken into account to the extent that they did not entail too dramatic a drop in the precision rate. The inclusion of base forms (V*B) and past tense forms (V*D) after make, for instance, does not fundamentally affect the precision rate (from 29.73% to 25.84%). Including the NN1 tag, on the other 13. According to some linguists (e.g. Hantson 1981: 159–161, Andersson 1985: 148–149, Altenberg 2002a), the use of known in [X make Y Vpp], as in (19), is adjectival rather than verbal. This seems to be supported, among other things, by the fact that it takes a to-complement, not a by-complement (make known to the seller). However, the fact that known can be intensified by means of well, like a verb (e.g. a well-known problem), shows that the status of known is at least debatable. Following grammarians like Kirchner (1952: 460–461), Scheurweghs (1959: 169) or Van Roey (1982: 85), it will be considered here as a past participle – while acknowledging its borderline nature.

39


hand, would entail retrieving all the sentences of the type make + NP (make a comment, make efforts, etc) and would therefore result in too poor a precision rate.14 In addition to the expected tags (VBI, VDI, VHI and VVI for infinitives; VBN, VDN, VHN and VVN for past participles; and, when relevant, VBG, VDG, VHG and VVG for –ing forms),15 the following tags were introduced in the query designed to retrieve causative constructions with get, have and make (causative constructions with cause did not present any tagging errors in SUB_ARTS): tags for base forms (VBB, VDB, VHB, VVB), tags for past tense forms (VBD, VDD, VHD, VVD), combination of singular noun tag and tag for base form of a lexical verb (NN1-VVB, VVB-NN1), combination of adjective tag and tag for past participle form of a lexical verb (AJ0-VVN, VVN-AJ0), and combination of tag for past participle form and past tense form of a lexical verb (VVN-VVD, VVD-VVN). These additions yield a very pragmatic method of extraction of the data, resulting in completely ad hoc queries, but queries which make the best of less than perfect material. Table 4. Examples of mistagged non-finite complements with causative make in SUB_ARTS Infinitive tagged as a base form: <w TO0>to <w VVI>make <w DPS>his <w AJ0>small <w NN2>figures <w VVB>appear <w AJ0>remote . Past participle tagged as a past tense form: <w PNP>it <w VM0>can <w VVI>make <w DPS>its <w NN1>presence <w VVD>felt Infinitive tagged as a singular noun: <w TO0>to <w VVI>make <w AV0>even <w NP0>London <w NN1>sound <w NN1>cosmopolitan Infinitive tagged with a combination of base form and singular noun: <w VVG>making <w AT0>the <w AJ0>surrounding <w NN2>objects <w VVB-NN1>fuse <w PRP>with <w AT0>the <w NN1>structure Past participle tagged with a combination of past participle form and unmarked adjective: <w AT0>the <w NN1>book <w CJT>that <w VVD>made <w PNP>him <w VVNAJ0>known , <w VBD>was <w AV0>almost <w DT0>all <w VVN>written

14. With a span of one to two words, for example, the precision rate equals 16.73%, as against 59.21% for a similar query not including the NN1 tag. 15. VB* tags apply to the verb be, VD* tags to do, VH* tags to have and VV* tags to lexical verbs.


Specific repair mechanisms were also used to retrieve causative constructions with get and have, which proved particularly resistant to precise extraction. For one thing, the explicit presence of a noun or pronoun was required somewhere in the central portion.16 For another, some tags were excluded in particular positions: the infinitive marker to (TO0), past participles (V*N) and –ing forms (V*G) immediately after get, in order to discard sentences such as (20) to (23), and the infinitive marker to and past participles immediately after have, in order to discard sentences such as (24) and (25).17 (20) We have got to get this ship under control.

(21) Getting to know the audience was important.

(22) Susan’s chosen metaphor of bringing fabula and sujet together is a characteristically biological one: “how could either come before the other, except as one twin happens to get delivered earlier? (23) We had a bite to eat and got talking. (24) she has to concede that her friend may be right.

(25) Some of these struggles have been inspired by base motives – desires to appropriate art to a national or political cause.

The resulting queries for the different causative verbs, formalised here by means of algorithms for ease of reference,18 are displayed in Table 5, together with their

16. This still allows for a great flexibility in the composition of the central element, since the noun or pronoun may be pre- or post-modified by any means, and items such as adverbs may precede or follow the noun phrase. Note, also, that BNCweb makes it possible to retrieve, say, all types of nouns (singular nouns, plural nouns, nouns neutral for number and proper nouns) in one go thanks to the generic tag “noun”. A similar tag covers all types of pronouns. 17. If, as Kimball (1973) and Martin (1981) claim, have got is a possible alternative to causative have, these cases could be retrieved by the query for get. It should be pointed out, however, that the corpus did not include any such example, and that the sentence Usually she’s got me cut her hair short in the elicitation test (exercise D.21) was rejected by almost 90% of the informants. The use of have got as a causative verb seems to become more acceptable when it is followed by a present participle complement, since He’s got me washing his car every week and pays me £2 for that (exercise D.22 in the test) was accepted by 68% of the informants. 18. The algorithms take the following shape:

lemma [figure-figure] TAG/TAG/TAG

where “lemma” stands for all the forms of the verb and the figures in square brackets indicate the range of positions that can be occupied by one of the following elements (tags). A backslash (\) signals that the following item(s) is (are) excluded in the position specified by the figures in square brackets.

41

42


recall and precision rates in SUB_ARTS. Table 6 compares the recall and precision rates before and after applying the repair mechanisms. It appears that the repair mechanisms make it possible to improve the recall rates by about 25% for each of the three verbs concerned. In terms of precision, they bring about a gain of 1.29% with have and 6.64% with get, but a loss of 3.91% with make. The gains in precision with get and have may seem insignificant when expressed as percentages, but in the case of have, for example, it means some 1,000 hits less to discard (and many more if one works with a larger corpus). As for the loss of precision with make, it is compensated by the gain in recall. Although the queries thus represent a good trade-off between precision and recall, they still involve a great deal of noise (over 98% of the hits with have, and almost 82% with get and 75% with make). The automatic stage with BNCweb therefore has to be followed by a manual stage of post-editing, where all these irrelevant hits are discarded. This “semiautomatic” method, while costly in terms of human work, exploits the strengths of automation and gets round its weaknesses. It also makes it possible to achieve the goal it set itself, namely the extraction from a given corpus of a maximum number of occurrences of causative constructions, since the queries have reasonable recall rates ranging from 83.33% with have to 100% with cause and get. In the next section, we will see how the semi-automatic method is operationalised, using BNCweb and querying the BNC-10 corpus. Table 5. Queries used and their recall/precision rates cause get

have

make

Query

Recall

cause [1- 10] TO0 get [1-6] noun/pronoun [2-7] VBB/VBD/VBG/VBI/VBN/ VDB/VDD/VDG/VDI/VDN/VHB/VHD/VHG/VHI/VHN/ VVB/VVD/VVG/VVI/VVN/NN1-VVB/VVB-NN1/AJ0VVN/VVN-AJ0/VVN-VVD/VVD-VVN \ [1-1] TO0 \ [1-1] VBG/VBN/VDG/VDN/VHG/VHN/VVG/VVN have [1-6] noun/pronoun [2-7] VBB/VBD/VBG/VBI/VBN/ VDB/VDD/VDG/VDI/VDN/VHB/VHD/VHG/VHI/VHN/ VVB/VVD/VVG/VVI/VVN/NN1-VVB/VVB-NN1/AJ0VVN/VVN-AJ0/VVN-VVD/VVD-VVN \ [1-1] TO0 \ [1-1] VBN/VDN/VHN/VVN make [1-7] VBB/VBD/VBI/VBN/VDB/VDD/VDI/VDN/ VHB/VHD/VHI/VHN/VVB/VVD/VVI/VVN/NN1-VVB/ VVB-NN1/AJ0-VVN/VVN-AJ0/VVN-VVD/VVD-VVN

100.00% 100.00% 100.00% 18.18%

Precision

83.33% 1.74%

98.36% 25.42%


Table 6. Comparison of recall and precision rates with and without repair mechanisms

get have make

Without repair mechanisms Recall Precision

With repair mechanisms Recall Precision

75.00% 58.33% 72.13%

100.00% 83.33% 98.36%

11.54% 0.45% 29.33%

18.18% 1.74% 25.42%

3.1.2.4 Automatic stage: Using BNCweb to query BNC-10 Using BNCweb, the method presented above was applied to BNC-10. This section describes the procedure, while also bringing to light a number of limitations of the system and showing how these can be overcome. Prior to the query proper, it is necessary to put together the subcorpus which will be searched, using the “create/edit subcorpora” option of BNCweb. The subcorpus is here defined via genre labels and files are added until they total the required number of words. Once the subcorpus has been created, the search can be run on it, by means of the queries defined in the pilot study (see Table 5). Two features of BNCweb (Version 2.0) have to be borne in mind when carrying out the search. First, the query cannot be performed on the subcorpus directly. It has to be performed on the whole BNC before it is possible to reduce the matches to a subcorpus. Second, the “tag sequence search” option can only be accessed once a search on a particular lexical item has been done. In other words, the program cannot look for all the sentences where, say, get is followed by an infinitive within a span of seven words, but it can first search for all the occurrences of get and then use the criteria defined in the “tag sequence search” option to further restrict the matches. The first step of the search, therefore, consists in retrieving all the instances of one of the causative verbs in the whole BNC. After the concordances have been displayed, the “subcorpus” option of BNCweb becomes available and allows the user to restrict the matches to those occurring in the selected subcorpus. Only then does it become possible to specify the pattern required in the “tag sequence search” option. In the case of get, for example, three restrictions have to be specified in the “tag sequence search” window: the presence of a noun or pronoun within a span of one to six words after get, the presence of one of the non-finite complements or mistagged complements within a span of two to seven words after get19 and the exclusion of all 19. It should be noted that the query does not specify that the (pro)noun should precede the non-finite verb, so that the matches include (unwanted) examples like (i), where a noun occurs within a span of one to six words, and an infinitive within a span of two to seven words, but where the noun actually follows the infinitive. i. My only training and instruction for the job was given by a detective chief inspector, who told us to “get out there and lock up thieves”.

43


the instances where get is immediately followed by the infinitive marker to, an –ing form or a past participle. Once the restrictions have been encoded in the system, the query may be submitted and the results are displayed on the next page.

3.1.2.5 Manual stage: Post-editing the concordances Among the concordance lines retrieved by the automatic search, some do not correspond to a periphrastic construction and hence have to be deleted manually. Consider (26) below. Although it contains a form of the verb get followed by a pronoun (you) within a span of one to six words and by a past participle (been) within a span of two to seven words, it is not a periphrastic causative construction, and should therefore be discarded in the post-editing process. (26) Salim presents himself in a light which requires the reader to be told that, although he himself has been making good, he is grieved, or affects to be, by the discovery that the slavish Metty has been getting on: “You’ve been very much getting on as though you’re your own man.”

Because of the ambiguity of certain structures (see Section 3.1.1.2), it is often necessary to have recourse to a larger context in order to determine whether a particular example is causative or not. This can be done in BNCweb simply by clicking on the causative verb itself. However, the context does not always make it possible to disambiguate the examples and sometimes, several interpretations remain plausible even after the construction has been situated within a wider context. When this was the case and one of the potential readings was the causative one, the sentence was kept. Sentence (27), for example, may be interpreted both as a causative construction and an experiential one (the author may have caused “The Prisoner” to be shut up in “Zenda”, or this may just have happened, simply through the process of writing and without the author really wanting it). Since both interpretations are equally plausible, and none of them is favoured by a larger context, the sentence was not discarded. (27) [Though sometimes interrupted by law work, I sat tight at the story, sometimes writing as much as two chapters a day.] I was only once seriously “stuck up”; I seemed to have got “The Prisoner” so tightly shut up in “Zenda” that it was impossible to get him out of it. [But that difficulty was in the end surmounted and, on the whole, the writing was easy and pleasurable.]  


3.1.2.6 Semi-automatic search of BNC-10 The method of automatic extraction described in Section 3.1.2.4 was performed on BNC-10 by means of BNCweb. It yielded a total of 54,468 matches for the four causative verbs, as detailed in the first part of Table 7. This automatic stage was followed by a stage of manual post-editing, aimed at discarding the irrelevant examples. That this task of post-editing is not an easy one appears from the elicitation test (see Section 3.2), where informants were presented with some ambiguous sentences and asked to choose between two paraphrases, one of them corresponding to a causative interpretation (exercise F). When presented with sentence (28) (item F4 in the questionnaire), for instance, 40% of the informants thought the sentence was causative (a), whereas 52% preferred the second paraphrase. In addition, three informants did not answer the question and one person considered both paraphrases suitable. With most of the other sentences of the exercise, the subjects seemed equally hesitant about which interpretation to choose. (28) Everybody says it does pay to have the house looking fresh. a. … to make the house look fresh. b. … to have a house that looks fresh.

While context may help determine the most likely interpretation, in some cases, such as the sentence above, figuring out the most likely interpretation is not a matter of context, but one of conceptualisation, which does not transpire from the text itself and could only be brought to light by questioning the author of the sentence. During the process of disambiguation and recourse to a wider context, some causative constructions were found which had not been retrieved by the automatic search, either because of a tagging error, e.g. (29) where fire is tagged as a singular noun, a non-canonical order (30) or a particularly long object (31), or because the different elements spanned a turn boundary (32). These constructions were added to the data. (29) It is the class of retinal images which make the perceptron fire (30) Now what shall I have done to my hair?

(31) it gives us chance then to get some more er <pause> er <pause> erm, reproaching done on the equipment (32) – Yeah, that’s it, or sit on there or get – Yeah. – somebody to stand with it.

45


Taking into account the sentences which were clearly causative and those whose interpretation as causative constructions was plausible, as well as the few relevant instances which were added during the disambiguation process, it was possible to collect 3,574 periphrastic causative constructions, as shown in Table 7. The comparison, in the same table, of the number of hits produced by the automatic search and the number of causative constructions obtained after the stage of manual post-editing reveals precision rates ranging from almost 70% with cause to only 2.47% with have (almost 33,000 hits manually checked to retrieve a mere 813 causative constructions). While this means a lot of manual work, it seems to be the price to pay if one wants to retrieve a complex grammatical phenomenon with a good recall rate (see pilot study), as long as large and delicately parsed corpora are not made available with appropriate software to query them. Table 7. Number of hits, number of causative constructions and precision rate cause make get have Total

No. of hits 286 5,640 15,592 32,950 54,468

No. of caus. constr. 200 1,251 1,310 813 3,574

Precision rate 69.93% 22.18% 8.40% 2.47% 6.56%

The causative constructions were then classified according to the pattern they exhibited (see Table 1). It should be pointed out that the data contained some “anomalous” constructions which had to be regularised to fit into the classification. Thus, get occasionally occurs with a bare infinitive (six instances), have with a to-infinitive (six instances), active make with a to-infinitive (four instances) and passive make with a bare infinitive (one instance), as illustrated by the following sentences: (33) And we’ve gotta (…) manipulate people to say well, how do we get them drink more? (34) now this is a very healthy old man with a very good income who could afford to buy any food he wants and the fact of the matter is his that he’s son is probably not, he doesn’t want to upset his old dad, and it’s handy to have him to come on Saturday for lunch and be done with it as it were, but I thought surely Bernard your own sense would tell you that nobody wants their old father and father-in-law every Saturday of their life, for lunch.

(35) What’s made you to think of that?


(36) I would have thought how stupid it was <pause> if I’d been made do it.

Such sentences are included among the “regular” constructions ([X get Y Vinf ] with [X get Y Vto-inf ], [X have Y Vto-inf ] with [X have Y Vinf ], etc) and considered as lapses, dialectal variations or, possibly, emerging features of the language.20 Similarly, constructions with ellipted effects, possible for [X get Y Vto-inf ], [X make Y Vinf ] and [X be made Vto-inf ], e.g. (37) and (38), count as full constructions.21 Table 8 shows the absolute frequency and relative frequency per million words of the ten causative constructions under investigation in BNC-10, from the most frequent one, [X make Y Vinf ], to the least frequent one, [X make Y Vpp]. These data form the basis of the analysis presented in the following chapters. They are, however, supplemented by experimental data, whose collection is reported on in the next section. (37) which I think is right, if you can afford to pay for it [health service] then you should be made, made to. (38) I’m telling you for a fact that Steven won’t touch you. (…) Well Steven won’t but he’ll get everyone else to. 20. Dialectal variation as an explanation for “irregular” complementation patterns is found in Kirchner (1952: 456) and Butters & Stettler (1986: 184). Other explanations are sometimes offered, such as rhythmical or metrical considerations (Poutsma 1926: 430, Visser 1973: 2266, Butters & Stettler 1986: 188n), distance between the causative verb and the effect (Poutsma 1926: 430, Kirchner 1952: 458, Andersson 1985: 149) or presence of a relative pronoun (Poutsma 1926: 430). Often, it is diachronic variation that seems to be responsible for the “irregularity” of the examples quoted. This is the case of the example from Coleridge’s (1798) The Rhyme of the Ancient Mariner quoted in Adamczewski & Delmas (1993: 23), or the examples given in Poutsma (1926: 429), Kruisinga (1931: 385) and Kirchner (1952: 386). The examples from the corpus seem to be mainly due to lapses, since they were usually rejected by the informants to which they were submitted in the elicitation test, with percentages ranging from 83% for They chose themselves to have the ruling family to be the Al Sabar family (exercise D.18) to 53% for What’s made you to think of that? (exercise D.16). One possible exception is [X be made Vinf ], which might perhaps be seen as an emerging feature of language, since I would have thought how stupid it was if I’d been made do it (exercise D.7) was rejected by 39% of the informants but accepted by 51%. 21. See e.g. Karlsen (1959: 197), Roggero (1979: 304), Hantson (1981: 171), Attal (1987: 828) or Huddleston & Pullum (2002: 1527) on elliptical causative constructions. It should be noted that the queries used for the extraction of the data are not designed to retrieve such constructions, since they rely on the presence of an overt non-finite complement. However, some elliptical constructions were retrieved, either accidentally by the query (in [37], due to the repetition, made is followed by a past participle, as required by the query for make) or through perusing the larger context of certain sentences.

47


Table 8. Absolute frequency and relative frequency per million words of periphrastic causative constructions

[X make Y Vinf ] [X get Y Vpp] [X have Y Vpp] [X get Y Vto-inf ] [X cause Y Vto-inf ] [X get Y Vprp] [X be made Vto-inf ] [X have Y Vinf ] [X have Y Vprp] [X make Y Vpp] Total

3.2

Absolute frequency

Relative frequency

1,122 815 671 366 200 129 98 72 70 31 3,574

114.4 83.1 68.4 37.3 20.4 13.2 10.0 7.3 7.1 3.2 364.6

Experimental data: Elicitation test

Corpora give access to naturally-occurring data, that is data produced in natural settings by people whose aim it is to communicate, not to produce a particular linguistic item. Experimental data, by contrast, are not produced in natural settings (typically, experiments take place in a laboratory) and, in the kind of test used here, the elicitation test, people are expected to provide a specific type of answer (the term “elicitation” comes from Latin elicere, i.e. to entice, summon, induce, provoke). The data collection, in other words, is “directed” (Senft 1995: 577). The next two sections describe the design and filling in of the questionnaire used for the elicitation test. Section 3.3 will show the relevance of experimental data and their complementarity with corpus data.

3.2.1 Designing the questionnaire A questionnaire is one of several possible methods to elicit data. The questionnaire used here (see Appendix 1) is explicit, in that the informants know what the object of the study is (the questionnaire starts with a broad description of causative constructions, as well as a couple of examples). As a result, it tends to reflect people’s conscious linguistic judgements (i.e. what they think is right or not), which, as pointed out by Greenbaum & Quirk (1970: 3), does not necessarily correspond to the facts of actual usage. Very often, these judgements are based on people’s explicit knowledge of their language and hence prescriptive


rules.22 The questionnaire, however, is designed in such a way that some questions (fill-in exercises) do not appeal to the right/wrong dichotomy and are therefore not influenced by prescriptive principles. In addition, informants are asked to answer as spontaneously as possible, giving “the first answer that comes to mind”. As advocated by Greenbaum & Quirk (ibid. 13), informants are requested to give information about their regional background, gender and profession. This information will only be used in case the data present very disparate results, which could be explained by one of the informants’ personal history. The questionnaire is made up of six types of exercises. It exists in only one version, since, for reasons that will become clear as we go along, the exercises have to follow a particular order that should not be changed. Exercise A, illustrated by (39), is what Greenbaum & Quirk (ibid. 5) term a “judgement – preference” exercise, where informants are asked to express their preference among a number of alternatives. (39) The policeman

a. b. c. d.

had the young boy apologise. made the young boy apologise. got the young boy to apologise. caused the young boy to apologise.

The context of the sentences was meant to test hypotheses coming from the literature and/or preliminary analyses of corpus data (Gilquin 1999, 2000, 2001), as shown in Table 9. For some sentences, several alternatives may be predicted, either because different authors predict different verbs in a particular environment (cf. Celce-Murcia & Larsen-Freeman 1983 and Goldsmith 1984 on power relations) or because elements in the sentence point to the use of different verbs (e.g. the combination of difficulty and service). The aim of this exercise is to determine which causative verb is preferred in a specific context. Since it contains complete causative constructions, it also makes informants familiar with this type of construction, which will help them for the other exercises. It is therefore necessary to place it at the beginning of the questionnaire. Exercise B is made up of four types of questions (B1, B2, B3 and B4), as exemplified in (40) to (43). In each of them, the informant is asked to complete the causative construction – this is what Greenbaum & Quirk (1970: 5) call a “performance – composition” exercise. In (40) the informant should supply a subject, in (41) an object, in (42) a non-finite clause and in (43) a by-clause. Since 22. This might influence the answers. Thus, one informant admitted that she had avoided the verb get in her answers because she had been taught “never to write the word get”. This, incidentally, reflects a “common prescriptive reaction” against the verb get (Johansson & Oksefjell 1996: 58).

49

50


Table 9. Hypotheses underlying exercise A of the elicitation test Context

Hypothesis

Make implies that the causer has power or authority over the causee (Celce-Murcia & Larsen-Freeman 1983: 481); have is used when one person’s wish is the other’s command (Goldsmith 1984: 119) The child [made/caused/had/got] his Have cannot be used when one person’s wish is mother (to) buy him a new toy. not the other’s command (Goldsmith 1984: 119) It was difficult to find someone to do it, but Get often collocates with words implying I finally [had/caused/got] the chimney difficulty/effort (Gilquin 2000: 37, Gilquin 2001); (to be) swept. have is normally used with reference to the service frame (Stefanowitsch 2001: 131) Get often collocates with words implying difShe is trying to [make/get/cause/have] us (to) be friends with her. ficulty/effort (Gilquin 2000: 37, Gilquin 2001) My sister [got/had] her hair cut for £10. Have is normally used with reference to the service frame (Stefanowitsch 2001: 131) I’m sorry if I [caused/had/made/got] you Make is often followed by a non-volitional (to) worry. effect (Gilquin 1999: 121) It is by torture that the policeman [had/got/ Coercion tends to be expressed by make caused/made] the man (to) speak. (Guierre 1959: 126) Phoebe doesn’t like shopping, so she hardly Have is normally chosen when a routine ever [gets/has] the shopping done. action is referred to (Celce-Murcia & LarsenFreeman 1983: 481); get is more frequent than have when the action is performed by the causer (Martin 1981, quoted by Celce-Murcia & Larsen-Freeman 1983: 482) Destruction of part of the cortex [makes/ Cause is typical of technical/scientific contexts causes/gets/has] cells in a corresponding (Chuquet & Paillard 1989: 170) part of the thalamus (to) die. The student worked late to [have/get] Get is more frequent than have when the action his geography essay finished. is performed by the causer (Martin 1981, quoted by Celce-Murcia & Larsen-Freeman 1983: 482); get suggests trouble or difficulty (Palmer & Blandford 1969: 197) She had her friend [wait/waiting] for her The –ing form expresses a durative aspect for two hours. (Hantson 1996: 26) The policeman [had/made/got/caused] the young boy (to) apologise.


the informants are not all linguists, the linguistic name of the element required is supplemented with a description and some examples. In order not to influence informants towards a particular type of word (e.g. an animate noun or a pronoun), different illustrations are provided.23 (40) ………………………………… got the bomb to explode. (41) She had ………………………………… playing. (42) You got them …………………………………………… . (43) I caused the chair to move by …………………………………………… .

The aim of B1 to B3 is to find out the sort of subject, object or non-finite clause that first comes to people’s minds – animate/inanimate, noun/pronoun, definite/ indefinite, etc for the subject and object; infinitive/past participle/present participle (as applicable), transitive/intransitive, volitional/non-volitional, nature of the object (if any) of the non-finite clause, etc. Each sentence is designed in such a way that any type of answer is possible. (40), for example, could theoretically be filled in with the terrorist (animate) or the jolt (inanimate). Similarly, one could get both a child and a record playing (41). And (42) could be completed with any of the three complements possible with get (infinitive, past participle or present participle), which could be transitive or intransitive, volitional or non-volitional, etc. It will be noticed that sentence 6 of exercise B3, reproduced here as (44), is slightly different from the others. All that the informants have to add in this sentence is a complement to the effect look. Depending on the kind of complement they provide, it will be possible to determine whether they spontaneously consider the verb as volitional or non-volitional in such a sentence. If the complement they choose is a prepositional phrase like through the window, look is volitional (the boy exercises his will to carry out the action of looking). If, on the other hand, they choose a predicative adjective like stupid, then look is used non-volitionally (the action happens independently of the boy’s will). (44) She made the boy look ………………………… .

The aim of B4, illustrated by (43) above, is to find out what distinction(s) people make between the various causative constructions (including the lexical causative construction, cf. move the chair vs. cause the chair to move), by asking them to specify how the causative action was performed (by …). The sentences are kept similar to each other in order to focus the informants’ attention on the 23. Despite this precaution and despite the great variety of stimuli presented in the test, an influence on the answers provided by the informants cannot be ruled out. This problem will be briefly returned to in Section 6.4.

51

52


only variation between the sentences, thereby forcing them to think about the nuances expressed by the different verbs. By requiring that the element following by should be a verb ending in –ing and by giving a couple of examples, it is hoped that answers such as “I made the doctor come by car”, “I had the doctor come by and examine me” or “I got the car repaired by 3 o’clock” will be avoided. Exercise C is another “judgement – preference” exercise (like exercise A). For each sentence, the informants have to decide who performs (performed, will perform) the action, as shown in (45). (45) My father is having all the locks changed. a. My father is changing all the locks. b. Someone other than my father is changing all the locks.

The aim of this exercise is to check whether the causer and the causee are seen as referring to the same entity or not in a number of sentences. Since ambiguity between the two readings solely arises in [X get Y Vpp] and [X have Y Vpp], only constructions of this type are used. As a result, this exercise should not immediately precede a composition exercise, as this might bias the answers towards past participle constructions. It is therefore placed after a composition exercise and before an evaluation exercise. The results of exercise C will make it possible to pinpoint the factors that seem to influence the choice between the two interpretations – e.g. context (you can pay a photographer to take photos, but you normally order napkins yourself), verb (get vs. have) or tense (has vs. is having). In exercise D, the informants are asked to rate the grammaticality of some sentences on a scale from 1 to 5 (“judgement – evaluation” exercise, see Greenbaum & Quirk ibid.), e.g. (46) What’s made you to think of that?

1

2

3

4

5

The sentences involve various problems, as illustrated in Table 10. Thanks to the informants’ ratings, it will be possible to determine to what extent the sentences are judged grammatical, ungrammatical or odd. Like exercise B, exercise E is a “performance – composition” exercise. It is more complex, however, in that informants are not merely required to complete a sentence, but to actually make up a whole causative construction, using one of the four verbs investigated (E.1 to E.4), e.g. (47) cause (or causes, caused, causing): ……………………………………...……

Since most of the preceding exercises present causative constructions, informants should by now have gathered what a causative construction is and should therefore be able to build one themselves. Their answers will show, for each verb, what


Table 10. Types of problems involved in exercise D of the elicitation test Problem

Example

Subclause negation Unusual complementation

Do you know what makes that not happen? We’ve got to manipulate people to get them drink more. Sooner or later we would have been had to do it. Let’s make pretend that we’re married. We’ll have to make us try and make our marriage work by living apart. Usually she’s got me cut her hair short.

Main clause passivisation Unusual lexical combination Use of reflexives Have got (= have) as a causative verb

type of causative construction first comes to their minds. Sentences E.5 to E.8, as illustrated by (48), test to what extent the combination of animate/inanimate causer and causee influences the choice of a particular causative construction. (48) The teacher/the student/leave/the room (+ one of the following verbs: cause/ get/have/make): ...………………………………………………………

Exercise F (“judgement – preference” exercise) comes last, as it contains sentences that have the structure of causative constructions but are not necessarily causative. Placing this exercise earlier in the questionnaire might confuse the informants as to what a causative construction is. The exercise contains questions such as (49). (49) They ought to have escalators going down in the shop. a. There ought to be escalators. b. They ought to build escalators.

The aim of this exercise is to check whether people think the sentences are causative or not, by offering two paraphrases, one of them implying a causative reading. The types of interpretations suggested are shown in Table 11. It will be noted that most sentences of the exercise contain a form of have (got), the most ambiguous verb among the four causatives investigated. Table 11. Types of interpretations possible in exercise F of the elicitation test Interpretations

Example

Causative and existential Causative and permissive Causative and lexical

She’s got friends coming for dinner. I’m not going to have the letters going to Ormskirk. This brief diversion into the cultural history of Germany has been made to emphasise the way a state may determine artistic production.

53

54


Let us note, finally, that most of the (complete) sentences presented in the questionnaire come from the corpus, sometimes in a slightly adapted version, except when the type of construction to be tested was not found in the corpus data and the aim, precisely, was to establish whether it was considered acceptable or not (e.g. passive use of causative have).

3.2.2 Data collection The informants all lived in London but came from various areas all over the United Kingdom, which resulted in a wide dialectal variety in the data. They also represented different social and professional backgrounds and included both linguists and non-linguists of different ages. Thanks to this diversity, the data were not biased towards a particular region, social or professional background. The informants were explicitly asked to answer the questions as spontaneously as possible, without consulting any grammar or dictionary, referring to someone else, or modifying their answers once they had completed the test. They were also informed that the aim of the test was not to judge them or their knowledge of English, but simply to study their intuitions as native speakers of English. In total, 50 questionnaires were completed, 21 by male informants and 29 by female informants. They originally came from such diverse places as London, Warwickshire, Staffordshire, North-West England and Wales, and occupied various jobs, from library assistant and lecturer in linguistics to occupational therapist, electrical contractor and musician. Among the answers, some had to be discarded because they did not correspond to what was expected. Sentences (50) and (51), for instance, are grammatical, but are not periphrastic causative constructions, as requested in exercise E. Similarly, the answers in (52) (exercise B2) and (53) (exercise B3) make sense, but the informant did not respect the instructions of the exercise. (50) I caused an accident by reversing into the road without looking. <EL:E1:003> (51) The dog got hotter and hotter in the car.

<EL:E3:003>

(52) She had never been playing.

<EL:B2.6:006>

(53) I made them bacon and eggs for tea.

<EL:B3.3:006>

Sometimes, no answer was provided or the writing was illegible. These data could not be taken into account either. The fact that a given question was regularly misunderstood or left unanswered, however, was seen as an indication that the construction illustrated was too complex or slightly unusual, a piece of information also relevant for the analysis.

3.3


Combining corpus and experimental data

The usefulness of corpora for the language sciences cannot be denied. Thanks to the advent of corpora, considerable progress has been made in the description of language, and the data they contain have made it possible to discover facts that could not have been discovered otherwise (Fillmore 1992, de Beaugrande 2002). Yet, as Čermák (2002: 270) remarks, “corpora must not be offered as a cure-all”. Despite the great advantages they present, they have their limitations. First, they are not exhaustive (Partington 1998: 146). The absence of a particular word or structure from a corpus, consequently, cannot be taken as evidence for its non-existence, since it could be a case of “accidental gap” – the word or structure does exist in the language, but happens not to be represented in the corpus. Second, corpora do not assess the degree of grammaticality of their data. As Fillmore (1992: 58) observes, they do not have asterisks indicating incorrect uses. The presence of an element in a corpus is therefore no guarantee that it is grammatical, since performance errors are always possible, especially in spoken language. Finally, the results of a corpus analysis very much depend on the quality of the corpus itself (Sinclair 1991: 9, Rundell 1996) – this is the “garbage in, garbage out” principle. If the corpus has not been compiled carefully, there is a risk that the generalisations made on its basis may not be valid. Because of these limitations (and others), several authors advocate the use of various sources of data to study language: corpora, but also introspection, elicitation and other types of experiments, surveys or interviews (Kennedy 1998: 8, Willems 2000: 150–151, Čermák 2002: 270). Johansson (1991: 313), for example, notes that [i]n spite of the great changes in the less than three decades since the first computer corpus, there is one way in which the role of the corpus in linguistic research has not changed. The corpus remains one of the linguist’s tools, to be used together with introspection and elicitation techniques. Wise linguists, like experienced craftsmen, sharpen their tools and recognize their appropriate uses.

It will be noticed that this quotation underlines the value of introspection in linguistic analysis. Introspection has often been undervalued, even denigrated by corpus linguists. Sampson (1980: 152), for instance, points out that “[s]peakers are often straightforwardly, and startlingly, wrong in their sincere convictions about even the most elementary facts of their own languages” (see also Quirk 1974: 167) and adds that linguists’ judgements about language are the most unreliable of all. In particular, introspection has been claimed to be inadequate to determine the frequency of a particular word or use of a word (McEnery & Wilson 2001: 15, de Beaugrande 2002), as well as its collocates, that is the words recurring in its immediate textual environment (Sinclair 1991: 116, Louw 1993: 173,

55

56


Stubbs 1996: 172). This bad reputation, however, is not totally justified. On the one hand, some psycholinguistic studies have demonstrated that people’s estimation of frequency may be reliable (e.g. Hasher & Zacks 1984), as may be their knowledge of collocates (e.g. McGee 2006, 2009). On the other hand, some sort of reflection is indispensable when conducting a corpus analysis, most notably to form hypotheses before the analysis and/or interpret the data in retrospect (Leech 1991: 74, de Beaugrande 2002).24 While many recognise the value of methodological pluralism, practical applications are still few and far between. When it comes to combining corpus and experimental data, it appears that psycholinguists regularly exploit the benefits of such a combination, but that corpus linguists tend to lag behind (Gilquin & Gries 2009). Although some corpus studies exist which use experimental data (e.g. de Mönnink 2000, Hoffmann 2006, Arppe & Järvikivi 2007), the importance and relevance of this combination seems to be largely underestimated among corpus linguists. It might be because, being based on the intuitions, actions, reactions, etc of (often linguistically naive) informants and produced in artificial settings, experimental data are seen as more subjective and less natural than corpus data. Yet, just like corpus data, they represent authentic data produced by real people. In addition, they fulfil functions which cannot be fulfilled by corpora, thus nicely complementing the corpus approach (see Gilquin & Gries 2009). They allow the study of phenomena that are too infrequent to be studied on the basis of corpus data alone (e.g. Meibauer et al. 2004). They make it possible to control for confounding or moderator variables in a systematic way (e.g. Swerts & van Wijk 2005). Using acceptability judgements, one may also make up for some of the weaknesses of corpora outlined above. Thus, against the non-exhaustiveness of corpora, acceptability judgements make it possible to investigate whether a structure not represented in the corpus is impossible or just accidentally missing (e.g. Pander Maat & Sanders 2001). They also give a way of assessing the degree of grammaticality of certain sentences found in the corpus. Unlike corpora, which show what is probable or not (performance), they show what is possible and what is not (competence), e.g. Thráinsson et al. (2007). Finally, experimental data may provide insights into the cognitive processes at work during the production and comprehension of language. This asset is most obvious in experiments studying measurable effects such as reaction times (e.g. Ellis & Simpson-Vlach 2009) or eye movements (e.g. Richardson et al. 2007), but may also be noticed in elicitation tests. 24. Some people call this ability to interact with the data and interpret them “intuition”, which they distinguish from “introspection”, the retrieval of relevant facts from one’s own mental lexicon (see Rundell’s post on the Corpora List in 2001: http://torvald.aksis.uib.no/corpora/20014/0080.html; see also Butler 2004: 150).


Several of these functions will be put to good use through the elicitation test. Exercise D, for example, makes it possible to determine the acceptability of a number of sentences, some of them from the corpus and others absent from the corpus. Exercise C seeks to investigate how the relation between the causer and the causee is conceptualised. In [X get Y Vpp] and [X have Y Vpp], it is not always possible, even with the help of a wider context, to establish whether the causer and causee refer to the same entity or not. While the ideal would be to conduct an interview with the author of the sentence to find out what s/he actually meant, informants’ views may bring to light the most likely interpretation, if not from the point of view of production, at least from that of comprehension. The first part of exercise E may be used to tackle the notion of prototypicality (at least from one angle). Geeraerts (1988a) considers two sorts of data as evidence for prototypicality, namely usage, as revealed in corpora, and introspection, representing language users’ conscious judgements. According to Geeraerts’s study, the insights gained into prototypicality by taking the two sources of evidence into account coincide to a large extent. But his plea for additional comparisons should encourage us to apply this methodology to other types of phenomena. By asking informants to produce the first causative construction they can think of, one may determine the cognitive salience of particular constructions or elements of constructions, i.e. establish which ones are more likely to attract our attention (see Schmid 2007: 120). Comparing these results with the most frequent constructions in the corpus data may then offer new insights into the nature of causative constructions and that of prototypicality, but also into the relation between language and mind, a hot topic among cognitive linguists. Finally, and more generally, the questionnaire will make it possible to validate the corpusbased results against corpus-external results, and hence improve the reliability of the findings thus derived. The combined use of corpus and elicitation data parallels the integration of corpus linguistics and cognitive linguistics described in the preceding chapter. The corpus data provide the basis for descriptive adequacy, while the elicitation data go some way towards giving the model some psychological plausibility (and at least offer additional evidence against which the corpus results may be validated). The resulting model, which will be called the Corpus-Cognition Integrated (CCI) model, may be illustrated by Figure 2. The lower part of the figure shows the types of data exploited: corpus and elicitation data. The corpus data are investigated by means of the tools and techniques of corpus linguistics, which is considered essentially as a methodology. The upper part of the figure represents the theoretical framework chosen for this study, namely cognitive linguistics. The analysis can proceed along two paths, as indicated by the central arrows: from theory to data (corpus-tested or top-down approach) or from data to theory (corpus-driven or

57

 58 


     

 







    

      







 







     





Figure 2. The Corpus-Cognition Integrated model

  



bottom-up approach). Ideally, the model involves constant to-ing and fro-ing between the two levels. This combination of corpus-tested and corpus-driven approaches makes it possible to uncover the syntactic, semantic, lexical and stylistic regularities of a particular linguistic item or construction. The lexical and stylis tic regularities mainly emerge from the corpus-driven approach, thanks to the techniques of corpus linguistics, whereas the syntactic and semantic regularities mainly emerge from the corpus-tested approach and the use of cognitive concepts. This is not a strict rule, as appears from the dotted line between the two ar between the two approaches. Howrows, which allows for some communication ever, it must be recognised that corpus linguistics has a limited access to syntax


and semantics (pattern grammar, as described in Hunston & Francis [2000], is an exception, but it tends to offer a rather fragmented picture of syntax; semantic prosody is another exception, but it represents only one small aspect of meaning). On the other hand, cognitive linguistics is more useful to investigate semantic and syntactic issues (with its notion of imagery, for instance), although phraseological and stylistic aspects are of theoretical relevance to cognitive linguists too.

3.4 Summary This chapter has presented the method of collection of the data that will be analysed in this study of periphrastic causative constructions. Corresponding to the twofold framework described in Chapter 2 and resulting from the integration of corpus linguistics and cognitive linguistics, the data consist in the combination of a subcorpus of the British National Corpus representing naturally-occurring speech and writing, and a questionnaire designed to elicit (parts of) causative constructions or judgements about such constructions. Because the quality of the analysis is very much dependent on the quality of the data, great care was taken when collecting the data. A pilot study was carried out in order to determine the most efficient method of extraction from the corpus, which involved repair mechanisms and manual post-editing in order to improve the recall and precision rates, and the elicitation test was prepared in such a way that each exercise relies on the preceding one but is not (or only minimally) influenced by it. The framework and the data come together in what has been called the Corpus-Cognition Integrated model, whose aim is to describe the syntactic, semantic, lexical and stylistic regularities exhibited by a lexical item (or construction) in authentic language, by considering both textual and cognitive aspects. Now that the background for the study has been introduced, we can turn to the analysis proper, which will be the subject of the next six chapters. In accordance with the Corpus-Cognition Integrated model, three chapters (Chapters 4 to 6) will have a theoretical construct as their starting points, thus adopting a more corpus-tested approach, whereas the other three chapters (Chapters 7 to 9) will begin with the data themselves, in a more corpus-driven perspective. Theory and data, however, will be closely intertwined throughout the six chapters.

59

chapter 4

Causative constructions in action The realisation of the action chain

At the basis of the causative construction lies a concept from cognitive linguistics which makes it possible to understand the structure of the construction: the action chain. The present chapter introduces the concept of action chain and shows how, combined with other cognitive notions, it can explain phenomena such as the identity between the causer and the causee, or the expression or not of the causing event. The linguistic realisation of the action chain is examined in the corpus and elicitation data, from which differences between the constructions start to emerge.

