Subjects in English: From Valency Grammar to a Constructionist Treatment of Non-Canonical Subjects 9783110589801, 9783110587258

252 91 3MB

English Pages 369 [370] Year 2018

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Subjects in English: From Valency Grammar to a Constructionist Treatment of Non-Canonical Subjects
 9783110589801, 9783110587258

Table of contents :
Preface
Contents
Abbreviations, symbols and frequently used labels
List of Figures
List of Tables
1. Introduction
2. Theory
3. Methodological Considerations
4. Clausal Subjects in active clauses
5. Analysis of subjects in passive clauses
6. Analysis of subjects in copular clauses
7. Extraposition
8. Existentials
9. Conclusions and perspectives for syntactic models
References
Index

Citation preview

Peter Uhrig Subjects in English

Trends in Linguistics Studies and Monographs

Editor Volker Gast Editorial Board

Walter Bisang Hans Henrich Hock Natalia Levshina Heiko Narrog Matthias Schlesewsky Amir Zeldes Niina Ning Zhang Editor Responsible for this volume Volker Gast and Niina Ning Zhang

Volume 321

Peter Uhrig

Subjects in English From Valency Grammar to a Constructionist Treatment of Non-Canonical Subjects

ISBN 978-3-11-058725-8 e-ISBN (PDF) 978-3-11-058980-1 e-ISBN (EPUB) 978-3-11-058728-9 ISSN 1861-4302 Library of Congress Control Number: 2018936674 Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. © 2018 Walter de Gruyter GmbH, Berlin/Boston Printing and binding: CPI books GmbH, Leck www.degruyter.com

| To my parents

Preface This book is a revised version of my doctoral thesis with the hardly marketable title Aspects of English Subjects – A valency-based analysis of item-specificity with a focus on clausal subjects, which I defended in 2013 at the Friedrich-AlexanderUniversität Erlangen-Nürnberg (FAU). The whole endeavour would not have been possible without the support of all the people whom I would like to thank in this preface. First and foremost, I am extremely grateful to my supervisor, Thomas Herbst, who was always there to discuss theoretical and methodological issues, gave great feedback, was extremely supportive and ever so patient with me when my dissertation fell out of focus again. Special thanks also go to the examiners, Stefan Schierholz, who was a critical reader and the expert on German linguistics, and Thorsten Piske, who was an examiner for my viva voce. I am very grateful to Dirk Siepmann, who made available a large portion of the corpus data on which this research is based, and to Gerold Schneider, who made available his parsed version of the British National Corpus to me. Thanks to Michael Klotz and Susen Faulhaber for many fruitful discussions on syntactic theory in general and valency theory in particular. I am indebted to the native speaker informants who went through the ordeal of judging all the strange sentences I found came up with or found in the corpora. In alphabetical order these are Jonathan Beard, Naomi Bishop, Anika Blacksmith, Amy Buer, Turlach O’Brion and Ned Reif. Special thanks go to the two people who endured by far the largest number of test items, Kevin Pike and David Heath. David also kindly proofread the entire manuscript and provided valuable feedback. Christina Sanchez-Stockhammer provided moral support, proofread parts of the manuscript and made excellent suggestions that helped me improve these parts. Though he was much less constructive in discussions of syntactic theory, Gerd Bayer still provided moral support and constant nagging, for which I am also grateful. Thanks to Thomas Proisl for help with the parsing of the corpora, to Besim Kabashi for always being able to provide suitable IT infrastructure and support when needed and to both for sharing their expertise and keeping me company during many enjoyable lunches. I also would like to gratefully acknowledge the compute resources and support provided by the Erlangen Regional Computing Center (RRZE), in particular by the High-Performance-Computing group, Thomas Zeiser, Michael Meier,

https://doi.org/10.1515/9783110589801-007

viii | Preface

Georg Hager and Gerhard Wellein. Moreover, I would like to thank the Studienstiftung des deutschen Volkes for their generous funding and exciting summer academies. Furthermore, I am indebted to Volker Gast, who provided very detailed and constructive feedback on an earlier version of the manuscript, and to Birgit Sievert, Julie Miess and Nancy Christ at De Gruyter for all their help and support in the process of turning the manuscript into a book. Heartfelt thanks go to my family. My parents, Anneliese and Michael Uhrig, have always believed in me and supported me on my way to the PhD. Together with my in-laws, Inge and Erhardt Förtsch, they made sure I found the time to finish the dissertation by babysitting and helping out whenever needed. Last, but definitely not least, my wife Sabrina deserves my gratitude for her patience, her moral support, and her love.

Contents Preface | vii Abbreviations, symbols and frequently used labels | xv List of Figures | xvii List of Tables | xix 1 1.1 1.2 1.3

Introduction | 1 Aims | 1 Brief account of the theoretical background | 1 Research questions and hypotheses | 3

2 2.1 2.1.1 2.1.2 2.2 2.2.1 2.2.1.1 2.2.1.2 2.2.1.3 2.2.1.4 2.2.2 2.2.2.1 2.2.2.2 2.2.2.3 2.2.2.4 2.3 2.3.1 2.3.2 2.3.3 2.4

Theory | 5 History of the term | 6 The beginnings | 6 Treatment in reference grammars over the past 100 years | 8 The role of the subject in generative grammar and in valency/dependency grammar | 16 Generative grammar | 16 Early generative grammar | 17 GB and Minimalism | 21 Generative Studies endorsing the subject concept | 25 Summary | 27 Valency models | 29 Basic tenets | 29 Criticism and multi-level approaches to valency | 31 The role of valency in grammatical models | 33 Note on the present study | 35 Towards a definition of subject in English | 36 Sentence/Clause | 36 Criteria | 37 Use of terminology in this study | 50 Summary | 52

3 3.1

Methodological Considerations | 53 English | 53

x | Contents

3.2 3.2.1 3.2.2 3.2.2.1 3.2.2.2 3.2.2.3 3.2.3 3.3

Data | 53 A Valency Dictionary of English (VDE) | 56 Corpora | 56 Representativeness | 56 Size and noise | 58 Annotation | 62 Native speaker interviews | 65 Acceptability/Grammaticality | 66

4 Clausal Subjects in active clauses | 71 4.1 Data analysis | 71 4.1.1 VDE data | 71 4.1.2 Corpus evidence | 73 Restrictions identified | 75 4.1.2.1 4.1.2.2 Problematic cases | 77 4.1.2.3 Summary | 78 Native speaker data | 78 4.1.3 Case study: modality and non-extraposed [to_INF] subjects | 81 4.2 4.2.1 Theoretical background | 82 Presentation of data | 83 4.2.2 4.2.3 Interpretation of the data | 86 4.3 Conclusion | 89 Analysis of subjects in passive clauses | 93 Short theoretical introduction | 93 Analysis | 96 Passive clause subjects as subset of active clause complements | 97 Passive clause subjects not found in the active clause | 100 5.2.2 5.2.2.1 Complement occurs only as passive subject | 100 5.2.2.2 Possible influence of other valency carriers in the clause | 105 5.2.2.3 Prepositional passive | 111 5.2.3 [to_INF] vs. [V-ing] | 115 5.3 Conclusion | 120

5 5.1 5.2 5.2.1

6 6.1 6.2 6.3

Analysis of subjects in copular clauses | 123 What counts as a copular clause? | 124 Theoretical treatment of copula verbs and copular clauses | 124 Data Analysis | 129

Contents | xi

6.3.1 6.3.1.1 6.3.1.2 6.3.2 6.3.2.1 6.3.2.2 6.3.3 6.3.4 6.3.4.1 6.3.4.2 6.4 6.4.1 6.4.2

Adjectives and extraposed clausal subjects | 130 [that_CL] but no [to_INF] | 131 [to_INF] but no [that_CL] | 134 Nouns and extraposed clausal subjects | 138 [to_INF] vs [that_CL] | 139 [V-ing] | 146 Case study: adjectives and non-extraposed clausal subjects | 149 Case study: nouns and non-extraposed clausal subjects | 153 Special case 1: of importance | 157 Special case 2: image | 158 Conclusion | 159 Summary of results | 160 Theoretical implications | 163

7 Extraposition | 169 7.1 Theoretical treatment | 170 Extraposition vs. dislocation | 173 7.1.1 7.1.1.1 The problem | 173 Evidence | 175 7.1.1.2 7.1.1.3 Conclusion for the present study | 177 7.1.2 Obligatory extraposition | 178 7.1.3 What is the subject? | 187 7.1.3.1 Survey of different positions | 188 7.1.3.2 Criteria for subjecthood revisited | 189 Summary | 199 7.1.3.3 7.2 Syntactic properties | 204 Extraposed to which position? | 204 7.2.1 7.2.2 [that_CL] with and without that | 206 7.3 Factors influencing extraposition | 209 7.3.1 Weight/Processing | 209 7.3.2 Information packaging | 212 7.3.3 Factivity | 214 7.3.4 Register | 217 7.3.5 Type of complement | 218 7.3.6 Type of valency carrier | 220 7.3.7 Valency carrier | 221 Other complements/valency pattern | 226 7.3.8

xii | Contents

7.4 7.5

Case study: tough movement | 228 Conclusion | 235

8 Existentials | 237 8.1 Phenomena covered | 237 8.1.1 Limitation to there-clauses | 237 8.1.2 Existentials vs locative there-clauses | 239 Survey of structures | 241 8.2 8.2.1 Bare existentials | 242 8.2.2 Extended existentials | 242 Relative clause extensions | 244 8.2.2.1 8.2.2.2 Infinitival extensions | 244 8.2.2.3 Participial extensions | 246 Extented existentials vs. bare existentials with postmodified or 8.2.2.4 complemented [NP] | 247 8.2.3 Presentational existentials | 251 What is the subject? | 252 8.3 Theoretical treatment | 263 8.4 8.4.1 Postverbal [NP] in extended existentials as subject of a small clause? | 263 Heavy [NP] shift | 264 8.4.1.1 8.4.1.2 Simplicity | 264 8.4.1.3 So-called ‘predicate restriction’ | 265 8.4.1.4 [NP]s which resist conversion into a clause | 266 8.4.2 Issues of agreement | 267 Further studies | 270 8.4.3 8.4.4 Proposed structure | 273 9 Conclusions and perspectives for syntactic models | 279 9.1 Data for linguistic analysis | 279 9.2 The concept of subject | 285 9.2.1 Problems of diverging subject properties | 285 9.2.2 Consequences for modelling ‘subjects’ | 288 9.2.3 A universal category? | 293 9.2.4 Summary | 294 9.3 Perspectives for grammatical models | 295 9.3.1 Item-specific selection of subjects | 295 Valency | 298 9.3.2 9.3.2.1 Valency patterns | 298

Contents | xiii

9.3.2.2 9.3.2.2.1 9.3.2.2.2 9.3.2.3 9.3.3 9.3.3.1 9.3.3.2 9.4

Long-distance dependencies and multiple valency carriers | 306 Tough movement | 306 Passives and small clauses | 311 Limitations | 313 The bigger picture | 314 Conflicting determinants | 315 A cognitive approach | 317 Summary | 327

References | 329 Index | 345 Appendix | available online

Note: Appendix available online https://doi.org/10.1515/9783110589801-222

Abbreviations, symbols and frequently used labels Symbols for linguistic units [to_INF]

a to-infinitive clause complement: To do so would confuse the others.

[that_CL]

a that-clause complement: That he turned out to be innocent really surprised me. in postverbal position also without the conjunction that: It’s a pity (that) he had to leave so early.

[V-ing]

an -ing-clause complement: Being with you has opened my eyes.1

[NP]

a noun phrase complement

[on_NP], ...

a particle phrase complement: He insisted on his right of way.

[...]

Sometimes, other symbols of this sort are used, e.g. [wh_CL] in I don’t know where she went. or [for_NP_to_INF] in It was stupid for him to do that. They should be transparent in principle. A full list of possible complements can be found in Herbst/Schüller (2008: 123f) or (in a slightly different notation) in the Valency Dictionary of English (Herbst et al. 2004: xv-xx).

Symbols for example status *

not acceptable

?

of questionable acceptability/dubious source/no agreement among native speakers

#

acceptable sentence but not in the given context or in a specific reading

|| 1 With -ing-forms, there is a gradient between clearly verbal and clearly nominal forms. The symbol [V-ing] is used for verbal forms, usually with a complement that shows that it is in fact verbal. Thus the underlined constituent in the following sentence is not classified as [V-ing] since the verb widen does not usually take an [of_NP] complement: [...] even in the countryside deep ploughing, ripping out of hedgerows and widening of roads have seemingly obliterated most traces of the ancient landscape. (BMT 32) https://doi.org/10.1515/9783110589801-015

xvi | Abbreviations, symbols and frequently used labels

Abbreviated literature CamG

The Cambridge Grammar of the English Language (Huddleston/Pullum 2002)

CGEL

A Comprehensive Grammar of the English Language (Quirk et al. 1985)

LGSWE

Longman Grammar of Spoken and Written English (Biber et al. 1999)

VDE

A Valency Dictionary of English (Herbst et al. 2004)

Sources of examples and data BNC

British National Corpus, XML Edition

AB1 234 and similar

example taken from the BNC; the first set of characters is the text ID, the number after the space is the s-unit (≈ sentence) ID within that text

HP7.123 and similar

example taken from a book of the Harry Potter series by J. K. Rowling; HP7 indicates volume 7; the number after the full stop is the page number

academic, britnews, fiction,

labels used for a set of additional corpora described in Section

newsmerge, newspapers,

3.2.2.1

popmerge

List of Figures Fig. 1: Phrase-marker for ‘sincerity may frighten the boy’ (adapted from Chomsky 1965: 65) | 17 Fig. 2: Structure of a sentence in the GB model (adapted from McCloskey 1997: 201) | 21 Fig. 3: Hierarchy of the grammaticalization of the argument relation (adapted from Jacobs 1994: 71) | 33 Fig. 4: Combined dependency/constituency diagram for “Pass your sister the meat.” (adapted from Matthews 1981: 91) | 34 Fig. 5: Varying scope of the term subject illustrated by Matthews (1981: 104; adapted) | 52 Fig. 6: Number of lexical units for the various types of clausal subject in the VDE | 72 Fig. 7: Plot of the distribution of modal verbs in the BNC and in the sample of sentences with [to_INF] subject | 84 Fig. 8: Distribution of modal elements among selected verbs occurring with [to_INF] subjects in the data collected for the present chapter | 85 Fig. 9: Distribution of modal verbs across all instances of selected verbs in the BNC | 86 Fig. 10: Complement block for complete (VDE 164) | 97 Fig. 11: Complement block for admit (A) (VDE 15) | 99 Fig. 12: Complement block for the general sense of remark (VDE 673) | 114 Fig. 13: Syntactic structure of (3) in Helbig’s model (adapted from Helbig 1992: 111) | 125 Fig. 14: Semantic structure of (3) in Helbig’s model (adapted from Helbig 1992: 111) | 126 Fig. 15: Representation of (3) in Welke’s model | 126 Fig. 16: Stanford typed dependencies representation of (3) | 127 Fig. 17: Analysis of a non-copular clause in Herbst/Schüller’s (2008) model | 163 Fig. 18: Analysis of a copular clause in Herbst/Schüller’s (2008) model | 163 Fig. 19: Possible structure for non-extraposed copular clauses | 164 Fig. 20: Modified structure for non-extraposed copular clauses | 167 Fig. 21: Complement block of appear (VDE 40) | 180 Fig. 22: Complement block for attract from the VDE (58) | 227 Fig. 23: Complement block for frighten from the VDE (334) | 228 Fig. 24: Relationships in example (73) according to word grammar (Hudson 1999: 201) | 267 Fig. 25: Graphical representation of (80) in Erdmann’s relational model (adapted from Erdmann 1976: 185) | 272 Fig. 26: Acceptance and rejection of test items for adjectives followed by a [that_CL] (see Appendix 19) | 282 Fig. 27: Feature structure for she in HPSG (Pollard/Sag 1994: 20) – © 1994 by The University of Chicago Press, reproduced by permission | 290 Fig. 28: Relevant features of walks in HPSG (Pollard/Sag 1994: 28) – © 1994 by The University of Chicago Press, reproduced by permission | 290 Fig. 29: Relevant features of existential there in HPSG (Pollard/Sag 1994: 147) – © 1994 by The University of Chicago Press, reproduced by permission | 291 Fig. 30: Relevant features of be as used in existentials in HPSG (Pollard/Sag 1994: 147) – © 1994 by The University of Chicago Press, reproduced by permission | 291 Fig. 31: Representation of the interrogative ‘yes-no-question’-construction (adapted from Herbst/Schüller 2008: 151) | 303 Fig. 32: Syntactic representation of (29) | 307 https://doi.org/10.1515/9783110589801-017

xviii | List of Figures

Fig. 33: Syntactic representation of the relevant part of (28) | 308 Fig. 34: Dependency representation of the relevant part of (30) | 309 Fig. 35: Two possible valency carriers for the subject in (31) | 310 Fig. 36: Two possible valency carriers for the subject in (32) | 310 Fig. 37: Two possible valency carriers for the subject in (33) | 311 Fig. 38: Possible dependency relations for (34a) | 312 Fig. 39: Possible dependency relation for (34a) | 312 Fig. 40: Simplified possible organization of NPs (adapted from Bybee 2007: 324) | 317 Fig. 41: Simplified possible organization of NPs (adapted from Bybee 2007: 325) | 318 Fig. 42: Aspects of a network representation of extraposed sentences | 320 Fig. 43: Non-extraposed and extraposed [that_CL] subjects together with the analogous structure with an [as if_CL] in a network representation. | 323 Fig. 44: Schematic representation of the frequency distribution of extraposed subjects with three adjectives (not to scale) | 326