4.1

   

 

The notion of action chain

An action chain reflects a transmission of energy from an entity (human being, animal, concrete object, abstract concept, etc) to one or several other entities, as represented in Figure 3. The “head” of the action chain, which corresponds to the energy source, transmits its energy, through contact, to a second entity. The latter is set in motion and transmits the energy it has received from the head to another entity, which itself comes into contact with yet another entity to which it transmits its energy, and so on until the energy reaches the “tail”, which consumes the energy and does not transmit it further (see Langacker 1991: 283 and Ungerer & Schmid 2006: 178–181). 

 









 Figure 3. Action chain (Langacker 1991: 283) 

 











62




  

                                      





       

























  























 

 



 









  

 





Figure 4. Linguistic realisation of different action chains



 The notion of action chain is important in the theory of cognitive linguistics, as it offers a model to describe clause structure. Consider the diagrams in Figure 4,  adapted from Langacker (1990: 221), and the sentences they stand for. While (a) illustrates the action chain in its simplest form, with no intermediary between the head, the Queen, and the tail, Snow White, the other diagrams represent threeelement action chains, with a transfer of energy from the Queen to Snow White via the apple. The diagrams in Figure 4 also show that different portions of an action chain may be “profiled”, that is foregrounded and given linguistic expression in

Chapter 4. The realisation of the action chain

the sentence.25 Thus, (b), (c) and (d) contain the same elements at the conceptual level. However, only in (b) are all the elements profiled and, accordingly, realised at the linguistic level. In (c), the apple, Snow White and the process of killing are profiled (as indicated by the bold lines), but the Queen and the process of using are not – they are “gapped”, and hence remain unexpressed. And in (d), only the tail, Snow White, and its change of state (i.e. its consumption of energy through dying) are profiled, yielding the sentence Snow White died. Interestingly, the action chain and the portions of it that are profiled may help predict the assignment of grammatical functions in a sentence (see Langacker 1990). In a two-element action chain like Figure (4a), the head, which corresponds to the role of agent, typically takes on the function of subject, since it is the initiator of the energy transfer and, consequently, the most prominent element. The tail, on the other hand, which corresponds to the patient, is less prominent and is therefore given the status of object. Naturally, it is possible to depart from this canonical word order, for example through passivisation (Snow White was killed by the Queen). In a three-element action chain like Figure (4b), the agent (the Queen) typically becomes the subject, while the patient (Snow White) becomes the object. The central element, the apple, has the role of an instrument, being the intermediate stage in the energy flow, and is normally expressed as an oblique complement. In case only part of the action chain is profiled, it is the head of the profiled portion (i.e. the linguistically expressed element that lies furthest “upstream” in the energy flow) that becomes the subject, whereas its tail (the element furthest “downstream”) becomes the object – although, here again, the elements of the action chain may be reordered through processes such as passivisation. In Figure (4c), where only the interaction between the instrument and the patient is profiled, the instrument (the apple) becomes the subject, and the patient (Snow White) becomes the object. When there is only one participant profiled, as is the case with Snow White in Figure (4d), it becomes the subject of the sentence.

4.2 The action chain of periphrastic causative constructions Using a periphrastic causative construction has the effect of “adding a link at the beginning of an action chain, thereby extending the scope of predication to include the original energy source” (Langacker 1991: 408). This is illustrated by Figure 5. 25. Talmy (1996) refers to this phenomenon as “windowing of attention”, which he opposes to “gapping”, that is the backgrounding of a portion by exclusion.

63


 



                                     



Figure 5. Effect of causativisation on the action chain

 While in (a) the Queen is presented as the head of the action chain, in (b) the  original source of energy, jealousy, is included in the action chain and becomes the head, whereas the Queen, as a result, moves further down the action chain.26  Another difference between (a) and (b) is that the action chain in (a) contains a single event (“killing”), whereas in (b), the periphrastic causative con struction, two (“causing” and “killing”). Sentence (b) may be further  it contains   illustrated by Figure 6, adapted from Langacker (2002: 255) (see also Beck    1996: 113). The action chain is construed as a sequence of separate, but causally related events, as indicated by the imbricate boxes. In the first event, the head  of the action chain, jealousy, uses its energy to initiate a process which affects  the Queen, and in the second event, resulting from the first one, the Queen kills    Snow White (the jagged arrow indicates the absorption of energy by the tail of  the action  chain).  26.  This process is often described in the literature in terms of valency extension (see e.g. Siewierska 1991: 28–29 or Dik 1997: 9): a predicate with x arguments may, through causativisation, include x+1 arguments, cf. The baby laughed vs. He made the baby laugh (1>2) or She changed  her mind vs. They made her change her mind (2>3). Note that an explanation in syntactic terms like Tallerman’s (1998), which refers to the demotion of the original subject and the introduction of a new subject, may be problematic. A causative construction such as We were made to leave the room, for example, has the same subject as its non-causative equivalent, We left the room. Within the framework of cognitive linguistics, on the other hand, this example may be explained by the fact that a link has been added at the beginning of the action chain (cf. Someone made us leave the room), but is not profiled and therefore remains unexpressed.


Figure 6. Action chain of Jealousy caused the Queen to kill Snow White

The action chain construed by a periphrastic causative construction will be schematically represented as in Figure 7, which also includes the terms that were already introduced in Chapter 2. The causer is the head, that is the initiator of the action chain, the source of energy that transmits its energy further. This transfer of energy is signalled by the causative verb and, together with the causer, it constitutes the causing event. By transmitting its energy to a causee, the causer instigates another event, the caused event, in which the causer is not directly involved. Thanks to the energy of the causer, the causee is set in motion and is thus able to transmit the energy it has received from the causer to another participant, the patient.27 This second transfer of energy is expressed by the effect. The patient represents the tail of the action chain, the place where the energy is absorbed, as 27. This description brings to light the dual nature of the causee in a periphrastic causative construction. Not only is it affected by the causer, but it is also the initiator of the effect. In other words, it is both the recipient of the energy flow (from the causer) and the transmitter of it (to the patient). The issue has often been discussed in the literature, especially by syntacticians, who note that this element has the function both of a subject and an object (e.g. Visser 1973: 2376, Van Ek & Robat 1984: 56). Among generativists, this dual status is typically explained by means of the distinction between deep and surface structures. While at the deep structure level a periphrastic causative construction is claimed to be biclausal and have two subjects (one for each clause), a transformation applies, the so-called “subject-to-object raising”, which changes the subject of the second clause into the object of the first clause, thus turning the biclausal construction into an essentially monoclausal one (Hendrick 1995: 334). This transformation arguably explains why both main clause and subclause passivisation exist in periphrastic causative constructions (although not all constructions accept both, and passivisation may result in a different meaning; see Huddleston & Pullum 2002: 1235ff.). The main clause passivisation of John made Helen laugh into Helen was made to laugh by John (Kastovsky 1973: 282) implies that Helen is the object of make and the structure is monoclausal, but the subclause passivisation of He had a specialist examine his son into He had his son examined by a specialist (Huddleston & Pullum 2002: 1236) suggests that a specialist functions as the subject of examine and the structure is biclausal.

65


indicated by the jagged arrow. With the exception of the patient (see below), all the elements are necessarily present in the action chain at the conceptual level –  although they may not all be realised at the linguistic level.

 









           





    construction  Figure 7. Schematic  action chain of a periphrastic causative   with a patient 

   The presence of a patient is not obligatory for a periphrastic causative construc tion to be well-formed. The sentence The Queen caused Snow White to die is a   periphrastic causative construction, although it involves only two participants,  the causer and the causee. The action chain of such constructions is slightly dif       may  ferent from a three-participant action  chain and be schematised as in Figure 8. The causer, thanks to the energy it releases, initiates the caused event, in which  the energy is consumed by the causee, in this case the tail of the action chain. The  jagged arrow represents the causee’s consumption of energy through the effect.           





     

  

 







 Schematic action chain of a periphrastic causative construction Figure 8. with no patient



It should be borne in mind, however, that even when it involves only two participants, a periphrastic causative construction still encodes an action chain which is


67

different from the action chain underlying a simple two-element action chain. This issue, in fact, brings us directly into the controversial question of the relation between periphrastic causatives and lexical causatives, as most notably illustrated by the heated debate over the equivalence (or lack thereof) between kill and cause to die.28 As appears from the action chains illustrated in Figure 9 and representing the  sentences The Queen killed Snow White and The Queen caused Snow White to die,  respectively, the major difference has to do with the conceptualisation of causation.

 

 

 

                                                                                                                

kill and cause to die Figure 9. Action chains with 

    While the first sentence construes causation as a single (complex) event, the   second sentence construes it as a sequence of separate (but causally related)   29 The diagrams in Figure 9 reveal another events (see Langacker 2002: 254).   difference, related to the  preceding one, namely the relation between the agent

    28. See e.g. McCawley (1968) for  an analysis in favour of such an equivalence and Fodor (1970)      or Shibatani (1972) for an analysis  against.   29. The existence of two events, incidentally, explains why temporal separation is possible in peri

phrastic causative constructions, but in principle excluded with lexical causative verbs. Compare John caused Bill to die on Sunday by stabbing him on Saturday with *John killed Bill on Sunday by stabbing him on Saturday (Fodor 1970: 433). See, however, Lemmens (1998: 23–24) for an authentic example where the lexical causative verb kill involves temporal separation between the kill-component and the die-component, and for an explanation of this apparent counter-example.

                   


and the patient. With the lexical causative, the agent (the Queen) directly effects some change of state (indicated by the jagged arrow) in the patient (Snow White). With the periphrastic causative, on the other hand, the role of the agent is to instigate a particular process (what we have called the caused event). In other words, the head of the action chain uses its energy, not to affect the tail directly, but to initiate a process that will affect the tail and in which it is not directly involved. Contact between the agent and the patient is, consequently, optional in the case of periphrastic causative constructions, whereas lexical causative verbs normally imply some direct contact between the two participants (see Radden 1992: 516). One cannot, however, exclude the possibility that the lexical causative verb and the periphrastic causative construction can be used to describe the same objective reality. But even then, the difference in the action chain would result in a difference in conceptualisation. A sentence such as The Queen caused Snow White to die gives more prominence to the “dying” element than The Queen killed Snow White. While the latter is primarily interested in the action performed by the Queen (she killed Snow White), the former also draws attention to what happened to Snow White (she died). Of course, “dying” inherently belongs to the meaning of “kill”, as appears from the ungrammaticality of *He killed her but she did not die, but by explicitly mentioning the death component, the speaker/writer chooses to construe it as more salient than it would otherwise be.30

4.3 Action chains in the corpus data We saw in the preceding section that the action chain underlying a causative construction may take one of two forms, depending on whether a patient is present or not. The linguistic realisation of the action chain, however, offers many more possibilities: the action chain may represent a literal or symbolic transfer of energy, some elements in the construction may be gapped and therefore remain unexpressed, two participants in the action chain may actually refer to the same entity, etc. The corpus data make it possible to examine the variety of linguistic realisations found in language. This is what this section sets out to show.

30. This may be linked to the phenomenon of iconicity, and more precisely the quantity principle, according to which more important (or semantically larger, or less predictable) information is given more coding material (see Givón 1994: 49–51). By being coded at the linguistic level, the death component is given more prominence than it would otherwise receive.


4.3.1 Different types of action chains 4.3.1.1 Literal vs. symbolic energy flow Although the notion of action chain is usually described in the cognitive literature in terms of transmission of energy through contact, the corpus data show that contact may be metaphorical and the energy flow, consequently, of a more abstract nature. Compare (54) and (55). In the first sentence, the head, the dehydrated plasma membranes, may be said to transmit energy, quite literally, to the cell walls, the tail of the action chain, which absorb this energy through the process of collapsing. In the second sentence, on the other hand, the energy flow is more symbolic, and the collapsing merely abstract. (54) The dehydrated plasma membranes adhere to the cell walls, causing them to collapse. (55) It has become common form to invoke the magic names of the French theorists, as if the names alone would cause a torpid academic establishment to collapse.

The proportion of literal and symbolic action chains in the corpus data is not directly quantifiable, as these are not clear-cut categories, but rather form a continuum from more literal to more symbolic energy flow. Thus, although (55) is more symbolic than (54), it is less so than (56), where the transmission of energy is hardly visible. Yet, the corpus data suggest that purely physical action chains are relatively rare in language, or at least, rarely expressed by means of a periphrastic causative construction. This will be confirmed in Section 6.3, where the billiardball model, a model based on a physical action chain, will be shown to account for an extremely small proportion of the data. (56) One of the paradoxical achievements of the computational approach is that it makes even very elementary processes seem complex while, at the same time, actually simplifying problems that were once thought complex.

4.3.1.2 Caused motion vs. caused rest Talmy’s (1986, 2000a) concept of “force dynamics”, which refers to “how entities interact with respect to force” (Talmy 2000a: 409), covers several related fields, among which “causing”, “letting”, “hindering” and “helping”. What is common to verbs of causing within this model is that the resultant state is the opposite of the intrinsic tendency expressed by what we have called the causee (and Talmy refers to as the Agonist). If the causee has an intrinsic tendency towards rest (or, more generally, inaction), the resultant state will be one of motion (or, more generally, action). If,

69


on the other hand, the causee has an intrinsic tendency towards motion, the resultant state will be one of rest. These two situations are illustrated respectively by (57) and (58), taken from Talmy (2000a: 418). In (57) the lamp’s intrinsic tendency towards rest is overcome by the ball, whereas in (58) it is the fire’s intrinsic tendency towards motion which is overcome and becomes a state of rest. (57) The ball’s hitting it made the lamp topple from the table. (58) The water’s dripping on it made the fire die down.

Although here again quantification is difficult, as some constructions do not really express any sort of action or inaction and the causee’s intrinsic tendency is sometimes impossible to determine, it may be said that, generally, it is a state of motion (or action) that results from the causing event, as in (59). Cases where the resultant state is one of rest, however, are possible, as illustrated by (60), where the caused event is the stalling of the open-loop system. (59) I’ve had him actually writing his words, doing letter formation. (60) This difference is important when the motor is producing maximum (pullout) torque, because any small increase in load causes the open-loop system to stall.

Interestingly, the examination of the data also reveals the existence of a number of causative constructions where the causee’s intrinsic tendency and the resultant state of the force interaction are the same. In (61) both the before and after states are actions and in (62) neither of them is. There is in both cases a change – from one course to another in (61) and from less young to younger in (62) – but the resultant state is not the opposite of the intrinsic tendency expressed by the causee (and so, strictly speaking, these changes would not qualify as causation in Talmy’s system). (61) As a consequence British Prime Ministers may have to work extremely hard to get them [Whitehall departments] to change course. (62) Right, okay so its [make-up] to make you look younger.

4.3.2 Length of the action chain As we saw above, the underlying action chain of a periphrastic causative construction may contain two or three participants. It contains two participants when the interaction is limited to a transmission of energy from causer to causee, as illustrated by (63), and it contains three participants when the exchange of energy is further extended to a patient, as in (64).


(63) It’s a wet month. (…) That <pause> that causes <pause> everything to grow! (64) And I had Mike ironing them [curtains] while I was putting ’em up, I said right stand here and watch me hang this curtain, so I’m showing him how to iron it, so while I’m upstairs putting ’em up he’s ironing.

The analysis of the corpus data reveals that 68% of the causative constructions in BNC-10 correspond to a three-participant action chain, as against 32% for the two-participant action chain. This preference for three-participant action chains may be explained by the fact that two-participant action chains can also be expressed by means of a simple transitive verb. Thus, instead of make think in (65), one could use the verb remind (yours reminded me of a story that I was told many years ago). Equivalents are not impossible for three-participant action chains, e.g. show them the colour instead of make them see the colour in (66), but these are less common than with two-participant action chains.31 (65) Actually yours made me think of a story that I was told many years ago on a coach trip over Dartmoor. (66) I just make them see the colour going up the stem of a plant.

This general trend, however, conceals the variation that exists between the different constructions (see Table 12). As expected, all past participle constructions correspond to a three-participant action chain, since subclause passivisation implies the presence of a patient. For some constructions, the two types of action chains are equally probable, though with a small preference for the three-participant action chain ([X cause Y Vto-inf ], [X have Y Vinf ] and [X be made Vto-inf ]). The remaining constructions show a marked preference for one type of action chain. The two present participle constructions, [X get Y Vprp] and [X have Y Vprp], are usually limited to two participants, as is [X make Y Vinf ]. [X get Y Vto-inf ], on the other hand, much more often involves three participants. These discrepancies are a first indication that the various causative constructions are used in different contexts (compare, for example, [X get Y Vto-inf ] and [X get Y Vprp], which are 31. This is not to say that equivalents are always possible for two-participant causative constructions. In fact, most of the instances in the corpus data do not seem to have any (obvious) corresponding transitive verb (e.g. make sb smile, have sb sit down, cause sb to applaud), which suggests that transitive verbs might be preferred when available and causative constructions might be reserved for cases where no transitive verb exists. Having less “competition” (i.e. fewer potential equivalents), three-participant action chains would have to be expressed by means of a periphrastic causative construction more often, hence the higher frequency of such constructions.

71


YV pp ] MA KE YV inf ] [X BE ma de V to [X inf ] MA KE YV pp ]

prp ]

YV

HA VE

inf ]

YV

27.8

85.1

0.0

40.3

64.4

0.0

66.0

40.4

56.3

72.2

14.9 100.0

59.7

35.6 100.0

34.0

59.6 100.0

[X

43.7

Three participants

[X

[X

HA VE

pp ]

Two participants

[X

[X

HA VE

YV

prp ]

GE T

YV

to inf ]

[X

GE T

YV

GE T

[X

CA USE

YV

to inf ]

Table 12. Length of the action chain (percentages)

[X

72

0.0

similar in terms of sentence organisation, with a causer followed by a causee, but display different preferences in terms of length of the action chain). This issue will be taken further in the next chapters.

4.3.3 Linguistic realisation of the action chain It was shown in Section 4.1 that the whole action chain need not be given linguistic expression, since the speaker/writer may decide to gap certain portions. For this reason, the length of the action chain as presented in the preceding section does not necessarily correspond to the action chain as it is expressed in the sentence. In (67), the construction conceptually involves three participants (you, the hairdresser and your hair), but only two of them are profiled (you and your hair). Similarly, (68) could accommodate two participants ([X make X-cells behave]) but only one (X-cells) is included at the linguistic level. (67) I just told them you’d had your hair cut really short.

(68) It is possible to make X-cells behave transiently (…) if the stimulus intensity is altered.

The data from BNC-10 show that, while a majority of the action chains construed by causative constructions involve three participants (see above), most of the time only two participants are realised linguistically (73% of the cases, as against 14% for three participants and 13% for one participant). As appears from Table 13, this is true of all the constructions except [X be made Vto-inf ], which prefers to profile only one participant. The past participle constructions, [X get Y Vpp], [X have Y Vpp] and [X make Y Vpp], follow the same tendency. Although they have the potential to express three participants, they usually do not take advantage of this feature and limit themselves to the expression of two participants, the


pp ] MA KE YV inf ] [X BE ma de V to [X inf ] MA KE YV pp ]

0.0 2.0 75.5 22.5

0.0 8.5 47.8 43.7

0.0 11.6 77.5 10.9

0.0 16.7 82.9 0.4

0.0 0.0 55.6 44.4

0.0 4.3 71.4 24.3

0.0 9.2 89.5 1.3

[X

prp ]

YV

YV

[X

HA VE

inf ]

YV

[X

HA VE

pp ]

[X

HA VE

prp ]

[X

GE

TY V

to inf ]

TY V

[X

GE

TY V

GE

[X

Zero participants One participant Two participants Three participants

[X

CA USE

YV

to inf ]

Table 13. Number of participants profiled (percentages)

0.1 12.5 67.2 20.2

0.0 60.2 39.8 0.0

0.0 19.3 71.0 9.7

causer and the patient. In other words, a sentence like (67) above is more likely to occur than one like (69), where the three participants (Frederick Barbarossa, Charlemagne and his anti-pope Paschal III) are profiled. (69) In 1165, Frederick Barbarossa had Charlemagne canonized by his anti-pope Paschal III.

Gapping is made possible by the structure of the sentence, e.g. the use of an impersonal structure such as it is possible to in (68) or, most often, the passivisation of the causative verb or the effect, as in (67). The passivisation of the causative verb may result in the gapping of the causer, e.g. (70). In [X be made Vto-inf ], this is the case in over 96% of the cases. This result confirms the finding made by several linguists (e.g. Svartvik 1966 for written English or Granger 1983 for spoken English) that the by-agent is usually deleted in passive sentences (80% in Svartvik’s data and 82% in Granger’s). This deletion can occur in cases where the agent is “recoverable from the context” (Granger 1983: 309), which may be the linguistic context, e.g. the suitors in (70), or the situational context, e.g. the government in (71). (70) In the old popular court, the suitors declared the law and gave judgement, or else passed judgement over to God by making one of the parties submit to an ordeal: to being thrown into a pool, for instance, or made to grasp a red-hot iron; if he sank, or was soon healed, he was innocent; otherwise, he was guilty. (71) Even the famine area was made to pay one-half of the supplemental tax levied for famine relief.

73

74


Poutsma (1926: 101) explains the use of agentless passives by “the desire of the speaker to avoid mentioning the primary participant in the action, because not clearly known or thought of no importance, or because involving the possibility of compromising him”. In (72) the writer does not reveal the identity of the causer, presumably because s/he does not know precisely who the causer is (we may only assume that it represents people who are in favour of the regime) or because s/he does not want to compromise this manipulator. In (73) it is irrelevant to the argument who could make the difference statistically significant. What matters is that it is possible to make it significant by taking a sufficiently large sample. (72) This “inverted causation” as it has been called, which is a major element of Marxist theory, is to be found in the theory of Asiatic production, in that the subjects of the Asiatic despot are made to believe that they can live because of the blessing of the god-king, the true guardian and shepherd of the community, while really it is he who is living off them. (73) Any observed difference between two groups, no matter how small, can be made to be “statistically significant” -- at any level of significance -- by taking a sufficiently large sample.

By contrast, in (74) it is important that the reader should know that it is a probable anti-Semite who made the causee seem anti-Semitic, as this suggests that the causer probably presented information about the causee in such a way as to reflect his/her own convictions (which were not necessarily the causee’s convictions). Hence the explicit mention of the causer as a by-complement – and the emphasis it thus receives in the sentence, since the agentful passive gives “prominence to the primary participant in the action, by mentioning it expressly at the end of the sentence” (Poutsma 1926: 101). (74) And now he has been put down, made to seem anti-Semitic, by a probable anti-Semite.

As already suggested, however, such sentences are extremely rare in the corpus data. It is also worth adding that, interestingly, all of them are two-element action chains, including a causer and a causee, but no patient. Thus, a sentence like He was made to read a book by his mother never occurs in the corpus.32 In the same way as the causative verb can be passivised, resulting in the possible gapping of the causer, the effect can be used in the passive voice, thus making the causee optional. This accounts for the high proportion of past participle

32. Such sentences, however, are not uncommon in grammars, e.g. The civilians were made to leave their homes by the soldiers and head for the border (Gusdorf & Lewis 2002: 531).


c onstructions profiling only two participants. Here again, various reasons can motivate the speaker’s or writer’s decision not to mention the causee. With [X have Y Vpp] and [X get Y Vpp], it is usually a matter of predictability. Because [X have Y Vpp] is predominantly used to refer to one person commissioning a professional to do something (see Chapters 7 and 8), the identity of the causee can usually be inferred thanks to one’s knowledge of the world. We know that it is the job of a manicurist to do somebody’s nails (75), the job of a hairdresser to cut hair, the job of a painter to paint walls, the job of a watchmaker to repair watches, etc. With [X get Y Vpp], the predictability mainly comes from the fact that the causer and the (implied) causee often refer to the same entity (see Section 4.3.5). Consider (76), where Jacqueline is an eight-year-old girl. It is quite obvious from the context that Jacqueline is both the causer and the causee, and a profiled causee (you get your dinner eaten by yourself), therefore, would be redundant. (75) [she took her (…)] to have her nails done. (76) Jacqueline you <pause> get your dinner eaten.

The causee may also be gapped because it is considered irrelevant. In (77), what is important is that the advantages and disadvantages of the strategy were made known by the United States. To whom they were made known is unimportant – and probably impossible to specify anyway. Likewise, the exact identity of the causee in (78) is not relevant to the situation described. All we need to be aware of is that there is some entity that may perform the effect, a piece of information that is provided, at the conceptual level, by the schema of the action chain. (77) Nevertheless, it is an approach employed by many countries: it was first used by the Soviet Union in 1951, but it is the United States with which this strategy is most associated and which has made known its advantages and disadvantages. (78) Such catalogues include accounts of relevant historical research, and may cause new original historical work to be done.

Although the causee in such constructions may be expressed by means of a bycomplement (or, sometimes, a to-complement), this is not often so, with only three cases for [X get Y Vpp] (0.4%), ten cases for [X have Y Vpp] (1.5%) and four cases for [X make Y Vpp] (12.9%). Here are a few illustrations: (79) Marriages of persons over that age, but under 18, are completely valid; and the only check on such marriages without the consent of parents or guardians is the difficulty of getting them celebrated by the clergyman or proper officer without making a false declaration, which involves penal consequences.

75

76


(80) This stands for Cooperative Awards in Science and Engineering and erm under this scheme, a company erm can have a problem tackled by a research student working in a university and erm a supervisor, and indeed in this case, the input, the financial input, by the company may be quite small, may only amount to a few hundred pounds. (81) As a result the function of second-level policy-making institutions such as Public Local Inquiries is seen in terms of merely affording an opportunity for those whose private individual rights and interests may be affected by a policy proposal, to make this known to those responsible for making the decision.

When no passivisation is involved, the causee is normally obligatory. Whereas in some languages a sentence like “He made notice that…” is perfectly acceptable (e.g. French Il a fait remarquer que…), most notably when the causee has indefinite personal reference (see Cannings & Moody [1978: 11] for French), it is ruled out in English. Yet, one such sentence was found in the data, namely (82), where the causee is left unexpressed. (82) I hate people who park on the ramp and causes to put yourself to a hill start.

Although the sentence is undoubtedly clumsy, as also appears from the incorrect verb agreement (causes instead of cause), the causee is grammatically predictable because of the use of a reflexive patient (yourself). It should also be noted that the causee is generic (you), which may remind one of some of the definitions found in dictionaries, where the implicit causee is supposed to represent anybody/anything, e.g. break: (7g) to cause to discontinue indulgence in a habit – used with of move: (2a) to cause to go or cause to keep on going continuously from one point or place to another <moved the flag slowly up and down as a signal> show: (1) to cause or permit to be seen (Webster’s Third New International Dictionary

of the English Language, Gove 1993)

In addition, the absence of the causee in an active construction may be due to fossilisation, as is the case with the constructions [X make do] and [X make believe], which can be said to have the status of idioms.33 The corpus contains 33. A few occurrences of the construction [X make pretend] were also found on the Internet, but the sentence Let’s make pretend that we’re married in the elicitation test (exercise D.19) was accepted by only 12% of the respondents.


t hirteen occurrences of make do and one occurrence of make believe, as illustrated by (83) and (84). This idiomatic use, incidentally, accounts for the one and only construction in the corpus where no participant at all is profiled (84). (83) Well I have in the past given them fifty each for Christmas but this year they’re going to have to make do with twenty. (84) This is direct experience, but it is not drama -- not until there is some pretence involved, some symbolic representation, some intention to make believe.

It should be clear from what precedes that different constructions allow for different linguistic realisations of the action chain. While in [X be made Vto-inf ] the causer can easily be gapped, in [X get Y Vpp] it is the causee that is optional. In [X make Y Vinf ], on the other hand, the causee is normally obligatory, expect in the idioms make do and make believe. The speaker’s or writer’s desire to profile or gap a particular participant, therefore, may explain the choice of a given construction – although we will see in the following chapters that there are many more parameters that guide this choice.

4.3.4 The causing event We have just seen that a participant may be profiled or gapped in the linguistic realisation of the action chain. Another element of the action chain that is optional at the linguistic level is the causing event. The existence of a causing event in all causative processes has been acknowledged by several philosophers (e.g. Vendler 1967, Lewis 1973) and linguists (e.g. Dowty 1972, Baron 1974, Shibatani 1976a, Talmy 1976, Wojcik 1976, Goddard 1998). As demonstrated by Vendler (1967), even sentences which seem to attribute the caused event (or, more generally, the effect – in the non-technical sense) to objects or people, e.g. The moon has an effect on the surface of the oceans, may be expanded so as to make the causing event apparent, cf. The moon’s attraction has an effect on the surface of the oceans. This may be illustrated by sentences (85) and (86). In (85) only the agent of the causing event, he, is profiled and nothing is said about what he does to make the causee laugh. In (86), by contrast, we know how the speaker sets about making the causee laugh, namely by telling jokes. In the former case the causing event is gapped, whereas it is profiled in the latter case. (85) He makes me laugh. (86) I always tell jokes that make them laugh.

77

78


The profiling of the causing event, however, appears to be the exception, rather than the rule, with only a quarter of profiled causing events in the corpus data. In its most explicit form, the causing event is expressed as a gerund and corresponds to the causer of the construction (87). Alternatively, it may take the form of a noun derived from a verb (88), a noun denoting an action or event (89), or a pronoun referring to any one of these, as in (90) to (92), where the referent of the pronoun has been italicised. (87) But there is, perhaps, one crucial objection to the constructivist’s claim that moving the focus away from mental representations and towards mental actions will make the mind-body problem appear to us as less of a problem. (88) A reduction of ozone would also cause more ultraviolet and visible radiation to reach the ground, so leading to a warming of the lower atmosphere and the earth’s surface. (89) So, although the events leading up to the Civil War made it seem desirable to Hobbes to write “De Cive” before the completion of “De Homine”, his finding this possible was not inconsistent with his conception of a three-part ordered system of “Elements of Philosophy”. (90) Vain Chantecler refuses to recognize that it is not his own crying of “Cocorico” which causes the sun to rise. (91) But it [this single observation] is certainly enough to make us want to persist with our exploration of this theory and to examine in some detail (in Chapter 7) the various possible explanations for the effect. (92) Those closest to him who should have been too frightened to behave as they did include his father, but they also include his sisters, who struck out for them selves in a fashion which has him siding with his father.

A pronoun may also stand for a complete clause or sentence, thus referring to a state or event, as is the case in the following examples: (93) But er <pause> i-- it was hot in there which made me feel sicky, but I enjoyed it Marg! (94) What made me laugh was this police woman saying, it annoys me, if you have a drink you shouldn’t drive (95) I prefer Beethoven, actually I’m not sure I do prefer Beethoven, but had you said Bach I would have preferred that, but I don’t see that that <pause> I mean you’re implying that that makes me better, or at least makes me think I’m better than somebody who likes Madonna, and that I don’t agree with.


(96) To provide the links, an extra field has to be added to the home bucket or home record, and an extra field to each synonym. This effectively reduces the file packing, and may also cause fixed lengths to become variable in length.

The causing event does not necessarily correspond to the causer of the construction. In (97), for example, it takes the form of a gerund clause (in suggesting the rule is an inalienable part of the language). Another, particularly interesting case, relatively common with [X cause Y Vto-inf ], is illustrated by (98), a sentence already quoted earlier. The grammatical subject – and causer – of the sentence is the dehydrated plasma membranes. However, since the use of an adverbial participle clause (supplementive clause) makes it possible to “give more specific details or additional information about the action in the main clause” (Biber et al. 1999: 832), it can be argued that the semantic subject of causing is the whole main clause, i.e. the dehydrated plasma membranes adhere to the cell walls. It is not the dehydrated plasma membranes that cause the cell walls to collapse, but the fact that these membranes adhere to the cell walls. (97) In suggesting the rule is an inalienable part of the language, Dale Spender ironically assists those she criticises in making their sexism look natural, when she ought to be exposing it as a cultural construct. (98) The dehydrated plasma membranes adhere to the cell walls, causing them to collapse; growing ice crystals rupture the cell walls and membranes and destroy cytoplasmic integrity.

When the causing event is not profiled, it has to be retrieved through “schema completion” (Mandelblit & Fauconnier 2000: 184), by making “inferences from the context which make probable what [the causer] does or is to produce a certain effect, or at least that he has done something whatever it is” (Magnusson 1987: 216, emphasis original). In order to infer the causing event in (85) above, He makes me laugh, one would need more information about the situation in which the whole causative process took place. In (99) and (100), it is knowledge of the world rather than knowledge of the situational context that is needed in order to retrieve the gapped portion. We know that it is not a film but watching it that gets you thinking and that it is not university in itself that makes students question themselves, but going to university, attending lectures, meeting other students, etc. (99) That JFK’s supposed to be very good you know, the film. (…) Yeah he said it’s very long. <pause> But i--, it’s <pause> it’s very good, it gets you thinking.

79


(100) I think candidates who are thinking of applying for entry in nineteen eightyfive, who are now say seventeen coming on eighteen, have got three or four years ahead of them at university, which is a considerable time, and if nothing else, university will make them question themselves, what their interests are, and they’ll introduce them to new subjects, new areas of study, and it may be a mistake to embark on a vocational course, and discover halfway through that actually it’s not what you want to do.

While profiled causing events occur in only a quarter of the causative constructions in BNC-10, Table 14 shows that they nonetheless represent a majority in two constructions, viz. [X cause Y Vto-inf ] (76.7%) and [X make Y Vinf ] (54.8%). In order to try and explain why profiling of the causing event is common in these two constructions, let us first see why it is not in the other constructions, taking the example of [X have Y Vpp]. As already suggested, [X have Y Vpp] is often used when the causer commissions a professional to do something. In other words, it tends to refer to an established “scenario” (see e.g. Goddard [1998: 70–71] on the concept of scenario), such as having one’s hair cut or having a car repaired, which most people are familiar with. The caused event, therefore, is sufficient to evoke the scenario and the event that brought it about. We know, for instance, that having one’s hair cut involves calling the hairdresser, making an appointment, going to the hairdresser, etc. Because all this is shared knowledge, mentioning these elements would only make the message unnecessarily long and redundant.

[X

CA US EY V to [X inf ] GE TY V to inf ] [X GE TY V prp ] [X GE TY V pp ] [X HA VE YV inf ] [X HA VE YV prp ] [X HA VE YV pp ] [X MA KE YV inf ] [X BE ma de V to [X inf ] MA KE YV pp ]

Table 14. Profiling and gapping of the causing event (percentages)

Profiled causing event

76.7

14.5

10.2

4.0

4.2

2.9

6.2

54.8

9.0

34.6

Gapped causing event

23.3

85.5

89.8

96.0

95.8

97.1

93.8

45.2

91.0

65.4

This is not to say that the causing event cannot be profiled when it is part of an established scenario. There are some contexts in which it is important to mention it. Consider (101) and (102), both of which involve the well-known scenario of having medical treatment done, but still make explicit reference to the transfer to the place of treatment (go into hospital, gone to a National Health hospital), which is part of the causing event. This is because, in both sentences, the place of treatment is crucial to the argument. In (101) the discussion centres on going to a hospital


vs. receiving treatment at home, whereas in (102) the contrast is between National Health hospitals and private hospitals. In addition, (102) profiles the request part of the scenario (with the request to…), which is of importance too, since Richard did not go to hospital on a doctor’s recommendation (he was not greatly hampered by his back problem and hence the operation was obviously not necessary), but at his own request – which seems to be quite acceptable to a private surgeon, but might be less so to the NHS. (101) Jane: Phillip: Jane: Phillip: Jane: Phillip: Jane: Phillip: Jane: Phillip:

Jane:

Phillip: Jane: Phillip: Jane:

That was the other thing she said, that he, he should have be-gone into hospital. Course he should. He tried to do it at home and that was They co-ridiculous! did he have the opportunity to go into hospital to have it done? Erm, well the I would have thought Yeah. so because erm I mean Paul said, with the blokes down in London and he, he ga-- gave them the pills and sa-- told them that they go back home and so he did. Yes but the, Julie said, and she’s right of course, you can’t have the erm <pause> strength of drugs to, to, to administer at home because obviously It wasn’t, it wasn’t you’ve got to be monitored on that sort of thing. You can’t be put into it wasn’t done right. a deep drugs and ni-- not have erm <pause> some medical staff <pause> monitoring you.

(102) Enid: I bet it doesn’t relate to an actual operation in a <pause> general hospital because you’ve got so many people wanting their rake off in a private one surely? I mean haven’t you? Patsy: Yes. Yes. Enid: That erm Patsy: Yes their profit Enid: Ya that’s right. Noel: rather tired <pause> bit fed up with private medicine at the moment. <pause>

81

82


Enid: Noel: Enid: Patsy: Enid: Patsy: Enid: Patsy: Enid: Patsy: Enid:

Patsy: Enid:

Oh because of Richard? I’m convinced that Richard you know Richard <pause> er er ? Mm? Had he gone to a <pause> National Health hospital Mm. with the request to have that back thing done for him Mhm. I’m convinced that at the age of seventy eight people do-they would have said don’t bother. Yes I’m sure they would have done. Ya. And er would have erm <pause> perceived that he wasn’t greatly hampered by it, but once you’ve got a surgeon who is going to get a fat fee <pause> he did say to him are you sure you want this done Mm. but Richard was sure he wanted it done <pause> and nothing came between him and that <pause> intention. And you see he’s, it’s going to finish him. He’s not going to get better.

Except when there is a particular reason for mentioning the causing event of a well-established scenario, as in the two examples above, this information is gapped because, as pointed out by Seuren (2001: 221), “[h]umans will do with the minimum necessary. What is already mutually known or understood need not be expressed”. This appears to be the case for a majority of causative constructions in BNC-10. If we exclude unusual scenarios, there do not seem to be several ways of, say, getting a car started or having a boiler going. The situation seems to be somewhat different for [X make Y Vinf ] and [X cause Y Vto-inf ]. In the former case, the caused event is not so often part of a well-established scenario and the predictability of the causing event, therefore, is less. As we will see later (Section 5.3.5), [X make Y Vinf ] often involves non-volitional effects, that is effects which are not directly dependent on the causee and hence are not controllable, such as laughing, crying or feeling (good, guilty, sorry, etc). Being uncontrollable, they do not follow strict rules of the type “X causes Y”. A given causing event may have different results in different contexts or with different causees and, more importantly for the present purposes, one and the same result may be achieved by different causing events. Example (103) shows that what makes the causee laugh is an unpredictable combination


of elements. Another person, however, might be amused simply by the jokes the causer tells. There are many ways of making someone laugh, and while this information may not be, strictly speaking, indispensable, its omission may result in some sort of “psychological gap” (Magnusson 1987: 218), since this is information that cannot possibly be retrieved through schema completion.34 (103) [But she makes me laugh though. (…)] The way she those things with like her face and stuff, I don’t know, just the way she stands, everything about her just makes me laugh.

To some extent, the same is true of [X cause Y Vto-inf ]. In (104) there are several ways in which the conventional allergists and general practitioners could have caused their patients to die. Specifying which of these possibilities is the actual causing event (using unsafe desensitisation) is therefore not redundant. (104) Using unsafe desensitisation, conventional allergists and general practitioners have caused 26 patients to die since 1957.

But the frequent profiling of the causing event with [X cause Y Vto-inf ] may also be explained by the genre in which the construction typically occurs, namely scientific and technical writing (see Section 9.2.2). The relation between cause and effect in this construction may be more systematic than is the case in [X make Y Vinf ], but it is a relation that is unlikely to be known to the reader (as opposed to the familiar scenarios found with [X have Y Vpp]). In (105) it is likely that the movement of the string in the direction of the arrow will always cause W to describe a tractrix and that a movement in a different direction would have another effect. However, this necessary link is not part of a well-known scenario and the writer may therefore not assume that the reader will be able to complete the schema and retrieve the causing event if this event is omitted from the sentence. In such circumstances, the mention of the causing event participates in the effort, prevalent in scientific and technical writing, to make matters as clear and precise as possible. (105) If a string is attached to load W which is free to slide without interference, then movement of the string in the direction of the arrow will cause W to describe a tractrix. 34. The other make constructions, [X be made Vto-inf ] and [X make Y Vpp], share the feature of predominantly taking non-volitional effects. [X be made Vto-inf ], however, puts the emphasis on the causee and, as already pointed out, gaps the causer, and consequently the causing event, most of the time. As for [X make Y Vpp], although it presents a very respectable proportion of profiled causing events (34.6%), it tends to focus on the result of the action rather than the action itself (as is the case with all past participle constructions, cf. Nosek [1965: 40] or Langacker [2002: 130]).

83

84


4.3.5 Identity between participants While the preceding sections have dealt with the profiling or gapping of the elements making up the action chain, this section discusses the possible identity between certain participants, using Talmy’s (2000a) concept of psychodynamics (see also Gilquin 2007). The emphasis will be on those causative constructions that display identity between the causer and the causee, which we will call “coreferential” constructions – although, as will become clear in what follows, the two participants need not be explicitly “referred to” at the linguistic level. Identity involving the patient is possible as well, but will only be briefly alluded to. As already noted, Talmy (1986, 2000a) includes causative constructions in his analysis of “force dynamics”, which describes the interaction of entities with respect to force. Force dynamics, as defined by Talmy, does not necessarily concern a physical phenomenon, as the interaction may also be of a more symbolic nature (cf. Section 4.3.1.1). This idea has been explored by Talmy (2000a) with the notions of “sociodynamics” (ibid. 438–440) and “psychodynamics” (ibid. 430–435). In the former case, the interaction involves social forces. In a sentence such as She persuaded him to come to the meeting, “he” has a tendency towards rest, but “she” exerts pressure on him to force him towards motion – and is successful in doing so, since the resultant state is an action. The physical contact between two objects with a transmission of energy, characteristic of the basic model of force dynamics, is here extended to “one sentient entity’s production of stimuli, including communication, that is perceived by another sentient entity and interpreted as reason for volitionally performing a particular action” (ibid. 438). In the case of psychodynamics, the participants are not two distinct entities, but two parts of a single psyche. In the sentence I refrained from yawning, the subject is presented as a divided self with opposing tendencies. As in physical force interaction, there is pressure towards the realisation of a certain act and resistance against performing it. One part of the self is characterised by a tendency towards action (i.e. yawning) and the other part by a tendency towards inaction (i.e. not yawning). Either of them may be stronger and determine the resultant state – in this case, a state of inaction. Lee (2001: 110) refers to the two parts of the human mind as the Subject and the Self and describes this division as follows:35 [O]ur conception of the human mind involves a distinction between two components – the Subject and the Self. The Subject is essentially the seat of our rational and moral judgments, whereas the Self is that part of our personality that interacts directly with the world. In the ideal situation, the Subject and the Self

35. See also Lakoff (1996), or Cottier (1991) who speaks of dissociation inside the subject.


are in harmony (compare She’s a very together person), with the Self acting in accordance with the directions of the Subject. However, the Self can escape such control and perform acts under its own agency (compare I couldn’t stop myself, I got carried away).