List of Tables Tab. 1: Cartography of subject positions according to Cardinaletti (2004: 136) | 24 Tab. 2: Additional corpora used in the present study | 57 Tab. 3: comparison of google.de and google.com for looking forward to doing/to do | 61 Tab. 4: comparison of google.de and google.com for looking forward to having/to have with and without a filter for region | 62 Tab. 5: Clausal subjects with verbs in the range a – b in the VDE | 72 Tab. 6: Clausal subjects of the [to_INF], [that_CL] and [V-ing] type with verbs in the range a – b in the VDE, augmented by corpus data | 74 Tab. 7: Distribution of modal verbs in the sample of sentences with [to_INF] subjects | 83 Tab. 8: Distribution of modal verbs in the BNC and in the sample of sentences with [to_INF] subject | 84 Tab. 9: Summary of restrictions on non-extraposed clausal subjects of active verbs | 89 Tab. 10: Summary of the findings for complete | 98 Tab. 11: Summary of the findings for admit (A) | 99 Tab. 12: Summary of the findings for teach | 101 Tab. 13: Summary of the findings for help | 102 Tab. 14: Summary of the findings for support | 102 Tab. 15: Summary of the findings for settle | 103 Tab. 16: Summary of the findings for read (γ) | 104 Tab. 17: Summary of the findings for blame | 105 Tab. 18: Summary of the findings for [to_INF] and [V-ing] with allow (A ‘permit’) | 117 Tab. 19: Summary of the findings for [to_INF] and [V-ing] with permit | 118 Tab. 20: VDE/Patternbank data for extraposed [to_INF] and [that_CL] subjects; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses | 139 Tab. 21: VDE/Patternbank data augmented with corpus results; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses | 140 Tab. 22: Data from VDE/Patternbank, corpora and native speaker interviews combined; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses | 142 Tab. 23: Results of corpus research and native speaker interviews on clausal subjects in canonical position of copular clauses with adjectives starting with h as predicative element | 150 Tab. 24: Results of corpus research and native speaker interviews on clausal subjects in canonical position of copular clauses with nouns starting with i as predicative element | 154 Tab. 25: Summary of restrictions on extraposed subjects in copular clauses with adjectives as predicative elements | 160 Tab. 26: Summary of restrictions on extraposed subjects in copular clauses with nouns as predicative elements; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses | 161 Tab. 27: Summary of restrictions on clausal subjects in canonical position in copular clauses with adjectives as predicative element | 162

https://doi.org/10.1515/9783110589801-019

xx | List of Tables

Tab. 28: Summary of possible restrictions on clausal subjects in canonical position in copular clauses with nouns as predicative element | 162 Tab. 29: Extraposition and non-extraposition of [to_INF] subjects with active verbs according to the VDE; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses | 169 Tab. 30: Summary of the criteria for subjecthood with extraposition | 200 Tab. 31: Position of extraposed [that_CL] subject | 205 Tab. 32: Frequency of canonical and extraposed variants in spoken and written texts (250 each) from Kaltenböck (2004: 132) | 217 Tab. 33: Proportion of extraposition by complement type, based on Erdmann (1987: 41); figures not exact | 219 Tab. 34: Frequency of non-extraposition compared with subject it-extraposition; figures from Kaltenböck (2004: 152), who also includes other types of complements | 219 Tab. 35: Proportion of extraposition by complement type and valency carrier, based on Erdmann (1987: 41, 92, 94, 101–103); figures not exact | 220 Tab. 36: Proportion of extraposition by complement type and valency carrier, based on Kaltenböck (2004: 152f) | 221 Tab. 37: Obligatory extraposition in active clauses in the VDE; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses | 222 Tab. 38: Basic clauses and their existential counterparts; constructed on the basis of CGEL (1404) | 243

1 Introduction In a nutshell, what the present study sets out to show is that subjects are influenced by more item-specific idiosyncrasies than has often been acknowledged in grammatical theory. It will also be shown that the notion of ‘subject’ might not deserve the central place in grammatical description and theory it often holds.

1.1 Aims The purpose of the present study is twofold. First it tries to give a descriptive account of subjects in English or, to be more precise, of elements that scholars have in some way called subject in a description or model of English grammar. The second aim is to elucidate the status of the subject both as a concept in grammatical theory and with regard to the question whether the subject is determined by item-specific arbitrary formal restrictions in the same way as postverbal complements of the verb have been shown to be.2 The descriptive account is based on the documentation provided by reference grammars, on individual studies of the phenomena in question and, most importantly, on data specifically collected for this study, both derived from corpora and from native speaker interviews. The descriptive part allows us to review and possibly challenge claims made in grammatical theory based on sound empirical evidence. It has to be made very clear that the present study is centred on English and that any observations and results regarding the grammar of subjects are confined to the English language. No claims concerning the universal applicability of the results will be made. As to the theoretical status of the subject, a short introduction is necessary in order to explain the motivation behind the research questions formulated in Section 1.3.

1.2 Brief account of the theoretical background The theoretical background of the present study is valency theory (see Section 2.2.2 for an introduction). There is a fundamental difference between the treat-

|| 2 See Faulhaber (2011) for a very thorough study of such phenomena including a review of the relevant literature. https://doi.org/10.1515/9783110589801-021

2 | Introduction

ment of subjects in valency theory and in many other frameworks, most notably generative grammar. In the generative tradition (Chomsky 1965, [1981] 1993) any semantically compatible subject (or, in later stages of the theory, any subject that realizes a certain theta role) can occur in a clause and the subject is thus not formally determined by the verb, i.e. it is not subcategorized for in the same way as postverbal elements are. Valency theory on the other hand treats subjects as valency complements of the verb in exactly the same way as postverbal elements, which in turn means that it could allow for item-specific restrictions on subjects. In the tradition of dependency theory (Tesnière 1953, [1959] 1965), many valency descriptions of German3 (e.g. Helbig/Schenkel [1969] 1973) treat the subject as an obligatory complement of the verb (similarly Welke’s 2011: 167ff “Grundvalenz” [‘basic valency’]), and Emons’ (1978) description of English generally adopts this position. In his model of valency in English in a generative transformational framework, Allerton (1982: 45) draws a distinction between surface subjects and valency subjects, a concept which is echoed and extended in a different framework by Herbst/Roe (1996), who investigate the necessity of clause elements at three different levels and show that, from the point of view of verb valency, the complement realising the subject is very rarely an obligatory complement. As obligatoriness is one of the two most important criteria for complement status in valency theory, the status of the subject as a complement of the verb would be in question if the second criterion were found not to be met either. This second criterion is that the formal realisation of a complement is determined by the verb. In the case of, say, the [on_NP] after insist, this is relatively easy to establish. In the case of subjects, the vast majority of which are straightforward noun phrases,4 it is much harder to confirm such a relationship.5 This is why most of the discussion in the present study is devoted to clausal subjects even though their frequency in the language is marginal com-

|| 3 Much of the valency research done on English is inspired by the extensive literature on German valency grammar. 4 It is not particularly relevant for the purpose of this study to discuss whether NPs are noun phrases or determiner phrases (DPs). For an overview of the arguments for and against the “DP Hypothesis” see for instance Hudson (1984, 1990), Huddleston (1984) or for a generative position Abney (1987) and Bernstein (2001). 5 As much of the research on valency grammar was done in the context of German linguistics it should be pointed out here that in German the subject [NP] is (usually) in nominative case and thus formally determined. On the other hand, one could argue that English pronouns are formally selected in the same way and in French, the language for which Tesnière (1953, [1959] 1965) originally devised dependency grammar, case also only exists in pronouns.

Research questions and hypotheses | 3

pared to that of [NP]s: There are practically no verbs in English that do not allow an [NP] subject, so if we want to find syntactic restrictions on the formal realisation of subjects, we will have to analyse sentences with non-[NP] subjects. While the different forms of clausal subjects may not be interchangeable for various other reasons than the valency of the valency carrier, their distribution overlaps enough to allow for such a study.6

1.3 Research questions and hypotheses The major question that this study sets out to investigate is whether the form of the subject is arbitrarily restricted by the valency carrier in the clause. We can split the question up into sub-questions and formulate hypotheses for each of these in order to make them more manageable. One chapter will be devoted to each of the phenomena in question. Hypotheses: 1. The form of the subject of a non-copular active declarative clause is determined by the main verb of the clause (discussed in Chapter 4). 2. The form of the subject of a passive declarative clause is determined by the main verb of the clause (discussed in Chapter 5). 3. The form of the subject of a copular declarative clause is determined by the predicative element (discussed in Chapter 6). The reason for distinguishing between non-copular and copular clauses7 is that in the latter, the predicative element seems to exert more influence on the subject than the copula verb (see discussion in Section 2.2.2 and, more specifically, at the beginning of Chapter 6).8 The reason for distinguishing between active and passive clauses is that subjects in the latter are often treated as derived from active clause ‘objects’ and that their degree of ‘governedness’ may thus differ from active clause subjects.

|| 6 Some researchers in the generative tradition may argue that differences in distribution between various clausal subjects are also due to different types of implicit arguments. In the present study, these are not covered since the valency model used (Herbst et al. 2004, Herbst/Schüller 2008) attempts to model verb complementation without null elements. 7 Copular clause is shorthand for a clause in which the main verb is a copula verb. 8 Basically this approach mirrors the distinction often made between verbal predicates on the one hand and adjectival and nominal predicates on the other (see for instance Mair 1990).

4 | Introduction

Furthermore, research carried out during the compilation of the Valency Dictionary of English (Herbst et al. 2004; henceforth VDE) suggests that subject extraposition does not only follow general rules related to processing constraints and information structure (see discussion in Section 7.3 for details), but that there is a lexical element to extraposition. This would mean that some verbs allow for a clausal subject to be extraposed or not, whereas some others may the clausal subject to be extraposed. The same may be true of adjectives and nouns as valency carriers in copular clauses, so we shall give the respective hypothesis a rather general wording: 4. Whether subject extraposition is obligatory, optional or impossible is determined by the valency carrier, other factors being the same. (discussed in Chapter 7) There is a fifth hypothesis that is only loosely related to the first four but is highly relevant for a discussion of what grammatical models of English should treat as subject and whether the term really makes sens in the description of English. To address this question, extraposed and existential structures, i.e. structures in which the identification of subjects is less straightforward since there are multiple candidates, are reviewed in some detail: 5.

The concept of subject in grammatical theory represents an amalgam of properties that can be mapped onto a single constituent in canonical clauses but less so in non-canonical structures (discussed mainly in Chapters 7 and 8).

Subject is both one of the most controversial notions in linguistics and the one most often taken for granted. (Bakker/Siewierska 2007: 141) ‘Subject’ is a concept which has been notoriously difficult for linguists to define. (Rothstein 2001: ix) But the identification of a subject seems unusually straightforward. (Matthews 2007a: 104)

2 Theory Subjects in English are peculiar. For most English sentences most linguists will agree on what the subject is, but few would agree on a definition.9 Having been in use for over 2,000 years in the study of language, logic and philosophy, the term is often taken for granted, but if it is not, it is highly controversial. In this chapter, a brief overview of the history of the term will be given, including an outline of how it has been applied to English over the past 100 years. As Jespersen observed, “[a]n exhaustive critical examination of everything that has been said by grammarians and logicians on this question would require a whole volume” (Jespersen 1924: 145). Not surprisingly, the situation has not improved since Jespersen made that statement over 90 years ago. In addition, the present study is meant not to be exclusively a review of the relevant literature. Thus the discussion of the history will have to be limited to the few prominent approaches mentioned below, although there are many more and any such selection has to be to some extent arbitrary. After the descriptive perspective, a discussion of generative grammar and of valency theory will be carried out since these are two diametrically opposed models with regard to the status of the subject and since the theoretical background of the present study is valency theory. In the remainder of the chapter, criteria for subjecthood mentioned in various descriptive and theoretical works will be presented as an approximation to a definition of the term. The review of criteria will not only be helpful in determining the subject status of elements in non-canonical structures such as existentials later in the study, it will also allow us to reflect on our own use of the term subject.

|| 9 Koya (1992: 24) makes a similar point. https://doi.org/10.1515/9783110589801-025

6 | Theory

2.1 History of the term 2.1.1 The beginnings The term subject stems from Latin subiectum, which is a translation of Ancient Greek hypokeimenon (ὑποκείμενον). Although the term was not coined by the philosopher Aristotle, his influential writings, above all the Categories, introduced the twofold division of subject and predicate that is still used by many grammarians today. According to his definition, a subject10 is “that of which everything else is predicated, while it is itself not predicated of anything else“ (Metaphysics 1028b36).11 It has to be noted that Aristotle relates categories in the language to categories in the world: In Topics I 9 and throughout the Categories, Aristotle is concerned to establish that the basic types of linguistic expression correspond to the basic categories of reality. Complete thoughts expressed in sentences are the primary vehicles of meaning, but these consist of smaller units of meaning, significant words and phrases. These units are significant just because the extralinguistic world shares a similar structure. Aristotelian categories are categories both of beings and of linguistic items. In addition, the distinction between substance and property in the world is reflected in the distinction between subject and predicate in a sentence. In both instances, the existence of a single unified entity, a concrete particular or a complete thought or assertion, is consistent with the analysis into the more basic ontological categories of substance and property (quality, quantity, and so forth) or the more basic linguistic categories of subject and predicate, because the relation between substance and property or subject and predicate is such that a single item results from their combination. (Modrak 2001: 43; her italics)

We can thus already understand Aristotle’s approach in the linguistic sense in which it is understood today: “Simple sentences are formed by combining a single subject and a single predicate” (Modrak 2001: 44). According to Modrak (2001: 44f), “[s]imple sentences may be combined by conjunction or disjunction || 10Not all translations use the term subject in the cited passage: “Now the substratum is that of which everything else is predicated, while it is itself not predicated of anything else.” (Ross’ translation; my italics), “The substrate is that of which the rest are predicated, while it is not itself predicated of anything else.” (Tredennick’s translation; my italics) or “What underlies is that of which other things are predicated while it itself is predicated of nothing further” (Bostock’s translation; my italics). Wedin (2000: 183f) reads subject and makes a case against the phrasing “everything else” in the translation. 11 By convention, the works of Aristotle are cited following the numbering in Bekker’s edition of the complete works, irrespectively of the actual edition used. This reference indicates page 1028, second column, line 36 of the Metaphysics. The translation of the citation is by Modrak (2001: 154).

History of the term | 7

to form complex sentences” in Aristotle’s theory. Simplicity/complexity is determined on semantic grounds, however, so it is clear that the terms do not denote formal categories here.12 Seuren (1999: 42ff), on the other hand, maintains that Aristotle left a terminological gap where we expect what we call subject today. According to him, the term hypokeimenon is used only to refer to the “thing or entity” (Seuren 1999: 42) in the real world: The property assigned (if it is an accidental and not a necessary property) is called symbebēkós (accidens in Latin). The sentence part expressing the property is called katēgoróumenon or katēgórēma. No specific term, however, is created for the sentence part that refers to the hypokeímenon. (Seuren 1999: 42; his italics)

Seuren claims that the terminological confusion between subject as “the thing the sentence is about” and “the sentence constituent we now recognize as subject” (Seuren 1999: 42) lasted for many centuries.13 He found that the term hypokeimenon is applied to the grammatical subject from the fifth century AD onwards and terminus subiectus in the grammatical meaning in the sixth century (Seuren 1999: 43f). Around that time, a disambiguation in Latin translations of hypokeimenon into suppositum (reference term) and subiectum (subject term) can also be observed, although it was never consistently applied (Seuren 1999: 44). In the 19th century, linguists became aware that what they called logical subject14 did not always coincide with the grammatical subject of a sentence: “Die Grammatik aber, werden wir jetzt zu zeigen suchen, verdreht gar oft das logische Verhältniß von Subject und Prädicat” [“Grammar, though, as we shall try to show, often reverses the logical relationship between subject and predicate.”] (Steinthal 1855: 198). Steinthal’s logical subject is more or less Aristotle’s hypokeimenon.

|| 12 See also Modrak (2001: 45f, 72 and 155f) for further discussion. 13 In the light of psychological subjects, notional subjects, deep structure subjects, initial subjects and other concepts that will be discussed below, this citation reveals a rather optimistic perspective on the matter. Indeed one might be inclined to argue that the confusion is still as bad as it ever was. 14 Seuren criticises Steinthal and others for using the term “logical”, which is, he claims, due to “the widespread misconception that the subject-predicate distinction originated in Aristotle’s logic” (Seuren 1999: 46). Seuren himself might be mistaken in this respect, though, as subject and predicate were established terms in philosophy in general and logic in particular at that time, even though Frege’s account of the issue was still to come and even though he is right to observe that the distinction between verbal and cognitive concepts was not fully developed (Seuren 1999: 48).

8 | Theory

Seuren (1999: 46ff) continues to outline how the term subject came to be used as a discourse-related notion around 1900. From that point of view, a change in contrastive stress results in a change of logical (or psychological – both terms were used for the same concept by different scholars) subject as in (1)

a. SCOTT wrote Ivanhoe. (Seuren 1999: 46) b. Scott wrote IVANHOE. (Seuren 1999: 46)

where the stressed elements are considered “’logical’ predicates [...] and the uncontrasted parts are, correspondingly, the ‘logical’ subject” (Seuren 1999: 46). Von der Gabelentz (1869, 1891: 365–373) extends the division by distinguishing between grammatical, logical and psychological subject, where the latter is probably best described by the term theme in work done on functional sentence perspective in the Prague School (e.g. Mathesius 1929). In fact, Halliday’s definition of theme in functional grammar (Halliday 1994: 30–33) specifically cites the psychological subject in traditional accounts. The logical subject is probably best described as semantic argument in many current models (see for instance Helbig 1992: 13f, Ágel 2000: 176–179) or core element in frame semantics (Fillmore 2007: 133; a concept that does not correspond exactly to the notion of argument in other approaches but is largely compatible).15 As we have seen the term subject has come to be pervasive in the syntactic literature even though there has never been much consensus as to its definition (see Seuren 1999: 53 for a similar point).

2.1.2 Treatment in reference grammars over the past 100 years It is, indeed, unfortunate that the grammarian has to use the word “subject,” which in ordinary language means, among other things, also ‘topic’ (‘subject matter’). (Jespersen 1924: 146f)

It is commonplace in works on linguistic theory to refer to concepts from the realm of “traditional grammar” and to then state in what way one’s own analysis offers advantages over the traditional account (e.g. CamG 208). Interestingly, it often seems to be taken for granted that there is one uniform model of tradi-

|| 15 Fillmore states that the “core vs. periphery distinction is analogous to, but not identical to, the distinction in Tesnière (1959) between actants and circonstants. Core elements are those which are necessary to the central meaning of the frame” (Fillmore 2007: 133).

History of the term | 9

tional grammar and that every linguist is familiar with it, so that references and citations are unnecessary. However, even with the few traditional works used for the present study, there is considerable divergence in terminology, definitions and the syntactic model, so that some key concepts will be briefly discussed here. The selection of reference grammars for the present chapter was based on the influence they have had on the study of English over the past 100 years. For instance, Jespersen’s grammatical descriptions of English16 had an enormous impact on subsequent studies and are still frequently cited today. Zandvoort’s Handbook of English Grammar (1957/1972)17 is a very traditional account of English grammar but was tremendously successful, particularly in the teaching of English at university level before the London School grammars (Quirk et al. 1972 and derived works) became available. Quirk et al.’s seminal Comprehensive Grammar of the English Language (1985; CGEL) will be used as the basis for many terminological decisions. When the Cambridge Grammar of the English Language (CamG; edited by Rodney Huddleston and Geoffrey Pullum) was published in 2002, its aim was to “bridge the large gap that exists between traditional grammar and the partial descriptions of English grammar proposed by those working in the field of linguistics” (CamG xv). Thus, while the book upholds a tradition of comprehensive and descriptively adequate grammars, it marks a deliberate departure from many traditional analyses “to take account of the progress that has been made by linguists in our understanding of English grammar” (CamG xv). While its strong descriptive basis and its use of established terminology justifies its inclusion among the other reference grammars, its grammatical model is inspired to a great extent by studies in generative transformational grammar and other theoretical frameworks. Of all the grammars used here, it offers the most comprehensive and detailed list of criteria for the identification of subjects and thus is the most important source for Section 2.3.2 below. The discussion will advance in chronological order and thus start with Jespersen’s model. His definition of subject at the grammatical level remains rather vague and contains vivid imagery in addition to dubious criteria: || 16 For the sake of simplicity, Jespersen’s writings are treated as uniform in this chapter although there are differences between his big grammar (1909–1949) and his 1924 and 1933 books used here. The divergence is of no particular relevance for our purposes, though. 17 In the context of this study, the monolingual edition was used. The first edition, with comparisons to Dutch, was published in 1945 in the Netherlands. Shortly after the publication of the English version it was already criticised for its “neglect of contemporary (American and post-Saussurian) linguistics” (Aarts 1987: 69).

10 | Theory

Clearly to understand what the word subject means in its grammatical application, it will be well to recur to what was said in the chapter on the three ranks. In every sentence there are some elements (secondary words) which are comparatively fluid or liquid, and others (primary words) that are more firmly fixed and resemble rocks rising out of the sea. The subject is always a primary, though not necessarily the only primary in the sentence;18 this amounts to saying that the subject is comparatively definite and special, while the predicate is less definite, and thus applicable to a greater number of things. (Jespersen 1924: 150)

What exactly definite means in a predicate and how special can be crafted into a grammatical test is not elaborated upon, but Jespersen observes that there is a general “disinclination to take as subject a word with the indefinite article” (Jespersen 1924: 154) unless the latter is an instance of a generic article. The easiest test for subjecthood according to Jespersen is “taking the verb in the form in which it is found in the sentence, and asking Who (or What) followed by it: Tom beats John. Who beats? Tom” (Jespersen 1933: 98). Problematic cases are discussed a few pages later: If we analyse sentences like He happened to fall. He is sure to turn up. He is believed to be rich,– there can be no doubt as to the grammatical subject: it is he. But notionally the matter is not so simple: we cannot in the usual way ask: “Who happened?” “Who is sure?” “Who is believed?” We must either complete these questions by adding to fall, etc., or else we must ask: “What happened?” “What is sure?” “What is believed?” We thus discover that the notional subject is really a complete nexus, in which he is the primary, and fall, turn up, and be rich respectively is the secondary (adnex). We may express this in an unidiomatic way by saying that the notional subject, which is thus split in two, is he-to-fall, etc. (Jespersen 1933: 107f)

Jespersen also describes sentences of the type (2)

The path was easy to find.

in which the subject of the sentence, The path, is the object of find at the same time, although he proposes no further theoretical account of such phenomena. || 18 In Jespersen’s terminology, primaries are similar to heads in many contemporary works: “In the combination extremely hot weather the last word weather, which is evidently the chief idea, may be called primary” (Jesperson 1924: 96ff). This makes his definition of subject even more difficult to grasp, as most grammarians (and probably also Jespersen himself) would analyse not only the primary weather in The extremely hot weather caused hundreds of deaths. (www.veryen.org) as subject but the entire underlined [NP].

History of the term | 11

In analogy to his treatment of the sentences in the long quote above, “to-findthe-path” would probably have to be regarded as notional subject.19 For copular sentences in which two nominal elements can be interchanged, Jespersen (1924: 153) offers a test to determine the subject of the sentence: For him, the subject and the predicative element in (3a) and (3b) are interchangeable. In these cases, sentences of the structure exemplified in (4a) and (4b) have to be formed. Since (4a) is acceptable and (4b) is not, it follows that Miss Castlewood is the subject in (3a) and in (3b). (3)

a. Miss Castlewood was the prettiest girl at the ball. (Jespersen 1924: 153) b. The prettiest girl at the ball was Miss Castlewood. (Jespersen 1924: 153)

(4)

a.

I look on Miss C[astlewood] as the prettiest girl at the ball. (Jespersen 1924: 153) b. *I look on the prettiest girl at the ball as Miss C[astlewood]. (Jespersen 1924: 153)

This test thus wins over the obvious positional criterion as concord and the whoquestion do not deliver distinctive results. It is doubtful whether this is a test of grammatical status, though, as it actually checks whether one element is predicated of the other and predication is a logical more than a grammatical concept. Subjects of the [V-ing] type are generally called gerund in Jespersen’s terminology. [NP_V-ing] and [NP’s_V-ing] are treated as gerundial nexus (1933: 327f). Jespersen claims that [NP_V-ing] is often preferred in subject position because the [NP] is felt to be the subject of the whole sentence and thus should be in common case. This line of argument fails to account for the fact that in nonsubject positions [NP_V-ing] is also more frequently found than [NP’s_V-ing]. However, for our purposes, the most modern and important statements in Jespersen’s writings are about what a subject is not. While he acknowledges that often the subject provides given information (“relatively familiar element” [Jespersen 1924: 145]) and the predicate new information, he stresses that “[t]his may be true of most sentences, but not of all” (Jespersen 1924: 145) and accordingly rejects a definition of the grammatical element subject based on information structure. His remarks on the positional criterion show a similar carefulness:

|| 19 See Section 7.4 for a more detailed discussion of such structures.

12 | Theory

It will be well to point out that word-order is not always decisive, though in many languages there is a strong tendency, and in English a very strong tendency, to place the subject first. (Jespersen 1924: 151)

To prove his point that the very strong tendency is not a hard and fast rule, Jespersen cites the counter-examples in (5) himself: (5)

a.

Great was his astonishment when he saw the results (Jespersen 1924: 151; his italics) b. A scoundrel is Tom. (Jespersen 1924: 151; his italics)

Furthermore, he also shows the limitations of a semantically defined subject: The subject cannot be defined by means of such words as active or agent; this is excluded by the meaning of a great many verbs, e. g. suffer (he suffered torture), collapse, as well as by passive constructions […]. (Jespersen 1909–1949, vol. 3: 206)

We can thus see that Jespersen is fully aware that such criteria are prototypical in character rather than defining criteria. Zandvoort, on the other hand, adopts a traditional semantic definition of subject without giving any indication of possible mismatches between the syntactic and the semantic level: Most sentences of more than one word consist of two nuclei, one indicating the person or thing about whom or which a statement is made (or a question asked), the other containing the statement or the question asked. The word (or words) indicating the person or thing referred to is (are) called the SUBJECT of the sentence; that (those) containing the statement (or the question) the PREDICATE. (Zandvoort 1972: 196)

He then goes on to list example sentences with italicized subjects, most of which are straightforward, apart from some without a finite verb in them (“He a gentleman!” [Zandvoort 1972: 196]), which would probably cause problems with some grammatical criteria (e.g. concord) but are generally of minor importance. Quirk et al. (CGEL 724–726) do not attempt to define the term subject; they do, however, list its characteristic features, which will be discussed in more detail in Section 2.3.2. It will become clear that these properties are meant to describe a prototypical subject and are neither necessary nor sufficient criteria. The model used by CGEL assumes a constituent structure (not necessarily with binary divisions) and makes use of established terminology.20 Furthermore,

|| 20 See for instance Herbst/Schüller (2008: 167–172) for a critique of the traditional term object as used in CGEL.

History of the term | 13

it uses syntactic, semantic and pragmatic criteria for the definition (or rather identification) of subjects, which is probably the most serious point of criticism from the point of view of modern linguistics, even though many concepts from theoretical linguistics (such as semantic roles) are used. The authors are aware of all the subtleties uncovered by linguistic research and of the many exceptions to their rules; since their description of the English language does not claim to be a theoretically stringent model, it cannot be overly criticised for providing prototypical features instead of a clear-cut definition. As mentioned above, CamG is the reference grammar that incorporates the most recent advances in theoretical linguistics as will become apparent in the following short overview of their grammatical model. Although Huddleston/Pullum adopt a constituent grammar approach to language, their grammatical model is similar to valency grammar (see Section 2.2.2 for details) in that they see the verb as central element in the clause, i.e. “ultimate head of the clause” (CamG 50). Within their constituent structure model Huddleston/Pullum postulate “heads and dependents” (CamG 24) in their phrasal categories, thus incorporating dependency grammar elements. In their terminology, the head of a clause is the predicate, and the head of a VP is the predicator.21 This is terminologically and conceptually much closer to Chomsky’s 1965 and [1981] 1993 models than CGEL is, as is their treatment of the subject as an external complement. They attribute a special status to the subject, which is a “type of complement that is clearly distinguished, syntactically, from others” (CamG 53). Subject and object are contrasted in the following way: [The subject] is an external complement in that it is located outside the VP. It is an obligatory element in all canonical clauses. The object, by contrast, is an internal complement and, as just noted, is permitted – or licensed – by some verbs but not by others. (CamG 53; their emphasis)

While this could be read to mean that subjects are not licensed by verbs, the treatment of subordinate clauses in subject position suggests the opposite: “The prototypical subject is an NP; all verbs (and VPs) allow an NP as subject, but some license a content clause as well” (CamG 957). The type of clause is also determined by the verb:

|| 21 Compare also Aarts/Aarts (1982), who modify the Quirk et al. (1972) model to contain a predicate which includes a predicator and predicator complements of various types, thus attributing a status to the subject similar to that in CamG.

14 | Theory

When a clause functions as complement in the structure of a larger clause, the verb of the latter determines what kind of subordinate clause is permitted – whether declarative, interrogative, or exclamative, whether finite, infinitival, gerund-participial, and so on: [13] i Whether we go abroad / *That we go abroad depends on the cost. [...] Example [i] shows that depend licenses an interrogative clause, but not a declarative, as its external complement [...]. (CamG 220; their italics)

In fact, Huddleston/Pullum argue “that what licenses a subject of this kind [i.e. content clauses] is often not the verb alone, but the verb in combination with one or more complements” (CamG 958).22 Examples (6) and (7) illustrate the issue: (6)

a. John left me confused b. John left me.

(7)

a. That he died with $ 1.6 million in the bank left me confused. b. *That he died with $ 1.6 million in the bank left me.

One could, however, argue for two different lexical units of the lexeme leave in the sense of Cruse (1986: 80) in this case, which then again would mean that it might just as well be the verb which licenses the subject. Whether this restriction is a syntactic phenomenon or rather a matter of semantic compatibility could also be debated, as the difference between (8a) and (8b) below is arguably analogous to the one between (7a) and (7b) above, even though an [NP] functions as subject in both examples. (8)

a. This story really left me confused. b. *This story really left me.

It is interesting to note that the centrality of the verb seems to apply to copular23 clauses as well, as can be inferred from their interpretation of (9)

That he was guilty was obvious. (CamG 23; their emphasis)

|| 22 Oppenrieder (1991: 259) takes a similar position for German subject clauses. 23 In CamG only clauses featuring the verb be are copular clauses. The general construction of this type – including those verbs that are called copulas in other grammars (e.g. seem, appear in CGEL) – are called complex-intransitive (CamG 218 fn. 6). The rigid approach makes it difficult to assess the status of postverbal elements as in “Sue Brown provides an excellent example of a woman who has achieved outstanding success in the world of business while bringing up a large family” (CamG 255), where the constituent following the verb is treated as an object and not as a predicative complement.

History of the term | 15

where they state that the underlined constituent “is centred, ultimately, on a verb (was)” (CamG 23). This is in contrast to another statement: Where the superordinate clause is complex-intransitive it is the PC [=predicative complement] that licenses the external complement: cf. Whether we go abroad / *That we go abroad is dependent on many factors. This is one of the factors that makes a PC more like a predicator than is [sic] an ordinary complement. (CamG 221 fn. 9)

The fact that the predicative element licenses the external complement does not mean that the latter is a complement of the former, though: (10) a. Kim is fond of animals. (CamG 252; their emphasis) b. Kim loves animals. (CamG 252; their emphasis) CamG explains that fond in (10a) “has the experiencer aligned with S [=subject], a complement of the verb be, rather than of fond itself” (CamG 252) and goes on to clarify the relationship of the elements in a footnote: Since Kim is a complement of love but not of fond, it follows that while love is bivalent (has two complements), fond is monovalent (has only one). However, as semantic predicates love and fond are both dyadic (have two arguments). (CamG 252 fn. 27)

Thus CamG allows for divergent syntactic and semantic levels. The assignment of a central function to verbs and of licensing abilities to complements of verbs may be sensible but is not explained in a detailed enough way to allow a full understanding of how the different mechanisms at work (e.g. valency, licensing) interact. Much as in CGEL, the subject in CamG is defined via a range of criteria (which are listed below), but CamG is careful to include only “grammatical properties” (CamG 236) and not semantic and pragmatic features. To sum up, the concept of subject in descriptive reference grammars of English has evolved towards a multi-faceted notion that is defined based on a range of features in the most recent works. These features will be discussed individually in Section 2.3.2 and an evaluation of such definitions is offered in Section 9.2.

16 | Theory

2.2 The role of the subject in generative grammar and in valency/dependency grammar 2.2.1 Generative grammar Noam Chomsky’s theory of generative transformational grammar (Chomsky 1957, 1965 and with major modifications [1981] 1993 and 1995) had revolutionary impact on the study of language, not only in the United States but throughout the world.24 While many of his original concepts and ideas were revised in later works or abandoned altogether,25 the model which evolved from Aspects of the Theory of Syntax (Chomsky 1965) or the one presented in Lectures on Government and Binding (Chomsky [1981] 1993) were so influential and were the basis of so many scholarly works cited in this study that they have to be introduced here – at least briefly.26 The earliest of Chomsky’s published books (i.e. Chomsky 1957) will be largely ignored, though, as Chomsky (1965) contains an extended and revised version of that model (see Sampson 2001: 141–164 for a very critical account of the influence of Chomsky 1957 on linguistic research and the linguistic community).

|| 24 Whether revolutionary is the right word in this context is a question that caused a serious and heated debate. After Kuhn’s publication of The Structure of Scientific Revolutions in 1962, many adherents of generative transformational grammar claimed that there had indeed been a ‘Chomskyan revolution’ in linguistics. Critics denied this and a battle of publications started, including for instance Winston’s (1976) “Did a (Kuhnian) Scientific Revolution Occur in Linguistics?” (Answer: No.) or Newmeyer’s (1986) “Has There Been a ‘Chomskyan Revolution’ in Linguistics?” (Answer: Yes.). Koerner (1989: 101–146) provides a critical discussion of the topic and of the relevant literature; Kibbee (2010) – a Festschrift presented to Koerner – offers recent perspectives on Chomskyan (R)evolutions. 25 Chomsky himself has become critical of some of these concepts over time: Well, looking at it now, I think the minimalist critique of the past 10 years has given substantial reasons to suppose that none of these things exist: d-structure, s-structure, and LF just don’t exist. There’s no X-bar Theory forming d-structure. Of all those levels, the only ones that remain are PF and the semantic interface. The others, the strictly internal linguistic ones – LF, d-structure, and s-structure – probably don’t exist. (Chomsky 2004: 152 – the cited passage is part of an interview recorded by Naoki Fukui and Mihoko Zushi on November 22, 2002) [LF: Lexical Form; PF: Phonetic Form] 26 See Lyons (1991) or Horrocks (1987) for a detailed account of Chomsky’s model.

The role of the subject in generative grammar and in valency/dependency grammar | 17

2.2.1.1 Early generative grammar In the so-called Standard Theory (i.e. Chomsky 1965), subjects play a minor part. A subject is “a grammatical function rather than a grammatical category” (Chomsky 1965: 68; his italics) such as an NP and as such is “redundant, since the notions Subject, Predicate, Main-Verb, and Object, being relational, are already represented in the Phrase-marker [reproduced as Figure 1 below] and no new rewriting rules are required to introduce them” (Chomsky 1965: 69).