This two-tier conception of the human mind explains why the causer and the causee may be identical or, put more simply, why it is possible to cause oneself to do something. Consider (106). Myself is the Self, the part that interacts directly with the world and can, say, sit down, talk with a friend, go for a walk or go shopping. I is the Subject, who examines the situation, decides what should be done about it, etc. In this sentence, the Self is reluctant to spare the time to talk with Neil because he wants to get ready to come away. But the Subject thinks that he should sit and give Neil time, for instance because Neil is a very good friend of his or because he has serious problems, and consequently the Subject forces the Self to sit and really give him [Neil] time (in Talmy’s words, the Subject manifests his tendency at the expense of the Self). (106) Neil, bring cup of tea in here and we’ll talk about it <pause> and I didn’t really want to spare the time because you could imagine that I wanted to get ready to come away <pause> but I I made myself sit and really give him time.

Note that dissociation between the Subject and the Self may also result from the Self escaping the Subject’s control and performing an action of his or her own free will. In (107) it is not a conscious decision on the part of the Subject to make the Self feel sick (as it is the Subject’s conscious decision in [106] to sit and take time). Simply, the Subject was presumably not in control when the action of eating (too much) took place. (107) But I ate myself silly on them last night and made myself feel sick so I’ll never eat them again now.

It should also be pointed out that identity between the causer and the causee does not necessarily require the two participants to be encoded at the linguistic level. Compare (108), taken from ICE-GB, and (109). ( 108) She made herself drink the rest of the coffee, though it was cold. (109) I’m never going to get it [old cup of coffee] drunk.

They differ in that the Subject and the Self are both expressed in the first sentence (she and herself), whereas in the second sentence only the Subject is mentioned (I). In both cases, however, there is a desire on the part of the Subject to drink the

85


coffee and resistance on the part of the Self to drinking it (because it is cold or old). The gapping of the Self (causee) in the second sentence should be seen in relation to the discussion of Section 4.3.3, where it was pointed out that irrelevant or redundant participants can often be left out in the linguistic realisation of the action chain. Obviously, being identical to the causer is a prime reason for the causee not to be profiled. Note, incidentally, that the expression make do, already mentioned earlier, may be understood as a construction with an ellipted causee identical to the causer. (110) could be paraphrased as “make yourself do [= manage] with what you can afford at the time”. (110) [You have to] Make do with what you ca--, what you can afford at the time.

The analysis of the corpus data reveals that with most constructions, the identity between the causer and the causee is a relatively marginal phenomenon, with proportions ranging from 0% to 12% (see Table 15). With [X get Y Vpp], however, this phenomenon accounts for over 63% of the examples.36 This peculiarity may be explained by the semantics of the construction, and more precisely the fact that it is often used when some sort of effort or difficulty is involved in the causative process (Section 7.3.2). Because the split between Subject and Self necessarily involves some sort of effort, since one part of the mind has to escape the control of the other part and perform acts under its own agency (see Lee’s quotation above), the get construction is particularly appropriate to describe such a state of affairs. In the following example, the difficulty associated with the split between the causer and the causee is clearly present, as appears from the underlined elements: (111) And she was gonna, aye trying to do her ironing because she said if she didn’t get it done there’d be all next week’s as well to do.

[X have Y Vpp] can express the same sort of difficulty when the causer is identical to the causee. To the question “What is the difference between I opened the door and I had the door opened”, Poldauf (1967: 33) answers that “[t]he have construction gives the statement an undertone of effort spent in performing the action, especially one necessary to overcome difficulties”, as illustrated by (112). However, because the notion of effort is not intrinsically part of the semantics of [X have Y Vpp], it is 36. This is all the more surprising since this phenomenon is hardly ever mentioned in the literature. Swan (2005: 201) very briefly refers to the possibility of using [X get Y Vpp] “to talk about completing work on something”, although he does not seem to consider this pattern a causative construction. Dieterich (1975) proposes a lengthier treatment of such constructions (which he sees as causative), but only deals with have.


Distinction

100.0 100.0

99.2


prp ]

YV

[X

HA VE [X

inf ]

YV

HA VE [X

pp ]

HA VE

[X

prp ]

[X

GE

TY V

inf ]

TY V

to -

[X

GE

TY V

GE

[X

[X

CA USE

YV

to inf ]

Table 15. Relation between causer and causee (percentages)

19.6 100.0 100.0

75.7

Identity

0.0

0.0

0.8

63.1

0.0

0.0

11.7

97.1 100.0 2.9

0.0

97.0 3.0

Ambiguity

0.0

0.0

0.0

17.4

0.0

0.0

12.6

0.0

0.0

0.0

less suited to the purpose than [X get Y Vpp], and the proportion of constructions with a causer identical to the causee is therefore lower (11.7%). (112) Mind you they have done a bit more to it since Wiggie’s been there in as much as she’s had a big curtain put across (…) All the way across. Because Steve couldn’t sleep.

It will be noticed that Table 15 contains a row for ambiguous cases. When both the causer and the causee are profiled, there can be no doubt about their co- referentiality, if only because the causee then normally takes the form of a reflexive pronoun, as in (106), (107) or (108) above.37 When the causee is not profiled, on the other hand, the construction is completely ambivalent with respect to the identity (or lack thereof) between the causer and the causee. Out of context, a sentence such as David had the house painted can mean either that David painted the house himself or that he asked someone else (a professional painter, for instance) to do it. Some sentences in the corpus contain linguistic clues that indicate the correct interpretation, e.g. (113) I mean if if if you’ve had your vehicle serviced and the sump plug hasn’t been put back in, then obviously you’ve got a perfectly legitimate claim against the person that’s done the work. (114) I couldn’t , I think what I do now, when I, I don’t look at anything and suddenly <pause> almost ready to <pause> have something done myself. 37. There is only one exception in BNC-10, viz. i. But he we’ll have to live apart, we’ll have to make us try and make our marriage work by living apart. However, this sentence was judged as not very felicitous or wholly unnatural and abnormal by 70% of the informants who took part in the elicitation test.

87

88


(115) And I said I’ll bring it round personally and er <pause> you know, get it switched on for her.

Sometimes, our knowledge of the world may help us determine the most likely reading of the sentence. We know, for instance, that it takes more than a minute to get a door handle fixed by someone else, say a professional, as one needs to contact that person, perhaps make an appointment and wait for him/her to come, and we may therefore presume that (116) refers to an action performed by the causer him- or herself. (116) Now wait a minute till I get the handle fixed.

Most of the time, however, one has to rely on a larger context to interpret such constructions. Consider the sentence in (117). Out of context it is not clear what get that done refers to, nor whether the construction is co-referential or not. The larger context reveals that the caused event consists in repairing a car, but we need an even larger context (and the metatextual information provided in the corpus) to understand that the speaker, Melvin, is a self-employed panel beater and that the construction, therefore, is most probably co-referential. (117) Yeah, well you can drive the Yugo for a little while until I get that done, I’ll drive the Sierra, if you want.

But while having access to a larger context definitely helps interpret the relation between the causer and the causee, some occurrences of [X get Y Vpp] and [X have Y Vpp] remain ambiguous, as indicated in Table 15. Here are some of the sentences that could not be disambiguated by means of the context: (118) Oh she said she’s gonna erm, part of the garden in Wales she’s having it made into like a rose garden or something ( 119) They’ve had their house refurbished in the middle of Oxford. (120) I wanna get you expelled.

(121) Gary wants me to find out a bit more about it cos he wants a new one [computer], another, well he wants a new one, cos we done a deal, get my bed, if we get the bedroom done (…) he’ll buy a PC, I said okay fair enough

Note that sharing the speaker’s situational context would not necessarily be sufficient to determine the correct interpretation. In (119), for example, the speaker may not be aware of who actually refurbished the house. This, however, is not essential to the success of the interaction, for what is important in such


c onstructions is the result of the action, not who performed it (see Ikegami 1989: 210 and Cottier 1991).38 In (122), the moment when the shopping is finished is more important than the actual process of doing the shopping.39 These ambiguous constructions, therefore, are often used in cases where the speaker merely wants to emphasise the result of the causing event, without having to specify who performed the effect. Consider for example (123), where the ambiguity of the construction enables the speaker not to compromise him- or herself too directly. (122) Anyway, I’m going to get ready or we’ll never get the shopping done before you go to work. (123) I’ll have to <pause> see if I can get some banana skins put on the stairs [to bump him off].

Another point that should be borne in mind is that interpretation also has a great deal to do with conceptualisation, as one and the same situation may be conceptualised differently by different people. In (124) it is essentially a matter of conceptualisation whether you consider that, as a customer, you book a trip yourself, or you ask a travel agent to book it for you with a supplier. In the same vein, the cheque in (125) may be conceived of as being put into the bank account either by the customer (simply through giving it to a bank clerk) or by the bank clerk, who performs this action, say, by introducing the correct data into his/her computer. Such differences in conceptualisation are not visible at the linguistic level and could only be detected by questioning the speaker about his/her utterance. ( 124) Have you got your hu-- your honeymoon booked up though now? (125) cos er, you have to wait three days for it, and then cheque’s gotta clear in the bank, it’s like our Mark he had a cheque put in his [bank account], he had to wait three days before it cleared

38. This, incidentally, may explain why, with have, stress normally falls on the past participle in co-referential causative constructions, while it falls on the causative verb in regular causative constructions (see Brinton 1994). 39. This focus on result may remind one of adjectival causative constructions like They made it clear (see Altenberg 2002a or Boas 2003). In such cases too the focus is on the result (see Poldauf [1967: 34] on the resultant force of adjectives) and it is the causer that performs the action, as appears from the paraphrase They clarified it. The parallelism between the two types of structures is very clear in (i), where the past participle is coordinated with an adjective.

i. Cos I can’t get her hair washed and clean now, look you’ve put some of that nasty stuff in.

89


Besides identity between the causer and the causee, there may also be identity between other participants in the causative construction. The patient may be identical with the causer (126) or with the causee (127). It is even possible for the patient to be identical both with the causer and the causee, as in (128), which could be paraphrased as “I’m getting myself to organize myself ”. (126) A child who is having difficulty in making herself understood may require a great deal of support; a few minutes later that same child dealing with a different topic may become relatively fluent and require a much lower level of adult participation. (127) But there seems no room for any notion of the decorous in Olson’s “objectism”, any more than there is room for it in the lawless world of the Cantos, or in Pound himself when he is without a master to translate, whose example makes him surpass himself. (128) Yes I’m going to the street and getting an early lunch and getting myself organized for <pause> going out.

This type of identity, however, is rare in the corpus, as Table 16 shows. It never occurs in half of the constructions studied here, and when it does occur, it is with a proportion of under 4%. The only exception is [X make Y Vpp], in which the patient refers to the same entity as the causer in 17.6% of the cases, as illustrated by (129). Sometimes, the link between the causer and the patient is looser, as in (130) and (131), where the patient may be said to “belong” to the causer. It is not uncommon to find references in the literature to this special link between the causer and the patient in [X make Y Vpp]. It is often suggested, however, that this link is an obligatory one (Andersson 1985: 149), whereas the corpus proves it to be optional, since in a sentence such as (132) the causer and the patient refer to different entities. (129) More generally, the issue became involved with widespread feelings of disquiet over the increase in the “visibility” of local government in Orkney as the Orkney Islands Council, under the impetus of new staff, new problems, new government legislation and new powers began to make itself felt throughout the islands. (130) Other molecular techniques are making their impact feclt [felt] in micro biology. (131) if you have any comments about the work of The Old Church and Society Committee, the work we do fo—fo—for Synod and in the provence <pause> please make those comments [= your comments] known <pause> to Catherine or to your district representative!


Identity

1.3

0.0


prp ]

YV

96.6 100.0 100.0 100.0 3.4

0.0

0.0

0.0

[X

HA VE [X

inf ]

YV

HA VE [X

pp ]

HA VE

[X

prp ]

GE [X

98.7 100.0

1.0

[X

99.0

TY V

to inf ]

TY V

GE

TY V

GE

[X

Distinction

[X

CA USE

YV

to inf ]

Table 16. Relation between patient and causer/causee (percentages)

98.2 100.0 1.8

0.0

82.4 17.6

(132) Oh I should think he [the clown Dimitri] makes money, yeah makes Switzerland known <pause> Like Gabriella Sabatini is a tremendous asset to the Argentine.

Before closing this section, let us say a few words about the place of such constructions within the frame of causation. Like “regular” causative constructions, coreferential causative constructions express a process in which a causer transfers energy to a causee, which may (but need not) transfer this energy to a patient. Co-referential constructions, however, differ from regular constructions in that the transfer occurs within one and the same participant (or, more precisely, from one part of a participant to another part of this participant) and not between distinct entities. Both types of constructions reflect the general schema of the action chain and may consequently be described as causative constructions, but they vary in their degree of prototypicality. According to Lakoff (1987: 54), a prototypical causative construction should contain “a single definite agent and a single definite patient”, the implication being that the two entities should be distinct from one another. In a co-referential causative construction, this condition is not fulfilled, so that the construction may be regarded as a less central member of the frame of causation than a construction where all participants are distinct from each other. We will come back to the question of prototypicality in Chapter 6.

4.4 Action chains in the elicitation data Following the Corpus-Cognition Integrated model, we also examined how action chains are realised in the elicitation data. A first interesting observation is that informants tend to profile elements of the action chain which are recoverable from the context (because they are part of an established scenario) and hence could be gapped. Consider (133) and (134). In both cases, the scenario makes it

91

92


clear who the causee is, and this participant could therefore have been gapped, resulting in a past participle construction (He got the leaking pipe fixed and Jan had Mark arrested). This tendency towards explicitness will be discussed again in Section 6.4, as it may also be viewed as a manner of producing more prototypical constructions. (133) He got the plumber to fix the leaking pipe.

<EL:E3:033>

(134) Jan had the police arrest Mark.

<EL:E4:018>

One exercise in the elicitation test was specifically devoted to the causing event in the action chain, namely exercise B4, where informants were asked to supply a byclause, thus making the causing event explicit. Their answers provide valuable insights into the relation between lexical causative verbs and periphrastic causative constructions. When completing the sentences “I moved the chair by…” and “I caused the chair to move by…” (exercise B4.1-2), some informants illustrated the distinction between direct causation (i.e. physical contact) and indirect causation (i.e. no physical contact) mentioned before. For example, one informant completed the first sentence with “pushing it with my foot”, and the second sentence with “tying a string to a leg and tugging gently upon it”. According to some linguists, indirect causation in the case of a periphrastic causative construction may take the form of magical powers (see Radden [1992: 516] or Goddard [1998: 268]). This is reflected in some answers, for instance “I moved the chair by pushing it to the left” vs. “I caused the chair to move by casting a magic spell”. Table 14 showed that [X cause Y Vto-inf ] is the construction with which the causing event is most often profiled. This is confirmed by the elicitation data, and more precisely exercise E. When asked to produce a causative construction with cause [E.1], two quarters of the informants made the causing event explicit, e.g. (135) Heating the test tube constantly caused the liquid to boil.

<EL:E1:018>

(136) The invasion caused the war to begin.

<EL:E1:022>

(137) I braked, causing my passenger to exit suddenly through the windscreen. <EL:E1:004>

And in the second part of exercise E, which consists in selecting one of the causatives to join a number of elements, some informants chose the verb cause but adapted the sentence so as to have the opportunity to mention the causing event. Thus, the teacher/the student/leave/the room [E.5] was turned into: (138) The students’ attitude caused the teacher to leave the room.

<EL:E5:002>


Exercise B4, already referred to above, is also enlightening with respect to the existence of a scenario in the action chain. It was argued earlier that causative have often evokes a well-established scenario and that the explicit mention of the causing event is therefore rarely necessary. This is corroborated by the elicitation data, which show that people find it difficult to even think of any causing event in have constructions. It is with have that the number of blanks is the highest: between 11 and 15 informants out of 50 left the have-questions unanswered, while the rate ranges from 0 to 9 with the other causatives. When the informants did provide a by-clause, the answers turned out to be surprisingly similar, cf. “I had the doctor come by… phoning his secretary/ringing him/making an appointment/asking him”, “I had the car repaired by… phoning the garage/leaving it at a local garage/taking it to a specialist/going to the mechanic”. The causing events proposed with the other causatives, by contrast, were more varied. Several exercises in the elicitation test concern the relation between the causer and the causee. A first general finding is that co-referential constructions are usually judged as grammatically acceptable, especially when the causee is expressed. Over 80% of the informants accepted I made myself sit in the armchair (exercise D.9), while some 52% accepted By the time we got here and started cooking something we had to rush to get it eaten (exercise D.13) – as against 27% who rejected it (the remaining 21% were uncertain). Another useful exercise to investigate co-referential constructions is exercise C, where informants have to decide who (the subject or someone else) performs the action in a decontextualised causative construction. The importance of context in disambiguating such constructions probably explains why this exercise was often pinpointed by the informants as particularly difficult. A number of informants rightly observed that some or all of the sentences were ambiguous and could have both interpretations. One person even noted that “most English speakers are not able to tell the difference between these answers”, and another one systematically ticked the two answers. Crucially, none of the ten sentences of the exercise could lead to a complete agreement among the fifty informants, with some of them, such as (139) and (140), obtaining an almost equal proportion of (a) and (b) answers. ( 139) Get the door shut. a. The addressee/interlocutor will shut the door. b. Someone other than the addressee/interlocutor will shut the door. ( 140) We’ve got to see if we can get all the fees paid. a. … if we can pay the fees. b. … if someone other than us can pay the fees.

93


These results underline the ambivalence of such constructions, especially out of context. They also suggest that conceptualisation might have an important role to play in the interpretation of these sentences. It is very likely that, with no context to resort to, informants rely on their own experience of the world to answer the test item in (141), that is whether they normally wrap up their Christmas presents themselves or have it done by shopkeepers. ( 141) You’ve got to get these Christmas presents wrapped up. a. You’ve got to wrap up these Christmas presents. b. Someone other than you has got to wrap up these Christmas presents.

One important finding from the corpus analysis is also supported by the elicitation data, namely the preference of [X get Y Vpp] for situations where the causer and the causee refer to the same entity. In exercise C, the get constructions obtained an average of 67.6% (a)-answers (i.e. the “causer = causee” reading), whereas the average for the have constructions was under 36%. In exercise A, another judgement – preference exercise, the two relevant test items, (142) and (143), made this tendency even more obvious: 96% of the respondents chose the get construction in the first case and 92% in the second. ( 142) Phoebe doesn’t like shopping, so she hardly ever a. gets the shopping done. b. has the shopping done. ( 143) The student worked late to a. have his geography essay finished. b. get his geography essay finished.

Finally, the answers to test item 8 in exercise B4 also confirm the association of get with co-referential constructions. When completing the sentence “I got the car repaired by…”, a number of informants made it clear that they assumed identity between the causer and the causee, e.g. “I got the car repaired by doing the work myself/fixing it myself/repairing the engine/working at it for hours”. The above shows that, by and large, the elicitation data corroborate the findings about the action chain made on the basis of the corpus data. The two types of data thus converge to demonstrate that, from the point of view of the action chain and its realisation, the ten constructions under investigation present important differences, which should be taken into account when describing the use of these constructions.


4.5 Summary This chapter has started from the theoretical construct of action chain. It has shown that the same action chain underlies all periphrastic causative constructions, but that the action chain may be realised differently at the linguistic level, depending on which elements are profiled and which are gapped. Not only does the action chain make it possible to identify intrinsic differences between the ten causative constructions under study (e.g. the focus of past participle constructions on the end result of the causative process), but a careful examination of the realisation of the action chain in the corpus data also reveals some interesting facts, most notably the absence of an explicit causing event in most get and have constructions (which we have attributed, in part, to the existence of a wellestablished scenario making the mention of the causing event redundant), and the frequent identity between the causer and the causee in [X get Y Vpp], two findings which are largely supported by the elicitation data. The next chapter also starts from a cognitive concept in an attempt to better describe and explain the phenomena observed in the corpus and elicitation data. This time, the focus will be on the syntax and semantics of causative constructions, and the starting point will be Goldberg’s (1995) “Principle of No Synonymy”.

95

chapter 5

The syntax and semantics of causative constructions The Principle of No Synonymy

In this chapter, we start from Goldberg’s (1995) “Principle of No Synonymy”, which predicts that the difference in structure between the causative constructions under study goes hand in hand with other differences (semantic, pragmatic, etc). Following this principle, the chapter seeks to identify the syntactic and semantic features that distinguish between the ten constructions, using two types of statistical analyses, namely a chi-square analysis, which considers each parameter individually, and hierarchical feature selection, which considers all the parameters together.40

5.1

No (more) synonymy

The principle according to which “a difference in syntactic form always spells a difference in meaning” was originally formulated by Bolinger (1968: 127). Since then, this “one form, one meaning” principle has been adopted in several theoretical frameworks, most notably in functional grammar and cognitive linguistics. In cognitive linguistics, it has been expressed among others by Goldberg (1995: 67) and her “Principle of No Synonymy”, which states that “[i]f two constructions are syntactically distinct, they must be semantically or pragmatically distinct” (see also Croft’s [2001: 111] “Principle of Contrast”). Langacker (1988: 10–11) introduces the notion of “image” to convey a similar idea. He argues that “[t]wo expressions may be functionally equivalent and serve as approximate paraphrases or translations of one another, and yet be semantically distinct by virtue of the contrasting images they incorporate”. The notion of “image” and the related notion of “imagery” are fundamental to cognitive linguistics. It is a pivotal claim of cognitivists that a speaker may choose from a wide array of 40. See Degand (2001) for another study that seeks to identify the constraints that play a role in the selection of different causative constructions (in Dutch and also including the study of some causal connectives and prepositions).


alternative “images” in order to describe a given scene, and that the choice of a particular expression or construction over another is meaningful. This “ability to mentally construe a conceived situation in alternate ways” (Langacker 1988: 7) is what is called “imagery”. The “Principle of No Synonymy” implies that the ten causative constructions under investigation must represent different conceptualisations of the causative situation. However, this is not what transpires from the literature, where these constructions are often regarded as (quasi-)synonymous. It is particularly remarkable in grammars. The focus tends to be on complementation (see below), leaving very little room for other issues (especially semantic ones), thus giving the impression that the constructions merely differ in their structure. Sometimes, the constructions (or some of them) are explicitly equated with one another. This is especially the case in generative grammars, which typically bring all causative constructions together as being derived from the same deep structure (see Brame 1976: 17). But other, more traditional grammars (including pedagogical grammars for foreign language learners) also fail to recognise the specificity of each construction. Visser (1973: 2269), for example, notes that [X have Y Vinf ] “might be apprehended as a synonym of make or get to, and consequently is often replaceable by them”, and Todd & Hancock (1986: 218) claim that in [X get Y Vpp] get is the equivalent of have, and that in [X get Y Vto-inf ] it corresponds to cause. This lack of differentiation is also apparent from the examples and paraphrases that are given in the literature, for example the equation of I’m getting a new house built with I’m having a new house built (Palmer 1974: 159), He gets his hair cut once a month with He has his hair cut once a month (ibid.), Why didn’t you have that suit cleaned? with Why didn’t you get that suit cleaned? (Eckersley & Eckersley 1967: 195–196), He must have his pen mended with He must get his pen mended (Allen 1959: 35) or This would cause the wall to collapse with This would make the wall collapse (Roggero 1979: 304). In Swan (2005) and Lester (2008), the same examples are used to illustrate the get and have constructions. Lester (2008: 221) also points out that “[g]et and have both mean to ‘cause somebody to do something’”. This failure to semantically distinguish between the constructions is not the only weakness of the treatment of causative constructions in the literature. More specific claims will be presented and commented on in the course of the following analysis, but for now let us highlight some general tendencies characterising the syntactic and semantic description of periphrastic causative constructions in the literature, with special emphasis on grammars. The syntactic description of causative constructions mainly focuses on two (related) issues, namely complementation and passivisation. The different non-finite complements that are possible with each causative verb are mentioned and the special case of make, which takes a bare infinitive in the active voice but a to-infinitive

Chapter 5. The syntax and semantics of causative constructions

in the passive voice, is considered. Sometimes, exceptions to the general rules of complementation and passivisation are given, for example the use of cause with a bare infinitive (Poutsma 1926: 429), a present participle (Visser 1973: 2383) or a past participle (ibid.), have followed by a to-infinitive (Kirchner 1952: 386) and (active) make followed by a to-infinitive (Adamczewski & Delmas 1993: 23) or a present participle (Visser 1973: 2346). Syntax is so pervasive in some of these grammars that it actually guides the organisation of the grammar itself. In the case of causative constructions, this results in causatives being dispersed into several sections depending on the complement(s) they take (see Section 10.2 for more details). Complementation is sometimes linked to semantics, as when linguists try to provide semantic explanations for why certain causatives take a bare infinitive whereas others take a to-infinitive (see e.g. Dirven [1989] and Méry [1990], or Givón’s [1994] binding scale of event integration, which relates syntactic integration between two propositions to a semantic bond between two events). The semantic treatment of causative constructions, when present at all, also includes issues such as the nature of the elements involved in the causative construction (animate or inanimate and agentive or non-agentive participants, volitional or non-volitional effect, etc), the ambiguity of certain constructions (see Section 3.1.1.2) and, more generally, the meaning of the verbs or constructions (with, among others, the famous discussion about the relation between the verb kill and the periphrastic causative construction cause to die, already referred to earlier). The problem with these semantic claims is that they sometimes contradict each other. A striking example is that of Givón (1993: 9), who claims that causative make only occurs with agentive causers and who consequently rules out sentences such as *John’s behaviour made Mary quit her job or *The political situation made Mary quit her job. Adamczewski & Delmas (1993: 23), on the other hand, give the following examples, where the causer of make is clearly not agentive: The cold always made his nose run; The smell of the roast beef made Canning realize that he was hungry. Such contradictions, together with the little information that is provided about the semantics of causative constructions, leave the reader in the dark as to how these constructions should be used and how they differ from each other.

5.2

From data to database

In an attempt to offer a more circumstantial account of the behaviour of periphrastic causative constructions, each causative construction retrieved from BNC-10 was categorised according to a number of parameters, listed in Table 17. The parameters amount to about 50 and concern the three possible participants

99


of a causative construction (causer, causee and patient), as well as its two verbs (causative verb and effect). Some additional information is also provided such as the reference of the sentence in the corpus or the genre in which it occurred. The values of these parameters were encoded in a database for each instance of a construction, using the program Microsoft Works Database. Only the syntactic and semantic parameters will be taken into account in this chapter (with particular emphasis on the latter). Table 17. Parameters of causative constructions investigated causer - expression - implied? - sem. category - event? - verbal? - animacy - (pro)noun - number - gram. person - definiteness - specificity

causative verb - tense - time - voice - gram. polarity - main clause?

causee - expression - sem. category - animacy - (pro)noun - number - gram. person - definiteness - specificity - = causer?

effect - lemma - sem. prosody - expression - dynamicity - volitionality - funct. category - voice - gram. polarity - transitivity

patient

various

- expression - sem. category - animacy - (pro)noun - number - gram. person - definiteness - specificity - co-referential?

- code - reference - medium - genre - remarks

Some remarks are in order about the way classification was carried out. First, it should be stressed that some categories are not applicable to all examples, and that others are represented more than once in a single sentence. (144) contains no patient, and (145) no causee. In (146), on the other hand, there are two effects, and in (147) two causers.41 Note that, whenever possible, information was nonetheless provided about unexpressed elements, for example the animacy of the causee in (147). (144) The regulation of population density can only be a consequence of migration, not the reason why natural selection causes the habit to evolve. (145) You’re getting innocent people, men and women killed, children killed and we’re expected to accept this. (146) Well that’ll make them sit up and think that’s for sure

41. Coordinated participants are treated individually, except for number. In (147) the subject of have would be described as an animate causer followed by another animate causer, but as one plural causer. This is meant to reflect the fact that the causer of the construction is multiple, rather than single.

Chapter 5. The syntax and semantics of causative constructions 101

(147) It’s not twenty quid, if Melanie and I have our hair permed, that’s, that’s two perms and two, two cuts

Second, recourse to a larger context than the concordance line was often necessary in order to determine the value of the parameter (for example in the case of pronouns). Yet, there were instances where the value of the parameter remained questionable, despite access to the larger context, which is partly due to the fuzzy nature of categories and their overlapping boundaries. In such cases, the most likely and sensible value was attributed, the key point being consistency in the classification of all similar instances. This approximation is made necessary by the quantitative approach adopted in this study. As rightly pointed out by McEnery & Wilson (2001: 77), “[q]uantification (…) entails classification. For statistical purposes, these classifications have to be of the hard-and-fast (socalled ‘Aristotelian’) type, that is an item either belongs in class x or it doesn’t”. This, they add, may “entail in some circumstances a certain idealisation of the data: it forces the analyst to make a decision which is perhaps not a 100 per cent accurate reflection of the reality contained in the data” (ibid.). This may be illustrated by the case of the volitionality parameter, which makes a distinction between effects involving the causee’s will and those happening independently of the causee’s will (see Section 5.3.5). Kemmer (2001) notes that there can be varying degrees of control involved in a situation. She gives the example of laugh and cry: [W]ith laugh and cry, the experiencer has some potential control, in that one could conceivably carry out such actions on purpose rather than as a non-volitional reaction to a stimulus, and in that the experiencer could also in general keep from carrying out the action if needed. (emphasis added)

However, as Kemmer herself admits, such events involve a lower degree of control than activities like “reading” or “walking”, and they will therefore be classified together with (more) non-volitional verbs like “see” or “suffer”. The last point to be made about the classification of the corpus data is that in some cases certain parameters turned out to be indeterminate as to their value. In (148), for example, while we can say that the causer is singular and pronominal, it is impossible to determine, beyond the simple animate/inanimate distinction, which semantic category it belongs to – concrete object or abstract entity. (148) Why, why what makes you say that?

On the basis of the database thus created, a statistical analysis was carried out, with the aim of identifying the categories that best distinguish between the ten causative constructions under investigation. The statistical analysis was twofold.


First, the parameters were examined one by one, and their significance for the different constructions was tested. Second, all the parameters were considered together and only those that made it possible to predict the choice of the constructions in a statistically significant way were retained. Both analyses may be described as “category statistics”, as opposed to “text statistics”, which simply measure the frequency of particular items or co-occurrence of items. We will see in what follows that this type of statistics brings to light interesting facts which could not have been revealed by means of simple text statistics.

5.3

Individual approach: Chi-square analysis

5.3.1 Methodology: Chi-square test with adjusted residuals In order to test the statistical significance of the different parameters (i.e. determine whether certain values of a parameter are distinctively associated with some constructions or whether the observed values occur with similar proportions in all constructions), a chi-square test was performed on each of them. The method that was used is the chi-square test in r x 10 tables, with adjusted residuals.42 The r x 10 contingency table contains ten columns, one for each of the ten causative constructions, and a number of rows corresponding to the number of values that may be taken on by the parameter (e.g. two for the animacy of the causer, which can be animate or inanimate, and three for its grammatical person, which can be a first, second or third person). Thanks to the chi-square result, the probability (p-value) can be determined, and hence the statistical significance (or otherwise) of the parameter. In order to identify the cells of the contingency table that are responsible for a significant overall chi-square value, one may compute the adjusted residuals, which measure the deviations of observed from expected frequencies in each cell (Haberman 1973, see also Everitt 1992: 46–48). Table 18 provides an illustration for the functional category of the effect (see Section 5.3.5 for a description of this category and the values it can take on). The first part of the table lists the observed frequency of each value of the parameter (material, mental or relational) in each of the ten constructions. The chi-square result amounts to 1,239 which, with a degree of freedom of 18, is highly significant (p till a game finished! (150) those of you had said yes, what, what, what experiences have you had that made you say yes?

As appears from Table 19, three constructions are distinctively associated with a subordinate verb, namely [X cause Y Vto-inf ], [X make Y Vinf ] and [X make Y Vpp]. The other constructions are either associated with a main verb or do not present 43. Blank cells in the tables mean that the construction is not significantly associated with the corresponding value (i.e. is neither attracted nor repelled).


21.4

prp ]

YV

+

–


30.6

HA VE

inf ]

YV

HA VE

pp ]

78.6

[X

30.0

69.4

+

[X

33.1

HA VE

prp ]

GE

TY V

inf ] to -

TY V

70.0

[X

– 29.0

66.9

[X

+ 50.5

GE

TY V

GE

+ 71.0

[X

Subordinate verb

– 49.5

[X

Main verb

[X

[X

CA USE

YV

to inf ]

Table 19. Status of the causative verb (distinctiveness and percentages)

– 64.7

62.5

35.3

37.5

–

– 56.1

32.3

43.9

67.7

+

+

any significant association in terms of the status of the causative verb. In the case of [X cause Y Vto-inf ], the association of the verb with subordination is to be related to another interesting feature of the construction, namely its frequent occurrence in adverbial participle clauses, which offer a convenient way of expressing the causing event (see Section 4.3.4), as in: (151) Heating collapses the blue spectral peak, causing the violet luminescence to shift towards red.

The study of the voice of the causative verb is limited by the fact that the search strings used to extract causative constructions would not make it possible to retrieve cases, if any, where causative get and have are used in the passive (except for cases where passive have would be directly followed by a bare infinitive or a present participle, but these did not occur in the corpus). This is because the repair mechanisms used to improve the precision rates with these two verbs involved excluding the infinitive marker to, past participles and –ing forms immediately after get, as well as the infinitive marker to and past participles immediately after have (see Section 3.1.2.3). It should be noted, however, that passivisation of causative get and have is often said to be impossible in the literature (e.g. Close [1975: 199], Andersson [1985: 144], Biber et al. [1999: 697] on get, and Quirk et al. [1985: 1207], Dixon [1991: 195], Wierzbicka [1998: 122] on have), and even those linguists who give examples of such a pattern admit that it is marginal at best (Cottier 1991). This will be largely confirmed by the elicitation data (Section 5.5). As for the other causatives, the voice of the verb is normally determined by the construction: [X make Y Vinf ] contains an active causative, whereas in [X be made Vto-inf ] the causative is in the passive. The only two constructions that show some variation in the voice of the causative verb are [X cause Y Vto-inf ] and [X make Y Vpp]. With [X cause Y Vto-inf ], there is just one instance of main clause passivisation (152),


which is not sufficient to make this a distinctive feature of the construction. With [X make Y Vpp], on the other hand, the passive use of the causative verb is statistically distinctive for the construction (although it represents a minority of some 19% in terms of frequency). An example is given in (153).44 (152) The most obvious effect of plate tectonics is that continents can be split and their components driven apart if a divergent plate margin becomes established beneath them, and can be caused to collide with each other along the lines of subduction zones, where mountain belts such as the Himalayas may thereby be generated. (153) The exercise was regarded by the then minister for health in Scotland, Michael Forsyth, as a piece of “action” research, the results of which would be made known at intervals during the evaluation.

The results for the tense (and non-finite forms) of the causative verb are detailed in Table 20.45 Here, we will single out a few interesting tendencies. [X cause Y Vto-inf ] appears to be associated with two tenses, namely simple present and present participle. The use of the simple present makes it possible, among other things, to describe universal truths, caused events that invariably happen as long as the causing event is performed in a particular way, e.g. (154). As for the present participle, it tends to occur in adverbial participle clauses like (155), which, as explained in Section 4.3.4, promote the expression of the causing event by establishing a link with the preceding clause (in the example, it is the fact that the ridge was breached in the middle during a storm that caused the harbour to revert to its former double spit state). (154) A simple extension is to provide a particular value of the tag field which does not specify a data format, and which always causes an error condition to be signalled whenever an operand is accessed with this value in the tag field. (155) The ridge survived until a storm in 1910, when it was breached in the middle, thus causing the harbour to revert to its former double spit state.

Get is the only verb that is associated with the imperative, both in [X get Y Vto-inf ] and [X get Y Vpp]. This shows that this causative is often used to exhort someone to do something, as in (156) and (157). In the case of [X get Y Vpp], the use of the imperative is often combined with co-referentiality (see Section 4.3.5), 44. See also Stefanowitsch (2001: 196–209) and Hollmann (2006) on the passivisability of English periphrastic causatives. 45. Tenses or non-finite forms which did not occur more than three times in any of the ten constructions were grouped together in an “other” category.


[X CA US E Y V [X to inf ] GE T Y V to inf ] [X GE T Y Vp rp ] [X GE T Y Vp p] [X HA VE YV inf ] [X HA VE YV prp ] [X HA VE YV pp ] [X MA KE YV inf ] [X BE ma de V to [X inf ] MA KE YV pp ]

Table 20. Tense of the causative verb (distinctiveness and percentages)

Simple present

+

–

–

–

–

+

+

36.5

11.2

12.6

9.3

11.1

12.9

7.0

36.7

29.6

12.9

0.5

1.1

1.6

1.9

0.0

4.3

5.5

2.3

0.0

3.2

–

+

+

+

–

6.0

1.1

22.0

14.7

4.2

7.1

13.4

1.6

5.1

9.7

–

+

–

+

+

9.6

23.6

11.3

30.0

24.6

17.8

13.3

9.7

0.2

0.0

3.2

1.0

3.2

Present progressive

+

Present perfect Simple past 13.5

11.1

Past progressive

+ 0.0

0.0

0.0

0.0

0.0

1.0 +

–

1.0

0.0

1.6

1.0

1.4

1.4

1.5

0.2

+

+

–

–

7.0

8.7

4.7

7.4

12.5

2.9

2.2

3.9

3.1

9.7

1.9

1.5

2.0

3.2

Past perfect Simple future

+

Conditional Imperative

+

0.0

+ 1.0

0.5

–

+

0.0

9.0

Bare infinitive

0.0

1.5

2.8

7.1

+ 1.6

+

7.9

– 0.0

2.9

+

–

1.3

3.2

0.0

–

–

+

3.2

12.5

15.6

10.2

13.4

9.7

12.9

8.3

6.6

36.7

0.0

0.0

0.5

0.0

0.6

0.0

1.4

0.4

0.2

0.0

0.0

–

+

+

+

+

–

–

7.5

34.7

26.7

36.1

17.1

27.3

19.3

2.0

22.6

1.4

0.0

0.7

1.3

1.0

3.2

Perfect infinitive To -infinitive Present participle

19.7

+ 10.0

– 0.3

0.0

0.1

whereby the hearer (causer) is also the person who is expected to carry out the action (causee). In (157), the context makes it clear that “you” is both the causer and the causee, which means that the sentence could be paraphrased as Shut the door. The use of a periphrastic causative construction, rather than a simple lexical verb in the imperative, may be viewed as a politeness strategy, since it allows the speaker to formulate the order in a more indirect way (see Brown & Levinson 1987: 132–144).


(156) [But she said it’s a bit too heavy to <pause>] carry so get your husband to come round and get it. (157) Get the door shut.

[X have Y Vpp] is the construction that is associated with the highest number of tenses. It is also one of the only constructions that distinctively occurs in the progressive form (present progressive and past progressive). This use is illustrated by (158) and (159) and emphasises the fact that the causative process described by have may extend through time. In (159), the building (or transformation) of the kitchen took so long that the causer actually had to stay in a caravan for a while. (158) Jack is having forecourt painted.

(159) Cos I remember erm what I do remember is when you were having it [kitchen] done, stopping in the caravan with you

Like cause, but unlike get and have, make is associated with the simple present. This is true of [X make Y Vinf ] and [X be made Vto-inf ] – although [X be made Vto-inf ] is also associated (and in fact more frequent) with the bare infinitive. While the use of the simple present with cause tends to refer to universal truths, with make it usually has other functions, as shown by the following two sentences: (160) God that [a good shot] makes me look like a good player that, doesn’t it? (161) At the same time, the people who create these contexts, the teaching practitioners, are often made to feel that their own experience as pedagogic providers is not given sufficient recognition but is, on the contrary, misprized as lacking in rigour: wisdom which leads to understanding is overridden by knowledge which leads to explanation.

5.3.3 The causer Although linguistically speaking, the causer need not start the causative construction (it never does, for example, in [X be made Vto-inf ]), it is the first element of the action chain at the conceptual level (see Chapter 4) and will therefore be treated first among the participants of the construction. We will investigate its form (noun, pronoun or clause), its grammatical person, its definiteness, its specificity, its animacy and its semantic nature. Table 21 shows the results for the form of the causer – noun, pronoun or clause. Nominal causers are distinctively associated with four constructions, viz. [X cause Y Vto-inf ] and the three causative constructions with make.