Fig. 1: Phrase-marker for ‘sincerity may frighten the boy’ (adapted from Chomsky 1965: 65)

In a hypothetical account, Chomsky proposes the definition of “Subject-of” in English “as the relation holding between the NP of a sentence of the form NP⁀ Aux⁀VP and the whole sentence” (Chomsky 1965: 69). While it is fully in line with the principles of a constituent structure grammar to model the subject function as a part-whole relationship, one page later the relationship can hold between an NP (or a noun, as there is no analysis offered) and a VP (or some constituent the exact status of which is not made clear) or between an NP and a predication: (7) (a) John was persuaded by Bill to leave [...] (c) what disturbed John was being regarded as incompetent In (7a), John is simultaneously Object-of persuade (to leave) and Subject-of leave; [...] in (7c), John is simultaneously Object-of disturb, Object-of regard (as incompetent), and Subject-of the predication as incompetent. (Chomsky 1965: 70)

18 | Theory

Thus, while Chomsky offers a tentative definition, he does not consistently follow it in his own discourse. Three pages later he also offers a definition of another relation, namely “Subject-Verb” as “the relation between the Subject-of a Sentence and Main-Verb-of the Predicate-of the Sentence” (Chomsky 1965: 73). After reviewing his hypothetical definitions, Chomsky comes to the conclusion that “it seems unnecessary to extend the system of rewriting rules” (Chomsky 1965: 74) to include information on the Subject of a sentence or the Subject-Verb relation. All these observations are related to the model’s grammatical deep structure and Chomsky (1965: 220f, Note 32) argues that the notion Subject-of cannot easily be extended to the surface structure and that a distinction of Topic/Comment parallel to Subject/Predicate could possibly be made at the surface level. Nonetheless, Chomsky seems to employ the traditional terminological distinction between grammatical (i.e. surface structure) subject and logical (i.e. deep structure, “underlying”) subject (e.g. Chomsky 1965: 23). For the research questions outlined in Chapter 1, it is particularly important to see that, “Verbs are not strictly subcategorized in terms of types of Subject NP’s or type of Auxiliary, apparently” (Chomsky 1965: 96). Whether finding counter-examples is enough to invalidate this claim is of course a matter of debate. The line of argument presented in the following short excerpt renders it incredibly difficult to reject any of Chomsky’s hypotheses on the basis of conflicting evidence within the generative framework: It must be borne in mind that the general rules of a grammar are not invalidated by the existence of exceptions. Thus one does not eliminate the rule for forming the past tense of Verbs from the grammar on the grounds that many Verbs are irregular; nor is the generalization that relates Manner Adverbials to passivization invalidated by the fact that certain items must be listed, in the lexicon, as conflicting with this generalization, if this turns out to be the case. In either the case of past tense or that of passivization, the generalization is invalidated (in the sense of “internal justification” – cf. Chapter 1, § 4) only if a more highly valued grammar can be constructed that does not contain it. It is for this reason that the discovery of peculiarities and exceptions (which are rarely lacking, in a system of the complexity of a natural language) is generally so unrewarding and, in itself, has so little importance for the study of the grammatical structure of the language in question, unless, of course, it leads to the discovery of deeper generalizations. (Chomsky 1965: 218)

It is easy in such a model to relegate any non-conforming behaviour to the lexicon,27 to the periphery28 or – where it becomes even less relevant – to perfor-

|| 27 The division of a grammatical model into a grammar and a lexicon component is one of the major points in generative grammar, although the distinction is much older. Thus Bloomfield is frequently quoted in this context: “The lexicon is really an appendix of the grammar, a list of

The role of the subject in generative grammar and in valency/dependency grammar | 19

mance.29 Stefanowitsch (2010) on the other hand argues that an item-based system can cover cases which follow general rules but that a rule-based system cannot handle idiomatic structures, which is a serious challenge to any model using the lexicon or periphery for exceptions.30 To sum up the 1965 position, subjects are “secondary notions that are definable on the basis of phrase structure representations of sentences” (Davies/Dubinsky 2001: 1), thus not much more than “epiphenomenal mnemonic terms” (Davies/Dubinsky 2001: 2). Nonetheless the notion of subject has lived on in generative thinking. A prominent indication of this is the introduction of the specified subject condition (SSC) in Chomsky (1973), which is of no particular importance here and was dropped in the 1981 model.31 It is noteworthy in this context that a condition is || basic irregularities” (Bloomfield 1935: 274). Bloomfield only covers morphology in the passage around the quote (e.g. that oxen as well as ox has to figure in the lexicon) and makes no mention whatsoever of unpredictable grammatical structures and the like, but his quote has often been understood to mean the same thing as the generative extension to such an idea of the lexicon, which is described with a vivid analogy by Di Sciullo/Williams: [The lexicon] contains objects of no single specifiable type (words, VPs, morphemes, perhaps intonation patterns, and so on), and those objects that it does contain are there because they fail to conform to interesting laws. The lexicon is like a prison – it contains only the lawless, and the only thing that its inmates have in common is lawlessness. (Di Sciullo/Williams 1987: 3) For a more detailed account of the evolution of the term lexicon and its use in different linguistic schools see Neef/Vater (2006), who put “special emphasis on structural and generative linguistics” (Neef/Vater 2006: 27), and Contini-Morava/Tobin (2000: xf). For a general overview of research on the lexicon in the generative tradition see also the other articles in Wunderlich (2006). 28 The observation is also made by Culicover/Jackendoff, who state that “’periphery’ tends to become a tempting dumping ground for any irregularity one’s theory cannot at the moment explain” (Culicover/Jackendoff 2005: 25). Note that the term periphery is an anachronism at this point in the discussion as it was introduced only in later works (see Chomsky [1981] 1993: 9ff, 1986: 147). 29 See Chomsky (1965: 4) for the definition of competence and performance. A brief discussion of the relevance of performance data to linguistic description will be carried out in Chapter 3. 30 Such a view was inspired by challenges to a dichotomy of lexicon and grammar on various unrelated grounds (e.g. Bybee 1995 on morphology, Sinclair 1991 on syntagmatic relations) and is in line with many recent theories that give up the distinction altogether. Goldberg (2006: 220) for instance agrees with Langacker in that “[l]exicon and grammar are not distinct components, but form a continuum of constructions” (Langacker 2005: 102). 31 Postal (1976: 170–179) gives an impressively large number of counter-examples and arguments against the SSC which lead him to conclude “that the SSC is an artifact, rather than a genuine principle of language or even a genuine principle of English grammar” (Postal 1976: 179).

20 | Theory

not a rule, which explains why in 1974 Chomsky himself subscribed to the statement that “within the theory of generative transformational grammar (TG) as understood by its original formulator, N. Chomsky, [… n]o transformational rule can refer to notions like subject, etc.” (Postal 1976: 151). As Postal goes on to show, not all of Chomsky’s followers had used the term with the same rigour and many transformational rules had been stated that had to rely on the notion of ‘subject’. Since Postal, one of the inventors of relational grammar (RG), firmly believes in syntactic relations such as subject, he is rather harsh in his criticism of the theory, although he is of course right in his observation of the lack of rigour: If it had been generally realized that TG in Chomsky’s sense precludes rules that refer to the subject, some of the work involving formulation of such rules might have been cast in the form of arguments showing the incorrectness of the theory. (Postal 1976: 152)

Most such rules also made no mention of the level at which the subject existed, which is particularly problematic if a grammar allows for deep structure and surface structure subjects, which leads to two possible referents of the term (namely John and Jill in (11)) in sentences in the generation of which transformations such as the passive have been involved. (11)

John had been shot twice by Jill.

Thus, Postal quite rightly demands that “[t]o be precise, any description in terms of subjecthood must specify the relevant level or levels at which the subject condition must be met” (Postal 1976: 156). McCloskey points out that most of the 1970s treatments which make use of the notion of ‘subject’ attribute a special status to it very much like ‘traditional’ grammar (and the opposite of many valency accounts, see Section 2.2.2) in that “the NP-daughter-of-S will be more prominent than any other argument position” (McCloskey 1997: 199). Since his own research on Irish rejects such a position, he claims that “the Aspects model is irredeemably English-specific” (McCloskey 1997: 200), which is what led some researchers to the false assumption that subjecthood could be regarded as a primitive of the theory of grammar. In the 1970s, adherents of relational or functional categories brought forward grammatical models as an alternative to mainstream generative transformational grammar. Relational grammar (Perlmutter/Postal 1974) and lexical functional grammar (Bresnan/Kaplan 1982; LFG) were the most prominent proposals. Their influence on Chomsky’s generative grammar remained limited, though.

The role of the subject in generative grammar and in valency/dependency grammar | 21

2.2.1.2 GB and Minimalism With the incorporation of X-bar theory into the model of generative grammar in Chomsky’s Lectures on Government and Binding ([1981] 1993), some of the criticisms made against an English-centric view of the supposed universal properties of deep structure were successfully countered (see McCloskey 1997: 200f). One of the major changes in the theory involved the abandonment of the category S, which was replaced by InflP (Inflection(al) Phrase/Inflectional Projection, often referred to as IP) as the highest projection in the sentence:

InflP Subj

Infl1 XP

Infl0

X X

Complement

Fig. 2: Structure of a sentence in the GB model (adapted from McCloskey 1997: 201)

Thus subjects now were the NP (or later DP) in the specifier position of IP, which allowed for cross-linguistic generalizations of no importance here. This position again treats the subject as being more prominent than all other arguments in the sentence. The new way of modelling sentences and their subjects also helped capture certain properties of subjects in terms of the generative model, for instance “the requirement that there be a structural subject” (McCloskey 1997: 202). The subject position can be assigned a theta role but does not have to (Chomsky [1981] 1993: 36) and it is the only element that has to be expressed even if no theta role has been assigned, so that “either the insertion of an expletive element or raising of a nominal from some lower position” (McCloskey 1997: 202) is required. In Chomsky ([1981] 1993), the term subject is widely used as a grammatical function and in various stages of the discussion. It is defined as GF [NP,S] ([1981] 1993: 42) as shown above but in a different context the term SUBJECT (in upper case, as opposed to subject) is described as the “subject of an infinitive, an NP or a small clause”32 ([1981] 1993: 209), which is a more depend-

|| 32 To illustrate the scope of this term, a few of Chomsky’s sentences are reproduced here:

22 | Theory

ency-like view of subjects. Despite the fact that the subject position may be assigned a theta role, it is still “a position not associated with a subcategorization feature of a lexical head” (Chomsky [1981] 1993: 36), i.e. the lexical entry of the verb does not contain formal specifications of the elements that can occur as subjects. In this respect, which is Chomsky’s answer to one of the research questions formulated in Section 1.3, the model is opposed to valency grammar as outlined in the next section.33 Still, regardless of the long discussion of subjects in Chomsky ([1981] 1993), they do not have an actual theoretical status – an observation also shared by McCloskey: The theory provides theoretical grounding for the observed prevalence of operations raising nominal phrases into “subject position” but it entails no reference, direct or indirect, to a theoretically primitive notion of subjecthood. (McCloskey 1997: 202)

By coincidence, the fact that Chomsky ([1981] 1993) uses the term subject far more than a hundred times although it is not a primitive of the theory and could be expressed in other terms is an indication of how difficult it is to discuss issues in linguistic theory without reference to (often preconceived) traditional notions everyone is (or believes to be) familiar with, even if the traditional term is just used as a kind of shorthand to establish reference. Davies/Dubinsky even claim that the use of the term in the “EPP [=Extended Projection Principle]34 elevated the notion ‘subject’ (albeit a configurationally defined notion) to the level of a UG determinant of core grammar” (Davies/Dubinsky 2001: 4).35

|| (ii) he wants (very much) [for JOHN to win] (iii) he believes [JOHN to be intelligent] (iv) [JOHN’s reading the book] surprised me (v) he considers [JOHN intelligent] Chomsky ([1981] 1993: 210) The capitalised element is always the SUBJECT. This is particularly interesting in the case of (iv) and (v) on which there is no agreement in the linguistic community. 33 In practice, valency grammarians often do not describe the subject either because it is taken to be uninteresting. 34 The EPP states that subjects are obligatory even if no theta role is assigned to them. While the issue is discussed in Chomsky ([1981] 1993: 38ff), the term was introduced slightly later: “I will henceforth refer to the Projection Principle along with the requirement that clauses have subjects as the Extended Projection Principle” (Chomsky 1982: 10; his emphasis). 35 Government Binding (GB) Theory is the theoretical background to many studies of individual phenomena related to subjects that were published in the 1980s, even if they challenge certain positions. Thus Napoli (1989) provides a study of predication theory based on GB even though she uses concepts that can be found in theoretically distinct approaches related to valency/dependency grammar (e.g. Napoli 1989: 7).

The role of the subject in generative grammar and in valency/dependency grammar | 23

In the years after the publication of GB theory, the notion of a “subject position” was severely challenged by two independent movements which could be regarded as theory-internal since they argued for changes in the model without opposing the general framework as such in the same way Relational Grammar and lexical functional grammar mentioned above did. The probably more influential of the two, the so-called “Internal Subject Hypothesis”, claimed – for reasons that cannot be discussed here (see McCloskey 1997: 203–225 for a very detailed account) – that subjects are actually base generated in the Verb Phrase (VP) and end up in their position only after a movement operation. This position is, among others, taken by Alexiadou/Anagnostopoulou (2001, 2007), who give detailed reasons why the subject (or, in some languages the object) has to be externalized from the VP. While this looks like a small albeit eccentric theoryinternal discussion, it has far-reaching consequences on the status of the subject: One of the properties of the Internal Subject Hypothesis is that it blurs in a fundamental way the distinction between internal arguments and the external argument (a distinction introduced by Williams in 1981). If all semantic role assignment is accomplished within lexical projections, then the structural difference between internal and external arguments is lessened. (McCloskey 1997: 220)36

McCloskey (1997: 220) cites an interesting idea by Marantz (1984: 23–31) in this context, namely that the subject is not an argument of the verb (whereas complements of the verb are) but an argument of the VP as a whole (which is echoed in CamG almost 20 years later, see the discussion above).37 Recent approaches, most imporantly Kratzer (1996), try to model this behaviour through the introduction of a further null element in the phrase structure tree which assigns the relevant theta role. The second challenge to the “subject position” proposed in Chomsky ([1981] 1993) is the so-called “Split Infl Hypothesis” which – as the name suggests – claims that “Infl has a complex internal phrase structure” (McCloskey 1997: 216). Thus, according to Pollock (1989), who first suggested such an analysis, there are reasons for splitting up the category ‘Inflection’ into an agreement and a tense component. This in turn – in combination with often binary branching –

|| 36 Since the GB model (and McCloskey, too) aims for universality, it should also be mentioned that research on German (Oppenrieder 1991: 351) claims that S and VP are identical and that accordingly the subject cannot be an external complement in German, either. 37 See Bresnan (1982: 348–353) for a critique of Marantz’ arguments (presented in his 1981 dissertation) and substantial counter-evidence.

24 | Theory

leads to more levels from which the subject can originate in the deep structure, and positions outside of VP but not in the final (surface) “subject position” have been suggested, among others by McCloskey (1996) on the basis of evidence from Irish, a VSO language.38 Subjects may in fact undergo a series of movement operations before they reach their final position.39 Similarly, according to Cardinaletti’s “specialization hypothesis” (1997: 50), there is more than one subject position, the choice of which seems to depend on the type of subject (weak/strong pronoun, pre-/post-verbal etc.). This approach is extended in Cardinaletti (2004) and culminates in the following “cartography of subject positions” (Cardinaletti 2004: 136): Tab. 1: Cartography of subject positions according to Cardinaletti (2004: 136)40

SpecSubjP

Subj° SpecAgrSP AgrS°* spec??P

DPs

Weak pronouns

Strong pronouns

pro

Weak pronouns (egli/esso)

Weak pronouns

??° spec?P

?°* SPECVP

DPs

DPs

Strong pronouns

Strong pronouns



Predicate DPs

It is noteworthy that Cardinaletti introduces a “subject-of-predication feature” (2004: 126), which is then elaborated into a “subject criterion” by Rizzi and Shlonsky (Rizzi 2006, Rizzi/Shlonsky 2006, 2007). The subject criterion – named in analogy to other criteria Rizzi introduced into minimalist syntax (e.g.

|| 38 It appears that no general agreement whatsoever has been reached as to the exact nature of these elements within the generative community. Thus whether the tense projection (TP) dominates the agreement projection or the other way round, whether there actually is an agreement projection ARGP or only ARG, whether there are two ARG projections, one dominating TP and the other being dominated by it, is still a matter of (heated) debate (see the introduction in Boeckx 2006 for details). 39 Falk (2006a), who has a background in lexical functional grammar (LFG), also rejects the notion of a single subject element on typological grounds: “Given the properties of mixedsubject languages, the conclusion that subject needs to be factored into two distinct grammatical functions, as we have done here, is inescapable” (2006a: 197). 40 There is also a cross-linguistic analysis based on Cardinaletti’s approach in Cardinaletti (2014).

The role of the subject in generative grammar and in valency/dependency grammar | 25

the Topic Criterion) – aims to replace Chomsky’s EPP. However, the basic idea that subjects are assigned theta roles and not subcategorized for seems to remain unchanged even in the latest studies.

2.2.1.3 Generative Studies endorsing the subject concept There are two studies worth mentioning here that are rooted in the generative paradigm but depart from it on the subject issue in that they endorse a subject function and regard it as a necessary element of the theoretical description rather than an epiphenomenon. One of the most frequently cited studies on the definition of subject is Keenan’s (1976) article “Towards a universal definition of ‘subject’”, which, as the name suggests, is based in the paradigm of universal grammar.41 Keenan produces a list of some 30 criteria for the identification of subjects and insists that “the subjecthood of an NP (in a sentence) is a matter of degree” (Keenan 1976: 307), which Pullum rejects as a “curious idea” (Pullum 1980: 1).42 As none of Keenan’s properties is both necessary and sufficient, he claims that “an NP in a b-sentence (in any L [=language]) is a subject of that sentence to the extent that it has the properties in the properties list below. If one NP in the sentence has a clear preponderance of the subject properties then it will be called the subject of the sentence” (Keenan 1976: 312; his emphasis). Thus Keenan is in some way in line with the gradient and feature oriented definitions given in the English reference grammars since he describes a prototype,43 but draws on evidence from a wide range of languages. In consequence, not all of his properties are of relevance to the definition of English subjects and some others can be rejected to some extent with the help of counter-examples. For instance, the independent existence of the entity referred to by the subject as opposed to the entities referred to by other NPs is asserted and illustrated using sentences such as (12a–

|| 41 Of course, this is not the pure Chomskyan breed of universal grammar, as Keenan argues for the need to include the concept ‘subject of’ (as the subject of a sentence, not a verb) in universal grammar “in order for the many universal generalizations which use this notion to be well defined” (1976: 305) – a position incompatible with the Chomskyan view that subjects are not primitives of the grammar but only derived notions. 42 For methodological reasons, Keenan only looks at basic sentences (“b-sentences”) which he defines as the most basic sentences expressing a “complete thought” (1976: 308). For him, passives are, “in general, less basic than the corresponding actives” (1976: 310). See Johnson (1977: 675f) for a critique of Keenan’s basic sentences and other aspects of his discussion. 43 See also Welke (2002: 137ff).

26 | Theory

c), where the subject “exists independently of the action or property expressed by the predicate”44 (Keenan 1976: 313): (12) a. b. c. d.

A student wrote a poem. (Keenan 1976: 313) He defined a term. (based on Keenan 1976: 313) She proved a theorem. (based on Keenan 1976: 313) The research project has emerged from a British Council Exchange Link between the Departments of Geography at the Universities of Keele and Zimbabwe. (HJ1 20312)

In (12d), however, the situation is reversed given that the British Council Exchange Link exists independently whereas the research project, which would have to be identified as subject (at the grammatical level and potentially others, too), does not. This would speak in favour of a verb-centred analysis where it is the verb which assigns semantic roles (and possibly semantic properties) to the clause elements and not a grammatical component that assigns these semantic properties independently of the verb used. Although Keenan identifies different groupings of his properties – e.g. case marking properties, semantic role, and immediate dominance – he seems to maintain a unity of the subject concept. For him, it appears, there is only one subject in a sentence. A discrimination of subjects at different levels (grammatical, notional, ... – see above) does not occur. In sum, Keenan’s article is probably less helpful for the definition of the subject in English than the reference grammars (CGEL and in particular CamG; see Section 2.3.2), mainly due to his attempt to cover all languages. With Predicates and their subjects, Rothstein (2001) presents her impressive model of subjecthood, which is deeply rooted in the generative tradition, even though she rejects the generative treatment of subjects: The central idea […] is that the subject argument is special because it is a structurally selected argument of a structurally defined constituent, and not a lexically selected argument of a lexical head (although a lexical head may constrain what can fill the subject position). (Rothstein 2001: x)

In her theory, whose principles were first put forward in 1983, the structurally defined constituent in the quotation is the syntactic predicate, which is unsaturated by default and needs a subject for saturation. This is a view which is strik-

|| 44 Keenan’s criterion echoes Aristotle’s original distinction between subject and predicate cited in Section 2.1.1.

The role of the subject in generative grammar and in valency/dependency grammar | 27

ingly similar to Herbst/Roe (1996) in some respect, as they also separate syntactic from lexical (i.e. valency) requirements for complements. Rothstein argues in the same vein that subjects are obligatory even if there is no thematic reason for them to be realised, i.e. if all semantic role requirements are already fulfilled. Thus in (13), the subject is needed “despite the fact that the VP looks like a thematically saturated constituent” (Rothstein 2001: x). (13)

It seems that this sentence needs a subject. (Rothstein 2001: x)

Contrary to established generative theory, she claims “that it is this syntactic definition of subject as ‘subject of a syntactic predicate’ which explains the properties which subjects in English have,” (Rothstein 2001: xi) – a radical departure from the idea that subjects are constituents in a tree that can be defined via their position in this tree and stand in a relation to the clause as a whole.45 Rothstein also rejects any definition of subject that is related to aboutness as “there is nothing about the world which makes it true that a proposition is ‘about’ a single element” (Rothstein 2001: 4) and she illustrates her claim with examples of the type mentioned in Herbst/Klotz (2002) as well: (14) a. John and Mary first met in 1989. (Rothstein 2001: 4) b. John first met Mary in 1989. (Rothstein 2001: 4) Of course she is nonetheless happy to “make the fairly non-controversial claim […] that the unmarked case in English is that the sentence subject gives the sentence topic” (Rothstein 2001: 8) without accepting a definition that uses the observation as a criterion.

2.2.1.4 Summary To conclude this subsection it seems fair to state that today it is even less clear what a subject is in generative grammar than it was at its outset.46 No uniform concept has been adopted, and the refinement and extensions of the supposedly universally applicable rules and positions have led to an immensely complex

|| 45 This claim is made more verbose later in the book: “’subject’ is subject of a predicate and not ‘subject of a clause’” (Rothstein 2001: 59) and “predication is a syntactic relation which is independent of theta-role assignment holding between a predicate and a non-predicate (its ‘subject’)” (Rothstein 2001: 60). The latter statement is interesting as it appears that not the verb but the entire predicate assigns theta roles, which is opposed to a valency view but in line with the standpoint taken by Marantz (1984) or – for certain constructions – by CamG (957f). 46 For a detailed model of subjecthood in a minimalist framework, see Mohr (2005).

28 | Theory

apparatus for the generation of even the simplest English sentences. Even though it is not a primitive of the grammar and a decreasingly sensible category in their model, generative grammarians would still identify the same element as surface subject in the vast majority of English sentences that most other accounts would, too. A good summary of the development of the term in generative grammar is in fact provided by McCloskey’s (1997) introduction: [I]n the tradition which extends from the “Standard Theory” through the “Extended Standard Theory” to “Principles and Parameters Theory” and then to the “Minimalist Program”, the notion of subject plays no formal role at all. Not only is “subject” not a primitive term in these theories, but in their most recent instantiations it is not even clear that there is any derived or defined notion which captures the traditional intuition of what a subject is (as there was, for instance, in the theory of Chomsky 1965). What we have seen, in a sense, is a progressive deconstruction of the traditional category “subject” so that the properties which are supposed to define it are distributed across a range of distinct (but derivationally linked) syntactic entities and positions. (McCloskey 1997: 197)

Davies/Dubinsky (2001: 8) point out that such proposals can be found ten or twenty years earlier in relational grammar. In relation to our research questions, we can thus state that subjects in generative grammar are not syntactically selected by the verb (i.e. not “subcategorized”), even though they may be assigned a theta role.47 Such a view is incompatible with the approaches presented in the following section. On the other hand, due to the Extended Projection Principle, dummy subjects48 can be provided by the grammar if no theta role for a subject is assigned. Such a position is shared by many valency grammarians and is relevant for the treatment of extraposed structures (see Chapter 7 for discussion).

|| 47 Before the incorporation of theta roles into the model, semantic selection was proposed as a means of restricting the use of any item as subject with any predicate: Predicates are not (Chomsky (1965)) subcategorized for subject NPs – every predicate of English takes a subject. Selection for complement types, however, extends to subjects, as the examples in (i) and (ii) show: (i) a. Who was shot in the fight last night is still unclear. b. *Who was shot in the fight last night is unusual. (ii) a. *For a man to be shot in a fight is unclear. b. For a man to be shot in a fight is unusual. Grimshaw (1979: 314) 48 In the generative model, so-called subject raising may take place instead.

The role of the subject in generative grammar and in valency/dependency grammar | 29

2.2.2 Valency models The terms dependency and valency are often treated as a necessary pairing. Both are usually attributed to Lucien Tesnière’s Eléments de Syntaxe structurale ([1959] 1965) and some of his earlier works, although the history of the terms is not as straightforward as it may seem.49 For a full theoretical background of valency/dependency theory see Heringer (1970, 1996), Welke (1988, 2011), Helbig (1992). For treatments of English, see Emons (1974, 1978), Allerton (1982), and both Herbst et al. (2004) and Herbst/Schüller (2008) as the immediate background to the present study. While Herbst/Schüller take a valency approach, their model is not a dependency grammar (see Section 2.2.2.3 for details). Furthermore, the concept of valency is often extended to noun and adjective valency. Thus see Herbst (1983) for adjective valency in English and Schierholz (2001) or Hölzner (2007) for noun valency in German. Teubert (2003) is a general overview of noun valency and Groß (2003) of adjective valency. Fischer (1997) offers a contrastive perspective on German and English verb valency.

2.2.2.1 Basic tenets To begin with, let us briefly introduce some basic concepts of valency theory, starting with the idea of valency itself: “The basic assumption of valency theory is that the verb occupies a central position in the sentence because the verb determines how many other elements have to occur in order to form a grammatical sentence” (Herbst et al. 2004: xxiv). Elements whose occurrence is specific for a given valency carrier are called complements; all other elements in a sentence are called adjuncts. The distinction between complements and adjuncts, though not always straightforward,50 is fundamental to valency grammar. Furthermore obligatory and optional complements can be distinguished.51 Allerton (1975, 1982) introduces a third option in the form of a contextually optional complement,52 which does not have to appear in a sentence if it is identifiable || 49 For a detailed account of valency and dependency in Tesnière’s works see Askedal (2003), who includes references to historical overviews as well. 50 See Somers (1987) for a thorough review of criteria and tests for complement status. 51 Adjuncts are by definition optional. Only few valency grammarians agree with Welke’s (1988) distinction between optional and obligatory adjuncts. 52 Allerton speaks of indefinite deletion and contextual deletion; the term contextually optional is preferred by Herbst, who says that “the use of the term deletion implies that the divalent use

30 | Theory

from the context, which is taken up by Herbst et al. (2004) and Herbst/Schüller (2008). The status of the subject in valency theory is that of a complement of the verb.53 In German valency theory, the subject in active declarative clauses is usually treated as an obligatory complement (e.g. Welke 2011: 55) because it has to be present in such clauses and has nominative case marking. Such an approach is also taken in VALBU (Schumacher et al. 2004), which can lead to the conclusion that active declarative clauses are apparently regarded as primary. In a similar vein, Welke (1988: 69f) states that the valency of the verb in an active declarative clause is the basic valency that is stored in the lexicon entry of the verb in question and that verbs used in passive sentences show a valency reduction.54 Herbst/Roe (1996) offer a different account: They argue against a derivation of passive clauses from active clauses (and thus of passive valency from active valency) and regard them as structural alternatives instead.55 If active and passive sentences are structural alternatives and if the complement that realizes the KILLER argument is not present in (15) below, it cannot be treated as an obligatory complement of the verb kill in (16) either since if we treat kill as the same verb in both sentences, anything that is obligatory from the point of the view of the verb would have to be present in (15), too: (15)

Hagrid’s roosters were killed! (HP2.216)

(16) a. Voldemort killed my parents, remember? (HP1.197) b. *Killed my parents, remember?

|| represents a primary pattern which is modified through a process of deletion, which is a view that we would not subscribe to. In particular, the term deletion seems inappropriate if it is taken to mean ‘verbatim recoverability’ because no such requirement holds in this case” (Herbst 1999). According to Helbig (1992: 99) and Heringer (1996: 162), contextual optionality is the default case of optionality and pure optionality is less common. 53 Tesnière already states that the subject should be regarded as “un complément comme les autres” (Tesnière [1959] 1965: 109). Wegener (1990: 153f) criticises this view for the description of German from a generative position on the basis of a variety of arguments, the two most important of which may be psychological reality (e.g. if we cite verbs, we use forms such as “give sb. sth.”) and economy (since it is highly uneconomical to specify for nearly every verb in the language that it takes a nominative complement). 54 Heringer (1996: 86) on the other hand assumes that the quantitative valency is constant in the active and the passive voice in German. 55 This treatment has to be seen in connection with their introduction of three levels of necessity discussed below.

The role of the subject in generative grammar and in valency/dependency grammar | 31

It follows logically that the fact that (16b) is ungrammatical cannot be due to Voldemort being an obligatory element of the verb kill in (16a). Accordingly, there must be some other requirement, which was named structural necessity by Herbst and Roe (1996). Thus the requirement that all finite active declarative clauses have to have a subject in English56 exists at a structural level independently of valency. In our example, kill has an optional complement that represents the killed being and another optional complement that represents the killer; nonetheless the killer, Voldemort, has to be present for structural reasons in (16). The third level of necessity introduced by Herbst and Roe is communicative necessity. At this level, elements that do not even have complement status can be obligatory. (17) a.

“Professor Dumbledore left ten minutes ago,” she said coldly. (HP1.194)

Thus ten minutes ago in (17a) might well be necessary at the communicative level if the sentence is taken as an answer to (17) b. When did Professor Dumbledore leave Hogwarts? even though it would have to be treated as an adjunct from a valency perspective.

2.2.2.2 Criticism and multi-level approaches to valency The fundamental distinction between complements and adjuncts presented in the previous section was seriously challenged in German linguistics from the 1980s onwards. While Helbig (1992) discusses the extension to logical, semantic and pragmatic levels the concept of valency had seen in the 1970s and accordingly distinguishes up to six levels of valency description, he still maintains the distinction between complements and adjuncts, even if he is fully aware that there are many issues and that the distinction thus may be a case of gradience (Helbig 1992: 94ff; see also Somers 1984 and 1987). Similarly, the model employed in the VDE makes use of valency slots (roughly Helbig’s logical level), semantic roles, a syntactic level and often lists semantic features of the slots in || 56 This position is in fact compatible with Chomsky’s Extended Projection Principle (see above) at first sight, but the perspective is of course different since in valency grammar the subject is analysed as a complement of the verb and Herbst/Roe only make a statement about the obligatoriness of complements, not about filling a non-theta slot with another element. See also the discussion by Matthews (1981: 275–278).

32 | Theory

the notes on meaning. Despite the problems of the complement/adjunct distinction discussed in detail in the front matter of the dictionary, the distinction is maintained. The most important reason for this choice is that both Helbig’s and Herbst et al.’s approaches are mainly targeted at lexicographic applicability, so didactic considerations have to play an important role in their models. From a more theoretical perspective,57 Jacobs (1994) criticizes the prevalent valency models58 and suggests a multi-dimensional concept of valency. His alternative model was taken up a in a wide range of publications (see Welke 2011: 57ff for references) and forms the basis of many analyses and valency models (for instance Ágel 2000: 167ff) even though it reads like a deconstruction of the concept of valency altogether. In fact, he compares the term valency to the terms subject and object in that they all are simplifications that combine several phenomena that often occur together but are independent of one another in principle into one idealized and overgeneralized concept (Jacobs 1994: 68). We shall use the revised version of Jacobs’ model hinted at in the postscript to his book to illustrate the multi-dimensional character of such a recent understanding of valency. According to his analysis, the identification of complements of a valency carrier is based on the following four independent relations59 – obligatoriness (“Notwendigkeit”; NOT) – logico-semantic argumenthood (“Argumenthaftigkeit”; ARG) – formal specification (“Formale Spezifität”; FOSP) – semantic specification (“Inhaltliche Spezifität”; INSP) Since ARG is a relation at the so-called logical level of linguistic description and INSP at the semantic level, the two most relevant relations for the present study – since it aims to be a study of syntax – are NOT and FOSP. Also, these two are the highest-ranking elements in a grammaticalization hierarchy proposed by Jacobs:

|| 57 It makes sense to follow Ágel here: “Um theoretische Probleme adäquat angehen zu können, müssen theoriebezogene und praxisbezogene Argumentation sauber auseinander gehalten werden“ [“To be able to tackle theoretical problems appropriately, theory-related and practice-related argumentation have to be strictly separated.”] (Ágel 2000: 171). 58 His original manuscript was written in 1986 and circulated widely before its publication in 1994, but his analyses and interpretations were just as valid then. The published version contains a postscript written in 1993, in which Jacobs modifies parts of his theory. Jacobs (2003) is a further modification. 59 Jacobs’ model originally contained seven relations (1994: 14ff); Ágel (2000: 167ff) distinguishes five, but the four presented here are what one might want to call a common core.

The role of the subject in generative grammar and in valency/dependency grammar | 33

{ NOT FOSP }

> INSP > ARG

Fig. 3: Hierarchy of the grammaticalization of the argument relation (adapted from Jacobs 1994: 71)

Given that in the previous section we showed that basically NOT (i.e. obligatoriness; related to not being able to occur freely) is one criterion for complement status and that FOSP (i.e. the formal determination of a complement by the valency carrier) is the other, Jacobs’ criticism is not so much a deconstruction of the idea of valency but rather a (necessary and important) wake-up call to remind linguists that a mixture of relations (and accordingly criteria) is responsible for the many problematic cases of distinctions between adjuncts and complements (see also Helbig 1992 for a discussion) and that the arguments exchanged by researchers were often at different levels or focused on different relations, which made a fruitful discussion impossible. Jacobs himself takes back some of his criticism in the postscript and accepts valency as a prototypical bundle of relations (1994: 71).

2.2.2.3 The role of valency in grammatical models While dependency models usually need some sort of valency concept, not all models that make use of valency can be classified as dependency models. Many German grammars are dependency models with a valency component, e.g. Welke (1988), Eroms (2000),60 Engel (2004). Eisenberg (2006) seems to tend more towards constituent structures as the sentence is the highest ranking unit in his graphs. For valency descriptions of English, the situation is different: Emons (1978) uses a constituent structure model based on earlier work by Heringer on German; Allerton favours a dependency model over constituency (1982: 22–25) but also proclaims that “[i]t goes without saying that a constituency relationship can hold between two valency structures” (Allerton 1982: 142), so it appears as if the two kinds of analysis are intertwined in some way. Fischer (1997) based his contrastive analysis of German and English on Engel’s dependency grammar for German but acknowledges the merits of constituent structure and makes use of it in one of his chapters. While not being in the tradition of valency grammar, Matthews (1981) has a certain closeness to valency and dependency approaches || 60 Eroms’ approach has to be classified as a dependency model although he represents S[entence] as the top node in his diagrams.