[X C

AU SE YV [X to inf ] GE T Y V to inf ] [X GE T Y Vp rp ] [X GE T Y Vp p] [X HA VE YV inf ] [X HA VE YV prp ] [X HA VE YV pp ] [X MA KE YV inf ] [X BE ma de V to [X inf ] MA KE YV pp ]

Table 21. Form of the causer (distinctiveness and percentages)

Noun Pronoun Clause

+

–

–

–

66.0

11.8

5.0

5.3

–

–

25.4

2.9

11.6

+

+

+

21.3 100.0 46.2

–

+

+

+

+

+

–

–

–

31.5

88.2

95.0

94.7

74.6

97.1

88.4

78.3

0.0

53.8

0.0

0.0

0.0

0.0

0.0

0.0

0.5

0.0

0.0

+ 2.5

ronouns are more distinctively associated with the get and have constructions, P except [X have Y Vinf ], which does not present any significant association. As for clauses, they are only distinctive for the cause construction (although they are not very frequent in this construction either), which again may be related to the tendency of this verb to make the causing event explicit, as in: (162) Increasing the number of reactive components in the filter stage to two, as in the simple low-pass L-C filter of figure 8.7(a), causes to reach a maximum rate of fall-off outside the pass band of 40 dB per decade of frequency.

The parameter of the grammatical person of the causer roughly divides the constructions into two groups (see Table 22): those that are distinctively associated with both first- and second-person causers, and those that are distinctively associated with third-person causers. To the first group belong the three get constructions and to the second group [X cause Y Vto-inf ] and the make constructions. In addition, [X have Y Vpp] is associated with first-person causers. From the percentages, it appears that [X get Y Vpp] is the construction that occurs most often with a first-person causer (46%). This may arguably be linked to another characteristic of the construction, namely its preference for situations where the causer and the causee refer to the same entity (see Section 4.3.5), as in: (163) I’d better go and get some work done Joyce!

It also turns out that [X cause Y Vto-inf ] almost exclusively occurs with third-person causers. First- and second-person causers are only instantiated once each:


–

+

+

+

0.5

24.8

26.9

27.1

+

–

–

–

99.0

33.4

36.1

26.6


prp ]

YV

[X

HA VE

inf ]

YV

HA VE

YV

+ 46.3

HA VE

pp ]

prp ]

GE T [X

[X

GE T

YV

to inf ]

YV

GE T

+ 37.0

[X

Third person

+ 41.8

[X

Second person

– 0.5

[X

First person

[X

[X

CA USE

YV

to inf ]

Table 22. Grammatical person of the causer (distinctiveness and percentages)

+

–

35.8

32.4

33.7

8.8

9.0

25.0

18.1

–

–

42.6

48.2

– 0.0

3.8

11.7

0.0

11.5

+

+

+

–

55.2

79.5 100.0 84.6

+

+

97.5

98.2

+

–

–

–

42.4

5.0

2.5

1.8


prp ]

YV

[X

HA VE

inf ]

YV

HA VE

HA VE

pp ]

GE

+

+

–

–

91.0

98.5

96.1

87.5

0.0

–

–

+

+

1.5

3.9

9.0

[X

[X

[X

TY V

prp ]

inf ] to -

TY V

GE

TY V

GE

+ 95.0

[X

Indefinite

– 57.6

[X

Definite

[X

[X

CA USE

YV

to inf ]

Table 23. Definiteness of the causer (distinctiveness and percentages)

80.8

12.5 100.0 19.2

Generic

91.0

91.2

– 8.4

3.2

9.0

8.8

prp ]

YV

HA VE

inf ]

YV

HA VE

pp ]

HA VE

GE

TY V

prp ]

inf ] to -

TY V

96.8


6.2

91.6

[X

52.2

+

[X

–

[X

93.8

+

[X

47.8

[X

+

[X

–

GE

TY V

GE

[X

Specific

[X

CA USE

YV

to inf ]

Table 24. Specificity of the causer (distinctiveness and percentages)

+

–

–

–

93.8

85.8

40.0

69.2

–

+

+

+

6.2

14.2

60.0

30.8

Chapter 5. The syntax and semantics of causative constructions

(164) It would off been o-- okay if I just been able to pull out and drive down the road because I had to pull out round the stationery vehicle as well, I would off taken too long getting out and I’d have caused the other one to slow down. (165) You’ve caused us to miss Red Dwarf!

Generally speaking, definite causers (166) are more typical of causative constructions than indefinite ones (167), as Table 23 illustrates. There are three exceptions, namely [X cause Y Vto-inf ], [X make Y Vinf ] and [X be made Vto-inf ], but only one of these presents a majority of indefinite causers, [X be made Vto-inf ], with five indefinite causers out of five expressed causers. [X have Y Vinf ] and [X make Y Vpp] do not display any significant association, which will be the case for several other parameters in what follows. (166) Do you really think that the Jolly Green Giant puts the lights on at two o’clock in the morning and gets them all to pick the corn? (167) But a concern with the political unrest preceding the Civil War caused the third part, De Cive, to be prematurely “ripened and plucked” before the first two, De Corpore (“On Body”, 1655) and De Homine (“On Man”, 1657).

The distinction between specific and generic causers may be exemplified by the following pair of sentences: (168) Er, my my natural mother er, actually had me adopted when I was a baby (169) The causal explanation must be that old age causes poverty, not that poverty causes people to be old; in a cross-tabulation of age by poverty, it is more natural to examine the proportion of each age group who are poor rather than the proportion of each income category who are old.

Table 24 shows that the constructions that are distinctively associated with generic causers are [X cause Y Vto-inf ] and the three make constructions, and that with [X cause Y Vto-inf ] and [X be made Vto-inf ] generic causers actually represent a majority. It also shows that this parameter is irrelevant for three constructions, namely [X get Y Vprp], [X have Y Vinf ] and [X have Y Vprp]. The next parameter, that of the animacy of the causer, is interesting in that only two constructions depart from the general tendency, as appears from Table 25. While most constructions are associated with animate causers, as in (170), [X cause Y Vto-inf ] and [X make Y Vinf ] are more typically found with an inanimate causer (91% and 52%, respectively), as illustrated by (171) and (172).

111


Animate Inanimate


prp ]

YV

[X

[X

HA VE

inf ]

YV

[X

HA VE

pp ]

HA VE

[X

prp ]

[X

GE

TY V

inf ] to -

TY V

[X

GE

TY V

GE

[X

[X

CA USE

YV

to inf ]

Table 25. Animacy of the causer (distinctiveness and percentages)

–

+

+

+

+

+

+

–

+

8.7

98.6

89.8

99.1

97.2

97.1

99.9

47.7

89.6

+

–

–

–

–

–

–

+

–

91.3

1.4

10.2

0.9

2.8

2.9

0.1

52.3

10.4

80.6 19.4

(170) Methodologically, it is not possible for any researcher to get the kids to talk with much sense of ideas since the question “Why?” to the smashing of milk bottles is one that is not possible for the boy to answer outside the context of the whole Saturday evening ... (171) If the public’s confidence in the safety record of the nuclear power industry faltered following the Three Mile Island accident, the accident at Chernobyl in the Soviet Union caused nuclear reactor safety to be completely re-examined. (172) and we also know that advertising reinforces smoking, it makes, makes people think that smoking is okay.

The association of cause with inanimate causers confirms a claim made by huquet & Paillard (1989: 170) – although, contrary to what a linguist like Inoue C (1992: 134) argues, (agentive) animate causers are possible, as illustrated by (173), where the adverb maliciously makes it clear that the causer is agentive. Similarly, the association of the get and have constructions with animate (or agentive) causers is regularly mentioned in the literature (e.g. Baron 1974: 320, Andersson 1985: 153, Belvin 1993: 64). However, it should be underlined that inanimate causers are not totally impossible with these two verbs, as appears from sentences (174) to (177). (173) By their statement of claim the council claimed, inter alia, (…) (ii) that in those issues of “The Sunday Times” the third and fourth defendants falsely and maliciously wrote and the first and second defendants falsely and maliciously printed and published, or caused to be written, printed or published of and concerning the council and of and concerning the council in the way of its discharge of its responsibility for the investment and control of the superannuation fund


(174) I said that’d do you good to be in business for a couple of years, that’d get you sorted out! (175) A pedagogy which aimed at teaching the functional potential of grammar along the lines I have described, would have to get learners to engage in problem-solving tasks which required a gradual elaboration of grammar to service an increasing precision in the identification of relevant features of context. (176) This is a male view of the matter, in which women are attacked, and which had me admiring these ugly sisters. (177) Both behaviourism and “humanistic” views would have us believe that to be affective is to be effective in setting up conditions for learning.

The significant association of [X make Y Vinf ] with inanimate causers, on the other hand, which is reinforced by the predominance of such causers in terms of frequency, is not only absent from the literature (although some linguists suggest that inanimate causers are not ruled out, see Adamczewski & Delmas [1993: 23] or Salkoff [1999: 91]), but it also flatly contradicts a claim such as Givón’s (1993: 9), which excludes the possibility of having a non-agentive causer with make (see Section 5.1). This finding says something about the general meaning expressed by the construction. The coercive meaning, which is often said to be typical of causative make (e.g. Guierre 1959: 126, Faure & Casanova 1968: 192, Tavernier & Etienne 1973: 40, Werner et al. 1990: 392, Li 1991: 347), is only possible with animate causers (though not necessary: compare [178], which is coercive, with [179], which is not). If most causers in [X make Y Vinf ] are inanimate, as in (180), then coercion cannot be the most common meaning of the construction. This will be confirmed by other results in this chapter. (178) Daddy’s gonna make me eat some tomato and, and cucumber (179) Coleridge’s development after 1798 was roughly parallel to that of Wordsworth, though he was always more concerned to make his opinions square with Christianity. (180) Transactions of this kind must make us pause before we condemn all landlords as Gradgrinds, or make too large assumptions about the nature of medieval serfdom.

Table 26 gives more specific information about the types of animate and inanimate causers that are distinctive for the different constructions. Following Lemmens (1998), causers were classified into six types: “human”, “human-like” (e.g. governments, agencies), “animal”, “organism”, “physical object” and “abstract


Human


prp ]

YV

[X

[X

HA VE

inf ]

YV

[X

HA VE

pp ]

HA VE

[X

prp ]

[X

GE

TY V

to inf ]

[X

GE

TY V

TY V

GE

[X

[X

CA USE

YV

to inf ]

Table 26. Semantic nature of the causer (distinctiveness and percentages)

–

+

+

+

+

+

+

–

8.2

96.2

89.0

98.3

93.1

94.3

98.5

45.5

67.2

67.7

0.5

2.5

0.0

0.9

4.2

2.9

1.3

1.9

1.5

9.7

–

+

+

Human-like

+

Indeterminate human

– 0.0

0.0

Animal

Physical object Abstract object

0.0

0.0

0.0

0.0

0.0

20.9

3.2

+ 0.0

Organism

0.0

0.0

0.8

0.0

0.0

0.0

0.0

0.3

0.0

0.0

1.4

0.0

0.0

0.2

0.0

0.0

0.0

0.1

0.0

0.0

+

–

–

+

9.7

0.3

3.9

0.1

0.0

0.0

0.1

7.9

0.0

0.0

+

–

–

–

–

–

–

+

–

79.2

1.1

6.3

0.5

2.8

2.9

0.0

40.4

7.5

19.4

–

+

1.0

0.0

0.0

0.0

0.0

0.0

0.0

3.9

3.0

0.0

+

Indeterminate object

–

–

–

object”, the last category including states, events and the like. The six categories are illustrated by (181) to (186). In addition, a category called “indeterminate human” and one called “indeterminate object” were created for cases where it was clear that the causer was a human or an object, but its exact nature could not be determined, as in (187) and (188). (181) Kelly’s been and had her hair dyed.

(182) And any school can get a lecturer to come out, subject to availability of course, to give one of these? (183) Male canaries sing at least in part to make females prepare for reproduction. (184) The autoantibodies cause the junctions to break down and the keratinocytes to separate (acantholysis) giving rise to the formation of blisters. (185) he reckoned that, that my blanket made him feel worse


(186) For history reveals, time and again, that while vertical thinking can bring our full intellectual powers to bear upon a problem and thus to consolidate a position, it is chance that causes us to stumble upon it (both the problem and its possible solution) in the first place. (187) Failure to comply with the SAO is a summary offence, as is failure to cause a pupil registered at a school to attend regularly (see section 39(1)). (188) I just, I tell you what, I, I know it sounds nasty, but I am glad that Penny is some way has had something that has made her realise what that place was all about.

As the table shows, most of the constructions that are associated with animate causers (see above) are actually associated with human causers (only [X be made Vto-inf ] is more strongly associated with the indeterminate category of human causers). As for the two constructions that are distinctively associated with inanimate causers, [X cause Y Vto-inf ] and [X make Y Vinf ], they are significantly attracted to both physical and abstract objects, but the latter are much more common than the former in terms of frequency. Finally, note the association of [X make Y Vinf ] with indeterminate objects, which may be accounted for by the use of the interrogative pronoun what as a subject (absolute frequency of thirty occurrences, as against one instance in [X cause Y Vto-inf ] and no instances in the get and have constructions). This use is illustrated by the following sentences: (189) What makes you think you’re not [a mental case]?

(190) He says I bet you’re from Yorkshire I said well what makes you say that?

5.3.4 The causee The second element along the action chain is the causee, which may be studied by means of the same parameters as those used to investigate the causer (form, grammatical person, definiteness, specificity, animacy, semantic nature). While the nominal/pronominal form of the causer is well distributed among the constructions (five constructions were shown to be associated with pronominal causers and four with nominal causers, see Table 21), the form of the causee shows a clear preference for nouns, with seven constructions being associated with nominal causers, against just one, [X make Y Vinf ], with pronominal causers (see Table 27). An example of the latter is given below. (191) So I, I made him swap places as, er, with Bryony and of course he wou-- he wasn’t sitting there!


Pronoun

–

–

17.8

50.7

52.7


prp ]

YV

inf ]

YV

YV

HA VE

[X

+

[X

+

47.3 100.0 57.5

[X

+ 49.3

[X

+ 82.2

HA VE

pp ]

HA VE

prp ]

GE

TY V

inf ]

TY V

to -

GE

TY V

GE

[X

Noun

[X

[X

[X

CA USE

YV

to inf ]

Table 27. Form of the causee (distinctiveness and percentages)

+

+

–

+

54.9

90.9

24.8

70.6

–

–

–

–

+

–

0.0

42.5

45.1

9.1

75.2

29.4

66.7 33.3

Second person Third person


– 7.6

0.0

26.3

–

–

–

0.0

4.9

5.3

+

+

+

98.2

91.9

YV

[X

HA VE

HA VE

YV

pp ]

HA VE

YV

GE T

YV

[X

– 3.2

prp ]

[X

8.5

GE T [X

– 1.8

inf ]

[X

8.2

prp ]

0.0

YV

to inf ]

YV

GE T

[X

First person

[X

[X

CA USE

YV

to inf ]

Table 28. Grammatical person of the causee (distinctiveness and percentages)

– 0.0

12.3

87.0 100.0 79.5

+

– 5.9

+

–

4.2

0.0

18.3

2.9

+

+

–

+

87.3 100.0 55.5

0.0 0.0

91.2 100.0


YV

prp ]

YV

[X

HA VE

inf ]

YV

HA VE [X

–

+

–

66.2

45.5

89.2

61.8

+

+

+

–

+

27.4

33.8

54.5

10.8

38.2

33.3

[X

–

72.6

+ 24.4

HA VE

[X

– 66.7

–

16.7

pp ]

prp ]

[X

GE

TY V

inf ] to -

TY V

75.6

+ 30.7

GE

TY V

GE

83.3

– 69.3

Indefinite

[X

Definite

[X

[X

CA USE

YV

to inf ]

Table 29. Definiteness of the causee (distinctiveness and percentages)

66.7 33.3


Like the form of the causee, the grammatical person of the causee presents strikingly similar tendencies across constructions, with the exception, again, of [X make Y Vinf ], which is distinctively associated with first- and second-person causees, as opposed to the other constructions which, when statistically significant, are associated with third-person causees (see Table 28). This opposition is illustrated by the following sentences: (192) I’m listening to what people are saying tonight, it’s it’s sort of making me feel a bit sick what they’re saying. (193) It makes you think of summer dunnit?

(194) T. Behrens is considerably baffled by this strange case, by the question of what it was that determined Ursula’s adherence to this programme, and of what it was that caused her to bring to an end her loving friendship with Justin.

[X make Y Vinf ] is also the odd one out when we consider the definiteness of the causee (Table 29), since it is the only construction that is distinctively associated with definite causees, e.g. (195). The other constructions, by contrast, are more distinctively associated with indefinite causees, as illustrated by (196) and (197). ( 195) She said oh it makes our house and garden seem so small. (196) The resultant loss of staff confidence caused important members of the team to leave. (197) When the offer first came up we’d had an even lower-budget series go out on a national US cable network but, apart from that critically-lauded effort, no TV experience.

Just like the causer, the causee in a causative construction may be specific or generic. Compare: ( 198) you will not get me to to separate myself from my colleagues. (199) The infant’s eye is elastic and so a raised intra-ocular pressure causes the eyeball to enlarge.

As emerges from Table 30, only two constructions are distinctively associated with generic causees, namely [X cause Y Vto-inf ] and [X be made Vto-inf ], two constructions which, it will be noticed, are also among those that are associated with a generic causer (see Table 24).


Generic

+ 95.4

+

–

–

54.6

11.9

4.6

25.0


prp ]

YV

YV

[X

HA VE

inf ]

YV

+

[X

91.8

[X

75.0

HA VE

pp ]

HA VE

prp ]

GE

[X

[X

+ 88.1

TY V

inf ] to -

TY V

GE

TY V

GE

– 45.4

[X

Specific

[X

[X

CA USE

YV

to inf ]

Table 30. Specificity of the causee (distinctiveness and percentages)

+

–

94.4 100.0 77.4

–

–

8.2

5.6

65.7

66.7

+ 0.0

22.6

34.3

33.3

–

+

36.6

99.5

+

–

+

–

65.2

8.3

63.4

0.5


YV

prp ]

YV

[X

HA VE

inf ]

YV

HA VE

pp ]

HA VE

–

+

–

78.1

55.6

99.7

70.8

+

–

+

+

–

44.4

0.3

29.2

51.5

0.0

21.9

[X

prp ]

TY V

[X

GE

TY V

inf ] to -

[X

GE

TY V

GE

+ 91.7

[X

Inanimate

– 34.8

[X

Animate

[X

[X

CA USE

YV

to inf ]

Table 31. Animacy of the causee (distinctiveness and percentages)

–

+

48.5 100.0

The animacy of the causee (see Table 31) also bears some similarity to the results for the animacy of the causer (Table 25), in that [X cause Y Vto-inf ] and [X make Y Vinf ], which are both associated with inanimate causers, are also distinctively characterised by inanimate causees, as exemplified by (200) and (201). Note that, in the case of [X make Y Vinf ], this is additional evidence that the main meaning of the construction cannot be coercion. (200) This causes the double stranded DNA in the sample to dissociate into two single strands as the hydrogen bonds, which hold the two strands together under physiological conditions, break down reversibly on heating. (201) But it becomes clear that Krashen is not thinking of theory in general , that is to say of a theoretical perspective on pedagogy, but of a theory in particular which can be applied directly; not, therefore, of the process of referring actual


problems to abstract ideas but the process of making practice conform to a preconceived conceptual pattern.

Three other constructions present the same feature of being associated with inanimate causees, viz. [X be made Vto-inf ], as well as the present participle constructions with get and have, illustrated in (202) to (204). (202) In The Waste Land the idea of a fertility cult in City or suburb is made to seem not irrelevant, but horribly absurd as in the scenes overheard in the pub or observed by Tiresias. (203) Couldn’t get these earphones working.

(204) Yes because y-- if you’ve used all your hot water <pause> you can’t have that boiler going <pause> for an hour or two can you?

The case of the present participle constructions is particularly interesting, as these constructions depart from the generally interpersonal interactions expressed by the other get and have constructions. It is also noteworthy that most of these inanimate causees refer to objects with some energy of their own, e.g. earphones in (203) and boiler in (204); other examples include computers, cars, tape recorders and dishwashers. This confirms the point made for causative get by linguists such as Goldsmith (1984: 123) or Wierzbicka (1998: 124), who allude to “objects perceived as being able to move (or effect their own kind of activity) through some internal energy-source” (Goldsmith 1984: 123), with the additional caveats that this phenomenon appears to be mainly restricted to the present participle construction with get (rather than being typical of get in general) and that it is also characteristic of the present participle construction with have. Now that we have the results for both the animacy of the causer (Section 5.3.3) and the animacy of the causee (see above), we can study the ten constructions using Talmy’s (1976) four-way classification of causative events.46 Inducive causation means that an animate causer acts on an animate causee; volitional causation corresponds to an animate causer acting on an inanimate causee; in affective causation an inanimate causer acts on an animate causee; and in physical causation an inanimate causer acts on an inanimate causee. Table 32 indicates that inducive causation, as illustrated by (205) and (206), is the most frequent type of causation in terms of frequency. It accounts for over 99% of the past participle constructions with get and have, and almost 91% of the occurrences of [X get Y Vto-inf ]. It also represents the most common type of causation with [X make Y Vpp], [X have Y Vinf ], [X be made Vto-inf ] and [X have Y Vprp]. 46. The interpretation of Talmy’s model will be Hollmann’s (2005), itself based on Croft (1991) and Verhagen & Kemmer (1997).


Volitional Affective Physical Other


prp ]

YV

[X

HA VE

inf ]

YV

–

+

–

75.0

51.4

99.5

33.1

[X

[X

HA VE

pp ]

HA VE

YV

prp ]

GE T

YV

[X

to inf ]

GE T

GE T

YV

[X

[X

Inducive

[X

[X

CA USE

YV

to inf ]

Table 32. Types of causation (distinctiveness and percentages)

–

+

–

+

7.3

90.7

30.7

99.1

+

–

+

+

–

+

+

1.6

8.0

59.1

0.0

22.2

45.7

0.3

14.4

33.3

+

–

–

–

–

–

–

+

–

28.3

1.1

5.5

0.4

2.8

2.9

0.2

38.5

1.5

13.8

+

–

–

–

–

–

+

62.8

0.3

0.5

0.0

0.0

0.0

14.0

7.6

0.0

+

+

–

+

+

11.7

12.0

0.1

9.5

22.3

–

4.7

– 4.5

0.1

5.2

8.2

+ 57.6

86.2 0.0

0.0

(205) And erm <pause> wha-- what happened was we had Martin er assessed by an independent tutor, a guy that teaches ( 206) Persevere in getting the child to comply without losing their temper.

However, it is volitional causation, e.g. (207) and (208), that is distinctive for the highest number of constructions, namely [X get Y Vprp], [X have Y Vprp], [X be made Vto-inf ], [X have Y Vinf ] and [X make Y Vinf ]. ( 207) But you’ve managed to get so many exciting things going. (208) I’m gonna make this work, I’m not rushing into it

The other two types of causation, by contrast, affective (209) and physical (210), are only distinctive for two constructions – which happen to be the same – viz. [X cause Y Vto-inf ] and [X make Y Vinf ]. (209) The present state of the hospice movement and continuing research on pain relief make us ask, “is optimum pain alleviation being sought?” (210) Productivity gains expected may be seriously constrained because of ingrained attitudes which cause human behaviour to change slower than the potential rendered by technology or even go in a direction directly opposed to the requirements of the system.


Interestingly, these four types of causation arguably correspond to varying degrees of directness (see Verhagen & Kemmer 1997: 71). Inanimate entities are believed to act directly on other things, hence the directness of physical and affective causation. Animate entities, by contrast, “can only act on animates via the intervening physical world, i.e. (…) one cannot reach into another person’s mind and directly cause him or her to do, feel, or think something” (ibid., emphasis original). This has as a consequence that inducive causation is the most indirect type of causation. Volitional causation, which refers to the interaction between an animate causer and an inanimate causee, is more direct than inducive causation, but less so than physical and affective causation. According to this interpretation of Talmy’s model, [X cause Y Vto-inf ] should be considered as the most direct construction, followed by [X make Y Vinf ] (for which volitional causation is also distinctive). The most indirect constructions would be [X have Y Vpp], [X get Y Vpp], [X get Y Vto-inf ] and [X make Y Vpp]. The other constructions may be said to occupy an intermediate position along the direct-indirect cline. Coming back to the nature of the causee alone, Table 33 specifies its semantic category. It reveals, among other things, that the get and have present participle constructions are associated with physical objects ([X get Y Vprp] is also associated with abstract objects, but the frequency is lower), and [X cause Y Vto-inf ] with all types of objects, including indeterminate objects. The presence of indeterminate objects with [X cause Y Vto-inf ] mainly results from the passivisation of the effect, with which it is not quite clear what the (implicit) causee refers to, save that it is an object, e.g. (211) A better method is to provide a special supervisor call instruction which causes a distinguishable interrupt into the supervisor to be generated. (212) The positive feedback holds the output at while capacitor C charges through resistor R until the potential at the inverting input exceeds that, at the noninverting input when the positive feedback causes the output to be driven into negative saturation, .

Another interesting feature, which did not emerge from Table 31, is the typical association of [X have Y Vinf ] with human-like causees. In this case, human-like entities mainly refer to imaginary people and fictional characters, as illustrated by the following sentence: ( 213) When Jane has Roche inspect the hut, the wild man, with his black face and his pigtails, has gone, leaving behind him “only a vague warm smell of old clothes, dead animals, grease and marijuana”.


Human


prp ]

[X

+

–

–

+

–

–

+

99.3

61.6

48.6

98.9

67.3

35.9

93.8

–

+

+

–

4.0

5.9

1.5

0.2

16.4

6.9

0.6

1.9

6.8

0.0

0.2

0.0

0.0

0.0

–

+

+

+

1.6

5.8

6.3

1.0

0.0 0.0

+

+

+ 0.0

0.0

+

0.0

0.0

0.0

–

2.5

1.3

0.8

0.0

0.0

0.0

0.0 –

+

1.5

0.5

0.0

0.2

0.0

0.0

0.0

1.1

+

–

+

–

+

–

+

24.7

6.5

40.5

0.0

38.9

0.3

14.4

13.6

+

–

+

–

–

+

+

28.8

1.3

21.4

0.2

0.0

13.6

36.9

0.0

–

–

0.0

0.2

0.0

0.0

Organism


YV

– 34.4

0.0

Abstract object

YV

+ 84.4

Indeterminate human

Physical object

[X

– 28.3

Human-like

Animal

HA VE

inf ]

YV

[X

HA VE

pp ]

HA VE

[X

prp ]

[X

GE

TY V

to inf ]

[X

GE

TY V

TY V

GE

[X

[X

CA USE

YV

to inf ]

Table 33. Semantic nature of the causee (distinctiveness and percentages)

+ 10.1

11.0 11.0

5.6

– 0.0

1.5

0.0

0.0

0.0

Such a construction may be explained by referring to the notion of mental space, that is a cognitive structure which is built up during discourse and to which situations are assigned appropriately (see Fauconnier 1985). The causer and the causee normally belong to the same mental space, and in most of the causative constructions examined so far, this was the reality space. In (213) the participants also belong to the same mental space, but this time a fictional one, since Jane and Roche refer to fictional characters.47 This type of phenomenon has been highlighted for have by Brugman (1996), who calls such uses “depictive” (see also Kirchner 1952: 387, Lukmani 1979: 265, Ikegami 1990a: 185). As Brugman aptly notes, the two entities may also belong to different mental spaces. In (214) and (215) the causer is part of a real-world mental space, while the causee belongs to a fictional mental space. Such sentences, Brugman argues, have the same syntactic and semantic properties as basic causative constructions, except that they involve 47. Note, incidentally, that human-like causers are also found in [X have Y Vinf ] (with a proportion of 4.2%), though they are not distinctively associated with the construction (see Table 26).


the building of a daughter space, triggered by a “space builder” (the detective genre and Pynchon in [214] and the notebooks and Dostoevsky in [215]). Because the causer acts on a causee located in a different space, constructions like these will be called “cross-space constructions” (see Lee [2001: 98] for another example of this type of phenomenon). (214) Drawing once again on the detective genre, Pynchon complicates the linear hunt for information, partly by rendering every detail as ambiguous as possible and partly by having Oedipa literally go round in a huge circle when she is pursuing an “underground” mail courier. (215) We recall that in the notebooks Dostoevsky has Raskolnikov reflect upon his crime and declare he had to commit it to achieve moral development and get himself out of the mess he was in.

Although this is rarely acknowledged in the literature (see Kirchner [1952: 457] for an exception), such constructions are also possible with [X make Y Vinf ] – with a slightly higher frequency, but a much lower proportion. An example is given in (216), which contains the space builder the author. (216) In this way the author (somewhat naively) avoids the anti-climax of making Roland fall victim to a Saracen sword.

5.3.5 The effect The effect of causative constructions may be considered from the point of view of transitivity, dynamicity, functional category (material, mental or relational) and volitionality. Table 34 gives the results for transitivity, with a distinction between transitive, intransitive, ditransitive, copular and other effects. Leaving the “other” category aside, we notice that, naturally, transitive effects are distinctive for the three past participle constructions. Intransitive effects, on the other hand, are distinctive for all the other constructions (except [X be made Vto-inf ], with which the feature is not statistically significant). Note, in particular, their association with [X get Y Vprp] and [X have Y Vprp], which actually present a large majority of intransitive effects (83% and 63%, respectively). Ditransitive effects are only distinctive for one construction, [X get Y Vto-inf ], and copular effects for two, [X make Y Vinf ] and [X be made Vto-inf ]. Some of these typical associations are shown below. ( 217) But they said like <pause> they will get his room done.


Intransitive

+ 99.6


prp ]

YV

[X

HA VE

inf ]

YV

HA VE

pp ]

–

+

–

–

+

54.5

26.0

99.9

24.2

36.9

96.8

[X

[X

– 9.7

HA VE

YV

prp ]

GE T

YV

[X

to inf ]

[X

GE T

YV

– 56.1

+

+

+

–

+

+

–

+

37.4

27.1

83.6

0.0

33.8

63.0

0.0

39.4

25.2

0.0

0.5

4.0

0.7

0.2

0.0

1.4

0.0

0.4

0.0

3.2

–

–

–

–

–

+

+

6.3

0.8

1.5

0.0

6.5

1.4

0.0

26.5

15.5

+

+

–

+

+

11.7

12.0

0.1

9.5

22.3

Ditransitive

+

Copular Other

GE T

– 44.1

[X

Transitive

[X

[X

CA USE

YV

to inf ]

Table 34. Transitivity of the effect (distinctiveness and percentages)

–

–

– 4.5

0.1

5.2

8.2

0.0 0.0

( 218) He had him playing yesterday, er, but they dropped Hughes. (219) so I shall definitely get him to give us a lift

(220) And in one respect it’s not quite erm as straightforward as I’ve made it appear, and perhaps after all I’m not wrong in making the claim I did (221) Psychologists are often made to feel the poor relations of biologists, because the latter have more facts at their disposal, and to put a lower value on psychological than on biological data and theories.

The effect of a causative construction may be dynamic or stative, an opposition illustrated by the following pair of sentences: (222) It is important for parents not to supplement the diet with high sugar and salt content snacks in an effort to get their child to eat. (223) The fact that the teacher finds that the child has spelt a word correctly in one context and then perhaps in a sentence or two later may misspell the word, makes the teacher think that this is a question of carelessness, where we know, in fact, that is a feature of dyslexia – that you may be able to get a spelling correct in one context and yet you will misspell it, as it were, a few moments later.


–

1.5

1.0

prp ]

YV

YV

HA VE

inf ]

YV

HA VE

pp ]

HA VE


–

4.5

[X

– 11.3

Stative

+

+

–

–

–

81.8

98.6

99.4

64.8

65.0

9.7

–

–

+

+

+

18.2

1.4

0.6

35.2

35.0

90.3

[X

99.0

[X

+

98.5

[X

+

95.5

[X

+ 88.7

Dynamic

[X

prp ]

GE

TY V

inf ]

TY V

to -

GE

TY V

[X

GE

[X

CA USE

YV

to inf ]

Table 35. Dynamicity of the effect (distinctiveness and percentages)

Table 35 shows that the only causative that is distinctively associated with stative effects is make, in all of its three patterns (see [223] above for an example of [X make Y Vinf ]). Particularly striking is the case of [X make Y Vpp], with which stative effects represent a proportion of 90%. This is illustrated in (224) and (225). (224) I’d have thought it would have been far in the interests of the people of Banbury and the children and parents of Banbury if they’d taken full advantage of the discussion on the tertiary college and had made their opinions known. (225) The group who face the most difficulty getting adequately trained are women doctors, and they made their voices heard at the conference despite the organisers admitting with startling candour that they had not anticipated that this issue would arise.

Causative constructions with cause, get and have are all characterised by a dynamic effect (although this is not statistically significant with [X cause Y Vto-inf ] and [X have Y Vinf ]). Stative effects do occur with these constructions, but not frequently. However, that such a combination is possible at all may seem surprising in the case of get and have, since it is sometimes claimed that these two causatives are not used with states (e.g. Baron [1974: 320] or Gillardeaux [1992: 100]; see also Andersson [1985: 155] on present participle constructions). This claim is challenged by corpus examples such as the following: ( 226) Well I’ll try, I’ll try and get Bob to remember the name of it (227) Yes, you’ve got it looking like the yard.


(228) “We are not for having any man turn sceptic, and disbelieve his senses; on the contrary we give them all the stress and assurance imaginable; nor are there any principles more opposite to scepticism, than those we have laid down.”

effects may be divided into material, mental and relational processes, which are the three main types of processes recognised by Halliday (2004: 179ff.). Material processes are processes of doing, involving an agent that does something, possibly to some other entity. Mental processes, or processes of sensing, include clauses of feeling, thinking and perceiving. As for relational processes, also called processes of being, they “serve to characterize and to identify” (ibid. 210). Table 36 indicates that material effects are distinctive for the largest number of constructions, namely the three get constructions and [X have Y Vpp], and that they represent a majority in terms of frequency in all constructions except [X make Y Vpp]. Some examples are given below: (229) My only regret on that particular occasion now is that I didn’t organise properly getting it videotaped, because it would have been a nice thing to keep, but as far as I could I kept to the traditions of mayoral ceremonial on those sorts of occasions. ( 230) If I get him to drive me round now I’ll remember which way to go. (231) I’ve already had it stretched once, it might break.

Mental effects come second and characterise [X have Y Vinf ], [X make Y Vinf ], as well as [X make Y Vpp] with which the percentage reaches a staggering 90%. This is illustrated by the following two sentences: (232) Now, obviously, you know, yo-- getting the menopause out into the open so that <pause> everyone can talk about it, exchange information that that <pause> it’s not seen as a as a taboo or something to be particularly fearful of by men or women <pause> isn’t going to be much good if it just makes everybody worry <pause> for er, for for the first half of o-- o-- o-- of their lives and, and then gibber through the second half! (233) Kirillov replies: “There is nothing secret that will not be made known. He said so.”

Relational effects only occur distinctively with two make constructions, viz. [X make Y Vinf ] and [X be made Vto-inf ], e.g. (234) and (235). (234) Right, you’ve got to make it sound really interesting, let me switch this thing off cos they don’t want to hear you read again I don’t suppose


–

–

+

10.8

7.2

1.5

1.1

–

–

–

13.1

4.5

6.0

0.5

prp ]

YV

YV

HA VE

inf ]

YV

[X

Relational

[X

Mental


68.8

[X

–

98.4

[X

+

92.5

[X

+

88.3

[X

+ 76.1

Material

HA VE

pp ]

HA VE

YV

prp ]

GE T

to inf ]

YV

GE T

YV

[X

GE T

[X

CA USE

YV

to inf ]

Table 36. Functional category of the effect (distinctiveness and percentages)

+

–

–

–

86.3

97.8

51.0

54.4

9.7

–

+

23.4

4.1

0.7

17.6

9.7

90.3

–

+

+

–

7.8

9.6

1.5

31.5

35.9

0.0

+

(235) The idea of a particular line can be “made to represent or stand for all other particular lines of the same sort”.

Finally, effects may be divided into volitional and non-volitional verbs. A volitional effect involves the causee’s will. It is, in other words, an effect that is done “on purpose”, as in (236). A non-volitional effect, on the other hand, happens independently of the causee’s will, as in (237). (236) She could have her Mum (…) call her at home <pause> to make sure, if anything had happened (237) Does it [glasses] make me look clever?

It is important to bear in mind that a verb is not intrinsically volitional or nonvolitional. Its volitional or non-volitional character also depends on the context. Describe, for instance, is volitional in (238), but non-volitional in (239), where the causee (W) is inanimate. Similarly, the verb jump is usually volitional, since jumping is normally something that one does on purpose, but it is non-volitional in the case of (240).48 (238) Again it is this incomprehension of social mobility which seems to have caused my ACC [Army Catering Corps] to describe an officer who took a degree and then left to become a solicitor as “a bad experience for the force” (and not a gain for society). 48. A comparison of the three sentences also underlines the importance of the type of causee in determining the volitionality of the effect (see also Jackendoff 1990: 129). More precisely, it turns out that inanimate causees can only occur with non-volitional effects, whereas animate causees can occur with both volitional and non-volitional effects.


–

+

34.3

98.8

+

–

+

–

78.0

11.4

65.7

1.2


prp ]

YV

YV

[X

HA VE

inf ]

YV

HA VE

pp ]

HA VE

–

+

–

–

–

57.1

49.3

99.3

26.1

34.0

9.7

+

–

+

+

+

50.7

0.7

73.9

66.0

90.3

42.9

[X

prp ]

[X

GE

TY V

inf ] to -

TY V

[X

GE

TY V

GE

+ 88.6

[X

Non-volitional

– 22.0

[X

Volitional

[X

[X

CA USE

YV

to inf ]

Table 37. Volitionality of the effect (distinctiveness and percentages)

(239) If a string is attached to load W which is free to slide without interference, then movement of the string in the direction of the arrow will cause W to describe a tractrix. ( 240) But er <pause> gosh, you , you made me jump you!

Table 37 reveals divergent tendencies in terms of the volitionality of the effect. While [X get Y Vto-inf ], [X get Y Vpp] and [X have Y Vpp] are distinctively associated with volitional effects (also in terms of frequency), this is not the case of [X cause Y Vto-inf ], [X get Y Vprp], [X have Y Vprp] and the three make constructions, which are all associated with non-volitional effects (and are predominantly found with such effects, [X make Y Vpp] coming first with a proportion of 90%). [X have Y Vinf ] presents no distinctive preference for either type of effect, but the high proportion of non-volitional effects (43%) in comparison with its get counterpart (11%) is worth underlining. These findings contradict a number of claims found in the literature, for example Babcock’s (1972: 32) and Mittwoch’s (1990: 113) assertion that non-volitional effects are impossible in [X be made Vto-inf ] or Harley’s (1998) claim that the effect in [X have Y Vinf ] should be volitional. In fact, non-volitional effects account for 66% of the data in [X be made Vto-inf ], e.g. (241), and, as already pointed out, they represent a relatively high percentage in [X have Y Vinf ] too, e.g. (242). (241) Mm, I can see them now the outside, they were like made, made to look like wood weren’t they the caravans? (242) I used to do that I’ll have you know.

More generally, the finding that a majority of the causative constructions (six out of ten) are preferably constructed with a non-volitional effect forces us to recognise


that, somewhat counter-intuitively, one usually causes somebody to do something that it not dependent on their will – which necessarily rules out a coercive reading. This, importantly, is also true of make, the causative which, of all causatives, is most often said to express coercion (see Section 5.3.3). In fact, three conditions are necessary to allow the expression of coercion (in the sense of “forc[ing] a person to do something by an exercise of influence, authority, or violence”, Terasawa 1985: 135), namely an animate causer, an animate causee and a volitional effect, as illustrated in (243) to (246) for each of the four causative verbs. If at least one of these conditions is not fulfilled, coercion cannot be expressed.49 (243) This may be possible through community representation and has been evident where neighbourhood groups have challenged the status quo and caused government to reappraise the practice of service delivery. (244) At least by the eleventh century every king expected to recruit a part of his army by paying mercenaries, or from knights who received a fee not in land, but in cash; though he did his best to make his great nobles provide contingents for which he did not have to pay, or (at least in the twelfth century) pay him in cash if they did not serve him in person. (245) At the sermon, Father McKenna introduced one of his altar boys, Colm by name, and got him to tell the story about how he and his altar boy teammates had recently beaten another altar boys’ team by eight goals to five. (246) In my own research into books and reading I have had classes of 15-year-olds write essays on the subject of how they would feel about working in a bookshop.

Table 38 presents the results for the different possible combinations of volitional/non-volitional effects and animate/inanimate causers and causees. The first line corresponds to the conditions that are necessary for the expression of interpersonal manipulation (see Givón 1975, Givón & Young 2001, Shibatani 2001) – and hence coercion. It varies from 3% with [X cause Y Vto-inf ] to 99% with [X have Y Vpp], and it is statistically distinctive for three constructions, viz. [X get Y Vto-inf ] and the past participle constructions with get and have.50 With 49. Note, however, that these conditions are not sufficient to express coercion. As pointed out by Shibatani (1976b: 251), in such contexts “there is a wide range of activity that the causer can perform in effecting the caused event: He may force the causee, he may persuade the causee, or he may gently suggest that the causee perform an act that constitutes the caused event”. 50. Note that, in the case of the past participle constructions with get and have, the (animate) causee is most of the time implied.