34 | Theory

and tries to combine dependency with constituent structure models, which results in diagrams of the following type:61 S[ V[pass]V NP[your

sister]NP NP[the meat]NP ]S

Fig. 4: Combined dependency/constituency diagram for “Pass your sister the meat.” (adapted from Matthews 1981: 91)

Such a view is in line with Welke’s position that sentences are always structured in both ways (Welke 2011: 23). Similarly, Herbst/Schüller (2008: 4–9, 11) prefer a constituent structure model over a dependency model since their step by step division of sentence elements enables them to account for certain pro-forms and other syntactic tests (e.g. Herbst/Schüller 2008: 7), which would be more difficult in a strict dependency model. However, they combine the grammatical aspect of constituency and the lexical aspect of valency by incorporating both into their analytical terminology, so that valency relations are present at the same time as a constituent structure. Thus he in (18) is treated as subject complement unit (SCU) because it is “the clause unit that functions as the subject of the clause and is realised by a complement of the governing verb” (Herbst/Schüller 2008: 165). (18)

And he conned me into posting his mail! (KCX 4990)

In copular clauses, some valency/dependency approaches treat the predicative element as the head (see de Marneffe/Manning 2008 for an applied perspective), some posit a combined head which consists of the copula and the predicative element (see for instance Heringer 1970: 180ff and the tentative proposal in Herbst/Schüller 2008: 141f) and some reject the idea and treat the verb as the highest ranking element (see for instance Engel 2004). The phenomenon will be discussed in more detail in Chapter 6. Concord is treated independently from verb valency by Herbst/Schüller (2008) while Eroms (2000: 183ff) sees it as an indicator of the relationship between the finite verb (i.e. the auxiliary whenever present and possibly the finite component of the lexical verb) and the subject in German, which leads him to claim that the subject is not a dependent of the lexical verb but of the conjugated verb. It is important to note that he accepts that the main verb has a valency || 61 See also Fischer (1997: 39) for alternative ways of presenting the same kind of information.

The role of the subject in generative grammar and in valency/dependency grammar | 35

slot that is filled by the subject but that he sees it as immediately dependent on the finite element (he also speaks of “Strukturvalenz” [‘structural valency’; Eroms 2000: 139] in this context). We can see that although valency and dependency are often treated as belonging together (particularly in German linguistics), this is not necessarily the case. Järventausta also supports this view by making clear that the principal difference between both approaches is that valency theory is a theory of the lexicon and dependency theory a theory of syntax (2003: 783).

2.2.2.4 Note on the present study Within Herbst’s model of valency theory (Herbst et al. 2004, Herbst/Roe 1996, Herbst/Schüller 2008), the research question is whether the (item usually called) subject is a complement of the verb or not. As briefly described above, the issue had not been raised in German valency theory62 because the relations called NOT (obligatoriness) and FOSP (formal specification) in Jacobs’ model were both analysed as being met. Given the introduction of different levels of necessity by Herbst/Roe (1996), one has to maintain that within Herbst’s model NOT is not met for many subjects in English. For German linguistic theory, this view is not necessarily a problem, since the nominative case markings that German subject [NP]s show actually indicate an FOSP relation and many German models then posit the same case for clausal subjects even though clauses do not inflect for case. For English, [NP]s do not offer substantial evidence for an FOSP relationship since they can occur with any verb and often do not show case markings,63 and that is the reason why the present study aims to show that an FOSP relation exists with the help of clausal subjects, the distribution of which is more complex than that of [NP]s.

|| 62 There are discussions of the problem, though. See for instance Järventausta’s telling title “Das Subjektproblem in der Valenzforschung” (‘The problem of the subject in valency research’) (2003). In fact, her questions are very much like those asked in this study. However, she also observes that in German dependency grammar models, the subject always has complement status: “Das Subjekt wird ohne Ausnahme zu den Aktanten gezählt” (‘The subject is counted among the actants without exception.’) (Järventausta 2003: 785). 63 Even if they show case markings, these could be analysed as properties of their position in the clause and would not have to be treated as assigned by the valency carrier. Due to the much less rigid word order in German, such an argument would not work there.

36 | Theory

2.3 Towards a definition of subject in English After the overview of theoretical approaches towards subjects presented above, the present section aims to describe in a more systematic manner what is understood by the term subject. It will present the criteria usually given for the identification and/or definition of subjects in grammatical descriptions of English and discuss their validity as well as their relevance for the present study.

2.3.1 Sentence/Clause For the definition of the term subject, it is necessary to refer to the entity in which the subject occurs. Pre-theoretically, this unit is called a sentence, and many grammarians make use of this term in one way or another. For Jespersen, a sentence is “a (relatively) complete and independent unit of communication” (Jespersen 1933: 106), which is a pragmatic notion we would call utterance today rather than a grammatical definition of sentence, so it is of limited use for us. Zandvoort’s grammar departed from this terminology over time, since he removed the term utterance from later editions but still focused on communicative units: An oral or written communication is made up of one or more units, each of which contains a complete utterance formed according to a definite pattern. Such units are called SENTENCES. (Zandvoort 1965: 195; his emphasis) An oral or written communication is made up of one or more units called SENTENCES. (Zandvoort 1972: 195; his emphasis)

He then went on to delimit sentences from one another by means of orthography and intonation. CGEL is more modern in this respect and insists on a separation of levels: More careful consideration of the relation between grammar, semantics, and pragmatics will require further distinction of terminology. Thus we may distinguish SENTENCE (a grammatically autonomous unit) from UTTERANCE (a unit which is autonomous in terms of its pragmatic or communicative function). (CGEL 78; their emphasis)

However, the authors of CGEL state that “[t]he CLAUSE, particularly the independent clause […], is in many ways a more clearly-defined unit than the sentence” (CGEL 47; their emphasis), so that the subject is treated as a constituent of a clause. We shall follow CGEL’s argument in the present study and distinguish between sentences and clauses where necessary. We shall also follow

Towards a definition of subject in English | 37

CGEL’s terminological distinction between a main clause (the whole of example (19)), a subordinate clause (solid line), and a matrix clause (dotted line). (19)

Most British people believed that, whatever their own prosperity, the fortunes of their country were bad and getting worse. (A66 1689)

The problem is, however, that CGEL does not offer a clear-cut definition of the term clause, either. Using a range of clause elements and example clauses, their concept of clause is elucidated via the clause structure and can probably be grasped best in the sense of Wittgenstein’s family resemblance (Wittgenstein 1953: 32). Their generalizations about clause structure (“The verb element (V) is the most ‘central’ element, and in all the examples above it is preceded by the subject (S)” (CGEL: 50)) are clearly prototypical in character, as indicated by the existence of clauses without subjects and verbless clauses, which are described later in the grammar. While Huddleston/Pullum’s definition of the clause as “a syntactic construction consisting (in the central cases) of a subject and a predicate” (CamG 44) is compatible with CGEL’s account, the authors reject the division of sentences into simple, complex and compound sentences because the distinction is based on a mixture of two dimensions, i.e. embedding and coordination (CamG 45). They define canonical (= basic) and non-canonical clauses to begin with in order to simplify the syntactic description and then “describe the rest derivatively” (CamG 46), but it is made very clear that this is “only a descriptive device” (CamG 48) and that any view which treats derivation as a psychological process has to be rejected. While CamG’s criticism is well-founded, it is largely irrelevant for our study, so the terms complex and compound will be used where necessary for easy readability. To sum up, while clause is the preferred term over sentence, its definition is prototypical in character and usually refers to subjects. Strictly speaking, this would mean that the concept cannot be used in a definition of subject without risk of circularity.

2.3.2 Criteria The first 10 criteria presented below follow the detailed list of “[d]istinctive grammatical properties of the subject” (CamG 236) given by CamG (236–239), from where the headings are borrowed as well. Criteria 11 to 17 are based on other works.

38 | Theory

1.

Category

According to CGEL a subject is “normally a noun phrase […] or a nominal clause” (CGEL 724). A restricted but not uncommon type of subject in the form of a prepositional phrase or other “adverbial form” (CGEL 736) is also possible:64 (20)

Will after the show be soon enough? (CGEL 736; their italics)

Similarly, CamG states that the “prototypical subject has the form of an NP” (CamG 236). They show that in those cases where it is the only NP in the clause or which functions directly in the clause, the form is a sufficient criterion. Subordinate clauses are mentioned as non-prototypical, as are other categories which “appear as subject only under very restrictive conditions” (CamG 236). 2.

Position

In CamG’s terminology, the default subject position is before the predicator, except for inverted constructions, which are discussed below. Even if other complements are fronted, the subject is the constituent before the predicator, although an adjunct can intervene between the two (CamG 236f). CGEL takes a similarly prototypical view when it states that the subject “normally occurs before the verb in declarative clauses” (CGEL 724). For some structures, such as the examples in (3) on page 11, the identification of the subject based on the positional criterion is difficult since they may be treated as cases of inversion.65 3.

Case

What little case marking left in English is relevant to the identification of subjects as the distinct subjective (CGEL; nominative in CamG) forms (I, he, she, we, they) are used in subject position of finite clauses and objective (CGEL; accusative in CamG) forms (me, him, her, us, them) in object position. CamG (237) rightly notes that case only plays a marginal role in identifying subjects. No mention is made of sentences of the type exemplified by (21a) which are often treated as the ‘correct’ version by purists – as opposed to the more frequent variant in (21b):66 (21) a. It is I. b. It is me.

|| 64 Newmeyer rejects an analysis put forward by Bresnan, who claims that such uses are elliptical and have in fact an underlying NP structure (Newmeyer 2003: 163). 65 See Patten (2016) for a recent detailed discussion of such structures. 66 See Palmer (1971: 15ff) for a detailed discussion.

Towards a definition of subject in English | 39

In the first sentence, the postverbal element is in nominative case but is usually regarded as the subject complement (Palmer 1971: 15) nonetheless. Given that the semantic structure and information structure of both is identical, it would be counter-intuitive, although theoretically possible, to argue that (21a) is an inverted structure with the subject following the verb whereas (21b) is the canonical structure. 4.

Agreement

As with case, agreement is of limited use in the identification of subjects as – except for the verb be – the only contrast in person and number inflection is “between 3rd person singular and any other person-number combination” (CamG 237) in the present tense and in finite clauses. To determine the subjecthood of elements on the basis of inflection, CamG suggests a commutation test, however. Thus replacing the personal pronoun They in (22a) with She to form (22b) leads to a change of agreement on the verb, indicating that They is indeed the subject in (22a). (22) a. They accept your proposal. b. She accepts your proposal. For English there are numerous exceptions to the basic rules of concord that are discussed in great detail in CGEL (755–767); for our purposes the issues with nominal relative clauses and with a conflict between grammatical and notional concord are of limited importance, though. For variation in concord with other types of clausal subjects see the discussion under uniqueness (criterion 10 below) and in Section 7.1.3.2.67 5.

Inversion

Subjects follow the predicator (operator in CGEL 724) in “closed interrogative main clauses and various other constructions” (CamG 237) and this again is useful as an indirect test for subjecthood. Sentence (22a) can be turned into the interrogative (22c), which confirms the subject status of They. (22) c

Do they accept your proposal?

|| 67 For clausal subjects in German, Oppenrieder (2006: 901) even wonders whether agreement exists or whether a neutral replacement form, i.e. 3rd person singular, is used.

40 | Theory

6.

Open interrogatives

According to CGEL “[i]n wh-interrogative clauses, subject-operator inversion also occurs except where the wh-element is itself the subject” (CGEL 725). Examples (23a–b) illustrate both cases: (23) a. What have you been telling him? (HP1.187) b. What has been going on? (HP1.82) This property can thus also be used as a test to determine the subject status of an element and is listed by CamG (238), too. 7.

Tags

CGEL (725) and CamG agree that “[i]nterrogative tags attached to a declarative clause contain a subject pronoun that agrees with the subject of that clause” (CamG 238). Tags can be used as a relatively reliable test as well, given that they also work if the canonical order of elements is reversed as in (24): (24) 8.

Even clearer is the second point, isn’t it? (CamG 238) Coordination

“Since its default position is external to the VP, the subject can enter straightforwardly into construction with a VP-coordination” (CamG 238). Examples of the type given in (25a) are presented as evidence: (25) a.

He turned there and waved his gun at us. (B11 156)

Indeed, coordination of the type exemplified in (25a) creates problems for shallow syntactic structures such as the default CGEL model outlined above as turned and there do not form a constituent of their own, being both immediate constituents of the clause in structurally simpler sentences such as (25b). (25) b. He turned there. CGEL solves the problem by speaking of a (“very common”) coordination of predicates (CGEL 948) despite the fact that the authors “find little need to refer to the predicate as a separate structural unit in the description of English grammar” (CGEL 79). CamG is consistent in this respect, but their model fails to explain why coordination can occur between groups of elements that do not have constituent status in their model, either, as in (26): (26)

[T]hey were working hard and long hours every day and they demanded and they deserved a higher wage. (HEM 272)

Towards a definition of subject in English | 41

The underlined structures in (26) have to be treated as coordinated given that they share the same direct object (“a higher wage”), but it would be hard to conceive of an analysis that treats the two as constituents in clause structure.68 9.

Obligatoriness

In CGEL, “[a] subject is obligatory in finite clauses except in imperative clauses, where it is normally absent but implied” (CGEL 725). The structure in (27) could be regarded as a (comparatively marginal) counter-example: (27)

Wish you were here. (CGEL 846)69

Another counter-example is diary style, which makes a passage such as (28) appear perfectly normal, despite the “missing” subjects. (28)

Went to school. Found it closed. In my anguish I had forgotten that I am on holiday. Didn’t want to go home, so went to see Bert Baxter instead. (Townsend 2003: 58)

Thus the obligatoriness of the subject has to be seen in relation to ‘normal’, ‘unmarked’ and possible written clauses in careful style. Similarly to CGEL, CamG states that “[i]n general, the subject is an obligatory element” (CamG 238) but also mentions exceptions such as certain clause types (imperative, non-finites) or casual style. The obligatoriness is then explained in more detail: Whether a clause has an object or not depends on the lexical properties of the verb (e.g. appear excludes one, while use normally requires one), but a subject is required in all canonical clauses. (CamG 238)

Without explicitly distinguishing levels of valency and structural necessity as done for instance by Herbst/Roe (1996; see also the discussion in Section 2.2.2 above), the authors of CamG nonetheless make clear that the obligatoriness of an element in subject position is determined by the clausal structure and grammatical rules for forming English sentences rather than by the valency of the verb governing the sentence. They deduce that this requirement leads to

|| 68 CamG finds an easy way out by distinguishing basic coordination from non-basic coordination (CamG 238). Since the criterion for this distinction is constituent status, though, the argument is circular. 69 For further information on sentences of this type see the section on “Block language” in CGEL (845–847).

42 | Theory

sentences in which it exists as a dummy subject without discernable meaning, echoing Chomsky’s Extended Projection Principle (see Section 2.2.1). In order to illustrate the obligatoriness of subjects, CamG makes use of what they call a “maximal finite reduction”70 (CamG 239) of a sentence. Example (29a) can thus be reduced to the maximally reduced finite variant (29c) in answer to (29b), which tells us that She and has are obligatory in sentence (29a): (29) a. Sue has eaten already. (CamG 239) b. Has Sue eaten already? (CamG 239) c. She has. (CamG 239) The problem with this criterion is that it is based on CamG’s canonical clause, and canonical clauses are the ones which contain subjects, so there is a certain circularity in the argument. Similarly, since CGEL defines clause structure with reference to subjects, so obligatoriness cannot be used as a criterion without circularity. 10. Uniqueness In CamG (239), only one subject is allowed per clause. This is in line with most other accounts, if we stay at the same level of analysis. So-called extraposed subjects are not treated as subjects in CamG and neither are what they call displaced subjects in existential clauses.71 CamG also extends uniqueness to mean that even coordination of subjects does not exist; the coordination in (30) is a coordination of NPs instead. (30)

Sir William and her Ladyship have been noticing things, I can tell you. (A0D 2300)

The approach is only possible because coordination in CamG is not restricted to elements of the same syntactic category, it rather occurs between “elements of syntactically equal status” (CamG 66), which could be phrases or clauses alike:72 (31) a.

She'll be arriving [tomorrow or on Friday]. (CamG 66)

|| 70 When CamG introduces the concept, it is referred to as “maximal finite reduction” (CamG 239), but the term “minimal finite reduction” (CamG 243) occurs as well, apparently for the same structure. 71 See Section 7.1 for more details on their treatment of extraposition and Section 8.3 for existentials. 72 This is contrary to the position taken in Chapter 13 of CGEL, which only allows coordination of NPs with NPs or of clauses with clauses.

Towards a definition of subject in English | 43

b. You must find out [the cost and whether you can pay by credit card]. (CamG 45) The criteria of which elements can be coordinated are discussed in the following paragraph: Coordinates must be syntactically alike, but the syntactic likeness that is required is in general a matter of function rather than of category. Thus in the clauses She’ll be arriving tomorrow and She’ll be arriving on Friday, the underlined phrases have the same function (adjunct of temporal location), and this makes it possible to coordinate them, as in [(31a)], even though the first is an NP and the second a PP. (CamG 66; their emphasis)

Relying on evidence from the functional level to justify a coordinated structure at the formal level is, however, slightly dubious – and it is definitely in stark contrast to Pullum’s otherwise quite harsh criticism of such arguments in general:73 People do not confuse butter knives with screwdrivers, even though occasionally someone who cannot find a screwdriver may use a butter knife to turn a screw. Yet in grammar people just cannot keep syntactically relevant categories or classes of words separate from the relational properties they have when used in particular constructions, and cannot keep either separate from meaning. (Pullum 2009: 255)

It is not surprising that we find such coordination between formally different elements in subjects, too: (32)

Ideas for innovation and adapting to change are more accepted by the small organisation culture. (G0U 2018)74

The authors claim that “agreement and tags treat the coordination as a whole as subject, not the separate parts” (CamG 239). However, there is counter-evidence in the context of clausal subjects: (33)

Checking high bills and dealing with tree roots are among the topics we cover. (CG5 47)

|| 73 It has to be noted, though, that Pullum is not the author of the cited passage in CamG. 74 Although one might argue that in this case innovation and adapting to change are coordinated prepositional complements, there is a reading which coordinates Ideas for innovation with adapting to change as can be seen in the sentence Adapting to change and ideas for innovation are more accepted by the small organisation culture.

44 | Theory

(34)

Entering the transcendent and feeling the beautiful ideas of creativity is to escape from the sad feelings of the child. (B1F 1497)

Whether two coordinated clauses take a plural verb or a singular verb (or the respective tag) seems to be related to the question of whether the two clauses are conceived as one unit of meaning (one action/event/entity/...) or as several ones. Thus uniqueness in the strict sense laid out by CamG is an at least partly questionable criterion. 11. Reflexive pronouns CGEL shows that reflexive pronouns used as “direct object, indirect object, subject complement, or prepositional complement” (CGEL 725) agree with the subject: (35) a. He killed himself. (HTU 5270) b. *He killed herself. They add that “[t]he same concord relation generally applies when the emphatic genitive my own, etc is used” (CGEL 725; their italics). 12. Passivization While CamG does not treat passivization as relevant to subjecthood, CGEL posits a regular relationship between the underlined and the italicized elements in the following examples: (36) a. His father shot him. b. He was shot (by his father). There is a systematic correspondence between active and passive clauses in that the direct or indirect object of an active clause becomes the subject of a passive clause while the subject of the active clause is either omitted or made the complement in a by-agent phrase[.] (CGEL 725)

13. Theme/topic CGEL notes that “[t]he subject is typically the theme (or topic) of the clause” (CGEL 726) with theme defined “in terms of position and prosody respectively” (CGEL 1362). As the prosodic criterion is limited to spoken language and thus difficult to verify in written corpora, the position criterion is the only relevant one for our study: “THEME is the name we give to the initial part of any structure when we consider it from an informational point of view” (CGEL 1361). Thus their approach is in line with Halliday’s definition of theme as “the point of departure for the message” (Halliday 1994: 34), but not all researchers would

Towards a definition of subject in English | 45

agree as to the treatment of theme and topic as basically the same.75 However, as the two often coincide (and often coincide with given discussed in the next paragraph) and given that they are less relevant to the description of clausal subjects, which often contain both thematic and rhematic information anyway, the distinction is of limited use for our purposes and need thus not be made. Since in English adjuncts quite often occur as the initial part of a sentence, counter-examples are not hard to find: (37)

After a while you fly out. (A05 738)

The underlined part is not the subject according to syntactic criteria, so we have to be aware that the match of syntax and information structure that CGEL proposes is an imperfect one. 14. Given information What is given and what is new information in a sentence is a matter of context and thus often cannot be determined by considering a sentence in isolation. While the subject prototypically (in the case of (pro-)nominal subjects) contains given information, clausal subjects express propositions of their own and thus as a rule contain at least some more or less new information. That the subject often contains given information is also acknowledged by Dowty: In an adequate linguistic description, greater relative degrees of connectedness to previous discourse, givenness, etc., must be explicitly specified as a semantic correlate of grammatical subject denotations (in English-like languages). (Dowty 1991: 564)

Again, CGEL makes use of an information-structural tendency to characterise typical subjects, but as a defining criterion for a syntactic element, the property is highly problematic. 15. Agentivity According to CGEL, “[i]n a clause that is not passive, the subject is agentive if the agentive role is expressed in the clause” (CGEL 726). If we take the classic definition by Fillmore, agentive is “the case of the typically animate perceived instigator of the action identified by the verb” (Fillmore 1968: 24; similarly CGEL 741). This is straightforward in many English clauses: (38)

My brother hit me with a stone. (C86 2832)

|| 75 The terms theme and topic are used interchangeably in CGEL, with the authors fully aware that some linguists use the term topic for what they call given (CGEL 1362).

46 | Theory

A problem arises in cases such as (39b) discussed in Herbst/Klotz (2002), where the grammatical subject in intransitive uses combines the referents that occur separately as agent and patient in transitive uses such as (39a): (39) a. He met her in Paris. b. They met in Paris. Furthermore, the word animate in the definition causes problems if we have clausal subjects since these usually represent facts, ideas, actions, … but not people (and neither do many nominal subjects), but CGEL does not discuss participants that are not realized by noun phrases in any detail (CGEL 740f). As we have seen above, Jespersen presents counter-evidence for equating subjects with agents (as in He suffered torture.). Furthermore, over 30 years of research on semantic roles has shown that any delimitation of semantic roles is highly problematic and to some extent arbitrary. The problem of the definition of semantic roles is illustrated in detail by Dowty (1991: 553f for agent), who proposes a distinction between only two prototypical roles he names protoagent and proto-patient (Dowty 1991: 572). In this model, the [that_CL] in (40) would be rather a proto-agent than a proto-patient: (40)

That they have cost me no more than my time makes such decisions much easier than if I had paid for them in gold. (A0G 1023)

Herbst/Schüller (2008: 158ff), by contrast, advocate a two-level approach to semantic roles by distinguishing between participant roles, which are verbspecific, and clausal roles, which are related to sentence structure. Subjects in active declarative clauses thus automatically receive an agentivityinterpretation.76 If we accept this model, we would have to do away altogether with agentivity as a criterion for subject status to avoid circularity.77 But even if we were to accept the criterion, it has to be borne in mind that all mappings of semantic roles and syntactic elements are to be seen as tendencies (as also illustrated in great detail by Faulhaber 2011), so we have to agree with Dowty:

|| 76 For German, Reis (1982: 182) shows that subjects receive agentive interpretation by contrasting converse pairs of verbs (such as buy – sell). Oppenrieder (2006: 911f) on the other hand claims that clausal subjects cannot be agents since – as opposed to [NP] agentive subjects – they do not allow instrument roles to co-occur with them. 77 Herbst/Schüller do not use agentivity as a criterion for subjecthood, so there is no problem of circularity in their model.

Towards a definition of subject in English | 47

In this paper I have been at pains to argue that, while the Proto-Agent/Proto-Patient opposition is CONNECTED to the grammatical opposition between subject and object, neither opposition is REDUCIBLE to the other; nor is the association of subject with (Proto-)Agent and object with (Proto-)Patient a necessary one. [...] The correlation of proto-roles with grammatical relations in English-like languages is only a TENDENCY, not an absolute, and it admits of quasi-violations (under relatively predictable circumstances). An example is the lexicalization of 'conflicting' pairs like like and please and 'counterexamples' like receive and undergo. (Dowty 1991: 610; his emphasis)

Thus we have to conclude that a semantic characterisation of subjects causes problems when used in a definition of the term. 16. Concord with subject complement In CGEL’s model, the underlined element in (41a) is called a subject complement, a term which roughly corresponds to predicative complement in CamG (217) and other models: (41) a.

Caroline is my sister. (CGEL 725)

CGEL claims that the subject “normally determines number of the subject complement when that is a noun phrase” (CGEL 725) and gives the opposition of (41a–b) as evidence. (41) b. Caroline and Vanessa are my sisters. (CGEL 725) Later in the grammar (CGEL 767) the authors show that there are “exceptions”, but if we take a closer look at the following example, it appears that the criterion is not about syntactic agreement but rather semantic compatibility: (42)

They were the perfect couple. (CH2 5905)

The plural subject here does not require a plural predicative element, it only needs a predicative element that expresses plurality semantically. While “grammatical plurality is independent of semantic plurality” in English (Koya 1992: 63), they do of course often coincide, which may have led CGEL to the formulation of this criterion. For some of the ‘exceptional’ cases CGEL offers an explanation: (43)

The younger children are a problem. (CGEL 767)

With the help of (43) and other, similar ones the authors claim that the subject complement “although nominal in form, has a characterizing function closer to that of an adjective” (CGEL 767). Again we can see that there is a strong tenden-

48 | Theory

cy in CGEL to mix formal and semantic features in the description of the language. Given the rather dubious character of this criterion, it will not be used in the subsequent chapters, also since agreement with the verb is usually easier to determine in sentences with a subject complement (since forms of be occur most frequently) and since it is the more reliable criterion. 17. Implied subject of non-finite subordinate clause According to CGEL, “[t]he implied subject of a subjectless nonfinite or verbless clause is normally identical with the subject of the superordinate clause” (CGEL 725). They use the following example for illustration: (44)

Susan telephoned before coming over. [‘…before Susan came over’]. (CGEL 725; their italics)

However, such a criterion is not applicable in the study of clauses the subjects of which are non-finite subjectless clauses. In that case no prediction can be made as to the implied subject’s referent. The following examples serve to illustrate that point: (45)

To hesitate would add to suspicions that the Fed cares more about staying on good terms with the White House than about price stability. (CR8 120)

The Fed (i.e. the Unites States Federal Reserve System) can be recovered as implied subject from the same sentence. (46)

To take the scalp of a major southern hemisphere team would be huge and would add to the achievement of beating the likes of France and England here this year. (newspapers)

Here a Scottish Rugby player talks about an upcoming match of his team, referred to by “we” in the previous sentence, so although the team does not feature in the same sentence, it can easily be recovered as the implied subject of the [to_INF] clause. In (47) below, the implied subject of “to hint at it” cannot be recovered verbatim, even in the context of the newspaper commentary it was taken from: (47)

There is, in royal circles, a superstitious terror of the word “abdication”, as if even to hint at it would begin the crumbling of the whole edifice. (newspapers)

Towards a definition of subject in English | 49

The most likely interpretation here is that the hinting at abdication by members of the aforementioned royal circles would have the negative effect stated. Example (48) appears to be a similar case since one might argue that the subject to be recovered is “widowers” or “the surviving husband”: (48)

Since “griefe” was entered as a cause of death in the rolls of mortality in the 17th century, it has been clear that widowers do worse than widows. To lose a wife causes a 40 per cent increase in risk for the surviving husband, compared to a third that figure for a wife in the same position. (newspapers)

However, another likely interpretation is that a kind of general subject – anyone basically – might be the subject. While this is hard to decide and may be a matter of debate, the general subject (i.e. ‘anyone’) interpretation is the only one possible in (49), which is given with extended context to illustrate that the referent of the implied subject cannot be inferred from the surrounding text: (49)

Although he is the only sprinter to have won Group races at two, three and four years, his supernatural speed never quite reaped him the rewards he deserved. Hence this why-the-hell-not crack at that elusive Group One victory. ‘It’s a bit like a 30-year-old winning £ 6 million on the Lottery and retiring. This is just a one-off – I wouldn’t want to keep the old lad in training after that. But he’s sound as a pound. This year’s sprinters lack a little bit in depth. And the owners think so much of him they just want to give him one more go at it.’ To stand close to a racehorse always brings a shock of delight, the more so when he is a proud, strong five-year-old coming to peak fitness. In the dim light of his box Mind Games’s coat still shines like Charbonnel and Walker melting in a late summer sun. His muscles are tightly packed like little crates crammed with books. (newsmerge)

In the passive, there is a further complication described by CamG (1435) which is described for examples (32) to (33) on page 119 and counters CGEL’s claim. Given that the criterion of implied subject does not work for clausal subjects and is irrelevant for the study of the clausal subject of a finite main clause, it will not be taken up in subsequent chapters.78

|| 78 A similar observation for German subject infinitivals (as opposed to object infinitivals) was made by Sternefeld (1985: 404f) and is followed up on by Oppenrieder (1991: 259f).

50 | Theory

18. Imperatives Falk adds a further criterion based on imperatives when he claims “that the addressee of an imperative is a subject” (Falk 2006a: 3) irrespective of the thematic role of the subject. Thus the addressee would be an agent in (50a), an experiencer in (50b), and a patient in (50c): (50) a. Go to school! (Falk 2006a: 3) b. Freeze, if that’s what you want! (Parent to child who refuses to put on a coat in freezing weather) (Falk 2006a: 3) c. Be arrested by the municipal police, not the state police! (Falk 2006a: 3) It would be more precise to reformulate Falk’s criterion to state that the argument of the verb that is not realized in imperatives79 is realized as subject in the canonical clause irrespectively of the semantic role that subject has. This criterion is listed for the sake of completeness, but since it only works for subjects one can address, it is of no further relevance to the discussion of clausal subjects in the present study and will thus not be drawn upon in the subsequent chapters.

2.3.3 Use of terminology in this study We are not free to define a notion like “subject of” in any way that suits our purposes. There is a large body of lore concerning the notion, and any proposed definition must at least largely agree with the traditional, and to some extent, pretheoretical usage of the term. (Keenan 1976: 306)

Marantz (1984) rejects Keenan’s view that a definition has to mirror pretheoretical usage and questions the feasibility of such a definition, echoing work done by Wittgenstein and Rosch: What makes linguists think that their notion of “subject” is any different from that of “chair” or “game”? If they cannot provide a definition or account of “game” that picks out

|| 79 There are of course imperatives with an overt realization of the addressee: Don’t you dare move a muscle! (HGT 3438) For German, Ágel argues that “subjectless” imperatives are not subjectless but that the subject is not realized at the macro-valency level (Ágel 1993: 43f). He assumes that the inflectional form of the German imperative is a micro-valency realization of the respective complement.

Towards a definition of subject in English | 51

all and only the entities considered to be games, they should not expect to discover a definition of “subject” that accounts for their intuitive concept of “subject”. (Marantz 1984: 2)

Marantz does not say that a definition is not possible, but he is right in stating that what exactly the scope of any definition is cannot be regarded as a criterion for judging the definition: “There can be no right definition of ‘subject’ or ‘passive,’ only a correct (or better) syntactic theory” (Marantz 1984: 3). It is thus obvious that for the purpose of the present study it is impossible (and probably unnecessary, too!) to find a definition that caters for all needs. Generally, the criteria presented by CGEL and CamG form the background for our use of the term subject, which has been shown in this chapter to be an amalgam of syntactic, semantico-logical and pragmatic features. In this study, the grammatical subject, i.e. the element defined by the grammatical criteria discussed above, will be in focus, so subject refers to it by default. For the logical or psychological levels, the terms logical subject and psychological subject will be avoided80 and we will find little need to refer to these entities at all. If we have to, the terms agent and theme (following Halliday 1994: 32) will be used instead.81 In problematic cases such as extraposition or existentials, the criteria listed above will be reviewed individually. As to the question of what a subject is a subject of, both constituency and dependency ways of thinking will be allowed. In fact it would be misleading to treat them as mutually exclusive, so we shall say that Harry in (51) is the subject of the clause, of both verbs, and of both predicates: (51)

Harry moved in front of the tank and looked intently at the snake. (HP1.25)

Such an approach is compatible with Matthews (1981), who states that “the notion subject should be looked at in two different ways. On the one hand, there is a subject as opposed to an object. […] On the other hand, there is a subject as opposed to a predicate” (Matthews 1981: 104f). He illustrates the two readings of the term subject with the following diagram:

|| 80 Halliday quite rightly insists that “[t]here is no such thing as a general concept of ‘Subject’ of which these [i.e. syntactic, semantic and logical subject] are different varieties. They are not three kinds of anything; they are three quite different things” (Halliday 1994: 32). 81 Halliday uses the term actor instead of agent.

52 | Theory

Fig. 5: Varying scope of the term subject illustrated by Matthews (1981: 104; adapted)

This is also in line with the distinction made by Croft (2001: 24) between the syntactic role subject (part-whole relationship), which he adopts, and the semantic relation subject (dependency subject-verb), against the existence of which he argues. For the purpose of the present study, the differentiation between these types in the various chapters appears unnecessary since the term subject will be used mainly to identify or refer to an element in a given clause. Thus it is sufficient if the term establishes successful reference. Whenever there is a risk of confusion (or where the status of the item is problematic), more specific terms such as extraposed subject (Chapter 7) or displaced subject (Chapter 8) will be used.

2.4 Summary It became clear in the present chapter that most attempts to define the notion of subject aim to characterise prototypical subjects with a range of criteria that cause problems in borderline cases because of cross-classification issues. Due to the long history of the term and its interpretation as either a semantic or a syntactic concept, its use in various schools of linguistics is extremely divergent and some theoretical approaches – most notable generative grammar – do not use the concept as part of their theoretical model. The term is still used, but only as a shorthand for ease of reference. The use of the tem subject in the present study follows this general idea, i.e. even though the term is used to establish reference to constituents/complements, no commitment to its theoretical status within valency theory is made. Section 9.2 will discuss the question of the usefulness of the term in more detail.

3 Methodological Considerations 3.1 English In the discussion so far, the word English from the title of the present study has been taken for granted, even though the matter is by no means simple, given the wide array of possible variation (see Crystal 2003, ch. 7 and 20–23). For our purposes, the the most appealing approximation in this regard is Standard English, despite the fact that there are good reasons to be cautious of the term, as is pointed out by McArthur: “A widely used term that resists easy definition but is used as if most educated people nonetheless know precisely what it refers to” (McArthur 1992: 982). For the purpose of this study, unless otherwise stated, English will be used to mean Standard English as loosely defined by Crystal as “a minority variety (identified chiefly by its vocabulary, grammar, and orthography) which carries most prestige and is most widely understood” (Crystal 2003: 110).82 Given that he also states that “[m]ore than anywhere else, SE is to be found in print” (Crystal 2003: 110), we can also justify the use of predominantly written corpora in the present study. Generally, it will be assumed that British Standard English and American Standard English have enough in common to treat them as Standard English in a uniform way (an approach also taken by CGEL 18f), so they will only be distinguished in this volume where there is a noticeable difference between the two.83 For most of our description, written English is the major source of information and thus the major mode that is discussed.

3.2 Data The question of what data is appropriate for syntactic research has been widely discussed in the linguistics community in recent years. There are two main camps which cluster loosely around the extreme positions on a scale, with those who see corpora as the only viable source of linguistic information at one end

|| 82 One further criterion mentioned by Crystal (2003: 110) is the fact that Standard English is used as an educational target, both within the respective English-speaking country and in foreign language teaching abroad. This role follows, of course, from its high prestige. 83 Trudgill/Hannah (2008: 4f) however prefer to distinguish between two English as a native language (ENL) varieties, British and North American English, with the former including also varieties spoken in New Zealand or South Africa.