[X make Y Vinf ], it represents a percentage of 16.4% and is the only combination that is not distinctive for the construction. The examples below illustrate the most frequent combination for each construction (which, in most cases, is also significantly distinctive for the construction). (247) “Writing, in short, does not “reproduce” a reality beyond itself, nor does it “reduce” that reality. In its new freedom, it can be seen to cause a new reality to come into being” (Hawkes’s italics). (248) he kept going on and on and on and he said do you think it was an error of judgment and he was trying to get Kinnock to say yes and he said yes it was an error of judgment the way he, he’d sort of said I think he must of said in the mid seventies, that if he got in power he’d get rid of all nuclear weapons (249) Yeah, I’ve just got it [fire] going granddad

(250) she said you know well it’s <pause> that’s how you’ve gotta take it stripped, for me to take that to get that <pause> done up by a restorer <pause> you’d have to take all the upholstery off, or quite a bit off

Inan. CAUSER / anim. CAUSEE / volit. EFFECT

Anim. CAUSER / inan. CAUSEE / n-vol. EFFECT

Inan. CAUSER / inan. CAUSEE / n-vol. EFFECT

[X

[X

[X

BE ma

de V to inf ] MA KE YV pp ]

inf ]

pp ]

MA KE YV

prp ]

YV

YV

[X

HA VE

inf ]

6.9

–

+

0.2

1.4

1.4

0.2

7.9

0.0

3.4

–

+

–

+

0.6

19.4

0.5

16.6

10.6

79.3

–

– 4.2

YV

–

47.0

4.7

– 1.1

HA VE

–

16.4

+ 98.5

+

pp ]

–

99.1

– 27.6

18.3

HA VE

prp ]

[X

GE

TY V

to inf ]

TY V

GE

GE

TY V

[X

+

45.7

+ 86.8

3.1 +

–

+

+

–

+

+

8.0

59.1

0.0

22.2

45.7

0.3

14.4

33.3

–

–

–

–

–

–

+

–

9.9

0.0

0.8

0.1

1.4

1.4

0.0

30.6

1.5

10.3

+

–

–

–

–

–

+

62.8

0.3

0.5

0.0

0.0

0.0

14.1

7.6

0.0

– 1.6

4.7

5.7

+

3.8

Inan. CAUSER / anim. CAUSEE / n-vol. EFFECT

– 55.6

– 3.1

Anim. CAUSER / anim. CAUSEE / n-vol. EFFECT

[X

CAUSEE / volit. EFFECT

[X

Anim. CAUSER / anim.

[X

[X

CA USE

YV

to inf ]

Table 38. Volitionality of the effect and animacy of the causer and causee (distinctiveness and percentages)51

0.0

51. Anim. = animate, inan. = inanimate, volit. = volitional, n-vol. = non-volitional.


(251) Yeah, they had them come and fit their kitchen and they really did pay thousands of pounds for it, and erm, there was a chip erm, one of the units on the top and they, they just fitted it like it, and my, their dad called them back and said, look you know, I’m paying thousands of pounds for this. ( 252) I’ve not had a blow lamp going this morning at all. (253) I’m having that delivered but the little items I bought myself. (254) Er I, I was going in the evening you know, doing the tailoring class but of course my <pause> illnesses have stopped me doing all of that and made me realize I can’t do it all. (255) “We [young doctors] are made to do things that are a waste of time in order to get accredited,” said one. (256) If you make it known that there’s stuff like that going on, if you get a band like that you can always just [put a bus on for some people that come down.]

5.3.6 The patient Very little will be said about the patient, as it appears to be the least interesting element of the causative construction from a statistical point of view, i.e. the one that discriminates least between the different constructions.52 Several of the parameters examined for the patient turned out to be among the least statistically significant (p-value above 0 using the chi-square calculator) and were therefore disregarded. The remaining parameters are shown in Tables 39 to 42. These tables reveal, among other things, that nominal patients are only distinctive for cause constructions, and clausal patients for [X make Y Vinf ] and [X be made Vto-inf ], as illustrated by (257) to (259), that [X get Y Vpp] and [X have Y Vpp] are characterised by definite and specific patients, as in (260) and (261), and that patients tend to be (physical or abstract) objects rather than human or human-like entities, e.g. (262) and (263). (257) The resulting competition probably causes the animals to occupy small but adequate territories which are vigorously defended by a monogamous pair. 52. One possible explanation for this is that causative constructions including a patient may be said to combine the frame of causation and the frame of transitive action. The characteristics of the patient in causative constructions, therefore, are arguably those of patients in transitive sentences in general, and hence show relatively little variation across constructions.


Noun

[X

[X

[X


prp ]

YV

YV

HA VE

inf ]

YV

HA VE

HA VE

YV

[X

– 48.7

28.6

46.2

64.3

–

43.4

61.9

+

–

56.3

26.2

– 5.0

pp ]

prp ]

GE T [X

[X

GE T

YV

to inf ]

YV

GE T

+ 89.0

Pronoun

[X

[X

CA USE

YV

to inf ]

Table 39. Form of the patient (distinctiveness and percentages)

Clause

60.0 35.0

50.7

43.6

64.9

+

–

–

49.3

21.4

18.9

–

+

+

0.0

35.0

16.2

– 6.0

5.1

7.1

0.2

11.9

5.0

55.9 44.1 0.0

23.9

28.6

–

+

18.9

40.5

[X

83.7


YV

prp ]

YV

[X

– 59.5

HA VE

inf ]

YV

[X

+

HA VE

pp ]

HA VE

70.0

[X

GE [X

81.1

+ 52.0

TY V

prp ]

TY V

inf ] to -

GE

71.4

– 48.0

Indefinite

TY V

[X

Definite

GE

76.1

[X

[X

CA USE

YV

to inf ]

Table 40. Definiteness of the patient (distinctiveness and percentages)

+

– 77.1

– 30.0

40.5

91.2

+

16.3

22.9

59.5

8.8

Specific

91.9 100.0 97.9

90.5

95.0

– 8.1

0.0

2.1

9.5

5.0


YV

prp ]

YV

[X

[X

HA VE

inf ]

YV

HA VE [X

HA VE

pp ]

[X

TY V

GE [X

+

+ 55.0

prp ]

inf ] to -

TY V

[X

GE

TY V

GE

– 45.0

Generic

[X

[X

CA USE

YV

to inf ]

Table 41. Specificity of the patient (distinctiveness and percentages)

+

–

93.6

82.1

–

+

6.4

17.9

– 86.5

73.5

13.5

26.5

+


+

+

prp ]

YV


–

–

–

5.0

3.4

4.6

0.0

8.8

+

Human-like

[X

11.9

[X

11.8

[X

14.3

+

HA VE

inf ]

YV

HA VE

pp ]

HA VE

prp ]

GE

TY V

to inf ]

TY V

GE

[X

14.5

[X

6.0

Human

[X

TY V

[X

GE

[X

CA USE

YV

to inf ]

Table 42. Semantic nature of the patient (distinctiveness and percentages)

+

0.0

0.4

0.0

0.0

2.4

5.0

0.1

0.7

0.0

2.9

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.9

0.0

0.0

1.3

0.4

0.0

2.9

1.0

0.0

0.0

1.0

0.0

0.0

0.3

1.1

0.0

2.9

+

–

–

–

28.6

35.0

71.4

18.9

8.1

2.9

Indeterminate human Animal Organism Physical object Abstract object

–

–

25.0

39.7

28.6

48.7

– –

+

–

+

+

+

68.0

44.4

57.1

36.4

57.1

55.0

22.8

73.9

91.9

79.4

0.0

0.9

0.0

1.3

0.0

0.0

0.6

0.4

0.0

0.0

+


+

(258) I think no mine are too thin, or maybe their not, and, and this, and seeing this in magazines, seeing it on television it makes me think that maybe I could have that (259) At the same time, the reader can be made to feel that, on closer inspection, the country’s politics might prove to be antics too. (260) I mean we have that on two in the morning to get this room warmed up don’t we have to put it down to one (261) Will you have your bridesmaids’ dresses made?

( 262) Mm they’ve had a special, they had a skylight blind made. (263) Calculating any extra resource using an index based on doctors’ own perceptions of what creates work seems the more valid approach as these will presumably be factors deterring doctors from working in underprivileged areas or causing them to limit access so as to avoid overwork.


5.3.7 Interim summary: Distinctiveness and frequency The tables in Appendices 2 and 3 summarise the distinctive and most frequent syntactic and semantic features of causative constructions. One finding that emerges from these results is that some constructions are more flexible than others, in the sense that they are associated with many different features. In particular, [X make Y Vinf ] turns out to be the most flexible construction, as it attracts the largest number of distinctive features (38 features).53 Next comes [X cause Y Vto-inf ] and its 33 features. The other constructions all have fewer than 30 distinctive features attracted to them. This difference in flexibility may be illustrated by means of the types of causation as defined by Talmy (1976) (see Section 5.3.4): while [X make Y Vinf ] is associated with three types of causation (volitional, affective and physical) and [X cause Y Vto-inf ] with two (affective and physical), the other constructions are associated with just one type. The frequency results (Appendix 3) confirm the versatile character of [X make Y Vinf ], since it is the construction that has the largest number of most frequent features with a proportion of under 50%, which means that, even though the features represent a majority, they have strong competitors in the other features. There is only one case where the feature is very clearly predominant, displaying a proportion of 95% or over, and this is for the voice of the causative verb, which can only be active in this structure. By contrast, a construction like [X get Y Vpp] has sixteen features with a proportion of 95% or over, which suggests that it tends to always occur with the same specific features and hence shows much less flexibility. On the basis of the tables in Appendices 2 and 3, we can also calculate the number of features shared by the different constructions. First, note that there is only one feature that is common to all constructions, namely the predominance of third-person causees (though with different proportions). In all the other cases, there is always at least one construction that differs from the others. This confirms a point made repeatedly in what precedes, namely that the ten causative constructions under study are not as similar to each other as the literature would have us believe. If we then consider the constructions two by two, we notice that their degrees of similarity vary a great deal. This appears from the two tables below. Table 43 shows the number of distinctive features that are shared by two constructions, i.e. cases where the two constructions either attract the same feature (+/+) or repel the same feature (–/–). The cells in grey indicate the number of distinctive features that are in opposition, i.e. are attracted to one construction and repelled 53. This seems to support Dixon’s (1991: 194) claim that make is the causative that has the widest meaning.


[X CAUSE Y Vto -inf] 11

[X GET Y Vprp]

18

12

YV

40

[X

HA VE [X

HA VE [X

18


prp ]

inf ]

YV

pp ]

HA VE

YV

45

[X

prp ]

11 28

GE T

YV

17

[X

to inf ]

[X

GE T

YV

GE T

33

[X GET Y Vto -inf] [X GET Y Vpp]

[X

[X

CA USE

YV

to inf ]

Table 43. Number of distinctive features in common and in opposition (grey shading)

11

8

8

5

3

9

12

47

26

19

14

2

0

13

26

15

16

13

15

5

57

36

24

0

14

13

3

2

13

24

10

13

57

36

21

16

10

6

48

23

[X HAVE Y Vinf]

11

17

14

12

[X HAVE Y Vprp]

16

26

33

20

20

[X HAVE Y Vpp]

10

41

26

57

13

22

[X MAKE Y Vinf]

42

7

15

5

10

12

6

[X BE made Vto -inf]

34

15

20

8

19

21

11

30

[X MAKE Y Vpp]

16

6

4

7

6

4

8

21

5 17

by the other (+/– or –/+). Table 44 gives the number of shared features in terms of frequency (rather than distinctiveness): given a particular parameter, can we say that two constructions predominantly exhibit the same feature (e.g. a majority of animate causers in both cases)? The number of shared distinctive features (Table 43) points to [X get Y Vpp] and [X have Y Vpp] as the most similar constructions. They have as many as 57 distinctive features in common, out of a grand total of 93 features. This pair is followed by [X get Y Vto-inf ] - [X get Y Vpp], with 48 distinctive features in common, [X cause Y Vto-inf ] - [X make Y Vinf ], with 42 features, and [X get Y Vto-inf ][X have Y Vpp], with 41 features. The constructions with the fewest distinctive features in common combine a get or have construction with a make construction (e.g. [X have Y Vprp] - [X make Y Vpp] and [X get Y Vprp] - [X make Y Vpp], with four shared features each) or the cause construction with a get or have construction (e.g. [X cause Y Vto-inf ] - [X get Y Vpp], with six features in common, and [X cause Y Vto-inf ] - [X have Y Vpp], with ten features in common). The figures for the distinctive features in opposition largely confirm these results. [X get Y Vpp] and [X have Y Vpp], which share the highest number of distinctive features, only present five features whose distinctiveness values are opposed to each other. Pairs of constructions combining get/have with cause or make, on the other hand, tend to have many distinctive features that are in opposition (up to 57 for [X make Y Vinf ] - [X get Y Vpp] and for [X make Y Vinf ] -



prp ]

YV

[X

HA VE [X

inf ]

[X

HA VE

YV

pp ]

HA VE

YV

[X

prp ]

GE T [X

to inf ]

YV

[X

GE T

YV

GE T

[X

[X

CA USE

YV

to inf ]

Table 44. Number of most frequent features in common

[X CAUSE Y Vto -inf] [X GET Y Vto -inf]

9

[X GET Y Vprp]

8

[X GET Y Vpp]

8

22

16

[X HAVE Y Vinf]

11

23

15

22

[X HAVE Y Vprp]

11

20

19

19

[X HAVE Y Vpp]

9

21

13

22

23

20

[X MAKE Y Vinf]

12

18

17

15

18

20

17

22 16

[X BE made Vto -inf]

15

14

12

13

16

16

14

12

[X MAKE Y Vpp]

11

18

13

17

20

19

18

16

13

[X have Y Vpp]). What the grey-shaded part of Table 43 also brings to light is the great similarity between [X have Y Vprp] and [X get Y Vprp], as well as between [X have Y Vprp] and [X have Y Vinf ]. Neither of these pairs presents any cases where a feature is attracted to one construction but repelled by the other. From the point of view of frequency (Table 44), we see that, again, it is a gethave pair that comes first, this time [X get Y Vto-inf ] - [X have Y Vinf ], with 23 features in common. In fact, there are only two features that are not shared by the two constructions in terms of frequency: the grammatical person of the causer (first-person causers represent a majority in the get construction, whereas thirdperson causers are predominant in the have construction) and the form of the causee (majority of pronouns in the get construction and majority of nouns in the have construction). Likewise, [X have Y Vinf ] and [X have Y Vpp] differ in only two features, namely the definiteness of the causee and the semantic nature of the patient. The pairs of constructions with the fewest number of shared features in terms of frequency (fewer than 12 features) all involve [X cause Y Vto-inf ], which suggests that it is the construction that is most dissimilar to the others. While these three ways of measuring the degree of similarity between the constructions (distinctive features in common, distinctive features in opposition, most frequent features in common) provide slightly different results, a number of interesting points still emerge from their comparison. First, it appears that the get and have constructions have a great deal in common, which seems to be consistent with claims to this effect in the literature (see Section 5.1). Second, a


difference in structure is not incompatible with a high degree of similarity (consider, for instance, the case of [X get Y Vto-inf ] - [X get Y Vpp]), but this, again, is mainly true of get and have constructions. Finally, the results also point to some degree of similarity between cause and make constructions, although this is less marked than for get and have. It should be emphasised, however, that these results merely concern the syntactic and semantic regularities of causative constructions. Some earlier results have already pointed to several differences between the constructions (for example the high proportion of co-referential causative constructions with [X get Y Vpp] as opposed to [X have Y Vpp]), and other differences will appear in the following chapters.

5.4 Global approach: Hierarchical feature selection In an attempt to consider all the parameters together, another, more global type of statistical analysis (category statistics) was performed on the basis of the corpus data, namely hierarchical feature selection in a linear regression (see Hocking 1976).54 Two types of selection were performed: backward elimination and forward selection. In both cases, the aim was to identify the features that best predict the choice of one construction over another. All the syntactic and semantic parameters dealt with in this chapter were included in the analysis. In addition, the following parameters were taken into account: number of words in the causer, in the causee and in the patient, grammatical polarity of the causative verb and of the effect, relation between the causer and the causee and between the patient and the other participants (identity or not), explicitness of the causer and of the causing event, lemma of the effect, medium and genre in which the construction was found. Also, the animacy of the causee and the volitionality of the effect were treated together, as they are correlated with each other (animate causees may be used with volitional or non-volitional effects but inanimate causees always occur with nonvolitional effects, see Section 5.3.5) and one single parameter was kept for the grammatical person of the causer and causee (higher for the causer, higher for the causee or equal). In the case of multiple elements per construction (two causers or two effects, for instance), the values of the parameters were either combined with one another (e.g. the coordination of two singular causees was encoded as a plural) or only the first element was taken into consideration (e.g. a volitional effect followed by a non-volitional effect was treated as volitional),

54. Thanks are due to Marie-Catherine de Marneffe for her help with this analysis.


so as to have just one value for each parameter per sentence. Finally, because the low frequency of some constructions in BNC-10 did not provide a solid enough basis for a statistical analysis such as this one, additional instances of the following constructions were extracted from the BNC and classified according to the same parameters: [X have Y Vinf ], [X have Y Vprp] and [X make Y Vpp]. In order to identify the features that best discriminate between the causative constructions, a backward elimination was first performed. The backward elimination removes, one by one, the features that least hurt the classification accuracy of the test data (McNemar’s test on 10-fold cross-validation accuracy), i.e. the parameters which do not help distinguish between the constructions in a statistically significant fashion. When the decrease in accuracy is no longer statistically significant (using a threshold level of statistical significance of p you [causer] ’d have to have the points [patient] fitted by them [causee] but <pause> you get your own (285) Well we [causer] ’re having <pause> a point [patient] put in the kitchen <pause> and one put in the bedroom. (286) So they [causee] should be made to bring their prices [patient] down.

With the other constructions, the proportion of iconic sequencing ranges from 75% with [X cause Y Vto-inf ] to 96% with [X have Y Vprp]). The low percentage with [X cause Y Vto-inf ] is mainly due to the relatively high frequency of passive effects in this construction (they account for 23% of the occurrences of the construction), which have the effect of demoting the causee and promoting the patient, as in (287), where the patient (changes in α) comes after the causer (a change of context) and before the causee (events on the immediately preceding trial). While this situation is not totally impossible with the other constructions, e.g. (288), the main reason for the non-iconic sequencing of these constructions is the absence of causer and/or causee (see also Section 4.3.3). This is the case, for example, with the idiom make do, exemplified in (289), or more commonly in a sentence like (290), where the causer is not made explicit. (287) This outcome can be predicted if it is allowed that a change of context [causer] causes changes in α [patient] to be determined solely or largely by events on the immediately preceding trial [causee] (i.e. if contextual change produced an increase in the value of γ in equation 3.2 above).


(288) oh you [causer] never, never hold back and make be known in God’s name the kingdom [patient]. (289) I [causer] just make do with what we’ve got.

(290) The dangers of disregarding, by ignorance or design, the essentially conditional nature of abstract models and of making data [causee] fit into preconceived categories are well attested in the theoretical domain.

The other two models of prototypicality, the billiard-ball model and the model of direct manipulation, are hardly represented in the corpus data. Direct manipulation accounts for only 5.29% of all the constructions, whereas there are just two instances of the billiard-ball model in the whole corpus (0.06%). The sentences illustrating the billiard-ball model are given in (291) and (292). One of them is a [X get Y Vto-inf ] construction and the other one a [X make Y Vinf ] construction. Surprisingly, while this model is associated with natural sciences (Lakoff 1987: 55) and is typically used to describe scientific experiments, it does not occur at all with [X cause Y Vto-inf ], the construction most commonly used in scientific and technical genres (see Section 9.2.2). Although this construction shares quite a number of features with the billiard-ball model, its causer tends to refer to an abstract entity, as mentioned in Section 5.3.3, which automatically disqualifies it for the billiard-ball model, which requires physical objects. ( 291) Yes, that should physically click on the pin to get the wire to connect. (292) It’s got a tube that would make the <pause> the other one go.

Direct manipulation occurs with [X get Y Vto-inf ] (29%), [X get Y Vprp] and [X have Y Vprp] (about 11% each), as well as [X have Y Vinf ] (almost 10%), [X make Y Vinf ] (4.5%) and [X have Y Vpp] (0.3%). Given the preference of cause for scientific and technical contexts, the absence of the model with [X cause Y  Vto-inf ] is understandable. For the other verbs, on the other hand, one would expect, on the basis of the literature, to find more instances of direct manipulation: get and have are often claimed to be used almost exclusively with animate participants (Belvin [1993: 64] even claims that inanimate causers are impossible with have), and make is regularly presented as implying coercion (e.g. Faure & Casanova 1968: 192, Werner et al. 1990: 392), which makes it an ideal candidate for expressing direct manipulation. Cases of direct manipulation do occur, as appears from the prototypical examples in (293) to (295), but they are relatively rare.58 58. Notice that, while the three examples exhibit all the features of the model, they express direct manipulation with different nuances: repeated insistence in (293), coercion in (294) and request in (295).

Chapter 6. Defining the causative prototype 157

(293) He’s always trying to get our Terry to go out with us.

(294) At least by the eleventh century every king expected to recruit a part of his army by paying mercenaries, or from knights who received a fee not in land, but in cash; though he did his best to make his great nobles provide contingents for which he did not have to pay, or (at least in the twelfth century) pay him in cash if they did not serve him in person. (295) she said well you’re intelligent, you can do this for me and she had me doing this like phone calls ringing people up

Interestingly, the corpus data reveal a large degree of overlap between iconic sequencing on the one hand and the models of billiard-ball causation and direct manipulation on the other hand. A causative construction that displays prototypical participants almost always displays prototypical ordering too (although the reverse is not true). There are just two exceptions in the data, namely (296) and (297), which reflect the model of direct manipulation, but whose participants are not ordered iconically. (296) Our period opens with the imperial coronation of 962; shortly after its close, in 1165, Frederick Barbarossa had Charlemagne canonized by his anti-pope Paschal III. (297) I had it [hoe] <pause> sharpened by Hector

Taking this overlap into account, it appears that the models of prototypical causation presented in the literature only account for some 45% of all the causative constructions (1,632 out of 3,574), which leaves about 55% of the data unaccounted for (i.e. exhibiting neither prototypical ordering, nor prototypical participants). These models, therefore, do not seem to be fundamental organising principles in naturally-occurring language – at least if we adopt a strict definition of prototypicality, requiring the presence of all the parameters. In the next section, we consider a looser approach, which examines the different parameters individually.

6.3.2 Loose validation Some linguists define the prototype as an item exhibiting the largest number of prototypical features (cf. Wittgenstein’s [1953] concept of family resemblance; see Rosch & Mervis [1975] for an application of this concept). Another way of testing the validity of the constructs described in Section 6.2.3 is therefore to examine the different parameters individually and establish an average score for each construction on this basis. The first model, that of iconic sequencing, does not lend itself well to such a methodology, as the model only truly makes sense when the


positions of the participants are considered in relation to each other. The other two models, by contrast, have features which are relatively independent of each other and can thus be examined individually. In this analysis, each feature was assigned a score between 0 and 10, based on the (rounded up) proportion of data exhibiting the feature in question. The scores of all the features were then averaged for each construction, yielding a value ranging from 0 to 10. The higher the value, the more prototypical the construction. It should be borne in mind that such a methodology places all the features on the same footing, while it may very well be that some of them should be attributed more weight than others. Lakoff (1982: 164), in fact, notes that the ten properties making up the model of direct manipulation “are obviously not all equally important”. Since he does not further develop this idea, however, the relative importance of the properties cannot be identified, and is therefore not taken into account in the calculation of the average scores. The same is true for the billiard-ball model. Table 49 gives the results for billiard-ball causation, obtained through the methodology described above. It will be remembered from Section 6.3.1 that, using a strict definition of prototypicality, only two instances of the billiard-ball model were identified in the whole corpus, one with [X get Y Vto-inf ] and one with [X make Y Vinf ]. With a looser approach, the results suggest a higher degree of pervasiveness of the model, which averages a score of 6.1 (out of a maximum of 10) across constructions. According to this approach, [X get Y Vprp] is the most prototypical construction (average score of 7), followed by [X get Y Vpp] (6.9) and [X have Y Vpp] (6.8). All the constructions except [X make Y Vpp] average a score of 5 or above. Yet, it should be pointed out that each of them fares pretty badly with respect to the parameter of physical object, which, if any, could probably be considered the core of the model (although, as noted above, no weight can be attributed to the parameters on the basis of the literature). The average scores across constructions for physical objects are the lowest (0.2 for causers, 1.4 for causees and 3 for patients). Parameters such as single participant, material effect or affirmative clause, by contrast, which belong to the model of direct manipulation too and are therefore less specific to billiard-ball causation, fare much better in the corpus data and hence tend to inflate the overall score of the model. The strict approach of Section 6.3.1 revealed a stronger presence of the model of direct manipulation than that of billiard-ball causation. The same tendency appears from Table 50, which lists the average scores of the different parameters of direct manipulation per construction and across constructions. The overall score of the model amounts to 7.6 (against 6.1 for the billiard-ball model) and all the constructions rate higher than 6. The highest score is with [X get Y Vto-inf ] (average of 8.8), followed by [X get Y Vpp] and [X have Y Vpp] (8.5 each). The attested dominance of [X get Y Vto-inf ] confirms the results obtained through a stricter


[X

BE ma

[X

[X

[X

de V to inf ] MA KE YV pp ] Ave rag e

inf ]

MA KE YV

pp ]

prp ]

HA VE

YV

YV

inf ]

[X

[X

HA VE

YV

pp ]

HA VE

prp ]

GE [X

[X

[X

TY V

to inf ]

TY V

GE

YV

GE

CA USE [X

TY V

to inf ]

Table 49. Frequency of the parameters of the billiard-ball model

single

7

8

6

7

7

7

7

8

10

4

7.1

specific

5

9

9

10

9

9

9

9

4

7

8.0

physical

1

0

0

0

0

0

0

1

0

0

0.2

single

6

7

8

10

7

8

6

7

6

5

7.0

specific

5

9

10

7

9

9

10

8

7

7

8.1

physical

2

1

4

0

1

4

0

1

1

0

1.4

single

7

8

9

8

8

6

8

8

7

5

7.4

specific

4

9

10

10

9

9

9

8

9

7

8.4

physical

2

4

3

5

3

3

7

2

1

0

3.0

EFFECT

material

8

9

9

10

7

9

10

5

5

1

7.3

Clause

affirmative

10

9

9

9

9

9

9

9

9

9

9.1

5.2

6.6

7.0

6.9

6.3

6.6

6.8

6.0

5.4

4.1

6.1

CAUSER

CAUSEE

PATIENT

Average

CAUSER

CAUSEE

de V to inf ] MA KE YV pp ] Av e rag e

inf ]

[X

[X

[X

BE ma

YV

YV

HA VE [X

HA VE [X

MA KE YV

pp ]

prp ]

inf ]

YV

pp ]

YV

HA VE

[X

[X

GE T

YV

prp ]

inf ] to -

GE T [X

8

6

7

7

7

7

8

10

4

7.1

6

10

10

10

9

10

10

9

0

8

8.2

human

1

10

9

10

9

9

10

5

9

7

7.9

single

6

7

8

10

7

8

6

7

6

5

7.0

definite

7

8

8

7

7

7

5

9

6

7

7.1

human

3

8

3

10

6

5

10

7

4

9

6.5

10

10

10

2

10

10

8

10

10

10

9.0

7

8

9

8

8

6

8

8

7

5

7.4

5

8

7

8

6

7

8

8

4

9

7.0

10

10

10

10

10

10

10

10

10

8

9.8

definite ≠ CAUSER/CAUSEE

Average

YV

7

definite

PATIENT single

Clause

GE T

single

≠ CAUSER

EFFECT

[X

[X

CA US E

YV

to inf ]

Table 50. Frequency of the parameters of the model of direct manipulation

material

8

9

9

10

7

9

10

5

5

1

7.3

volitional

2

9

3

10

6

5

10

3

3

1

5.2

affirmative

10

9

9

9

9

9

9

9

9

9

9.1

6.3

8.8

7.8

8.5

7.8

7.8

8.5

7.5

6.4

6.4

7.6


methodology, which seems to suggest that this construction has a special status in terms of prototypicality. For the other two constructions, on the other hand, the strict approach highlighted a very low, or even null degree of prototypicality, which is in contradiction to the results of the looser approach. If we observe the average scores of the individual parameters across constructions, we notice that none of them is lower than 5, which is different from billiard-ball causation. Another difference with the billiard-ball model is that the parameters of direct manipulation show much more variation from one construction to another. The volitionality of the effect, for example, scores an average of 10 or 9 with some constructions ([X get Y Vpp], [X have Y Vpp], [X get Y Vto-inf ]), but 1 or 2 with others ([X make Y Vpp] and [X cause Y Vto-inf ]). As a result, some opposing tendencies sometimes cancel each other out. Compare [X cause Y Vto-inf ] and [X get Y Vpp]. While the former has a score of 10 for the parameter “causer ≠ causee” and a score of 2 for the volitionality of the effect, the scores of the latter construction are opposite: 2 for “causer ≠ causee” and 10 for the volitionality of the effect. Such differences disappear if we add the scores of the two parameters together to obtain an average score. As this example shows, the global scores per construction may thus hide important internal variations.

6.4 Cognitive validation of the models of prototypical causation Now that we have tested the linguistic validity of the three prototypical models described in Section 6.2.3 by examining their frequency in the corpus data, we may test their cognitive validity. This validation will rely on the results of the elicitation test, and more precisely the exercise requiring the subjects to produce the first causative construction that comes to their minds for each of the verbs cause, get, have and make (exercise E.1-4). This type of exercise seems to be particularly suitable to identify the uses that are the most prototypical (in the sense of cognitively salient) for, as Geeraerts (1988a: 218) argues, these uses, because of their salience, are more likely to pass the threshold of informants’ conscious attention and hence appear in sentence production experiments such as this one. In addition, the technique of the “first-come-to-mind” item has regularly been used in psychology to investigate prototypicality (see Ungerer & Schmid 2006: 41). Despite the appropriateness of the exercise for the present purposes, however, it should be underlined that this analysis is exploratory and its conclusions, therefore, tentative. For one thing, although 50 questionnaires were completed, the exercise in question was often left unanswered (no fewer than 28 cases with cause and get, 27 with have and 26 with make), and not all constructions were represented in the answers ([X be made Vto-inf ], [X make Y Vpp] and [X have Y Vprp]


were never produced by the informants). For another, it might be argued that the causative constructions presented in the preceding exercises resulted in a priming effect and influenced the types of sentences the subjects came up with – although the constructions proposed throughout the test were quite varied and one may hope that the influences of the stimuli cancelled each other out and that priming effects were therefore minimised. Table 51 summarises the results for the three models of prototypical causation, viz. iconic sequencing, billiard-ball causation and direct manipulation, adopting a strict approach.59 It appears from these results that overall, the proportions of prototypical constructions in the elicitation data are higher than in the corpus data (compare with Table 48). What is particularly striking is that iconic sequencing represents 80% in the elicitation data (as compared to some 45% in the corpus data). This result is to be related to subjects’ preference for infinitive constructions (less than a quarter of the data are not of this type) and their strong tendency to make participants explicit. A closer examination of the data reveals that the elements are organised iconically even in cases where it does not sound very natural. Consider the following (elicited) sentences, already quoted earlier: (298) He got the plumber to fix the leaking pipe.

<EL:E3:033>

(299) John had the police arrest Mark.

<EL:E4:018>

Because the causees in these sentences are predictable, it would be more normal, in authentic language, to leave them unexpressed and use a past participle construction instead, yielding: He got the leaking pipe fixed and John had Mark arrested. Table 51. Salience of the models of prototypical causation Iconic sequencing N % [X cause Y Vto-inf ] [X get Y Vto-inf ] [X get Y Vprp] [X get Y Vpp] [X have Y Vinf ] [X have Y Vpp] [X make Y Vinf ] Total

22 14 4 0 9 0 24 73

100.00 100.00 100.00 0.00 100.00 0.00 100.00 80.22

Billiard-ball model N % 2 0 0 0 0 0 0 2

9.10 0.00 0.00 0.00 0.00 0.00 0.00 2.20

Direct manipulation N % 0 8 1 0 3 0 6 18

0.00 57.14 25.00 0.00 33.33 0.00 25.00 19.78

59. Because the figures are quite low, only the global approach of Section 6.3.1 has been applied to the elicitation data. The approach examining each parameter individually (see Section 6.3.2) has not been attempted.


The billiard-ball model represents a slightly higher percentage in the elicitation data than in the corpus data (2.2%, as against 0.06% in the corpus data), but only accounts for two sentences, as is the case in the corpus. It is worth underlining that both sentences are with [X cause Y Vto-inf ], the construction which, as noted above, is intuitively most closely connected with the billiard-ball model: (300) The rain caused the river to flood.

<EL:E1:026>

(301) The rain caused the roof to leak.

<EL:E1:032>

Direct manipulation, like the other two models, is more common in the elicitation data than in the corpus (19.78% vs. 5.29%). Some examples are given below: (302) I will get my sister to ring you.

<EL:E3:021>

(303) The teacher had him leave the class.

<EL:E4:025>

(304) She made the young lad eat his porridge.

<EL:E2:050>

In both the corpus and elicitation data, however, it is with [X get Y Vto-inf ] that the model is most common – although it is twice as common in the elicitation data as in the corpus data. The case of [X make Y Vinf ] is an interesting one. While direct manipulation accounts for only 4.55% of this construction in the corpus, it represents 25% in the elicitation data, which means that it is over five times more common in elicited language than in naturally-occurring language.

6.5 Explaining the discrepancies The analysis of the preceding sections has revealed the existence of important differences between the theoretical constructs described in the literature, the corpus data reflecting linguistic frequency and the elicitation data corresponding to cognitive salience. In what follows, we try to explain these discrepancies, first comparing the theoretical models from the cognitive literature with the empirical evidence, and then comparing the corpus data with the elicitation data.

6.5.1 Theoretical constructs and language data Although the comparison of the corpus data and the elicitation data reveals that the latter come closer to the three models of prototypicality than the former, there are still important divergences between the prototypical models described in the literature and the constructions attested in (spontaneous or elicited) language data, which somewhat challenges the validity of these theoretical constructs.


iven that these constructs do not seem to have any empirical basis, it would G be quite possible that they are linguistically and cognitively invalid and merely reflect their authors’ introspective judgements. However, other explanations may be put forward to account for these discrepancies. One possibility is that the three models used here are not suitable for the description of periphrastic causative constructions. It will be reminded that these models identify prototypical causation and that the phenomenon of causation may be expressed through a wide variety of linguistic devices, which may not all display the same prototypical features. While Lakoff (1987: 54–55) mentions the phrase cause to die when defining direct vs. indirect causation, he gives as typical examples of the model of direct manipulation the sentences Max broke the window and Brutus killed Caesar, both of which contain a lexical causative. On the basis of these two examples, and taking the reasoning one step further, one may argue that Lakoff ’s model describes prototypical transitivity, rather than prototypical causation.60 Causation has often been compared, sometimes confused, with transitivity (e.g. Croft 1994) and definitions of prototypical transitivity such as the one found in Kemmer & Verhagen (1994: 126) are remarkably similar to the models of prototypical causation that have been used here, the main difference being the presence or not of an intermediate participant:61 [The prototypical transitive (or two-participant) event] has an agentive participant, that is, a highly individuated entity capable of volition, and volitionally exerting physical energy on a second participant, which is also a highly individuated participant. This participant absorbs the energy, whereby it undergoes a change of state that would not have taken place without the exertion of energy. The effect on the second participant is direct, that is, there are no observed intermediaries such as a third participant; the effect is complete; there is physical contact between the two participants; and this contact is seen as giving rise to the change of state.

The similarity (or confusion) between prototypical transitivity and prototypical causation is further emphasised by an earlier work of Lakoff ’s who, in 1977, characterised the model of direct manipulation (with minor differences) as that of “prototypical agent-patient sentences” (Lakoff 1977: 244). Such problems of characterisation of the models may account for the discrepancies emerging from the 60. Thanks are due to Maarten Lemmens for this suggestion. 61. In fact, Kemmer & Verhagen (1994) themselves draw a parallel between simple two-participant clauses (e.g. I ate the cake) and causatives with an intransitive effect (e.g. I made Mary cry), as well as between simple three-participant clauses (e.g. I gave Mary a flower) and causatives with a transitive effect (e.g. He had the servant taste the food).


analysis: if the models of prototypical causation found in the literature do not concern causation per se, applying them to causative constructions is bound to lead to a number of differences. Another, more radical explanation for these discrepancies is that there is no such thing as prototypical causation. Hampton (1981) showed that the prototype structure of abstract concepts cannot always be demonstrated, and causation might be one of those abstract concepts that has no prototype. In this case, the literature would seek to describe, and we would try to track down in language data, something that simply does not exist – and hence cannot be found.

6.5.2 Corpus and elicitation data Whether we consider that the models found in the literature are valid descriptions of prototypical periphrastic causative constructions or not, the analysis carried out in this chapter has also revealed discrepancies between the corpus and elicitation data, and these require explanation too. It has often been assumed in the cognitive literature that there is a direct link between linguistic frequency and cognitive salience (or entrenchment, which can be described as a kind of enduring salience, see Tuggy 1993: 279). Thus, Langacker (1987: 100) claims that “[a]n event (…) becomes more and more deeply entrenched through continued repetition”. This idea has been elevated to the status of principle, Schmid’s (2000: 39) “From-Corpus-to-Cognition Principle”, which posits that “frequency in text instantiates entrenchment in the cognitive system”. In other words, the most frequent item in language is believed to correspond to the most salient one in the mind. It cannot be denied that frequency influences many aspects of language (see, for example, Ellis 2002 or Bybee 2006). However, some voices have risen to challenge the particular claim that the frequency studied by corpus linguists should necessarily reflect the salience studied by cognitivists. Sinclair (1991: 36) notes that “for many common words, the most frequent meaning is not the one that first comes to mind”, and some recent studies which compare corpus data (as a resource for identifying frequency) and elicitation data (as a resource for identifying salience) have come to the conclusion that the two types of data do not always overlap (e.g. Roland & Jurafsky 2002, Nordquist 2004, 2009, Gilquin 2008b, Shortall 2007). Shortall (2007) shows that subjects who are asked to use there in a sentence tend to produce constructions with a concrete noun, such as There are three books on the table (60% in Shortall’s elicitation data), whereas the most frequent use of there in corpus data from the Bank of English is with an abstract noun (59%), as in There is evidence to suggest that… Similarly, Nordquist (2004) points out that in authentic language the adjective small


is highly frequent with nouns indicating quantities (amount, piece, proportion, etc). Her elicited data, however, contain only five instances of the word amount (12%). In the remaining cases, small modifies different nouns such as car, animal or room. In fact, Schmid himself has recently started to question the validity of the link between frequency and entrenchment (Schmid forthcoming). There are at least two principles that may account for differences between corpora and elicitation (see also Gilquin forthcoming a). The first one has to do with the context of production. While elicitation data are produced in experimental settings, out of context, corpus data are produced in natural settings and are part of a larger (situational and textual) context. Elicitation data exhibit what Roland & Jurafsky (2002) call an “isolated-sentence effect”, with a tendency to use a set of default referents, including first-person subjects, VP-internal NPs anaphorically related to the subject and predictable head nouns for the direct object. The results of the above-mentioned studies also suggest that people favour the concrete over the abstract in single sentence production tasks, which might be due to the primacy of the concrete in neural representations (see MacLennan 1998). Connected discourse, by contrast, seems to be more abstract and is governed by phenomena such as grammaticalisation or idiomatisation. The presence of idioms and other phraseological units, precisely, is another factor that distinguishes corpora from elicitation data. Language relies heavily on multi-word expressions that do not need to be reassembled every time they are used (Sinclair 1991, Stubbs 1993) and this is reflected in corpora, which contain a large proportion of them. By contrast, the psycholinguistic nature of multi-word expressions prevents them from playing a crucial role in elicitation data. Multiword expressions are claimed to be stored holistically in the mind and independently of the words they are made up of (Wray 2002), so that subjects prompted for a word access just that, not the holistically stored expressions that contain this word (see also Nordquist 2004). Therefore, more idiomatic constructions like makes you wonder or I’ll have you know, which occur in the corpus data (see Section 7.4), are unlikely to be elicited by the stimuli make and have. All this confirms Wray’s (2005) point that corpora and intuition (which is at work in elicitation) draw on different pools of linguistic knowledge, and explains why our corpus and elicitation data differ with respect to the models of prototypical causation proposed in the cognitive literature. The above principles also partly account for the fact that the elicitation data come closer to these models than the corpus data. The models of prototypical causation found in the literature represent concrete situations anchored in everyday life. They refer to definite entities and material processes. The default referents favoured in isolated (i.e. elicited) sentences (see above) correspond better to these requirements. By contrast,


the causative constructions produced in natural context tend to be more abstract (consider, for example, the typically abstract causer in [X cause Y Vto-inf ] or the predominance of mental effects in [X make Y Vpp]) and are characterised by a high degree of idiomaticity (see Chapters 7 and 8). These characteristics of elicited vs. natural language, in fact, even point to a possible explanation of why the theoretical constructs of causation described in the cognitive literature are what they are. Since linguists (apparently) rely on their introspection to establish these constructs, it is only natural that they should be influenced by principles similar to those which guide “the man on the street”, hence granting a special status to the coercive meaning of [X make Y Vinf ] or the physical use of [X cause Y Vto-inf ], for example, while these senses are not particularly common in naturally-occurring language. The discrepancies discovered between the theoretical constructs and the results of the empirical analysis, as well as between the corpus and elicitation data underline the complex nature of the concept of prototypicality in linguistics. While this study suggests that prototypes should be identified empirically, as otherwise they might lack any (linguistic or cognitive) validity, the question remains as to what criterion should be chosen to identify them: frequency (as attested in corpus data), salience (as evidenced by elicitation experiments), or yet other criteria that have not been investigated here such as the historically oldest meaning of a word or its earliest acquired sense. This choice will depend on one’s definition of prototypicality and it is therefore crucial to clarify what we mean by prototypicality. However, it might also be that prototypicality can only be defined by means of a combination of criteria, each of which would contribute to the prototypical nature of the prototype: the more criteria an item satisfies, the more prototypical it is. Such a multifaceted conception of prototypicality would support Geeraerts’s (1988b) claim that prototypicality is itself a prototypical notion, with fuzzy boundaries, as well as central and more peripheral instances.

6.6 Summary This chapter has compared three models of prototypical causation found in the cognitive literature against language data. One of these models involves the iconic sequencing of the elements making up the causative construction, whereas the other two models, billiard-ball causation and direct manipulation, have to do with the nature of these elements. The comparison of these theoretical constructs with corpus and elicitation data reveals that they do not account for a particularly large proportion of language use. The model of iconic sequencing is the most common of the three models in the corpus, while billiard-ball causation


is the least frequent one. [X get Y Vto-inf ] turns out to be the most prototypical construction with respect to the model of direct manipulation, both in the strict approach (considering the combination of all the parameters) and in the looser approach (considering the parameters individually). The proportion of prototypical constructions is higher in the elicitation data, but the relative importance of the three models is the same, with iconic sequencing coming first, followed by direct manipulation and then billiard-ball causation. In the elicitation data [X get Y Vto-inf ] is the most “manipulative” construction (as it is in the corpus data) and [X cause Y Vto-inf ] is the only construction that occurs with the billiard-ball model. The analysis of the prototypicality of periphrastic causative constructions challenges both the (linguistic and cognitive) validity of the theoretical constructs found in the literature and the assumption that frequency (as attested in corpora) may be equated with salience (as evidenced by experimentation). It also suggests that the identification of prototypes in linguistics requires an empirical basis, but that the choice of this basis depends on the way one defines the concept of prototypicality. After three chapters where the approach has been essentially corpus-tested, starting from a theoretical concept and observing the data from this angle, the following chapters will be more corpus-driven. This time, the starting point will be the data, which will serve as a basis for theoretical generalisations.

chapter 7

Lexical co-occurrence in causative constructions

The lexical dimension of causative constructions is often neglected in the literature. Yet, it will appear from this chapter and the following one that this aspect is well worth studying, and in fact makes it possible to tell the constructions apart quite well. In this chapter, we examine the collocates (and semantic prosodies they evoke) as well as the word clusters typical of causative constructions, both in the corpus data and in the elicitation data. Lexical co-occurrence in the effect slot of causative constructions will be the subject of Chapter 8.

7.1

The co-textual approach

Corpus linguists do not study words “in splendid isolation” (Teubert 1999). They study words in context, or rather, in co-text (i.e. the surrounding text of a word, see Tognini-Bonelli 2001: 87), in the belief that this will help them understand the meanings of words better. The investigation of co-text has been introduced in language analysis relatively recently and, as Willems (2000) points out, largely under the impetus of corpus linguists. Because such phenomena are often perceived at a subliminal level, it is claimed that they “cannot be reliably retrieved by introspection” (Sinclair [1991: 116]; see also Louw [1993: 173] or Stubbs [1996: 172]).62 Corpora, on the other hand, present words in their linguistic environment. Text retrieval software typically displays search words in KWIC (keyword-in-context) format – and the more specialised tools make it possible to list the words recurring frequently within a particular distance to the right or to the left of the target word. The term “collocation” as it is used in corpus linguistics has to do with the preferential lexical company kept by a lexical item (see Sinclair [1991: 109–121] or Tognini-Bonelli [2001: 162]). It refers to the “tendency [of two or more words] to 62. See, however, McGee (2006, 2009), who demonstrates that frequent collocates can be more easily retrieved by means of introspection if they are part of a bare dyad (i.e. are not embedded in a larger chain).


co-occur within a few words of each other in running text” (Stubbs 2001: 24).63 A “collocate” represents any word exhibiting such a tendency. The characteristic collocates of words, as Stubbs (1996: 172) observes, “show the associations and connotations [these words] have, and therefore the assumptions which they embody”. They are thus useful pointers to the meanings of words. Sometimes, collocates group together to form a definable semantic set. Following Stubbs (1995a: 25), this will be referred to as the “semantic prosody” of a word (see also Partington 1998: 68). As underlined by Stubbs (1995b: 249), semantic prosodies can show varying degrees of generality, from the very general (positive vs. negative semantic prosodies) to the more specific (e.g. the field of care with the verb provide [Stubbs 1995b]).64 Co-text may also take the form of word clusters, that is uninterrupted strings of words which recur frequently in a corpus. Finally, the typical patterning of a word may also be grammatical, in which case the term “colligation” is used (see Tognini-Bonelli 2001: 163). Very little has been said in the literature about the typical collocates (or semantic prosodies, for that matter) of causative constructions. Sparse indications can be found as to what verbs tend to be used in the effect slot of some of the constructions, but because collocates in the effect slot will be the focus of the next chapter, this literature will be reviewed later. As regards collocation and semantic prosody in general, the only causative verb that has been treated in some detail is cause. Magnusson (1987: 216–217) and Stubbs (1995a: 27–28, 1995b: 247) have shown that this verb has a predominantly negative prosody, preferring the company of unpleasant words such as accident, death or trouble (Stubbs 1995a, 1995b). It should be emphasised, however, that both linguists are mainly interested in the lexical use of the verb. In fact, Magnusson (1987: 217) notes that out of a total of 139 object constructions with cause, about 110 have an unambiguously negative prosody, but that “[e]specially the NP + infinitive construction reveals a number of cases in which everyday language carries no negative associations” and takes on the “neutral meaning referring to the scientific search for explanations”. Similarly,

63. Note that the term will be used here as a statistical notion, a “statement about probabilities” (Stubbs [1996: 173]; see also Hoey [1991: 7–8] or Partington [1998: 16]). Herbst & GötzVotteler (2007: 211) refer to this as the “sandy beaches-type” of collocation. They distinguish it from the “guilty conscience-type” of collocation, in which “the combination is significant because it is established or institutionalized (…) and somehow unpredictable on the grounds of the meanings of the words” (see Cowie [1994: 3169] for an example of this use of the term “collocation”). 64. Sinclair (1996) calls this phenomenon “semantic preference”, as distinct from “semantic prosody”, which he limits to collocates performing the same pragmatic function (e.g. a negative pragmatic load; see also Louw 1993). Hoey (2005) uses the general term “semantic association”.

Chapter 7. Lexical co-occurrence in causative constructions 171

Stefanowitsch & Gries (2003: 239n) suggest that “the claims of a negative semantic prosody do not necessarily apply to [infinitive constructions with cause]”. As for Stubbs (1995a), he gives a few examples where cause is followed by a neutral or positive infinitive (e.g. caused him to smile). Other, anecdotal remarks found in the literature include the association of causative get with words such as finally, try (Cottier 1991), manage (Ikegami 1989: 207), can’t and be difficult to (Stefanowitsch 2001: 154), or the association between the past participle construction with get and body parts or clothes, e.g. I got my boots mended (Deutschbein & Klitscher 1959: 136). In the next section, we show how the collocates of the different causative constructions were extracted from the corpus, and in Section 7.3 we use these results to try and draw up the “collocational profile” of each construction.

7.2

Methodology

Although BNCweb (Version 2.0) offers great possibilities for the post-processing of concordances, including collocational search, these options could not be accessed here because of the form of the output and the limitations of the SARA server. More precisely, the “delete hits” command of BNCweb, which would have made it possible to eliminate the non-causative constructions from the query output and hence provide a “clean” basis for the collocational analysis, was not available, since it only applies to concordances of under 1,000 hits. Therefore, use was made of WordSmith Tools to establish the collocational profile of the different constructions. For each of the ten causative constructions under investigation, a new file was created which included all the relevant hits extracted from the corpus. In order to distinguish the causative function of cause, get, have and make from their other, non-causative functions, all the causative uses were replaced by the label “CAUS” (for “causative”). Thus, in (305) only the second occurrence of have was replaced by “CAUS” and, hence, taken into account in the collocational search. (305) It says that they have a right to have their own needs assessed, the wishes of the family and other carers <pause> erm, have to be taken into account <pause> carers have to be involved in drawing up local plans

In addition, following a method proposed in Altenberg & Granger (2001), all the inflected forms of a causative verb were replaced by the same label (in this case, “CAUS”), since the collocate display of WordSmith Tools does not work with lemmatised entries. This made it possible to group together the collocates of the


different inflected forms of the causatives.65 Only those collocates that occurred within a span of up to five words to the left or to the right of the causative verb were taken into account – a distance that is advocated by several linguists, among whom Church & Hanks (1990).66 Semantic prosodies were identified on the basis of collocates recurring at least five times within this span. However, words occurring with a lower frequency were considered too in so far as they were relevant to the semantic prosody for, as rightly emphasised by Stubbs (1995b: 249), “[w]hat is significant is the summed frequency of semantically related items”, and to this, even less frequent collocates may contribute. Word clusters were also identified by means of WordSmith Tools, using the same system of one individual file per construction and special labelling of the causative uses of the verbs. This time, however, since we were interested in the exact repetition of given strings of words, the distinction between the various inflected forms of the causatives was maintained. This was done by adding the label “CAUS” to the verb (rather than replacing the verb by “CAUS”), e.g. “CAUSmake”, “CAUSmakes”, “CAUSmaking” or “CAUSmade”. Only word clusters with a minimum frequency of three occurrences were taken into account in the analysis.

7.3

Collocational profiles in the corpus data

This section presents the main results of the collocational analysis, which consisted in examining the collocates of each construction and identifying the semantic fields around which these collocates cluster (i.e. their semantic prosodies). Note that the complete lists of collocates will not be presented, just those that correspond to the semantic field under discussion. Note also that, although the lists of collocates might include verbs occupying the effect slot, these will be the subject of the next chapter and will therefore not be particularly focused on at this stage.

7.3.1 [X cause Y Vto-inf ] It was pointed out in Section 7.1 that cause is the only causative that has been treated in some detail in the literature (though the focus has very much been 65. As a result, it was not possible to investigate whether, as is sometimes claimed in the literature, the different inflected forms of a lemma have different collocates (see e.g. Stubbs 2000 or Newman & Rice 2006). 66. In the examples, however, even collocates occurring beyond a span of [−5] to [+5] will be underlined, even though they were not taken into account in the automatic collocational analysis.


on the transitive construction). Yet, the preferential lexical company kept by [X cause Y Vto-inf ] is in fact among the weakest, together with [X have Y Vinf ], [X have Y Vprp], [X be made Vto-inf ] and [X make Y Vpp]. Only 38 words occur within a span of five words to the left or to the right of cause with a frequency of five or more occurrences, and if we apply a stop list which eliminates grammatical words from the results (Steyaert 2003), a mere four collocates remain: one,67 people, system and become. Still, there is a semantic field that emerges from the data, at least if one takes less frequent collocates into account. This field is suggested by the collocate system, as illustrated in (306) and (307). Like system, other collocates of cause point to a technical and scientific context, as appears from Table 52.68 Some examples containing this type of collocates are given in (308) to (310). (306) “end” causes the status of the system to be changed from REPLAY to INACTIVE. (307) At this point the excitation switches almost instantaneously from phase A to phase B and the motor now produces a positive torque, causing the system to accelerate towards the phase B equilibrium position at 8=2n/3p. (308) An input pulse at the noninverting input of such a sign and amplitude as to reverse the input signal to the open-loop operational amplifier, will cause the output to switch over to the opposite saturation level. (309) For example, I perceive this flower in front of me because light which has been reflected from it impinges on my retina and causes trains of nerve impulses to travel along the visual pathways to the visual cortex. (310) Magnetic saturation causes the torque/current and torque/position characteristic to depart from the “ideal” linear and sinusoidal relationships, but this problem can be counter-acted by adjusting the winding current levels to give uniform mini-steps.

As regards the negative vs. positive semantic prosody of the construction, it turns out that about a quarter of the examples refer more or less clearly to a negative situation, e.g. (311) and (312), a mere two percent refer to a positive situation, 67. Although one may be considered a grammatical word, it is not included in Steyaert’s stop list. 68. All the words that potentially belong to a semantic field have been included in this table and the following ones, although it is possible that not all the occurrences of a given word are used with the special meaning related to the semantic field. The word instruction(s) in Table 52, for example, could be used in a non-technical sense in some of its occurrences.


Table 52. [X cause Y Vto-inf ] and the semantic prosody of technical and scientific terminology system torque output switch accumulate data granules inhaling input instructions ions motor pressure radon variation accelerate accumulating adhere aluminium amplifier amplitude analysts autoantibodies bicarbonate bytes

6 4 3 3 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1

calibre cell cells chemical Chernobyl chloride climatic compaction components computer corpus cortex cylinder dielectric dishabituator DNA electromagnetic engine experimentation formatted gravitational hardware instruction luminescence magnetic

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

mechanism mercury molecules ocular orbit osmotic ozone phenotypic phosphate pixels polarization radiation refraction rejuvenation retina rotor saturation sinuosity spectral submergence tidal tractrix triggering triggers vector

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

e.g. (313), and the remaining examples, that is a large majority, simply describe a neutral situation, as in (314) or (315). These results, by and large, confirm the suggestions made in the literature that the negative prosody of cause might be limited to its transitive use (see Section 7.1). (311) Then, the artistic aims of members of the group were quite varied, and some personal animosities caused the alliance to break up. (312) Using unsafe desensitisation, conventional allergists and general practitioners have caused 26 patients to die since 1957. (313) The famous passage about the madeleine, the little cake whose associations call back the forgotten scenes of the narrator’s childhood, would have caused the Hartleian in Wordsworth to applaud.


(314) Following one stimulus rather than another with food, for example, is likely to establish associations that will cause the subject to approach the former stimulus rather than the latter when both are available. (315) How can a series of fixed instructions cause the computer to come up with a random sequence of results?

7.3.2 Get and have constructions It was demonstrated in Section 5.1 that the literature often fails to acknowledge the specificity of the different causative constructions. This is particularly true of causative get and have, which are most of the time described as interchangeable (e.g. Allen 1959: 35, Eckersley & Eckersley 1967: 196, Palmer 1974: 159, Todd & Hancock 1986: 218). And indeed, the analysis in Chapter 5 has demonstrated that the constructions share a large number of (syntactic and semantic) features. We will also see in Chapter 9 that with respect to style, which is one of the only differences sometimes recognised in the literature between the two verbs, causative get and have are not so dissimilar. In fact, the main feature that distinguishes the two verbs has to do with collocation, as we will see in this section. According to some linguists, causative get tends to collocate with words such as finally, try or manage, which all suggest effort (see Section 7.1). Other linguists associate the verb with the idea of effort or difficulty, although they do not refer to any specific collocates. Palmer & Blandford (1969: 197) claim that “get suggests that there may be slight trouble or difficulty in arranging for the action to be performed” (see also Ikegami 1990a: 188, Swan 2005: 200). Stocker (1990: 225) refers to the idea of effort with respect to the infinitive construction, and Cottier (1991) speaks of resistance on the part of the causee, which results in difficulty to perform the effect. The idea of persuasion, which implies the same sort of difficulty, is also often associated with causative get, at least in the construction [X get Y Vto-inf ] (e.g. Goldsmith 1984: 122, Murphy 1985: 110, Todd & Hancock 1986: 218, Berland-Delépine 1990: 172, Declerck 1991: 490). All this is confirmed by the collocational analysis, which brings to light a large number of collocates contributing to the semantic prosody of effort or difficulty. This is true of all the three constructions involving get. However, because this tendency is strongest with [X get Y Vpp], this construction will serve as the main illustration in what follows. Table 53 lists some of the collocates of causative get (in the past participle construction) participating in the semantic prosody of difficulty, and the following sentences are examples taken from the corpus, where the relevant collocates have been underlined:


( 316) So I take it Laura’s got her sinus problem sorted out did she? (317) Yes I’m trying to get it done, but I am getting pressure from the year group as well (318) In cases where women (and the children they directly control) collect forest products of one type or another, it is they who are most aware of any deterioration in resource availability, but who are often an inaccessible but important target group in agricultural extension efforts to get conservation methods accepted. Table 53. [X get Y Vpp] and the semantic prosody of difficulty sorted 99 try 18 12 trying sort 6 worry 6 settled 5 able 4 hope 3 problem 3 properly 3 doubt 2 heavy 2 help 2 helps 2 invalidity 2

managed maybe afford banned banning barrier complex dangerous difficult difficulty efforts hard hopefully hoping hurry

2 2 1 1 1 1 1 1 1 1 1 1 1 1 1

impossible panic panicking persuaded pity possible sanctions sorry stressful tried trouble unable worried

1 1 1 1 1 1 1 1 1 1 1 1 1

Next to the very frequent collocates sorted (which implies the existence of a problem), try and trying (which both suggest effort), it will be noticed that many other words occurring only once or twice in the immediate environment of [X get Y Vpp] also evoke the semantic prosody of effort or difficulty. In the case of [X get Y Vto-inf ] and [X get Y Vprp], it is mainly less frequent collocates that have to be examined in order to find semantically relevant collocates (which is to be linked to the lower frequency of these two constructions in the corpus), but these collocates point to the same semantic prosody as that of [X get Y Vpp], with words such as attempt(s), ineffective, risk or struggle. Here are some illustrations: (319) Attempts to get parents to reduce calorific intake, if the cause of the obesity is psychogenic, are doomed to failure; possibly this is why these families are often so difficult to treat. (320) I don’t know how that man <pause> managed to get the two edges to meet <pause> but he did a wonderful job on it!


(321) We’re not picking on you we’re just trying to get a conversation going here like, but it’s very hard when you just sit there and say nothing.

Only the most explicit and obvious collocates of difficulty have been listed in Table 53. However, many more words (even grammatical words) somehow convey the same idea. This is the case, for example, of all (123 occurrences in all the get constructions) and the related words everything (22), lot (14), much (14) and whole (5). What is common to these five words is that they all refer to a large quantity, a meaning which in itself may suggest difficulty. Consider (322). The first part of the sentence implies that there were many things to do but that only a few of them were actually done. As regards the second part, there is clearly more effort involved in washing all the cups than in washing just a couple of them. In (323) it is the contrast between the time available (only a very short time) and the mass of things to organise (everything) that brings out the idea of difficulty. (322) I haven’t got much done, I’ve got all the cups and things washed, but oh Catherine’s just phoned me (323) we’d only a very short time to get everything organized

Negation, which is well represented among the collocates of causative get (19 never, 11 couldn’t, 5 can’t, etc), may also entail difficulty. In (324) one may deduce from the combination of never and properly that the causer finds it difficult to operate the machine. In (325) the negation implies sheer impossibility for the causer to get the caused event performed. (324) I can never get it [this machine] to work properly.

(325) I just couldn’t get them to change their minds you know they just wanted you there

The word else may be used as in (326) or (327), where it indicates a condition, and hence an additional difficulty for the causative process to take place. Conditional clauses are also quite common in causative constructions with get (about 10% of the data, to be compared with a proportion ranging from 5% to 6% with the other causatives) and they fulfil the same function as else, see example (328). ( 326) Can we get some breakfast, or else we’ll get nothing done today at all. (327) I’m only showing you once darling and then you’ll have to do it or else I’ll never get me bingo done ( 328) If I don’t get on and do it <pause> I’m never gonna get it done.


Even simple and apparently insignificant words like please or bloody may evoke a difficult situation by suggesting annoyance at not being able to get something done. This is illustrated by the following sentences: (329) Get your homework done Paul please.

(330) And have you ever put a tree <pause> a Christmas tree in a tub and packed earth round it and got the bloody thing to stand up straight?

Finally, words having to do with time, present in over 35% of the concordance lines with get (as against 24% with have and only 1% with cause and make), may also imply difficulty, either because they suggest that an activity requires a great deal of time (331) or because they point to a tight schedule (332). Such temporal indications may come in addition to other collocates belonging to the semantic field of difficulty, as appears from (332), which also contains the expression trying their best. (331) See by the time I get to that stage it’s too late to do any thing to send off (…) by the time I get it [the crossword] done (332) They’re trying their best to get it done before Christmas

Among the 1,310 causative constructions with get, no fewer than 683 make explicit reference to some sort of difficulty, that is over half of the examples. This does not take account of sentences such as this one, where the notion of effort is present, but not explicitly mentioned: (333) One peasant walked 15 versts to get Yakovlev to act on a building designated for a school that was still occupied by an influential peasant.

While, as appears from what precedes, the semantic prosody of effort/difficulty is very strong with causative get, it is not the case with causative have. Taking all the three have constructions into account, the list of collocates involving effort or difficulty is virtually limited to six occurrences of whole, four occurrences of able, much, never, problem and try, two occurrences of possible and success, and one occurrence of attempt, hoping and risk, a list which does not stand comparison with the collocates of get expressing the same idea. Have, instead, displays another semantic prosody, which, like the semantic prosody of effort/difficulty with get and partly for the same reason (the frequency of the different constructions), is most obvious in the past participle construction. What is particularly striking when one goes through the list of collocates of causative have is the number of words referring to the field of hairdressing: cut (82 times), hair (72 times), permed (23 times), shaved (13 times), trimmed (6 times), etc. This field is illustrated by (334) and (335).


(334) Mind you, I think I’ll have my hair shaved off.

(335) well she’s growing the back of it <pause> and having it permed and highlighted.

The prevalence of such collocates is admittedly linked to the type of discourse in which the constructions mainly occur, viz. (female) conversations, where the topic of hairdressing may be said to fulfil a phatic function, just like small talk about the weather would do. However, the presence of these collocates is also indicative of a more general semantic prosody, which may be described as that of service, in the sense of providing people with something they need or want, usually against payment. Several subfields belong to this semantic prosody, in addition to the field of hairdressing, for example the field of (beauty or medical) care, with collocates like nails, manicured, hospital or doctor, the field of technical maintenance, with collocates like serviced, repaired, damaged or faulty, or the field of house maintenance, with collocates like decorated, painted, carpet or roof. An example of each subfield is given in (336) to (338). All in all, over 70% of the constructions with have refer to a service. (336) Lee’s er <pause> paid for <pause> Liz to go and have her nails manicured tomorrow. (337) Well I’ve had the mower <pause> serviced

( 338) she had stairs knocked in, she’s had all spindles done up the stairs

As mentioned above, a service often entails payment.69 Money is another field which collocates with causative have and contributes to the semantic prosody of service. It is explicitly mentioned in some 12% of the constructions with have. Examples of words belonging to the field and recurring in the immediate environment of the verb are: quid, pounds, pay, cost, price, bank, capital, cheaper, worth. This is illustrated by the following sentences: (339) Gonna cost about sixteen quid to have it repaired.

(340) another cleaning offer, half price <pause> should we have these carpets cleaned? 69. It is perhaps this payment that establishes a hierarchical relationship in have constructions within which the causee cannot refuse to do what the causer asks of him (see Goldsmith 1984: 119, Wierzbicka 1998: 121) and where compliance is therefore taken for granted (Duffley 1992: 72). See also Hollmann (2003: 78), who claims that with have, the causer has the causee in his/her “sphere of control” and can therefore apply his/her authority in order to bring about the caused event.


(341) I mean the thing is it went wrong and they had to pay for it, okay it’s only a tenner to have it done, but it, if a CD player, I mean if the carriage goes, you know, sixty, seventy quid, you know I’m buying these electrical things, it’s, because it’s a burg lock, I can buy them, get them a warranty with it an’ all, then it’s took the burden off them.

To give an idea of the extent of the semantic prosody of service, Table 54 lists the most frequent and clearest collocates of [X have Y Vpp] that participate in this semantic prosody. The other two constructions, [X have Y Vinf ] and [X have Y Vprp], also have collocates that refer to a service (e.g. build, constructing, home, electrician, plumber, job) but these are less frequent, especially in the case of the present participle construction. Besides, the two constructions have other semantic prosodies in addition to the frame of service, which will be briefly discussed below. Table 54. [X have Y Vpp] and the semantic prosody of service cut 80 72 hair built 25 permed 23 house 20 shaved 12 11 cleaned changed 10 checked 10 fixed 10 serviced 9 tested 9 removed 8 pounds 7 repaired 7

work hospital knocked quid trimmed decorated door extension face painted replaced tiles altered bath blood

7 6 6 6 6 5 5 5 5 5 5 5 4 4 4

carpet disconnected fitted garage gas highlighted home kitchen legs money roof room waxed

4 4 4 4 4 4 4 4 4 4 4 4 4

Before discussing these additional semantic prosodies, however, it must be stressed that the semantic prosodies of effort/difficulty and service associated with causative get and causative have, respectively, merely represent preferences. These, however, may be overridden by other considerations. The sentence in (342), for example, refers to a causer commissioning somebody to do something for him/her against payment, which is the type of context in which have is normally used. However, this notion of service seems to be overridden by the notion of difficulty suggested by a hell of a lot of, which triggers the use of the verb get. Conversely, (343) mentions the existence of problems, which could lead to the use of get, but in this case the reference to a service and payment is apparently stronger and guides the choice of have.


(342) Well they would move them round <pause> this fella himself, but it’s gonna be a hell of a lot of money to get them to do that. (343) We are misled time after time after time when people have problems like they’re having their vehicle serviced, it’s put a claim in for a water pump, and have the cost of their service paid for.

This point may be further illustrated by examples (344) to (346), the first two of which come from ICE-GB. (344) Ironic, since fashion has gone full-circle and kids actually ask to have their hair cut short now. (345) Alternatively, they might rebel and become violently opposed to short hair, refusing to allow their children to get their hair cut. (346) I suppose I, I better go and get [my] hair done. - Yes, you’d better had, it’s twenty

All three examples refer to the field of hairdressing, typically associated with causative have. Yet, only the first sentence actually contains the verb have. The other two sentences use a form of causative get. The explanation for this is that only (344) presents the service of cutting hair as “a matter of course” (Palmer & Blandford 1969: 197) or a routine job (Celce-Murcia & Larsen-Freeman 1983: 481). Additionally, the causing event in this sentence involves communication (kids ask to have their hair cut short), which linguists like Brugman (1996: 33) or Wierzbicka (1998: 120–121) view as characteristic of causative have. In (345), on the other hand, there is resistance on the part of the parents, as appears from the words rebel, violently opposed and refusing, and hence difficulty for the children to perform the caused event. In (346), two elements may be said to influence the choice of get. First, the temporal indication (it’s twenty), combined with the expression I better, suggests some urgency to act. Second, the context reveals that I is both the causer and causee of the sentence (a case which, it will be remembered from Section 4.3.5, favours the use of get) and that no service is therefore involved. Interesting too and particularly enlightening is example (347), where the speaker changes his/her mind as to what causative to use. The context here is one of service (booking an appointment), for which the most typical verb is have. And this is indeed the verb that the speaker first chooses. However, the notion of difficulty that is implied in the sentence (if I could, if I can), as well as the co-referentiality between the causer and the causee, justifies the use of get – and the speaker’s subsequent correction. (347) Probably to the one at Staffhill if I could have an, if I can get an appointment booked.


While the semantic prosody of service is also present in [X have Y Vinf ] (though to a lesser extent than in [X have Y Vpp]), the infinitive construction exhibits another semantic prosody which, albeit not particularly common either, is worth mentioning. This is the semantic prosody of fiction, as exemplified by (348) and its collocates plot and fiction. Other collocates participating in the same semantic prosody are: Dostoevsky, Snoopy, books, notebooks, page and reading. Although these collocates have a low frequency of one or two occurrences only, taken together they point to a semantic field which is to be related to the depictive use of [X have Y Vinf ] (cross-space causation, see Section 5.3.4) and is therefore relevant to the description of the construction. (348) Among other casualties in this process will be plot: “the plot having disappeared, it is no longer necessary to have the events of fiction follow a logical, sequential pattern (in time and in space)” (Federman 1975: 810).

As for [X have Y Vprp], which is even less common with the semantic prosody of service, it appears to collocate with words referring to machinery (see Table 55), as in the following examples: (349) you better not have that tape working, is it on?

(350) They have the batteries going and all.

Table 55. [X have Y Vprp] and the semantic prosody of machinery working work fridge playing running automatic batteries

7 3 2 2 2 1 1

boiler camera computer cooker dishwasher ferry ironing

1 1 1 1 1 1 1

lamp machine phone recording springs tape van

1 1 1 1 1 1 1

Interestingly, this semantic prosody is shared by [X get Y Vprp], although in the latter case the semantic prosody of effort/difficulty usually remains quite strong. Table 56 lists some of the collocates of [X get Y Vprp] referring to the semantic prosody of machinery, and a couple of examples are given in (351) and (352). Notice, in (351), the idea of difficulty suggested by old and see if I can. (351) Yeah, probably if you want me to get that old mower going I ought to go up to Woods and <pause> see if I can get a new drive belt. (352) No, they’ve got a st—they’ve got the diggers moving backwards and forwards, yeah.


Table 56. [X get Y Vprp] and the semantic prosody of machinery tape running moving working machine

8 7 4 4 3

printing recorder diggers earphones mower

2 2 1 1 1

play playing ringing television wire

1 1 1 1 1

7.3.3 Make constructions As is the case with the constructions dealt with in the preceding section, the strength of semantic prosodies with causative make tends to vary with the frequency of the construction. Although the three causative constructions with make appear to share the same semantic prosody, it is stronger with [X make Y Vinf ] than with [X be made Vto-inf ] and [X make Y Vpp]. The results given in this section, therefore, will mainly come from the collocational analysis of [X make Y Vinf ]. A close look at the collocates of causative make reveals the existence of two main semantic fields, which sometimes overlap with one another, namely the field of the physical self (bodily experiences, physiological processes, appearances, etc) and that of the mental self (inner sensations, feelings, qualities, etc). The former includes for instance appearances, head and bleed, as in (353) and (354), and the latter includes wonder, feel and guilty, illustrated in (355) and (356). Example (357) combines the two fields by referring to the body and the inner sensation of suffering. (353) Henna for their hair and erm, the men put oil on their bodies to make themselves look better, you know, more attractive to the female’s and, so I think erm, we’ve always strived to improve ourselves our appearances. (354) scratching his bloody head, and he’s made his head bleed. (355) I wonder if it [too much vitamin C] makes your brain go doolally though (356) There appeared no reason to suspect suicidal intent; rather, the therapist understood the overdose as being intended to evoke sympathy from Ann, to make her feel guilty about leaving home and to put pressure on her to return. (357) First, it meant that “no man [was] punishable or [could] be made to suffer in body or goods except for a distinct breach of law established in the ordinary legal manner before the ordinary courts of the land”.


Table 57 focuses on [X make Y Vinf ] and lists the most frequent collocates referring to what may be summarised as “bodily and inner sensations”. The list is limited to the collocates occurring five times or more, but it is important to bear in mind that less frequent collocates also include many such words, for example four occurrences of heart, ideas and love, three occurrences of awkward, clever and feeling, two occurrences of attractive, disease and friendly, and one occurrence of hope, insane and looks. These are just a few examples. The whole list of collocates contains dozens of them. With the other two constructions, the collocates referring to bodily and inner sensations are less frequent, but still present. Examples for [X be made Vto-inf ] include feel (seven occurrences), seem (four occurrences), look (three occurrences), believe (two occurrences) and want (one occurrence), whereas examples for [X make Y Vpp] include known (nine occurrences), felt (five occurrences), understood (two occurrences) and opinions, silent, views, wishes (one occurrence each). Table 57. [X make Y Vinf ] and the semantic prosody of bodily and inner sensations feel 160 laugh 154 look 105 think 91 know 70 wonder 37 appear 30 seem 27 want 21 sick 16 see 15 jump 14 sound 14

nice 12 thought 12 cry 10 head 9 wanna 9 guilty 8 grow 8 old 8 ache 7 cold 7 cough 7 sleep 7 sneeze 7

uncomfortable cringe die drop fall funny smile eyes mind realize tired

7 6 6 6 6 6 6 5 5 5 5

The collocates of make show that this causative is oriented towards the human being in all his/her dimensions (mental as well as physical), describing as it does the different ways in which s/he may be affected. It may even be added that this human being is predominantly the speaker, as appears from the unusually high frequency of the collocate me. It represents a frequency of 348 occurrences with make, while the agentive form I occurs 225 times. By contrast, causative get and have prefer the company of the agentive form I (517 and 274 occurrences, respectively) and exhibit a relatively low frequency of me (76 and 41 occurrences, respectively). With cause, the list of collocates merely includes five occurrences of the form I. One may argue, therefore, that while get, have and make, unlike cause, are all speaker-oriented, the first two causatives mainly consider the speaker in


his/her agentive role, whereas make tends to focus on his/her affected role (see also Sections 5.3.3 and 5.3.4 on the analysis of the grammatical person of the causer and causee, the results of which go in a similar direction).

7.4

Word clusters in the corpus data

While both collocational analysis and word cluster analysis deal with the preferential lexical company kept by a word, the former investigates the individual items surrounding the target word, whereas the latter studies recurrent strings of items including the target word. Word cluster analysis therefore makes it possible to highlight more fixed expressions than is the case with collocational analysis. The word cluster analysis performed by means of WordSmith Tools on the causative constructions shows that two-word clusters are mostly irrelevant (e.g. makes me, get it, to have) and that very long word clusters are either nonexistent (the longest clusters with cause consist of three words, and those with get consist of five words) or of little interest (the eight-word, seven-word and, to a very large extent, six-word clusters with have and make correspond to lyrics repeated several times, e.g. [358]). (358) buy you diamond ring my friend if it makes you feel alright <pause> I’d get you everything my friend if it makes you feel alright <pause> I don’t care too much for money, money can’t buy me love <end of voice quality>

Therefore, only three- to five-word clusters are considered here. The discussion of the results of the word cluster analysis will remain brief for two reasons. First, even though we restrict ourselves to three- to five-word clusters, there are still a number of word clusters that say very little about the construction itself. Second, some word clusters simply support the conclusions drawn on the basis of the collocational analysis and therefore do not really offer any new insights. What this section does is outline the main results of the analysis and focus on the few word clusters that seem to emerge from the data as formulaic expressions, likely to be stored holistically in the mental lexicon (see Wray 2002). The only word cluster that is of some potential interest with [X cause Y Vto-inf ] is will cause the, which is the most frequent word cluster including cause, with a frequency of seven occurrences. This word cluster is of some interest because it appears to refer to (future) predictions that are part of the description of an experiment, as in (359). This may be linked to the genre in which the construction typically occurs, namely scientific writing (see Section 9.2.2).


(359) If keys are greater than 155 bytes in length, the track index will exceed one full track in extent, and maximum length keys of 256 bytes will cause the cylinder index to take up as much as 28 per cent of the second track on the cylinder as well.

The word clusters involving causative get confirm the pervasiveness of the semantic prosody of effort/difficulty brought to light by the collocational analysis. With each of the three get constructions (but most strikingly with [X get Y Vpp]), a number of word clusters are found that emphasise this idea, e.g. trying to get (14 occurrences in [X get Y Vto-inf ] and 10 occurrences in [X get Y Vpp]), try and get (13 occurrences in [X get Y Vpp] and 11 occurrences in [X get Y Vto-inf ]), get it all (15 occurrences in [X get Y Vpp]), I can’t get (3 occurrences in [X get Y Vpp]), if you get (4 occurrences in [X get Y Vprp]). Some more elaborate (but less frequent) word clusters are: trying to get him to, got it all sorted out (or worked out) and see if I can get. Examples are given in (360) to (363). (360) I’ll try and get Bob to find the <pause> the number of it. (361) I don’t know whether they’ll have managed by Christmas to get it all done. (362) Well, it’ll be nice if you get a club going at the youth club, because they won’t use it in the morning (363) we were trying to get him to sell the Guinness fifty P a pint because it was Saint Patrick’s day

The results for [X get Y Vprp] contain another word cluster worth mentioning, namely get it going (13 instances of get it going and 6 instances of got it going), as illustrated by (364) and (365). It may be related to the formulaic expression get somebody going (Kirchner 1952: 225n), e.g. (366) – although it should be emphasised that in almost 80% of the occurrences of [X get Y going], the causee is inanimate, thus reflecting the general tendency of the construction to prefer inanimate causees (Section 5.3.4). ( 364) You’re not making a fortune at first, but once you get it [bar] going (365) the <pause> erm boy didn’t feed him properly so it’s all sort of skin and bone <pause> so we went and brought his pony and erm, we sort of fed it up and got it going again. (366) what got me going really was the fact that he had to go down and actually <pause> do a design of the


Another formulaic expression that is not particularly frequent in naturally-occurring language is get the ball rolling, which is often listed in idioms dictionaries (e.g. Cullen et al. 2000: 8), but occurs only once in BNC-10: (367) What, what he’s, what he’s had to say is, is, is to, is to get the ball rolling and get some money, and then we can start looking.

The most striking observation in the list of word clusters for [X have Y Vpp] is the recurrence of phrases having to do with hairdressing, which corroborates the results of the collocational analysis. These word clusters include have it cut (17 occurrences), had it cut (16 occurrences), had his hair cut, she’s had it permed and to have her hair done (4 occurrences each). Compare: (368) Yeah my hair was that long when I had it cut.

(369) Oh, cos Joelle was trying to, trying to work out if she can afford to have her hair done and go to the cinema.

[X have Y Vprp] does not give rise to any word cluster with a frequency of three or over. As for [X have Y Vinf ], its word clusters include have a dream come true, which corresponds to lyrics appearing several times in the corpus. More significantly, they also include two word clusters which look like formulaic expressions, viz. I’ll have you know (8 occurrences) and would have us believe (3 occurrences), as illustrated by (370) and (371) respectively. I’ll have you know usually takes the form of an interpolated clause, as in (370), but sometimes it is integrated into the main clause, as in (372). As for would have us believe, it is made up of the auxiliary would meaning “would like to”, followed by causative have.70 What is common to these two word clusters is that they contain a non-volitional effect (know and believe), a type of verb which, as was shown in Section 5.3.5, is more typical of make than of [X have Y Vinf ]. (370) I did twenty six thousand I’ll have you know!

(371) Now, we didn’t have a recession in nineteen eighty six, but I would suggest that augers rather better than the mongers of gloom and doom would have us believe for nineteen ninety one. 70. Opinions differ as to whether the use of have preceded by the auxiliary would should be considered causative. Jespersen (1940: 287), Van Ek (1966: 82) and Duffley (1992: 72) posit that have in this case is synonymous with want (or would like) and “no longer evokes an effectual exercise of control” (ibid.). By contrast, Poldauf (1967: 39) and Andersson (1985: 150) both suggest that have is still causative in meaning when preceded by would. For Visser (1973: 2266), the explanation for this particular use lies in the meaning of would which, in this case, is not an auxiliary, but a notional verb standing for “want, wish”.


(372) I’ll have you know I’ve got a full frigging set of it.

It should also be pointed out that both get and have constructions contain vague word clusters such as get something done (9 instances), get things done (5 instances), get somebody to (3 instances) or have it done (39 instances), as illustrated in (373) and (374). This vague language is to be related to the register in which these constructions are most commonly found, namely spontaneous conversation (see Channell 1994 on vague language and its association with informal conversation). (373) I thought ooh, I’ll get something done in the garden today. (374) and he’s got to pay forty five quid to have it done [= re-connect the telephone].

The list of word clusters for [X make Y Vinf ], finally, reveals a large number of strings of words that contain the same effect, but differ in the form of the causative or the causee. Thus, fourteen word clusters with a frequency of three or more occurrences are a variation of [X make Y feel], e.g. makes you feel (26 times), make you feel (17 times), made me feel and make me feel (15 times each) or makes me feel (9 times). Similar clusters arise with verbs such as laugh or look, and show that what is of real interest here is not so much the word cluster itself, but the recurrence of the effect, which will be studied in Chapter 8. Other word clusters, however, deserve particular attention, because they seem to be good candidates for the status of formulaic expressions. One of them is makes you wonder (27 occurrences). It may take the pronoun it as a subject (16 instances of it makes you wonder), as in (375), but the subject is often ellipted, as is the case in (376). While it may be followed by a subclause, it may also be used on its own (compare [376] with [377]). (375) it makes you wonder really, really where they get the background for planning doesn’t it? (376) Makes you wonder how much erm ours is worth, not as much as that obviously but er (377) Gracious, makes you wonder doesn’t it?

Another interesting word cluster is make do with (8 instances), e.g. (378), which is sometimes preceded by have to (3 instances), as in (379). It is often regarded as a fixed expression (see Strang 1968: 171), due to the fact that it is one of only two [X make Y Vinf ] constructions that may be used with no causee.71 71. The other one involves the verb believe, but is not represented in the list of word clusters.


( 378) He is requested to make do with a sum of money. (379) Not nice fat rolls like I usually get, he’ll just have to make do with this won’t he?

Similarly, but for other reasons (having to do with its metaphorical meaning), make ends meet is also considered a fixed expression, listed, for instance, in the Chambers Guide to Idioms (Cullen et al. 2000: 55). It is present in the corpus, four times as make ends meet, and three times as making ends meet and to make ends meet. An example is given in (380). ( 380) And we’re really having to struggle to sort of make ends meet.

7.5

The co-textual approach in the elicitation data

The results of the elicitation test confirm several of the findings made on the basis of the corpus data. Technical and scientific terms triggered the use of cause rather than the other causatives, both in exercise A.9 (381) and in exercise E.6, where the subjects were required to join the elements “the explosion”, “the temperature” and “rise”. In the first case, 88% of the informants chose the verb cause, and in the second case over 90%. (381) Destruction of part of the cortex [makes/causes/gets/has] cells in a corresponding part of the thalamus (to) die.

The idea of difficulty also seems to be quite well associated in people’s minds with the verb get. When asked to complete test item A.4 (382), which contains the collocate try to, 58% of the respondents opted for get (as against 36% for make and 6% for have). (382) She is trying to [make/get/cause/have] us (to) be friends with her.

It is also relevant that, in exercise E.7, where the informants had to join the elements “the woman”, “the car” and “start”, many of those who chose the verb get spontaneously added clues indicating effort or difficulty, e.g. the auxiliary could not (added by ten of the twenty-eight informants who opted for get), was able to or by pushing it. This is to be contrasted with the situation for make, with which one of the only additions was used her keys to, which evokes no particular problem. The elicitation test also included two items referring to the frame of service, namely A.3 and A.5, reproduced here as (383) and (384).


(383) It was difficult to find someone to do it, but I finally [had/caused/got] the chimney (to be) swept. (384) My sister [got/had] her hair cut for £10.

The first test item has this peculiarity, that it exhibits two semantic prosodies, normally associated with different causatives, namely the reference to a service (sweep the chimney), typical of have, and the idea of difficulty (it was difficult to, finally), typical of get. In this case, a small majority of the informants (52%) preferred the verb get, which seems to indicate that for them the idea of difficulty overrode the frame of service. In (384), which describes a service and makes explicit reference to payment, 58% of the subjects chose have, but 4% accepted both have and get. Finally, test item B1.5 made it possible to investigate the degree of entrenchment of the idiom get the ball rolling (i.e. the degree to which it is stored as a routinised unit in the mind, see Schmid 2007), by asking the informants to supply a subject to the phrase … got the ball rolling. It will be remembered from the preceding section that although this expression is included in some idioms dictionaries, it occurs only once in the corpus (367). Yet, despite this low frequency, it appears from the elicitation test that this expression is quite well entrenched, since only eleven of the forty-nine informants who provided an answer to this question completed the sentence with a causer suggesting a literal interpretation of get the ball rolling, e.g. The footballer got the ball rolling. By contrast, in twenty-five cases the sentence had clearly been interpreted as an idiom, e.g. The first speaker got the ball rolling. This discrepancy between elicitation and corpus data, however, should not come as too much of a surprise, since Hoffmann & Lehmann (2000: 24) have demonstrated that “very infrequent exposure is enough to establish familiarity”.

7.6

Summary

This chapter has shown that, while this is a neglected area in the literature, the lexical aspect of periphrastic causative constructions is an important one. More precisely, it appears that causative verbs keep strong preferential lexical company and tend to occur in “patterns of repeated phraseology” (Scott 2008). Unlike the semantic and syntactic analysis of Chapter 5, which has underlined the specificity of each individual construction, the collocational analysis has revealed that semantic prosodies are largely valid across the different constructions related to one and the same causative verb (although to varying degrees). Thus, the semantic prosody of bodily and inner sensations characteristic of [X make Y Vinf ] also applies to [X be made Vto-inf ] and [X make Y Vpp], although the association is weaker with these two constructions, probably as a result of their lower frequency.


The collocational analysis, however, has brought to light important differences between the causatives, especially between get and have, which could not be distinguished properly on the basis of the semantic and syntactic features alone. It now turns out that what discriminates best between the two verbs is their semantic prosodies, namely the semantic prosody of effort and difficulty with get, and the semantic prosody of service (often against payment) with have. Finally, although the lexical preference is weaker in this case, [X cause Y Vto-inf ] appears to be attracted to more technical and scientific terms. All these results are confirmed by the word cluster analysis, which also suggests potential candidates for the status of formulaic expression, likely to be stored holistically in the mental lexicon, for example makes you wonder, make ends meet, get it going or I’ll have you know. Overall, the results of the elicitation data corroborate the corpus analysis. It is interesting to note that some of the results presented in this chapter seem to provide an explanation for findings made earlier. To give but a couple of examples, the association of get with the semantic prosody of difficulty may explain why [X get Y Vpp] is often chosen to express identity between the causer and the causee, since, as demonstrated in Section 4.3.5, such co-referential constructions normally involve the Subject’s overcoming of the Self ’s reluctance to act. As another example, consider the recurrent expressions I’ll have you know and would have us believe. It was noted in Section 5.3.5 that [X have Y Vinf ] was surprisingly frequent with non-volitional effects. It now turns out that this is mainly due to the presence of word clusters with know and believe. The next chapter will also help explain some of the characteristics highlighted earlier. Like this chapter, it adopts a fundamentally corpus-driven approach and examines the lexical dimension of causative constructions. The focus, however, will now be on the effect slot of causative constructions, which will be investigated by means of the technique of multiple distinctive collexeme analysis.

chapter 8

Collexemes in the effect slot

It was demonstrated in the hierarchical feature selection of Chapter 5 that the non-finite verb occurring in the effect slot of causative constructions is of paramount importance in the choice of a construction. It therefore deserves a chapter in its own right, dealing with the lexical preferences displayed by the effect. For this purpose, the method of automatic extraction of collocates used in the preceding chapter is not suitable, since it does not distinguish between the collocates occurring in the effect slot and those occurring elsewhere in the sentence. Therefore, another method will be employed here, namely “collostructional analysis”, and more precisely “multiple distinctive collexeme analysis”, a technique aimed at investigating the interaction between a construction and words in a given slot of this construction. By comparing the different constructions, the technique makes it possible to highlight the (classes of) verbs that are most distinctive for one or the other construction (its “collexemes”), and hence to say something about the typical meaning of each construction. Following the Corpus-Cognition Integrated model, this corpus analysis will be supplemented by an examination of the relevant elicitation data.

8.1

Periphrastic causative constructions: Always safe?

Considering that little attention has been paid to the collocates of periphrastic causative constructions in general (see Section 7.1), it is not very surprising that the question of whether causative constructions exhibit any lexical preferences with respect to the effect slot has been largely ignored in the literature. It is widely accepted, if only implicitly, that periphrastic causative constructions are “always safe” (Stocker 1990: 61) and that causatives such as cause, get, have and make may therefore be combined with any verb to express causation. Paradoxically, the few linguists who do attempt to describe the effect slot in terms of lexical preferences tend to focus on the least frequent construction, [X make Y Vpp]. According to Van Ek & Robat (1984: 327), the non-finite verbs occurring in this construction denote “the exercise and recognition of influence in the widest sense”, a hypothesis which seems to be confirmed by the lists of possible verbs provided by other linguists (Van Ek 1966: 118, Van Roey 1982: 84,


Berland-Delépine 1990: 173): established, felt, heard, known, obeyed, respected, understood and valued. Very little is known about the effect of the other constructions. Van Ek (1966) discusses the effect of all the ten patterns investigated here, but in most cases his description is very general, referring to broad categories like “activity” or “state”, rather than to specific lexical items. Moreover, he makes no distinction between the causative and non-causative uses of the patterns. This is not a problem for a pattern such as [X make Y Vinf ], which is not ambiguous and for which the verbs mentioned (feel, laugh, look, seem, smile and think) can therefore only concern the causative construction. For a pattern such as [X get Y Vprp], on the other hand, the claim that verbs of motion such as coming or going are common should be treated with caution here, since this lexical preference could be limited to the non-causative uses of the construction.72 Another attempt worth mentioning is Francis et al.’s (1996: 306) list of frequent noun phrases and verbs in [X get Y Vpp] and [X have Y Vpp], reproduced in Table 58. It should be pointed out, however, that no real distinction is made between the two constructions, with only seven phrases presented as solely characteristic of have (in bold in the table). Table 58. Frequent noun phrases and verbs in [X get Y Vpp] and [X have Y Vpp] (Francis et al. 1996: 306)

have a limb amputated get/have something overhauled get/have your teeth capped have someone paged get/have a job costed get/have your hair permed get/have your hair cut get/have your ears pierced get/have your house decorated get/have something printed get/have your windows double-glazed have your stomach pumped have a tooth extracted get/have something remade get/have a prescription filled get/have something repaired get/have something fixed get/have your house rewired have someone followed get/have your car serviced get/have yourself immunized have someone tailed have a boil lanced get/have yourself vaccinated get/have something made get/have something valued get/have something mended get/have your legs waxed get/have an animal neutered get/have something printed up get/have a job costed out get/have a tooth taken out get/have your house done up get/have a washing machine plumbed in

72. In fact, most of the examples given by Van Ek are of the experiential type, e.g. But I suppose you get dozens of women throwing themselves at your head.

Chapter 8. Collexemes in the effect slot 195

Apart from these patchy indications, nothing is known about possible lexical preferences in the effect slot of periphrastic causative constructions, which reinforces the feeling that any verb may occur in these constructions and that the distribution of these verbs over the constructions is purely random. As will appear from the results of the multiple distinctive collexeme analysis, however, this is far from being the case, as the effect slot in periphrastic causative constructions shows strong lexical preferences, which tend to be specific to each construction.

8.2 Collostructional analysis and the technique of multiple distinctive collexeme analysis Collostructional analysis is a set of quantitative corpus methods developed by Gries and Stefanowitsch, and relying on the notion of collocation and the framework of construction grammar. Its aim is to measure the association strength between a construction (in the construction grammar sense of the term) and the lexical items occurring in one (or more) slot(s) of this construction. Note that the idea of slot is what distinguishes collostructional analysis from collocational analysis. While collocational analysis examines all the words occurring within a specified span around the node, collostructional analysis limits its investigation to the words occurring in a particular slot of a construction. Collostructional analysis can therefore be seen as an extension of collocational analysis, taking account of the various types of grammatical constructions recognised by construction grammarians (see e.g. Croft 2001: 17). “Collexemes” refer to those lexical items that are attracted to a particular construction, while “collostructs” correspond to those constructions that are associated with a given lexical item. A “collostruction” is a combination of collexeme and collostruct. Collostructional analysis includes three different techniques:

i. collexeme analysis, which studies one slot in a particular construction, e.g. the V slot in the [X think nothing of Vgerund] construction (see Stefanowitsch & Gries 2003); ii. (multiple) distinctive collexeme analysis, which studies one slot in two (or more) similar constructions, e.g. the verb in the ditransitive and todative constructions (see Gries & Stefanowitsch 2004a); iii. covarying collexeme analysis, which studies the interaction between two slots in a particular construction, e.g. V1 and V2 in the [X V1 Y into V2gerund] causative construction (see Gries & Stefanowitsch 2004b).


Each of these techniques relies on statistical calculation which takes into account the frequency of occurrence of a particular word in a given construction, but also the word’s frequency in other constructions and the construction’s frequency with other words (Gries et al. 2005). The calculation involves a measure of statistical significance, which makes it possible to distinguish between significant attraction (or repulsion) between a word and a construction on the one hand and random co-occurrence on the other. Collostructional analysis thus provides results which are “more comprehensive and more precise” (ibid. 652) than those obtained by means of a simple frequency analysis. Gries et al. (2005, forthcoming) also show on the basis of experimental data that collostructional analysis is psycholinguistically more reliable, outperforming frequency as a predictor of speakers’ behaviour, both in terms of production (sentence-completion tasks) and comprehension (reading tasks). The technique of multiple distinctive collexeme analysis is particularly appropriate for the study of periphrastic causative constructions, as it allows the comparison of several related constructions and the identification of the words that are most distinctively associated with each of them (the “distinctive collexemes” of the construction). In addition, it can safely be used to analyse the ten causative constructions under investigation since, as is required by the technique, these constructions are functionally and semantically near-equivalent. They all result from the process of causative formation (Siewierska 1991: 28), they involve the same participants at the conceptual level (Section 4.2), and they can be said to share the same basic meaning, namely that of indirect causation (as opposed to lexical causative verbs, which arguably express more direct causation, see among others Chomsky [1972: 72n], Baron [1974: 327], Kemmer & Verhagen [1994: 121] and Wolff [2003]).

8.3 Lemma-based vs. sense-based approach Typically, collostructional analysis relies on lemmatised lists of words occurring in one (or more) slot(s) of one (or more) construction(s). No distinction is made between the different inflected forms of a word or between its different senses. In the case of periphrastic causative constructions, the form of the verbs occurring in the effect slot has to be disregarded, as taking this into consideration would preclude comparison between constructions taking a different type of non-finite complement (e.g. [X make Y Vinf ], [X get Y Vpp] and [X get Y Vprp]). Word


sense, on the other hand, is not only a possible, but also a desirable addition to the input data, as will be shown presently.73 While words (or lemmas) are commonly taken as basic units in the sentence, some linguists recognise a more refined level of analysis, one where a word is associated with a particular sense. The verb see, for instance, corresponds to one word, but several senses, viz. “use one’s eyes”, “understand”, “consider”, “meet”, etc. Cruse (1986: 49) calls “lexical unit” the combination of a word with a specific sense. The relevance of lexical units, as opposed to words, as basic linguistic units is supported by several approaches. It is one of the central tenets of lexical semantics that the different senses of a word may behave differently from a syntactic point of view. Levin (1993) shows that, as a verb of contact, touch belongs to the “touch” verbs, together with verbs like caress, pat, prod or stroke, and as such, cannot be used with a resultative phrase (*She touched the window open). As a psych-verb, on the other hand, it belongs to the “admire” verbs with e.g. adore, like, treasure or worship, a class of verbs which do allow resultative phrases (It touched me to tears). In the Firthian contextual theory of meaning, it is assumed that “the formalisation of contextual patterning of a given word or expression is (…) relevant to the identification of the meaning of that word or expression” (Tognini-Bonelli 2001: 4). In other words, different senses of a word tend to occur in different (co-)textual environments (e.g. declare in the sense of “state clearly” in the Brown Corpus and in the sense of “authorise payments” in the Wall Street Journal, see Hare et al. 2004)74 and with different collocates (Yarowsky’s [1993] “one sense per collocation” principle), e.g. words such as life and animal for plant in the sense of “living thing” and words such as manufacturing, equipment and employee for plant in the sense of “factory” (Yarowsky 1995). The relevance of lexical units has been emphasised in the field of psycholinguistics too, with scholars like Hare et al. (2003) or Roland & Jurafsky (2002) suggesting that subcategorisation information is encoded in the mind in relation to verb senses, not verbs in the aggregate.

73. See also Gilquin (in preparation), on which the analysis presented in this chapter is based. For a lemma-based multiple distinctive collexeme analysis of causative constructions, see Gilquin (2006b). 74. See also Gale et al. (1992), who claim that there is a very strong tendency for multiple uses of a word to share sense in the same discourse. Krovetz (1998), however, claims that, outside encyclopaedia articles, which form the basis of Gale et al.’s study, it is not uncommon for words to be used in the same document with different senses.


There is a good case for arguing that sense is relevant to collostructional analysis as well, at least in the case of periphrastic causative constructions. To illustrate this, consider the following pair of sentences: (385) Right get that tape going Paulie, quick.

( 386) Well I suppose that’s what made him go unconscious really.

While the same verb is used in both constructions, the lexical units are different – go “operate” in (385) and go “become” in (386). In a lemma-based collostructional analysis, this difference would go unnoticed, since the various senses of a verb are lumped together. In a sense-based analysis, on the other hand, it might be possible to show that, say, [X get Y Vprp] is strongly associated with the “operate” sense of go, whereas [X make Y Vinf ] is more attracted to the “become” sense of go. Because such sense differences are potentially significant, a sensebased approach will be preferred to the usual lemma-based approach, and verb senses, rather than verb lemmas, will form the basis of the multiple distinctive collexeme analysis.

8.4 Methodology Collostructional analysis involves line-by-line inspection of the concordances and manual identification of the slot one is interested in. In the present case, this coding was performed as part of the database creation (see Section 5.2), resulting in a list of lemmas occurring in the effect slot of the ten different periphrastic causative constructions. In accordance with the sense-based approach advocated above, these lemmas were then tagged for sense. The senses were taken from the online version of the Cambridge Advanced Learner’s Dictionary (CALD), where different senses of a word have different entries and are summarised by a guideword in upper case.75 Semantically tagging the data involved selecting the appropriate sense of the verb on the basis of the context and appending the corresponding

75. The online CALD does not make any clear distinction between related and unrelated senses. Thus, say in the sense of “speak” and say in the sense of “give information” have separate entries, as do start in the sense of “begin” and start in the sense of “move suddenly”, although the senses are clearly more related in the former case than in the latter. Note that phrasal verbs were considered as different types and hence tagged individually. Phrasal verbs are those indicated as such in the online CALD, including not only verbs like sort out or write down, but also verbs like laugh at or wait on. Spelling variants such as realise and realize or co-operate and cooperate were grouped together.


guideword to it (e.g. go_operate, go_become). Sometimes the verb had only one sense in the CALD (e.g. seem or sneeze) and was therefore not tagged. The analysis itself was performed by means of Coll.analysis 3, a program for the R statistical package developed by Gries (2004) and suitable for the different techniques belonging to the family of collostructional analysis (see Section 8.2). For a sense-based multiple distinctive collexeme analysis, the program requires as input a file listing the constructions in one column and the verb senses occurring in their effect slots in the other column, with one line per corpus example. On this basis, it calculates the observed frequency of each verb sense in each construction, as well as its expected frequency (the frequency that would be expected if verb senses were distributed in a way corresponding to the frequencies of the constructions), and performs a binomial test to establish the probability of a particular observed frequency given the expected frequency. This probability is then log-transformed and receives a positive sign when the verb sense occurs more frequently than expected in the construction (i.e. is attracted to the construction) and a negative sign when it occurs less frequently than expected (i.e. is repelled by the construction).76 The resulting value, the “distinctiveness value”, indicates the degree of association between the verb sense and the construction: the higher the absolute value, the stronger the association strength, with a threshold level of statistical significance of 1.30103 (p she stood up and she said, I’m gonna give my talk about cats <pause> so I said, fine. (395) Not those kind of relaxants but something just to relieve the tension and make her feel calmer. (396) But the Hillsborough Agreement made Mrs Thatcher look credible and constructive over Northern Ireland in the eyes of world opinion, and especially in the United States.


Table 62. Distinctive collexemes for [X make Y Vinf ] Collexeme (n) 1 laugh (110) 2 feel_experience (130) 3 look_seem (82) 4 wonder_question (31) 5 think_consider (31) 6 appear_seem (26) 7 seem (28) 8 go_become (19) 9 want_desire (15) 10 think_reason (21) 11 do_acceptable (14) 12 work_effect (21) 13 sound_seem (11) 14 jump_suddenly (10) 15 think_imagine (11) 16 realise (8) 17 feel_opinion (12) 18 happen_existence (12) 19 meet_touch (8) 20 ache (6) 21 ask_question (6)

Distinctiveness 53.86 51.78 34.79 15.71 13.30 10.91 10.25 9.63 7.60 7.15 7.09 6.59 5.57 5.07 4.64 4.05 3.86 3.86 3.24 3.04 3.04

Collexeme (n) 22 cringe (6) 23 cough (5) 24 last_continue (5) 25 refer_describe (5) 26 smile (5) 27 sneeze (5) 28 worry_problem (4) 29 have_possess (5) 30 cry_tears (7) 31 wait (7) 32 say_speak (14) 33 drop_fall (3) 34 go_say (3) 35 have_experience (3) 36 look_see (3) 37 see_understand (3) 38 sleep_resting state (3) 39 stand out_noticeable (3) 40 die_stop (5) 41 vanish (5) 42 conform (4)

Distinctiveness 3.04 2.53 2.53 2.53 2.53 2.53 2.03 1.89 1.88 1.88 1.54 1.52 1.52 1.52 1.52 1.52 1.52 1.52 1.47 1.47 1.45

What is common to these three verbs is that they refer to processes which are not directly dependent on the causee, which confirms the association of [X make Y Vinf ] with non-volitional verbs (Section 5.3.5). In (395), for example, the causee’s feeling calmer is not under her control (although the taking of medicine is). Non-volitional collexemes have been highlighted in Table 62. It will be noticed that they represent a large majority and that the most distinctive collexemes are all of this type – which rules out coercion as a distinctive meaning of the construction. Different subclasses may be distinguished among these nonvolitional collexemes, the most prominent ones being verbs of impression (e.g. look_seem, appear_seem and seem), verbs of mental processes (e.g. think_consider, wonder_question and realise) and verbs of physiological processes (e.g. ache, cough and cry_tears), as illustrated by (397), (398) and (399), respectively.81

81. Of these three subclasses, only the first one (verbs of impression) may also be used with inanimate causees, as in (397).


(397) This made the accident appear reasonable, something which even they could have done. (398) It makes you think of summer dunnit?

(399) Oh poetry <pause> <sigh> some of those poems that Bon had in her Touchstones book, especially the ones, the war, about the war, I read some of those and it made me cry, it was so sad.

To some extent, [X be made Vto-inf ] is characterised by non-volitionality too (see also Section 5.3.5). This can be seen from Table 63, where non-volitional collexemes are in bold. However, the proportion of such collexemes is much smaller than is the case with [X make Y Vinf ], and their distinctiveness values are, as a rule, lower. Next to processes taking place independently of the causee’s will, as in (400) and (401), the construction is also associated with collexemes that do involve volition, e.g. (402) and (403). (400) Innocent wives and children and other dependents are made to suffer when the state imprisons thousands of working-class men for crimes which are often insignificant compared with corporate crimes. (401) One law for the rich and another for the poor, as the two systems can be made to seem, are laid down together in a book which commemorates a desertion, on the author’s part, of the rich for the poor. ( 402) Why shouldn’t, why shouldn’t we be made to pay for a health service? (403) There are people out there who’ve got problems that are bigger than the average household can deal with and erm I think, therefore, we’re looking for more institutionalized forms of help and folks like this particular character -I almost said his name -- almost need to be made to accept that kind of help.

This should not be taken to mean that [X be made Vto-inf ] predominantly expresses coercion, despite some linguists’ claim that the construction is merely acceptable in such cases (e.g. Babcock 1972: 32, Mittwoch 1990: 113). Coercion, or at least interpersonal manipulation, is possible, as appears from (402) or (403), but it is not very common. It should also be underlined that most of the collexemes in Table 63 which potentially involve volition, such as account for_explain, adhere to or cooperate, often occur in the data with inanimate causees, which excludes volition, and hence coercion: (404) J. B. Watson in 1914 and other behaviourist thinkers (Hull, Skinner) believed that an acquired behaviour element, the conditioned reflex for example, could be made to account for all behaviour, because such an element could be


Table 63. Distinctive collexemes for [X be made Vto-inf ] Collexeme (n)

Distinctiveness

1 mean_express (3) 2 perform_do (3) 3 pay_buy (5) 4 depend (2) 5 work_effect (4) 6 suffer_feel pain (2) 7 feel_opinion (3) 8 seem (4) 9 believe (2) 10 accept_take (1) 11 account for_explain (1) 12 adhere to (1) 13 allow for (1) 14 answer for_ responsible (1) 15 apply_relate (1) 16 argue_reason (1) 17 ascend_importance (1) 18 attend_notice (1) 19 beat_movement (1) 20 begin (1) 21 bring down_reduce (1) 22 carry_have (1) 23 contribute (1)

4.67 3.69 3.10 2.64 2.27 2.14 2.12 1.94 1.83 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56

Collexeme (n)

Distinctiveness

24 correspond_match (1) 25 cooperate (1) 26 disassemble (1) 27 enable (1) 28 encapsulate (1) 29 fail_not succeed (1) 30 give_stretch (1) 31 grasp_hold (1) 32 have_receive (1) 33 intensify (1) 34 observe_watch (1) 35 oscillate (1) 36 predict (1) 37 represent_describe (1) 38 rest upon_necessary (1) 39 reassemble (1) 40 sell off (1) 41 signify_mean (1) 42 stand for_represent (1) 43 swim_move in water (1) 44 yield_produce (1) 45 work_activity (2)

1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.56 1.37

treated as a “building block in theory much in the same way that nineteenthcentury physicists used “atoms” to build up a theory of matter.

(405) This paper discusses four respects in which processes embodying individual treatments must be made to co-operate with one another if anaphoric expressions are to be interpreted correctly.

Because these collexemes, therefore, do not really challenge the meaning of “causing a process that is not directly dependent on the causee”, most obvious with collexemes such as suffer_feel pain or seem, it is this meaning that will be said to be quite distinctive for [X be made Vto-inf ], with the caveat that the association is much looser than with [X make Y Vinf ]. In addition, it must be stressed that, generally speaking, the passive construction displays much less marked lexical preferences in the effect slot than its active counterpart, as appears from a comparison of the top distinctiveness values (53.86 for [X make Y laugh] and 4.67 for [X be made


to mean_express]). In fact, of all the causative constructions investigated here, [X be made Vto-inf ] is the one that has the lowest distinctiveness values overall.

8.5.4 [X get Y Vpp] and [X have Y Vpp] As already mentioned, the get and have past participle constructions are often equated with one another in the literature, and it was demonstrated in Chapter 5 that they share quite a number of syntactic and semantic features. The collocational analysis, however, revealed important differences (see Chapter 7), and so does the collostructional analysis, as will be shown presently. Table 64 lists the distinctive collexemes for [X get Y Vpp]. What is striking in this list is that the three most distinctive collexemes are all semantically tagged with the guideword “deal”: do_deal, sort out_deal and sort_deal, as illustrated below. (406) I’m only showing you once darling and then you’ll have to do it or else I’ll never get [my] bingo done. ( 407) So I take it Laura’s got her sinus problem sorted out did she? (408) Right, well we’ll get it all sorted for then and we’ll try and arrange a programme whereby we tie it up with Loretta or whoever that erm mom and I and Beth coming up to you and perhaps we can go out for lunch, yeah?

These three collexemes belong to a larger semantic field, together with verbs like book_arrange, set up_prepare and line up_organise, which can be broadly described as having to do with organisation (see collexemes in bold in Table 64). This class of verbs is to be related to the semantic prosody of effort/difficulty that emerged from the collocational analysis (Section 7.3.2). The idea of organisation implies a previous state where things were disorganised, and the change from a disorganised state to an organised state involves, if not difficulties, at least effort. This undertone is particularly clear with sort (out)_deal and work out_problem, whose meanings necessarily entail the existence of a problem or difficult situation, as is the case in (407) above. With finish_complete, the difficulty consists in reaching the final stage of an action, which is often underlined by the mention of a deadline in the sentence, as in (409). Note that do is regularly used in the same sense, e.g. (410). (409) I bet I’m not gonna get this finished in time now.

( 410) I wanted to get it done today, I didn’t want to go back again


Table 64. Distinctive collexemes for [X get Y Vpp] Collexeme (n) 1 do_deal (93) 2 sort out_deal (72) 3 sort_deal (13) 4 finish_complete (17) 5 do_make (31) 6 do_perform (97) 7 dress_clothes (9) 8 wash_clean (13) 9 do_clean (49) 10 organise_arrange (8) 11 fix_repair (12) 12 organise_system (7) 13 book_arrange (7) 14 do_cook (6) 15 change_clothes (4) 16 dry (4) 17 write down (5) 18 deliver_take (7) 19 start_begin (8) 20 insure (4) 21 cook_eat (3) 22 send off_post (3) 23 set up_prepare (3) 24 sort_order (3) 25 warm up_heat (3) 26 work out_problem (3) 27 kill_death (4)

Distinctiveness 48.32 43.71 8.47 8.33 7.09 6.51 5.87 5.50 4.80 4.36 4.15 3.75 3.19 2.65 2.61 2.61 2.57 2.42 2.07 1.99 1.96 1.96 1.96 1.96 1.96 1.96 1.60

Collexeme (n)

Distinctiveness

28 develop_process film (4) 29 start_work (3) 30 backdate (2) 31 blow_send out air (2) 32 brush off_clean (2) 33 change_money (2) 34 chuck out (2) 35 clear_not blocked (2) 36 clear up_place (2) 37 copy (2) 38 cover_place over (2) 39 do up_fasten (2) 40 expel (2) 41 feed_give food (2) 42 fix_arrange (2) 43 line up_organise (2) 44 lock_fasten (2) 45 mend (2) 46 mix up_untidy (2) 47 patch_piece of material (2) 48 pick up_lift (2) 49 shift_get rid (2) 50 shoot_weapon (2) 51 tile (2) 52 tune_engine (2) 53 use up (2)

1.60 1.43 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30 1.30

Besides these collexemes referring to the planning and arranging of some process, [X get Y Vpp] also appears to be attracted to verbs describing daily actions (in italics in Table 64), like dress_clothes or cook_eat in the following examples: (411) Get yourself dressed now Kyle.

( 412) So I’ve gotta get them [dumplings] cooked before Terry comes home.

This can be linked to the identity between causer and causee that characterises many [X get Y Vpp] constructions (see Section 4.3.5). Given that “getting something done” often implies that one does it oneself, it is only natural that verbs describing daily actions should be distinctive for the construction, since such actions are typically carried out by the causer him- or herself.


A comparison of Table 64 with Table 65, which lists the distinctive collexemes for [X have Y Vpp], shows that only three collexemes are distinctive for both [X get Y Vpp] and [X have Y Vpp], viz. do_clean, do_make and do_perform. It also reveals that, while do_make has a higher distinctiveness value in [X get Y Vpp] than in [X have Y Vpp] (7.09 against 1.99), do_clean is much more distinctive for [X have Y Vpp] than for [X get Y Vpp] (21.53 against 4.80). Do_clean refers to the process of “cleaning, tidying or making something look attractive”, as in have one’s hair/nails done. It is one of several collexemes for [X have Y Vpp] that refer to a service (in bold in the table), the most distinctive one being cut_knife. Collexemes such as cut_knife, perm, build_make, put down_kill or service_repair are all actions that one normally commissions a professional to do: a hairdresser to cut or perm one’s hair, a builder to build a house or a wall, a vet to put down an animal, a mechanic to service one’s car, etc. The collostructional analysis, therefore, supports the results of the collocational analysis (Section 7.3.2) by highlighting the importance of the frame of service for [X have Y Vpp], but it goes one step further in identifying the lexical units that are likely to occur in the effect slot of the construction. Here are a few typical examples: (413) Tomorrow would be quite nice because I’m having my hair done tomorrow. (414) I’m not sure, but I think it was Mr that had that bungalow built. (415) Charlie is on his last legs, has been for years and, might as well have him put down, as that Nick keep saying, I think I’ll have to have him put down he, when he takes him for a walk he collapses.

Earlier we reproduced Francis et al.’s (1996: 306) list of typical examples of the two constructions just reviewed. It now turns out, however, that these examples mainly describe services and are therefore more typical of have than of get. Verbs from the frame of service are possible in [X get Y Vpp], but they are not characteristic of the construction and mainly occur in cases where the presence of get is required for other reasons, especially reasons of semantic prosody. Remember (344) and (345), repeated here as (416) and (417) for convenience’s sake. Although both of them contain a verb referring to a service (cut), the second sentence suggests difficulty in carrying out the action, as shown by the italicised elements, which justifies the use of a get construction. (416) Ironic, since fashion has gone full-circle and kids actually ask to have their hair cut short now. (417) Alternatively, they might rebel and become violently opposed to short hair, refusing to allow their children to get their hair cut.


Table 65. Distinctive collexemes for [X have Y Vpp] Collexeme (n) 1 cut_knife (54) 2 do_clean (72) 3 perm (22) 4 build_make (20) 5 do_perform (89) 6 put down_kill (8) 7 put_operation (7) 8 shave (8) 9 clean_dirty (11) 10 cut off_remove (6) 11 put on_equipment (6) 12 service_repair (7) 13 put in_equipment (9) 14 remove_take (7) 15 test (8) 16 put_move (9) 17 replace_change (5) 18 trim_cut (5) 19 cut off_stop (4) 20 put up_build (4) 21 repair (7) 22 take_act (8) 23 take out_remove (5) 24 check out_examine (3) 25 highlight_hair (3) 26 knock down_destroy (3) 27 put up_fix (3) 28 shape_form (3) 29 wax_substance (3) 30 make_produce (9)

Distinctiveness 25.57 21.53 16.12 11.66 7.91 5.86 5.13 4.99 4.45 4.40 4.40 4.30 4.06 3.73 3.48 3.09 2.96 2.96 2.93 2.93 2.92 2.60 2.49 2.20 2.20 2.20 2.20 2.20 2.20 2.06

Collexeme (n)

Distinctiveness

31 do_make (18) 32 paint (4) 33 tape_record (4) 34 alter_change (4) 35 take off_remove (3) 36 connect_join (3) 37 change_different (10) 38 send_post (4) 39 add (2) 40 amputate (2) 41 assess (2) 42 baptise (2) 43 bring back_return (2) 44 chop off (2) 45 comb_hair (2) 46 disconnect (2) 47 invest (2) 48 lift_end (2) 49 make_cause to be (2) 50 mount_fix (2) 51 pierce (2) 52 put down_stop (2) 53 recognise_accept (2) 54 rip (2) 55 sharpen_able to cut (2) 56 staple (2) 57 thread (2) 58 check_examine (4) 59 decorate_make attractive (3) 60 set_get ready (3)

1.99 1.89 1.89 1.89 1.66 1.66 1.62 1.59 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.47 1.36 1.33 1.33

8.5.5 [X get Y Vprp] and [X have Y Vprp] Turning to the present participle constructions with get and have, it appears from the collostructional analysis that the two constructions attract essentially different collexemes, but that the same two classes of verbs may be highlighted, namely verbs of movement (in bold in Tables 66 and 67) and, secondarily, verbs of position (in italics in Tables 66 and 67). Movement can be literal, with collexemes such


as go_leave and come_move, or metaphorical, with collexemes such as go_operate and move_progress. Compare: ( 418) I’ll see if I can get an old, a walking club going with them, you know? (419) I’ll burn them in the garden. On the top patio. Have a little fire going.

Note that the recurrence in the corpus of [X get Y going] might be related to the existence of the formulaic expression get somebody going (Kirchner 1952: 225n), exemplified in (420), but that the relatively low frequency of this expression in the corpus (Section 7.4) and the presence of other verbs of movement among the collexemes of the construction point to a more general trend. (420) Yet in the end they were, they were coming round they were singing, dancing, clapping and they got all us going as well Table 66. Distinctive collexemes for [X get Y Vprp] Collexeme (n) 1 2 3 4 5 6 7 8

go_operate (43) go_start (16) go_develop (8) run_operate (5) move_progress (3) go_be (3) talk (5) stand_place (3)

Distinctiveness 47.71 18.30 9.37 4.88 4.33 3.74 3.33 3.06

Collexeme (n) 9 call_name (2) 10 vote (2) 11 work_operate (3) 12 go_leave (2) 13 come in_enter (3) 14 come_move (4) 15 come up_computer (1) 16 face_turn towards (1)


Table 67. Distinctive collexemes for [X have Y Vprp] Collexeme (n) 1 2 3 4 5 6 7 8 9

go_operate (11) work_activity (6) go on_happen (2) hang_fix at top (2) stick out_go beyond (2) play_game (2) go_start (3) cry_tears (2) admire (1)

Distinctiveness 7.62 7.32 3.41 3.41 3.41 2.65 2.05 1.81 1.71

Collexeme (n) 10 dig out_get out (1) 11 drill_regular activity (1) 12 go on_operate (1) 13 guard (1) 14 iron (1) 15 panic (1) 16 side with (1) 17 have_eat (1)



Verbs of position, as illustrated by (421) and (422), are less common than verbs of movement and obtain lower distinctiveness scores. In addition, they tend to belong to the periphery of the category of causative constructions, with a meaning coming close to that of the existential construction (see Section 3.1.1.2 on existential constructions). (421) I got it standing on that brick <pause> so it would drain. (422) Well you’d have to have it [fridge] sticking out

8.5.6 [X get Y Vto-inf ] and [X have Y Vinf ] The collexemes of [X get Y Vto-inf ] and [X have Y Vinf ] display quite low distinctiveness values, compared with the other constructions. Moreover, they do not really fall into one or two clear semantic classes, as is the case for the constructions examined up to now. As Tables 68 and 69 show, they include some verbs of movement (e.g. go_move, come_move, come in_enter) and more general verbs of action like phone or do_perform for [X get Y Vto-inf ], and make_perform or practise_train for [X have Y Vinf ]. Interestingly, however, they also include verbs less directly related to action, which sets the two constructions apart from their past participle and present participle counterparts. In the case of [X get Y Vto-inf ], the collexemes to be singled out are talk and say_speak, two verbs of communication, and agree and comply, two verbs of agreement. These classes are illustrated by (423) and (424), respectively. Taken together, they suggest that the “elicitation of words or agreement” is one of the distinctive meanings of [X get Y Vto-inf ]. (423) [I like him when he’s interviewing] because he gets them to talk, he doesn’t impose his views on other people (424) This is one of the difficulties, getting everyone to agree on erm standards for transmission and standard formats for erm for letters and erm the like.

With [X have Y Vinf ], the presence of non-action verbs is slightly more obvious, as appears from Table 69, with two such verbs coming first and second in the list of the most distinctive collexemes of the construction, namely know_information and believe. Other similar verbs are: disbelieve, imagine, reflect_think and consider_possibility. These verbs belong to the same class of non-volitional verbs as were shown to be distinctive for [X make Y Vinf ] and, to a lesser extent, [X be made Vto-inf ]. However, the collexemes of [X have Y Vinf ] seem to be limited to mental processes. In addition, they often occur in (semi-)fixed expressions (see also Section 7.4). The three instances of believe are used with the auxiliary would,


Table 68. Distinctive collexemes for [X get Y Vto-inf ] Collexeme (n)

Distinctiveness

1 get_obtain (10) 2 help_easier (10) 3 go_move (18) 4 come_move (13) 5 talk (10) 6 sign_write (5) 7 say_speak (11) 8 take_go with (4) 9 phone (4) 10 do_cause (3) 11 pick up_collect (3) 12 stay_not leave (4) 13 come in_enter (5) 14 carry_transport (3) 15 have_do (3) 16 provide_supply (3) 17 make_perform (5) 18 give_provide (4) 19 agree (2) 20 collect_gather (2)

9.65 9.65 6.76 6.13 5.09 4.09 4.05 3.86 3.20 2.90 2.90 2.44 2.41 2.33 2.33 2.33 2.04 1.96 1.93 1.93

Collexeme (n) 21 comply (2) 22 give_do (2) 23 look_examine (2) 24 look_think (2) 25 remember (2) 26 ring back (2) 27 undertake_do (2) 28 go out_leave (3) 29 call_telephone (2) 30 come up_move towards (2) 31 get_cause (2) 32 take away_remove (2) 33 fill in_write (2) 34 dig_move earth (2) 35 drive_use vehicle (2) 36 go down_move down (2) 37 do_perform (39) 38 buy_pay (3) 39 stop_finish (3)

Distinctiveness 1.93 1.93 1.93 1.93 1.93 1.93 1.93 1.50 1.49 1.49 1.49 1.49 1.49 1.49 1.49 1.49 1.39 1.33 1.33

Table 69. Distinctive collexemes for [X have Y Vinf ] Collexeme (n) 1 know_information (7) 2 believe (3) 3 come_change (3) 4 make_perform (3) 5 go out_leave (2) 6 go_move (4) 7 come_move (3) 8 run_operate (2) 9 accompany_go with (1) 10 challenge_invitation (1) 11 convert (1) 12 disbelieve (1) 13 follow_obey (1) 14 imagine (1) 15 insert (1) 16 inspect (1)

Distinctiveness 9.30 3.53 3.34 2.66 2.07 1.92 1.79 1.76 1.68 1.68 1.68 1.68 1.68 1.68 1.68 1.68

Collexeme (n) 17 interview (1) 18 practise_train (1) 19 put through_telephone (1) 20 reflect_think (1) 21 report_tell (1) 22 take_write (1) 23 undergo (1) 24 come in_enter (2) 25 come_leave (1) 26 continue (1) 27 have_eat (1) 28 receive_get (1) 29 consider_possibility (1) 30 declare_express (1) 31 put_move (2)

Distinctiveness 1.68 1.68 1.68 1.68 1.68 1.68 1.68 1.68 1.39 1.39 1.39 1.39 1.39 1.39 1.37


Table 70. Distinctive collexemes for [X cause Y Vto-inf ] Collexeme (n) 1 be_description (8) 2 become_be (5) 3 see_consider (3) 4 apprehend_understand (2) 5 approach_come near (2) 6 bulge (2) 7 collapse_fall (2) 8 deposit_leave (2) 9 differ (2) 10 evolve (2) 11 fall_lower (2) 12 leave_go away (2) 13 overestimate (2) 14 reach_level (2) 15 rise_move up (2)

Distinctiveness 5.61 4.84 3.67 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45 2.45

Collexeme (n) 16 slow down (2) 17 describe (2) 18 generate (2) 19 separate (2) 20 give_cause (2) 21 increase (2) 22 lose_not have (2) 23 switch_change (2) 24 shift_move (2) 25 fall_come down (2) 26 move_position (3) 27 appear_present (2) 28 suffer_feel pain (2) 29 switch_device (2)

Distinctiveness 2.45 1.99 1.99 1.99 1.99 1.99 1.99 1.99 1.70 1.70 1.61 1.50 1.50 1.34

as is the verb imagine, and the pronoun us, as in (425). As for know, all its occurrences are in the expression I’ll have you know, e.g. (426). One distinctive meaning of [X have Y Vinf ] that emerges from the analysis, therefore, is that of “elicitation of a mental response”, but mainly in idiomatic uses. (425) Both behaviourism and “humanistic” views would have us believe that to be affective is to be effective in setting up conditions for learning. (426) They’re clean I’ll have you know, look how shiny they are.

8.5.7 [X cause Y Vto-inf ] [X cause Y Vto-inf ] is one of the constructions whose collexemes have the lowest distinctiveness values (together with [X be made Vto-inf ]). Yet, its two most distinctive collexemes display strong semantic cohesion, in that they are both copular verbs: be_description and become_be. Compare: (427) The levees caused the areas between to be very subject to flooding and consequently peat formation was further favoured. (428) The result is a massive loss of water, sodium ions, chloride ions and bicarbonate ions, causing the patient to become dehydrated and acidotic (acidosis being an increase in the acidity of the blood).


The list of collexemes in Table 70 also includes a number of verbs of movement (in italics in the table), e.g. approach_come near, collapse_fall or fall_lower: (429) Even where the coast does not face directly towards the incoming waves refraction causes the waves to swing round and approach the shore less obliquely than they were when further out to sea. (430) The resulting rotation of the crank about its fixed pivot causes the horizontal portion to rise and fall, lifting link member D and with it the vertically constrained rod member E.

While this characteristic seems to relate the cause construction to [X get Y Vprp] and [X have Y Vprp], it should be emphasised that with [X cause Y Vto-inf ] the verbs of movement tend to be more specific (compare swing round in [429] with going in [418]), thus revealing the common association of the construction with scientific and technical contexts (see Section 9.2.2). Grouping together the copular verbs be and become and the verbs of movement, we may conclude that [X cause Y Vto-inf ] is most distinctively used to express the process through which a transformation or a specific movement is caused. This is confirmed by several of the collexemes of the construction which, though not reaching the threshold level of statistical significance, also contribute to the meaning of the construction, e.g. accelerate_move faster, explode_burst, fluctuate, turn_change direction, grow_increase.

8.6 The effect slot in the elicitation data One of the exercises of the elicitation test (B3) aims at eliciting the effect of causative constructions, by asking the subjects to complete the following sentences: (431) The smoke caused them to …………………………………………… (432) You got them …………………………………………… (433) I made them …………………………………………… (434) Shall we have them ……………………………………………? (435) They were made to …………………………………………… (436) She made the boy look …………………………

The number of answers is not large enough to perform a multiple distinctive collexeme analysis, but some interesting tendencies may nevertheless be observed in the results.


The answers to (431), The smoke caused them to…, may be divided into two main groups: those having to do with the difficulty of breathing (fifteen occurrences of cough, seven occurrences of choke, one occurrence of asphyxiate and suffocate) and those having to do with movement, and more precisely escape (seven occurrences or run, run away or run out, six occurrences of leave, two occurrences of evacuate, one occurrence of flee, go away and retreat). While the causer provided (the smoke) clearly influenced the answers, it is interesting to note that the verbs of movement, which were shown in the collostructional analysis to be distinctive for [X cause Y Vto-inf ], are present in the results of the elicitation test. It should be underlined, however, that only one of these verbs actually belongs to the list of distinctive collexemes of the construction (Table 70), namely leave_go away. Sentence (432), You got them…, allows the subjects to produce any of the three get constructions, viz. [X get Y Vpp], [X get Y Vprp] or [X get Y Vto-inf ]. Of the five past participle constructions that were produced, one includes the verb finish_complete and one the verb dress_clothes, both of which are distinctive for the construction according to the corpus data (Table 64). The three remaining verbs (tell off, sack_dismiss and fire_dismiss) did not emerge at all from the collostructional analysis. Of the eight present participle constructions that were elicited, only three include a verb of movement: go (twice)82 and run around (once), one of which (run around) does not belong to the list of distinctive collexemes of the construction (Table 66). Finally, very few of the 31 infinitive constructions produced contain a verb corresponding to the elicitation of words or agreement (two occurrences of admit_accept and one occurrence of tell_speak) and none of these verbs is actually distinctive according to the collostructional analysis (Table 68). The next test item (433), I made them…, exclusively elicited infinitive constructions, no past participle constructions. The most frequently produced effects are eat and leave (four occurrences each), neither of which belongs to the class of non-volitional collexemes shown to be distinctive for [X make Y Vinf ] (Table 62). In fact, most of the effects produced by the subjects suggest a coercive meaning, e.g. get up, go away, sit down and tidy. A mere seven answers refer to a non-volitional process, namely laugh (three times), feel_experience (twice), cry_tears and weep_cry, and only four of the verbs distinctive for the construction in the corpus are also found in the elicitation data: laugh, feel_experience, cry_tears and wait. Like the test item with get, the have test item (434), Shall we have them…?, allows for all three types of have constructions: [X have Y Vpp], [X have Y Vprp] or [X have Y Vinf ]. However, for none of the three constructions is there a high

82. The context is too vague here to determine the exact sense of the verb go.


egree of correspondence between the list of the most distinctive collexemes d (Tables 65, 67 and 69) and the answers given in the elicitation test. Most of the effects in the elicitation data occur only once and none of them is repeated more than twice.83 The semantic classes of verbs highlighted in the collostructional analysis are largely absent from the subjects’ answers (possible exceptions are collect and deliver as verbs of service with [X have Y Vpp], and stand as a verb of position with [X have Y Vprp]), and the only distinctive collexeme found in the elicitation data is come_move for [X have Y Vinf ]. As was pointed out in Section 8.5.3, [X be made Vto-inf ] displays the lowest distinctiveness values of all the causative constructions investigated here, which should not encourage the use of distinctive collexemes in the elicitation test. In fact, only two verbs mentioned in Table 63 are given as answers to the test item in (435) They were made to…, namely pay_buy and swim_move in water. Curiously enough, the most commonly elicited effect, with eight occurrences, is a verb that does occur in the corpus but whose distinctiveness value is not statistically significant, viz. look_seem. The last item of exercise B3 (436) was meant to elicit the most likely sense of the verb look in [X make Y Vinf ]. Table 62 showed that one of the most distinctive collexemes of the construction is look_seem, as in (437). Look also occurs with the meaning see, as in (438), but it is less likely, as evidenced by the lower distinctiveness value of this collexeme (1.52, as compared to 34.79 for look_seem). When asked to complete the sentence She made the boy look…, however, the subjects did not seem to be influenced by this difference in distinctiveness, since just over 51% of them chose a complement involving the volitional reading of the verb (look for the ball, in her eyes, into the telescope, at the board, etc) while the remaining 49% obviously favoured the non-volitional reading of look (look silly, embarrassed, upset, like an angel, etc). (437) They probably though[t] that your personality was so special that it would it would actually have made everybody else look rather stupid. (438) I don’t know what made me look in the bag

All in all, it turns out that the answers provided to exercise B3 of the elicitation test show relatively little overlap with the results of the collostructional analysis. The semantic classes that were shown to be distinctive for a particular construction are not always represented, and few of the distinctive collexemes are found in the elicitation data. It should be noted, however, that the subjects were not asked to 83. The effects that are repeated twice are: come, come over, eat and visit for [X have Y Vinf ], and execute for [X have Y Vpp].


produce what they considered the most typical effect, but simply to give the first answer that came to their minds.84 For more indirect tasks, such as judging the degree of acceptability of sentences, it looks as if informants might have a better feeling of what is a natural collostruction. Thus, only 16% of the subjects accepted sentence (439) [= D.14], which contains the effect paid, a verb not statistically distinctive for [X make Y Vpp] and not part of the semantic class of verbs denoting recognition of influence that was shown to be distinctive for the construction. (439) Child benefit is the only benefit which is in fact available to women, but a few years ago the government tried to change that and to make it paid to the man.

8.7 Summary This chapter has investigated the lexical form of the effect, using the technique of multiple distinctive collexeme analysis, which examines all the verbs (or rather, in this case, verb senses) occurring in similar constructions and identifies those that are more distinctive for one or the other construction. Most of the ten causative constructions under investigation display strong lexical preferences in the effect slot, which confirms the importance of this parameter for the choice of construction, brought to light by the hierarchical feature selection (Section 5.4). The collexemes with the highest distinctiveness values occur in [X make Y Vinf ], [X get Y Vpp] and [X get Y Vprp], whereas the collexemes with the lowest distinctiveness values occur in [X be made Vto-inf ], [X cause Y Vto-inf ] and [X have Y Vprp]. The following classes of verbs have been shown to be particularly distinctive: copular verbs and verbs of (specific) movement for [X cause Y Vto-inf ]; verbs of communication and agreement for [X get Y Vto-inf ]; verbs of mental processes for [X have Y Vinf ]; verbs of organisation and daily actions for [X get Y Vpp]; verbs belonging to the frame of service for [X have Y Vpp]; verbs of (literal or metaphorical) movement and, less frequently, verbs of position for [X get Y Vprp] and [X have Y Vprp] (with, typically, an undertone of difficulty in the former construction); verbs of impression, mental and physiological processes for [X make Y Vinf ] and, less distinctively, for [X be made Vto-inf ]; verbs denoting the recognition of influence on the part of the causee for [X make Y Vpp]. The meanings of the constructions as they can be read off from these classes of collexemes are shown in Table 71. Note that this time, the corpus results are hardly reflected in the elicitation data, which may be related to the instructions that were given to the informants.

84. McGee (2006, 2009) obtained better results by asking his informants, English language teachers, to give what they considered the most frequent collocates of a word.


Table 71. Meanings distinctively associated with periphrastic causative constructions Construction

Distinctive meaning

Example

[X cause Y Vto-inf ]

“To cause a transformation or specific movement”

[X get Y Vto-inf ]

“To elicit words or agreement”

[X get Y Vprp]

“To set an object in motion (also metaphorical) or, less frequently, set it in a particular position” [usually involves difficulty] “To organise an activity in difficult circumstances or under a tight schedule, or to carry out a daily action oneself ” “To elicit a mental response” [mainly idiomatic]

That’s causing traffic to be slow moving around that junction at the moment. John Allen (…) was able to give a great deal of energy to getting opposing groups to talk to each other. Yeah, probably if you want me to get that old mower going I ought to go up to Woods and see if I can get a new drive belt.

[X get Y Vpp]

[X have Y Vinf ] [X have Y Vprp]

[X have Y Vpp] [X make Y Vinf ]

[X be made Vto-inf ]

[X make Y Vpp]

“To set an object in motion (also metaphorical) or, less frequently, set it in a particular position” “To commission someone to do something” “To cause a process that is not directly dependent on the causee” (especially impressions, mental and physiological processes) “To cause a process that is not directly dependent on the causee” [less marked]

“To exercise some sort of influence”

Cos I want to try and get things sorted out round there. They’re clean I’ll have you know, look how shiny they are. I better have the dishwasher going.

Alan’s gonna go and have his hair cut. They just make you feel so inadequate.

Innocent wives and children and other dependents are made to suffer when the state imprisons thousands of working-class men for crimes which are often insignificant compared with corporate crimes. The May 1990 elections provided voters with an opportunity to make their views known.


As was the case in the preceding chapter, some earlier findings may be reinterpreted in the light of the lexical analysis. Thus, it was shown in Section 5.3.5 that the effect of the make constructions is predominantly non-volitional. It now becomes clear that this is largely due to the recurrence of verbs such as feel, laugh, think or wonder. We also saw in the same section that [X make Y Vpp] is strongly associated with mental effects. It now transpires that these correspond to just a few verbs, essentially know and understand. With respect to [X get Y Vprp] and [X have Y Vprp], it appeared in Chapter 5 that the two constructions, unlike the other get and have constructions, present a majority of intransitive effects. This is the case of most of the verbs of movement which, according to the collostructional analysis, are typical of the two constructions. The study of the collexemes in the effect slot, therefore, makes it possible to refine the results of the semantic and syntactic analysis. After two chapters dealing with the lexical regularities of causative constructions, we now turn to an analysis of their stylistic regularities. More precisely, we will assess the possible impact of register on causative constructions, both at the macro level (frequency) and at the micro level (semantics, syntax and lexis).

chapter 9

The influence of register on causative constructions

In this chapter, we take advantage of the stylistic information encoded in the BNC to study variation in the use of periphrastic causative constructions. We investigate the frequency of the constructions according to medium and genre, as well as the distribution of the syntactic, semantic and lexical features of the constructions (see Chapters 5, 7 and 8) along the speech/writing dimension. The reliability of these comparisons is guaranteed by the composition of BNC-10, which contains samples from the two ends of the speech vs. writing cline and whose genres were carefully identified for the World Edition of the BNC (Section 3.1.2.1). Note that, because the elicitation test was not designed to provide information about register variation (although one may perhaps claim that the results of an experimental task come closer to spoken language than to written language, see Baayen et al. 2006), the influence of register in the elicitation data will not be dealt with.

9.1

Studying variation

Willems (2000) mentions two topics that have been introduced in language analysis mainly thanks to the use of corpus data, namely co(n)text and variation. Co(n)text may be related to the study of collocation and other phraseological phenomena, as described in the preceding two chapters. Variation, on the other hand, refers to the differences language displays when one compares various registers.85 One of the most influential representatives of the study of language variation is Douglas Biber, who investigated the interaction of a number of linguistic features and was able, on this basis, to identify five dimensions, which he sees as “fundamental parameters of linguistic variation among English texts” (Biber 1988: 199). The study of variation has also made its way into grammars, e.g. the Longman Grammar of Spoken and Written English (Biber et al. 1999) which 85. Following Kress (1993), “register” will be used as a superordinate term covering, among other things, medium (or mode, i.e. speech vs. writing) and genres (defined here as the divisions found in the BNC World Edition, see Section 3.1.2.1).


distinguishes between four registers in its description of language (conversation, fiction, newspaper language and academic prose), and it is now widely accepted that, say, spoken English is different from written English, or legal English is different from business English. If we now turn our attention to the influence of register on causative constructions, that is their stylistic behaviour, it appears that, like the literature on the lexical preferences of causative verbs (see Chapters 7 and 8) and probably for the same reason (viz. the general introspection-based approach), the literature dealing with the stylistic features of periphrastic causative constructions is rather thin. Paradoxically, the verb that has attracted the most attention from a stylistic point of view is the verb that most often tends to be ignored in the literature on causatives, namely cause. Vinay & Darbelnet (1995: 197), in their book on comparative stylistics of French and English, observe that [X cause Y Vto-inf ] “is considered formal and obsolescent. As a direct equivalent to the French construction [faire + infinitive] it appears awkward”. They add that the construction is possible in legal discourse, but not limited to it, as shown by a comparison of the following two sentences, the second of which Vinay & Darbelnet borrow from George Orwell: (440) … in witness whereof I have hereunto … caused the Seal of the United States of America to be affixed. (441) It is the same motive that caused the Malaya jungle to be cleared for rubber estates.

Other linguists underline the formal character of cause. Kruisinga & Erades (1960: 325n) note that cause “occurs in literary English only”, and BerlandDelépine (1990: 173) and Deutschbein & Klitscher (1959: 136) describe its use as “mannered” and “refined” respectively, in contrast to the infinitive constructions with get, have and make, and the past participle constructions with get and have, which Deutschbein & Klitscher (ibid.) judge as “common in colloquial language” (my translations; see also Visser 1973: 2256, Hantson 1981: 152, Stocker 1990: 227). Chuquet & Paillard (1989: 170) and Magnusson (1987: 206), whose studies are corpus-based, are more precise and argue that cause typically occurs in scientific/technical discourse and informative prose. By contrast, the informal status of get is often underlined, both in its general use (Tavernier & Etienne 1973: 39’) and with a causative meaning (Frank 1972: 310, Girard 1978: 74, Attal 1987: 491), a characteristic which is often presented as the sole difference between get and have (Thomson & Martinet 1980: 107, Murphy 1985: 92, Carter & McCarthy 1999: 48, Lester 2008: 331). Some indications are also given regarding the distinction between American and British English. Palmer & Blandford (1969: 197), for example, note that the

Chapter 9. The influence of register on causative constructions 225

use of [X have Y Vinf ] is more typical of American than British English (see also Deutschbein & Klitscher 1959: 136, Attal 1987: 490, Berland-Delépine 1990: 174, Swan 2005: 209). Since the data used in this study all represent British English, however, this type of variation will not be considered here. Finally, it should be noted that the Longman Grammar of Spoken and Written English (Biber et al. 1999), which relies on corpus data representing several registers (see above), merely provides sporadic information about the stylistic distribution of causative constructions, pointing out, for example, that most of the passive verb complements in conversation are controlled by have and get (e.g. Oh, has Kathy had her hair done? or I’d never get my letters written anyway) (ibid. 940) or that [X have Y Vinf ] is attested relatively rarely in both American and British English (ibid. 708).

9.2 Frequency according to register 9.2.1 Medium We saw in Chapter 3 that the ten causative constructions under investigation exhibit (sometimes radically) different frequencies in BNC-10. In addition, each of these constructions displays differences in frequency when one compares the spoken component of BNC-10 (BNC-10S) and its written component (BNC-10W). This is illustrated by Figure 10, which shows the constructions in ascending order of (overall) frequency and distinguishes their relative frequencies in speech and writing. All the differences between speech and writing are statistically significant, at a minimum level of 0.05 (but most of them are significant at the 0.001 level). Three constructions are more frequent in writing than in speech, namely [X make Y Vpp], [X be made Vto-inf ] and [X cause Y Vto-inf ]. The remaining constructions, on the other hand, are more common in speech than in writing. It is interesting to note that some constructions may be relatively infrequent overall, but count among the most common constructions in a particular medium (speech or writing). The most striking example is that of [X cause Y Vto-inf ]. While its overall frequency is far behind that of a construction like [X make Y Vinf ], it is the second most common construction in writing, just after [X make Y Vinf ]. Although the speech/writing dichotomy should be distinguished from the formality scale, the composition of BNC-10 makes it possible, with some caution, to extrapolate from the frequency of the constructions in speech and writing to their degree of formality. Academic prose, which makes up BNC-10W, is arguably more formal than the spontaneous conversations included in BNC-10S.

  226 Corpus, Cognition and Causative Constructions 



          

eY

k

[X

ma

]

p

Vp [X

ve ha

Y

]

rp

Vp

[X

e av  h

Y

V in

[X

f]

b

e ad em

f]

in

V to

[X

t ge

]

Y

rp

Vp

f]

in

V to

[

Y e Y t us ge a c [X X 

f]

in

V to

[X

ve ha

Y

]

p

Vp

[X

ge

t

Y

f]

]

p

Vp

[X

eY

V in

k ma

Figure 10. Relative frequency per million words of causative constructions according to medium



           

On this basis, we may tentatively conclude that the most formal construction is  [X cause Y Vto-inf ], which would confirm what is found in the literature (see Section 9.1). It may not be the most frequent construction in writing (as pointed out  above, [X make Y Vinf ] is more frequent), but it is the construction whose ratio between speech and writing is the highest, amounting to 1:12 (to be compared  with approximately 1:3 for [X be made Vto-inf ] and [X make Y Vpp], the other two constructions that are more frequent in writing than in speech). At the other end of the formality continuum is [X get Y Vpp], which, with its high frequency in speech and its speech/writing ratio of 44:1, qualifies as the least formal construction. While this supports the claim made in the literature that get is an informal verb (see also the other get constructions, which are much more common in speech than in writing), it should be emphasised that formality is not what primarily distinguishes get from have. The have constructions, like the get constructions, are more frequent in speech than in writing, and their speech/writing ratio is quite high (e.g. 15:1 for [X have Y Vpp]) – though admittedly not quite as high as some of the get constructions. Also, while causative get is significantly more frequent than have in (informal) speech (highly significant at the 0.001 level), the difference between the two verbs in (formal) writing is not significant. All this suggests that causative have (and in particular [X have Y Vpp], as is clear from Figure 10) is relatively informal, just like causative get. As for causative make, it presents mixed results. For one thing,








          

eY

k

[X

ma

]

p

Vp [X

v ha

eY

]

rp

Vp [

v ha X 

eY

V in

[X

f]

f]

b

]

rp

-in

to Y eV et ad g [X em

Vp

f]

-in

V to

[


f]

-in

V to

[X

ve ha

Y

]

p

Vp

[X

ge

t

Y

f]

]

p

Vp

eY

V in

k

[X

ma

Figure 11. Relative frequency per million words of causative constructions according to genre (speech)





two constructions are more common in writing than in speech ([X make Y Vpp]  and [X be made Vto-inf ]),86 whereas the third one, [X make Y Vinf ], displays the opposite tendency. For another, [X make Y Vinf ] is the most frequent construc tion both in speech and writing, and its speech/writing ratio is among the lowest of all the causative constructions. This seems to point to [X make Y Vinf ] as a stylistically more neutral construction, which confirms the versatile character of the construction as it emerged from the syntactic and semantic analysis (see Section 5.3.7).

9.2.2 Genres Next to the distinction between speech and writing, a distinction may be drawn between the genres making up the spoken and written components of BNC-10. This section examines the frequency of the causative constructions according to genre, starting with the genres represented in BNC-10S. BNC-10S includes samples from three genres, namely spontaneous conversations (conv), broadcast TV and radio discussions (disc) and live sports 86. This may partly be explained by the fact that passives tend to be more frequent in writing than in speech (Biber et al. 1999: 476). Note, however, that [X get Y Vpp], for instance, also involves passivisation (of the subclause), yet is far more common in speech than in writing.


c ommentaries and discussions (sport). Figure 11 shows that causative constructions are very infrequent in live sports commentaries and discussions (the main exception being [X get Y Vprp], which is about as frequent in “sport” as in “conv”),87 and this genre will therefore not be dealt with any further. The comparison between spontaneous conversations and broadcast TV/radio discussions, on the other hand, becomes quite interesting if, as we did for the speech/ writing distinction, we tentatively extrapolate from genre classification to formality scale. It can reasonably be assumed that the genre of spontaneous conversations is slightly less formal than that of TV and radio discussions. As evidence for this, one need only consider the following two extracts, taken at random from the conversations and broadcast discussions, respectively: ( 442) M You enjoyed yourself in America B Eh? M did you? B Oh I covered a nice trip, yes M Oh very good B saw Mary and Andrew and M Yes, you did B in fact the whole family was together for Mary’s wedding M Oh very nice, very nice, yes. It’s horrible B It is horrible isn’t? M Have you been busy? B Yes M Yes, oh. Jim’s been for a, this afternoon at the Hart and Straw Club oh, not very well, we erm, we stopped going after Christmas because we had bad chests both of us B Oh M both cold and it’s hard going that three hours in the morning, you know (443) PS000 Let’s see what this hundred women make of the question, do men hate women? Button one, and button two for no. Ge-- generally speaking, I mean as far as you can tell, living as long as you have done how would you answer that question? And in this hundred, sixty seven say no, but thirty three which is exactly a third, say yes! Who said yes, and why?

87. This may be related to the use of the construction to describe movement and, less frequently, position (see Section 8.5.5), two meanings which may be useful when commenting on a sports event.


PS001 I think misogyny is is very ancient, it’s as old as man. It comes from Augustine Equinus it’s come right down through literature and history. And perhaps, today we would not say that men actually hate women, but there aren’t <pause> isn’t much obli--, er evidence of their respect. They’re not willing to erm, recognise women as their equals. PS000 Why? PS001 Well I think it’s been a mans world, they, they’ve had a wonderful time! Why, why should they give it up? PS002 I think erm <pause> like the main issue is that men are really afraid of woman and really PS000 Mhm. PS002 afraid of woman’s power and the only way they know how to combat that is is to put out this kind of <pause> this hatred or or this misogyny in in the language in the way <pause> they behave towards women. But I think it’s there, underneath it. Like fear, in trying to keep women down so that women can’t discover their power, because women are just so powerful! And we’re only just beginning to realise that.

If this assumption about the degree of formality of spontaneous conversations and broadcast discussions is correct, then the results of Figure 11 nicely support the suggestions made above on the basis of the comparison between speech and writing. The three constructions that are more frequent in broadcast discussions than in spontaneous conversations, namely [X make Y Vpp], [X be made Vto-inf ] and [X cause Y Vto-inf ], are precisely those constructions that were shown to be more typical of writing than of speech and hence were argued to be more formal than the others. The remaining constructions, by contrast, are more frequent in spontaneous conversations than in broadcast discussions, which arguably underlines their less formal nature. It also appears from Figure 11 that [X make Y Vinf ] is the construction that is found most often in the two genres and with very similar frequencies (non-significant difference), which seems to confirm the stylistically more neutral character of this construction. Note that two other constructions do not display any significant difference between spontaneous conversations and broadcast discussions, viz. [X have Y Vprp] and [X have Y Vinf ]. Their frequencies in the two genres, however, are much lower than is the case with [X make Y Vinf ]. In BNC-10W, six genres are represented, corresponding to different domains in academic prose: humanities (hum), medicine (med), natural science (nat), politics, law, education (pol), social and behavioural sciences (soc), and technology, computing, engineering (tech). This composition does not make it possible to draw conclusions about formality, since all the genres may be assumed to


be equally formal in character, and the analysis must therefore be restricted to a description of the domains that turn out to be favoured by the constructions. Figure 12 reveals a great diversity in the distribution of the constructions in writing according to genre, but some interesting tendencies emerge nonetheless. It appears that [X cause Y Vto-inf ], which was shown earlier to be one of the most frequent constructions in writing, is especially frequent in the genre of technology, computing, engineering and that of natural science. What is common to these two domains is that they are very technical and concern the exact sciences, which seems to corroborate Chuquet & Paillard’s (1989: 170) statement that cause is typically found in scientific and technical writing. By contrast, a construction such as [X make Y Vinf ] (or [X be made Vto-inf ], though with a lower frequency) favours more socially oriented genres, viz. humanities, and social and behavioural sciences. The presence of [X have Y Vinf ] in the humanities is also worth underlining, as it may be related to the depictive use of the construction (Section 5.3.4) and its association with terms referring to fiction (Section 7.3.2), which may be expected to occur in literary genres.88

88. A brief remark is in order concerning the particularly low frequency of causative constructions in the medical domain. While the automatic extraction revealed a relatively large number of [X have Y Vpp] constructions in this part of the corpus, most of these turned out to be experiential, not causative. Compare the following two examples:

i. Before their attack 356 (23%) patients had been prescribed treatment at step 2, 275 (77%) of whom remained at step 2 or lower after the attack, and 81 (23%) had their treatment stepped up according to recommended guidelines.

ii. I just find it very difficult to understand paying over two thousand for, er to have your tummy bulge removed.

While both sentences refer to medical treatment, they differ in the perspective adopted and the conceptualisation of the situation. In (i) the situation is presented from the point of view of the medical staff, and the people receiving treatment are seen as patients (both in the medical sense and the linguistic one), as indicated by the word patients and the presence of the passive voice. They are not even treated as individuals, but as a group of anonymous patients, as evidenced by the use of percentages. In sentence (ii), by contrast, the perspective is that of the (implied) subject you. If the operation takes place, the person will admittedly be “affected” by it, but it will be on the person’s own initiative (s/he will have phoned the doctor, made an appointment, paid for the operation, and so on). In (i) the subject merely experiences an event and is affected by it, whereas in (ii) it has the active function of a causer, who initiates the caused event. Because the texts included in the “medicine” sample come from books or periodicals written by and for medical experts, relating experiments, case studies or research work, the have constructions found in these texts tend to be of the first type (experiential constructions), and true causative constructions are correspondingly rare.
















         

eY

k

[X

ma

]

p

Vp [X

ve ha

Y

]

rp

Vp

[X

e av  h

Y

V in

[X

f]

b

e ad em

f]

in

V to

[X

t ge

]

Y

rp

Vp

f]

in

V to


[

f]

in

V to

[X

ve ha

Y

]

p

Vp

[X

ge

t

Y

f]

]

p

Vp

[X

ke

ma

Y

V in

Figure 12. Relative frequency per million words of causative constructions according to genre (writing)



  9.3 Semantic and syntactic features in speech vs. writing  Causative constructions do not only display differences according to register at  the macro level (frequency), but also at the micro level (semantics, syntax, lexis). This section focuses on the semantic and syntactic differences that are statistically significant (as calculated by means of a chi-square test) between speech and writing. Notethat, given the differences in frequency highlighted above, some  constructionsare too infrequent in one of the two media to allow for a reliable comparison.  [X get Y Vprp] and [X have Y Vprp] are found only three times in  89 BNC-10W and  will therefore not be dealt with here.  The semantic and syntactic comparison between speech and writing proceeds  in two steps. First, we deal with the differences that are common to all (or most)   causative constructions. While such differences are interesting, they are mainly    a reflection of the intrinsic differences that exist between the two media and the situations in which these media are used. They do not really help distinguish be tween the constructions, which is what the second step of the comparison does,  by describing the differences that are specific to one or two constructions only.                 89. For the same reason, the comparison of the syntactic, semantic and lexical features of the

constructions across genres will not be undertaken. 


9.3.1 Shared differences Conversation takes place in a shared context (Leech 1998). This explains a number of features exhibited by all or most periphrastic causative constructions in speech, as opposed to writing. Thus, there is a strong tendency for the participants in a causative construction to be more often definite in speech than in writing (this is especially clear in the case of the causer). Similarly, more participants are expressed as pronouns (rather than nouns) and are specific (rather than generic) in speech as compared to writing. In addition, inanimate patients in speech predominantly refer to concrete objects, whereas inanimate patients in writing typically describe abstract phenomena. All this has to do with the fact that spoken language, by involving a shared context, relies on shared knowledge (Burns & Joyce 1997: 13). People tend to talk about entities that are definite and specific, often surrounding them (hence concrete rather than abstract), and which may consequently be referred to by means of pronouns. These context-related differences are illustrated by the following two sentences, the first of which comes from BNC-10S and the second one from BNC-10W. (444) Shall I get her to get them <pause> tomorrow?

(445) Extreme values of temperature and the presence of fog can play a role in causing pollution concentrations to reach health-threatening levels (figure 7.5).

The dialogic nature of spoken language (ibid.) explains another difference found between the two media, namely the grammatical person of the causer. In speech, first- and second-person causers are relatively frequent. In writing, on the other hand, they are less common and tend to be supplanted by third-person causers. This difference is also illustrated by the above pair of sentences. Finally, explicit causing events, as in (446), are more frequent in writing than in speech. This difference may be accounted for in two ways. One possible explanation is that speech resorts to established scenarios more often than writing and therefore does not need to make the causing event explicit (see the discussion in Section 4.3.4). In (447), for example, the causing event need not be expressed for the hearer to imagine the event that led to the finished perm. The mention of the caused event is sufficient to evoke the whole scenario. In (446), by contrast, making a dog salivate is not a conventional scenario, and there are many ways of carrying out this caused event (of which the sounding of a bell is probably not the most usual one).


(446) Soon Pavlov could make his dog salivate just by sounding the bell, even without bringing its food. (447) She had it [her hair] permed didn’t she?

Another reason for the higher proportion of explicit causing events in writing may be that the scientific genres represented in BNC-10W need to be more precise about the process which led to a particular caused event. Compare (448) with (449) and (450). While in (448) it is important for the writer’s argumentation to make the causing event explicit, so as to show why the housewife and the motorist are empiricists, in (449) and (450) the process involved in getting the air conditioning unit working or getting the car started is not crucial (only the result is) and can therefore easily be dispensed with. (448) Most housewives and many motorists are empiricists. The housewife who knows that when her electric iron ceases to work she must replace the fuse (let us say it is a cartridge fuse in a 13-amp plug) is an empiricist. She observes that by taking out the old fuse and putting a new one in she has caused the iron to work again. Similarly, the motorist who discovers that he should pull out the choke button a particular amount so as to get the car to start on a cold morning is also an empiricist. (449) yeah it is air conditioning unit in Smith’s in Staines and they had to whatever happens they had to get it working for the next day (450) Yeah, she’d been to college but she couldn’t get her car started, she had to get her dad to drive her in <pause> and pick her up.

9.3.2 Specific differences Next to the differences which are shared by most of the ten causative constructions under study, there are also differences between speech and writing which are limited to a few constructions and must therefore be attributed to the constructions themselves and the way they behave in speech or writing, rather than to the mere nature of the two media as was the case in the preceding section. Table 72 summarises and illustrates those differences when they result in real divergences, i.e. when a construction displays a majority of features A in speech vs. a majority of features B in writing. All the differences are statistically significant at the 0.001 level.


Table 72. Main semantic and syntactic differences between speech and writing Parameter

Speech

Writing

Co-referential 64% Did you get your homework done in the lunchtime?

Distinct 53% We submitted our paper (or a shortened version) to six journals (…) but were unable to get it published.

Intransitive 50% I had Elsie go on a Wednesday night.

Transitive 86% How much simpler to station an interviewer at a busy part of a town and have her interview people who are passing by.

Animacy of causer

Animate 53% That bloody Williams bloke makes me puke.

Animacy of causee

Animate 77% What made you watch Oscar?

Inanimate 69% Similar considerations make me sometimes decline to examine students from foreign parts. Inanimate 51% It’s difficult to avoid making his suicide sound too purposeful.

[X get Y Vpp] causer / causee

[X have Y Vinf ] Transitivity of effect

[X make Y Vinf ]

[X be made Vto-inf ] Animacy of causee

Volitionality of effect

Animate 83% You’ve just been made to sit in this room for three hours. Volitional 65% Why shouldn’t we be made to pay for a health service?

Inanimate 61% The evidence is made to fit, instead of being used to test, the theory. Non-volitional 75% We are made to feel that the reversed meaning is wrong.


A first difference is that co-referential [X get Y Vpp] constructions, with identity between causer and causee, are more likely to occur in speech than in writing. In fact, out of the eight co-referential constructions found in BNC-10W, four actually reproduce spoken language, e.g. (451) SERGEANT. Aye, the auld plumbing is desperate for bringing on migraines. What happened, did he not get it [plumbing] finished over the weekend or samething? (FN 2/11/87, p. 2)

The second difference concerns [X have Y Vinf ] and the transitivity of the effect. It turns out that the effect is predominantly intransitive in speech, but predominantly transitive in writing. This difference in transitivity affects the length of the action chain, as described in Chapter 4. A transitive effect, by introducing a patient, expands the action chain, which now includes three participants and hence becomes more complex. One could argue that the avoidance of longer (and more complex) action chains in speech is due to its unrehearsed and spontaneous nature (Burns & Joyce 1997: 13). However, it should be borne in mind that the difference in transitivity (and length of action chain) highlighted for [X have Y Vinf ] is not shared by all the constructions (or at least, all the constructions that allow both transitive and intransitive effects). It can therefore not be seen as a mere reflection of the speech/writing dichotomy. The remaining differences shown in Table 72 all concern causative make, which reinforces the image of this verb as particularly versatile (see Sections 5.3.7 and 9.2). We first notice a tendency for its participants to be animate in speech and inanimate in writing (this is true for the causers and causees of [X make Y Vinf ] and the causees of [X be made Vto-inf ]). While at least one other causative construction displays a higher proportion of inanimate participants in writing than in speech (see below), the two make constructions are the only ones that actually have a majority of inanimate participants in writing and a majority of animate participants in speech. Another difference between speech and writing, this time limited to [X be made Vto-inf ], is that the effect of the construction is predominantly volitional in BNC-10S, whereas it is predominantly nonvolitional in BNC-10W. This is in part explained by the proportion of inanimate causees, since the presence of an inanimate causee entails a non-volitional effect. However, that this is not the only explanation appears from the example given in Table 72 for the non-volitionality of the effect in writing, where the causee (we) is animate. What precedes shows that medium may influence the use of certain causative constructions, also at the micro level of syntax and semantics. Differences such as those mentioned in Table 72 should therefore be taken into account


when analysing corpus data, as the composition of the corpus may affect the results of the analysis. In the present case, the results reported in Chapter 5 for [X get Y Vpp], [X have Y Vinf ], [X make Y Vinf ] and [X be made Vto-inf ] may be said to be biased by the difference in frequency of the four constructions according to medium. However, it should be emphasised that such radical differences are very rare in the corpus and that most of the time the differences between speech and writing do not result in real divergences (e.g. a majority of animate causers in speech, as opposed to a majority of inanimate causers in writing), but simply in slightly different tendencies (e.g. more animate causers in speech than in writing, but a majority of animate causers in both media). Differences which do not affect the regularities of causative constructions as described in Chapter 5 are discussed in what follows. Table 73 lists some of the other significant differences that are found between speech and writing. As just pointed out, these do not invalidate the results of Chapter 5. To give but one example, inanimate causers were shown to be rare in [X get Y Vto-inf ] (Section 5.3.3). The stylistic analysis reveals that this is true of both speech and writing, but that inanimate causers are significantly more frequent in writing than in speech: they represent 8.3% of all causers in writing, as against 0.3% in speech, a difference which is statistically significant at the 0.001 level, as indicated in Table 73. It should also be noted that some features occur too rarely in one of the two media to be considered. Stative effects in [X have Y Vpp], for example, as illustrated in (452), are more frequent in writing than in speech (6% vs. 0.2%) but the number of occurrences (three and one, respectively) is so small that statistical significance cannot be reliably tested. (452) There was a general belief that the nation should remain “strong” and have that strength publicly acknowledged.

Let us briefly review the main findings of Table 73. First, we notice a tendency for animate (or human) causers to be more frequent in speech than in writing (this is the case of [X get Y Vto-inf ] and [X have Y Vinf ]). This is to be related to the predominance in [X make Y Vinf ] of animate causers in speech and inanimate causers in writing (see Table 72) and might be accounted for by the preference for animate subjects in conversation (Biber et al. 1999: 378). Note, however, that not all causative constructions show such a significant difference between speech and writing when it comes to the animacy of the causer, which makes this phenomenon at least partly construction-specific. Another tendency that emerges from Table 73, and more particularly the results for [X get Y Vto-inf ] and [X make Y Vinf ], is the association of speech with dynamic/material effects and the association of writing with stative/mental/relational effects – even though dynamic


Table 73. More semantic and syntactic differences between speech and writing Parameter

S (%) W (%) Example

p

[X get Y Vto-inf ] 99.7

91.7

0.3

8.3

96.9

86.5

3.1

13.5

91.1

69.2

4.3

26.9

97.8

85.2

6.7

32.1

1.1

8.9

98.9

91.1

Dynamic effect Stative effect Material effect Relational effect

67.0

57.0

33.0

43.0

53.4

42.2

29.2

39.5

patient expressed patient unexpressed

22.4

29.9

77.6

70.1

Animate causer Inanimate causer Dynamic effect Stative effect

Material effect Mental effect

So I said to him shall we get Gary or somebody similar to do some Artexing in here She studied the use of directives – speech that gets someone to do something. He’s gonna get his brother to phone up tonight. In each case, the child must not only find a way of getting the adult to notice the object, but she must do this in such a way that the adult is aware of what she is doing and why she is doing it. Get him to make you pizza, brilliant, I'm not joking This session aims to get participants to consider the subject of alcohol in ways they had not thought of in the past.

p

Corpus, Cognition and Causative Constructions (Studies in Corpus Linguistics)

Perspectives on Corpus Linguistics (Studies in Corpus Linguistics)

Corpus Linguistics

Lexis in Contrast: Corpus-based Approaches (Studies in Corpus Linguistics)

Lexis in Contrast: Corpus-based Approaches (Studies in Corpus Linguistics)

Collocations in a Learner Corpus (Studies in Corpus Linguistics)

Corpus Approaches to Grammaticalization in English (Studies in Corpus Linguistics)

Keyness in Texts (Studies in Corpus Linguistics)

Researching Specialized Languages (Studies in Corpus Linguistics)

Statistics for Corpus Linguistics

An Introduction to Corpus Linguistics (Studies in Language and Linguistics)

Small Corpus Studies and Elt: Theory and Practice (Studies in Corpus Linguistics)

Discourse In The Professions: Perspectives From Corpus Linguistics (Studies in Corpus Linguistics, SCL 16)

Corpus

Lexicology and Corpus Linguistics (Open Linguistics)

Corpus Linguistics in Literary Analysis: Jane Austen and her Contemporaries (Studies in Corpus and Discourse)

Corpus

Metadiscourse in L1 And L2 English (Studies in Corpus Linguistics)

Corpus and Context: Investigating Pragmatic Functions in Spoken Discourse (Studies in Corpus Linguistics, Volume 30)

Trends in Teenage Talk: Corpus Compilation, Analysis and Findings (Studies in Corpus Linguistics)

Textual Patterns: Key Words And Corpus Analysis in Language Education (Studies in Corpus Linguistics)

Corpus linguistics around the world

Corpora and Language Teaching (Studies in Corpus Linguistics)

English Corpus Linguistics: An Introduction

University Language: A corpus-based study of spoken and written registers (Studies in Corpus Linguistics)

Corpus linguistics 25 years on

Corpora and Language Learners (Studies in Corpus Linguistics)

Keyness in Texts (Studies in Corpus Linguistics (SCL))

Structural Nativization in Indian English Lexicogrammar (Studies in Corpus Linguistics)

Corpus Linguistics: Method, Theory and Practice

Corpus Linguistics: Investigating Language Structure and Use

Corpus, Cognition and Causative Constructions (Studies in Corpus Linguistics)

Perspectives on Corpus Linguistics (Studies in Corpus Linguistics)

Corpus Linguistics

Lexis in Contrast: Corpus-based Approaches (Studies in Corpus Linguistics)

Lexis in Contrast: Corpus-based Approaches (Studies in Corpus Linguistics)

Collocations in a Learner Corpus (Studies in Corpus Linguistics)

Corpus Approaches to Grammaticalization in English (Studies in Corpus Linguistics)

Keyness in Texts (Studies in Corpus Linguistics)

Researching Specialized Languages (Studies in Corpus Linguistics)

Statistics for Corpus Linguistics

An Introduction to Corpus Linguistics (Studies in Language and Linguistics)

Small Corpus Studies and Elt: Theory and Practice (Studies in Corpus Linguistics)

Discourse In The Professions: Perspectives From Corpus Linguistics (Studies in Corpus Linguistics, SCL 16)

Corpus

Lexicology and Corpus Linguistics (Open Linguistics)

Corpus Linguistics in Literary Analysis: Jane Austen and her Contemporaries (Studies in Corpus and Discourse)

Corpus

Metadiscourse in L1 And L2 English (Studies in Corpus Linguistics)

Corpus and Context: Investigating Pragmatic Functions in Spoken Discourse (Studies in Corpus Linguistics, Volume 30)

Trends in Teenage Talk: Corpus Compilation, Analysis and Findings (Studies in Corpus Linguistics)

Textual Patterns: Key Words And Corpus Analysis in Language Education (Studies in Corpus Linguistics)

Corpus linguistics around the world

Corpora and Language Teaching (Studies in Corpus Linguistics)

English Corpus Linguistics: An Introduction

University Language: A corpus-based study of spoken and written registers (Studies in Corpus Linguistics)

Corpus linguistics 25 years on

Corpora and Language Learners (Studies in Corpus Linguistics)

Keyness in Texts (Studies in Corpus Linguistics (SCL))

Structural Nativization in Indian English Lexicogrammar (Studies in Corpus Linguistics)

Corpus Linguistics: Method, Theory and Practice

Corpus Linguistics: Investigating Language Structure and Use

Recommend Documents