54 | Methodological Considerations

and those who regard corpora as a complete waste of time at the other end.84 Both extreme positions are questionable: It is undoubtedly true that corpus research can lead to insights which are hardly visible to the naked introspective eye. Sinclair’s analysis of naked eye (Sinclair 2004: 30ff) is a case in point as are the descriptions of the behaviour of the verb cause presented in Stubbs (1995), Klotz (2000: 189–195) and Stefanowitsch/Gries (2003: 220ff). Corpus analysis can thus help to identify details which may be missed in a study based purely on introspection.85 Nonetheless, some of the reservations against corpus analysis expressed by its critics are not completely unfounded. One argument against corpora as sources of linguistic information is that they are skewed: Any natural corpus will be skewed. Some sentences won’t occur because they are obvious, others because they are false, still others because they are impolite. The corpus, if natural, will be so wildly skewed that the description would be no more than a mere list. (Chomsky in a debate at the Third Texas Conference on Problems of Linguistic Analysis in English 1958, published in Hill 1962: 159)86

Chomsky’s example that there are bound to be more instances of “I live in New York” than of “I live in Dayton, Ohio” (Chomsky in a 1964 discussion, cited by Halliday 1991: 30) in a corpus of American English may be correct, but this observation does not render grammatical information from the corpus useless – after all, both sentences follow the same grammatical pattern.87 But Chomsky does not trust frequency information on grammatical items either: Third, the notion “grammatical in English” cannot be identified in any way with the notion “high order of statistical approximation to English.” (Chomsky 1957: 16)88

|| 84 Fillmore’s (1992: 35) satirical depiction of “corpus linguists” and “armchair linguists” has still some truth in it today. See also Aarts (2000: 5ff) for the different positions. 85 According to Chomsky, advances in linguistic theory can only be achieved if the right questions are asked (see the interview in Aarts 2000: 5f), so he takes the opposite perspective, i.e. that people are going to miss out on the interesting part of linguistics if they study corpora. 86 Even with the larger corpora produced from the 1990s onwards, Chomsky’s view on corpora has not changed, which can be seen in the following exchange taken from an Interview in the year 1996: Interviewer: What is your view of modern corpus linguistics? Chomsky: It doesn’t exist. (Aarts 2000: 5) 87 And Mair quite rightly remarks: “The corpus is not used to limit the database but to provide data of superior quality” (Mair 1990: 3; his emphasis). 88 Leech comments on the problem as follows:

Data | 55

This leads us to the most critical problem of corpus analysis for the present study: there is a problem with negative evidence. It is almost commonplace in corpus linguistics research to stress that by its very nature, the corpus cannot provide negative evidence.89 Yet Stefanowitsch argues convincingly that it is possible by statistical measures “to distinguish between constructions that did not occur but could have (these could be referred to as ‘accidentally absent’), and constructions that did not occur and could not have (these can be referred to as ‘significantly absent’ structures)” (Stefanowitsch 2006: 62). However, clausal subjects are not frequent enough to allow the use of this methodology (see Stefanowitsch 2006: 72). Although it is an interesting fact as such if a construction is so rare that it does not occur in a corpus of the size of the BNC, it may still be perfectly acceptable, as, for instance, a [to_INF] subject with force:90 (1)

In some cases, the writer may not wish to rely on truth as a defense, even though the alleged libelous statement is true, because to do so would force him to reveal a confidential source. (academic)

In order to shed more light on constructions that are missing or rare in the BNC (which may also be performance errors), larger but less carefully collected corpora were consulted (see Section 3.2.2.1 below) and native speaker informants were interviewed and asked for acceptability judgements (see Section 3.2.3). The aim was thus to combine both methodologies in order to obtain the best from both worlds: items that native speakers may not have been able to come up with from the corpora and positive and negative evidence on constructions not frequently represented in the corpora, which is an approach similar to the one Johansson describes: In 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

|| However, Chomsky in his turn could not have conceived, in the 1950s, of a corpus of 500 million words capable of being searched in a matter of minutes or hours. While it is unlikely that foreknowledge of such a phenomenon would have changed Chomsky’s view of corpora at that time [...], we can see, in historical retrospect, how the availability of vastly increasing computer corpus resources has enabled syntactic and lexical phenomena of a language to be open to empirical investigation on a scale previously unimagined (Leech 1991: 13). 89 See for instance McEnery/Wilson (1996: 9). 90 See also Greenbaum (1988: 83ff) for a similar view.

56 | Methodological Considerations

elicitation techniques. Wise linguists, like experienced craftsmen, sharpen their tools and recognize their appropriate uses [...]. (Johansson 1991: 313)

We will come back to the question of different types of data in Section 9.1, where the problems encountered in the data collection for the present study will be discussed.

3.2.1 A Valency Dictionary of English (VDE) For much of the research done in this study, Herbst et al.’s Valency Dictionary of English (2004) provides a starting point. Based on the Cobuild Corpus and supplemented by native speaker data, it provides a thorough description of the syntactic combinatory potential of more than 500 verbs, more than 500 adjectives and more than 250 nouns. At the outset, the focus of the dictionary was on postverbal complements, which – due to the limited size of the Cobuild Corpus when the project was started – was of course a sensible choice. The information on subjects (in particular on non-extraposed ones) present in the dictionary was added later on the basis of native speaker intuition and thus has to be treated with more caution than the rest of the dictionary. Nonetheless, the dictionary makes available a wealth of information on which lexical items take clausal subjects that could be exploited for the present study. For this purpose, the release of the Erlangen Valency Patternbank (Herbst/Uhrig 2009) based on the VDE was particularly helpful since it allows a systematic search of patterns across valency carriers and thus provides a fresh and easy-to-access perspective on the VDE Data.

3.2.2 Corpora 3.2.2.1 Representativeness The most important corpus used for the present study is the British National Corpus (BNC; completed in 1994), whose limitation to British English has to be taken into account. In terms of representativeness of its population, “British English”, the BNC is excellent within the limits of feasibility.91 It contains 90 % written and 10 % spoken language. For American English, no such corpus was

|| 91 See Burnage/Baguley (1996) for a portrayal of the BNC; for general information on corpus design see Clear (1992), Atkins/Clear/Ostler (1992), Biber (1993).

Data | 57

available at the beginning of this study. With the American National Corpus project still short of funding, the only available alternatives of sufficient size were commercial Internet search engines, which face problems of representativeness since the World Wide Web represents all varieties of native and nonnative usage in completely obscure proportions, additionally distorted by search engine spam. Thus special care was taken to double-check Internet examples with the help of native speakers more carefully than results from the BNC. A further set of opportunistically sampled and not publicly available corpora was used for the present study: Tab. 2: Additional corpora used in the present study

Corpus label Content

Rough size (in million words)92

academic

journal articles, doctoral theses, academic text archives found on the Internet93

100

fiction

TV transcripts, fiction and fanfiction found on the web, some 260 novels

popmerge

acts of parliament, legal documents, 1911 Encyclopaedia 312 Britannica, web-derived corpus on transportation, collection of encyclopaedia entries, ...

newsmerge

British newspapers

270

britnews

British newspapers

162

newspapers

British newspapers

692

We can see that there is a strong bias towards newspaper language simply because it is easily available. The representativeness of these corpora should thus not be overestimated94 despite claims by Schierholz (2001: 97f) that (German) newspaper language is sufficiently representative of the standard language since it contains the general vocabulary of the language and is not limited to a particular domain, since there are many authors, which assures considerable heterogeneity and since there is a whole range of topics, including even some

|| 92 The numbers are very conservative in that punctuation marks, brackets, and the like were not counted as separate words. 93 A detailed description of the academic corpus is given in Siepmann (2005: 26ff). 94 There is also some small overlap between the newspaper corpora.

58 | Methodological Considerations

technical language. While Schierholz’ arguments sound convincing, they are not backed up by English corpus data. For instance, the four registers used in most analyses by LGSWE (conversation, fiction, news and academic) often show considerable differences, no matter which phenomenon is under scrutiny.95 This observation is of course not meant to devalue the data in our newspaper corpora – since the corpora are used mainly in order to find relevant examples and only occasionally to make statements about their distribution, the effect of the limited representativeness should not adversely affect our results to a great extent (see Schierholz 2005: 7–11 for further discussion). Both British and American usage is present in the academic, popmerge and fiction corpora and some portions of them may well include non-native use since determining authorship is of course a major problem on the Internet.96 The native speaker interviews should, however, have filtered out most unacceptable uses (see Section 3.2.3 below but also Section 9.1 for a discussion of related problems). In the course of this study, the Corpus of Contemporary American English (COCA) by Mark Davies became publicly available, which is a huge improvement over simple newspaper corpora and search engines, but is still considerably less balanced than the BNC.97 Since it was published after a large portion of the data collection in the non-parsed corpora was completed and since it is not available for download or in a parsed version, it could not be taken into account to the extent it would have merited. A further set of corpora often used for syntactic research is the Brown Corpus (see Francis 1965) family, all members of which are also very well balanced but include only 1 million words each and were thus too small to be of help for most of the questions addressed in the present study.

3.2.2.2 Size and noise We cannot expect that a corpus, however large, will always display an adequate number of examples of the phenomena relevant to a particular topic, especially when the phenomena occur relatively infrequently. (Greenbaum 1988: 83)

|| 95 See for instance LGSWE (578) for noun phrase modification, LGSWE (358–397) for verb complementation, LGSWE (476) for the passive voice, or LGSWE (487ff) for modal verb distribution. 96 See for instance Crystal (2011: 30ff). 97 Considering that COCA was constructed by Davies on a relatively small budget (compared to the BNC), it is a tremendous achievement and surprisingly well balanced, of course.

Data | 59

Greenbaum’s statement was made at a time when the 7 million word Cobuild corpus98 was state of the art, and when he cites studies using the 1 million word Brown corpus and other corpora of that size. While it is still valid for the BNC (completed in 1994), which is larger by 2 orders of magnitude, the problems of size and the introduction of noise discussed in this section are slightly different for even lager corpora. While the following passage will draw mainly on lexical material for illustration, we would expect grammatical constructions to behave alike and thus the problems to be highly relevant for the corpus data on the rare phenomena examined in this study. It is interesting to note that Zipf’s law (see Manning/Schütze 1999: 23ff for a discussion) is also valid for a 100 million word corpus, so a rough 50 % (52.8 % according to Hausser’s 2001: 292 count) of its types are hapax legomena, i.e. words that only occur once.99 Some of these are typing errors (acieved, fivbe), errors in automatic text processing (scientific/medical, scientist.), proper nouns (Becel, Schweinfurth) and so on, but many of them are derivations (yak-like, beaverish) or genuine words, such as anglicism, aspheric, audiophile. Statistically speaking, most of these words are over-represented in the corpus. They are included by pure coincidence, and many others which occur with roughly the same frequency in the language are not. Grammatical patterns seem to abide by the same distributional rules, as observed by Sampson (2007: 7–10) in a study of noun phrase structure. Similar conclusions can be drawn from the research on the Valency Dictionary of English (Herbst et al. 2004) and the follow-up project Erlangen Valency Patternbank (Herbst/Uhrig 2009), where a handful of patterns occur with a wide range of verbs and a huge number of patterns occur only with one verb. Many of the grammatical constructions which had to be checked during the research for this study are quite infrequent, so their non-occurrence or low frequency within the corpus (if one is actually able to find them, see Section 3.2.2.3 below) cannot tell us very much about their acceptability. This can only be remedied by larger corpora to a certain extent and that is why the additional set of corpora presented in the previous section were used, too. However, very large corpora face other problems: The research branch of the commercial search engine Google released a list of n-grams (n=1 to n=5) based on a 1 trillion word

|| 98 Within the Cobuild project, a larger unbalanced “reserve corpus” of about 20 million words was also created in the 1980s (see Renouf 1987). 99 This of course depends on the definition of word. Using Kilgarriff’s BNC frequency lists, we can confirm Hausser’s figures, but if we take a closer look at the list of hapax legomena, we can see that many of them are numbers or due to segmentation errors.

60 | Methodological Considerations

corpus of English web pages, which is of course not particularly helpful for research on clausal subjects due to the 5-word window (commas, hyphens, etc. count as words, too). What is interesting to note, though, is that the researchers who produced the n-grams set a cutoff point if a word did not occur at least 200 times. If one takes a look at the words which occur exactly 200 times, there are many foreign words (Abfertigungsstelle, Abituriententag), numbers, proper names, unusually capitalized words which occur more frequently in their regular form (ASSIDUOUS, BoomRang, BRiDe) and a lot of words with tokenization errors (F.Javier, Fast-Twitch/pathology) or just curious strings (FebruaryMinutes16, FBgn0014020), but there are still some rare words, which actually do only occur 200 times in the data, such as konimeter.100 Thus in the light of Internet data, where the corpus which is the basis for Google’s n-gram data set is 1 million times (!) larger than the Brown Corpus, one has to rethink to what extent Greenbaum’s assumption cited at the beginning of this section still holds since it may be that the corpus contains everything one is looking for but lots of “junk”, too. This is due to a problem which occurs wherever people try to amplify weak signals: By amplifying the low-frequency items (which would correspond to small details in a digital picture or a weak signal in audio transmission), we also amplify the noise, which in our case would be performance errors, typing errors, non-native usage and the like. The World Wide Web, which is in turn much bigger than 1 trillion words (as of mid-2008, a rough size of 1 trillion pages — not words! — were indexed by Google),101 probably contains pretty much every verb within the top 500 verb list with all its possible complement types, but it contains lots of “impossible” complement types (“noise”), too. As there is no editing process, frequent errors may occur more often than infrequent “correct” constructions,102 which makes it necessary to have all examples found via Internet search engines doublechecked by native speaker informants (as mentioned above). This issue is likely to become more problematic to the extent to which anyone can post on the In-

|| 100 In the meantime, another set of n-grams has been made available by Google which is based on 361 billion words of running text from the Google Books archive. Due to the fact that it is created on the basis of edited text, the noise level is much smaller. See Michel et al. (2011) for an overview. 101 Figure taken from Google’s Blog: http://googleblog.blogspot.com/2008/07/we-knewweb-was-big.html, which also contains a short discussion of what counts as a web page. (Accessed 18 February 2009) 102 At the lexical level, it is easy to illustrate the problem: The misspelling pronounciation occurs 104,937 times in Google’s n-gram list, whereas geminate occurs only 21,776 times and anglicism even only 5,952 times. (All combinations of upper and lower case were checked.)

Data | 61

ternet. In social networks such as Facebook or in the microblogging sphere (with Twitter in the lead), people write many things they would not write in careful ‘official’ contexts such as school assignments, official letters and the like.103 However, most grammatical description is based on carefully produced language and does not take arbitrary variation into account (unless it is particularly pervasive as in the case of non-concord of existentials discussed in Chapter 7).104 In addition, it is also important to check the origin of the website in question and to use the same search engine site for comparisons. A search for “looking forward to doing” and “looking forward to do” on google.de and google.com led to the following results:105 Tab. 3: comparison of google.de and google.com for looking forward to doing/to do

Website

Region

“looking forward to doing”

“looking forward to do”

google.de

any

778,000

1,060,000

google.com

any

776,000

357,000

Although both sites claim to search the entire web without regional limitations (and with all filters switched off), google.de gives an almost three times higher frequency for the variant “looking forward to do” – which most native speakers of English would classify as “incorrect” – than google.com. Surprisingly, this is not the case for “looking forward to having” vs. “looking forward to have”:

|| 103 See also Crystal (2011: Chapter 3) for a discussion of the linguistic analysis of Twitter data. 104 Sampson claims that “what linguists are primarily interested in is language as an aspect of human beings’ natural behaviour, including spontaneous speaking, rather than a wellbehaved subset of ordinary language that has been promulgated for ‘official’ purposes” (Sampson 2007: 28). This insight has only partially reached the creators of corpora, though, and not all linguists might agree. 105 All figures are the results of queries executed on 17 March 2009.

62 | Methodological Considerations

Tab. 4: comparison of google.de and google.com for looking forward to having/to have with and without a filter for region

Website

Region

“looking forward to having”

“looking forward to have”

google.de

any

1,050,000

1,360,000

google.com

any

1,050,000

1,360,000

google.com

United States

853,000

96,600

What we can see here, though, is how important it is to increase the proportion of native speaker data in web searches if we want reliable results valid for anything close to “Standard English”. Thus we can see that with greater size of corpora, there comes more data but also a higher level of noise. Where the ‘signal’ is so infrequent that it remains hidden in the noise, native speaker informants are needed to distinguish between the relevant data and the noise.

3.2.2.3 Annotation In corpus linguistics, the term annotation is used for the process of adding information which is not part of the text, or the result of that process (see, for instance, Garside/Leech/McEnery 1997). The most basic types of annotation include metadata at the text-level, such as the source of the text or its author, and linguistic annotation at the word-level, such as the word class and possibly the lemma. For our purposes, the latter information is extremely useful, as it lets us refine corpus queries in order to reduce the number of unwanted hits in the results. Lemmatisation is helpful in that it allows us to query all word forms related to a lemma in one go, so we can avoid queries in which we specify all alternatives manually such as “rise|rose|risen|rising” (which would also deliver instances of the nouns rise and rose) if we are looking for the verb rise. Part-ofSpeech (PoS) tagging is one step more refined: A search for to as a word form results in 2,599,205 hits in the BNC whereas a search limited to to as TO0 (toinfinitive-marker) delivers 1,585,812 hits. The form that behaves in a similar way (632,241 as a conjunction compared to 1,119,521 without word class restriction). This sounds too good to be true – and indeed it is. Part-of-Speech taggers (i.e. the computer programs which assign word class tags to words) are not 100 % reliable. “The error rate of state-of-the-art taggers is between 2 and 5 %” (Schmid 2008: 547), but even these impressive numbers mean that up to every 20th word is assigned the wrong word class. And of course, the errors are not equally distributed: A PoS tagger makes only few mistakes on the word the,

Data | 63

which is used as a determiner in the vast majority of cases. This means on the other hand that the error rate for ambiguous and accordingly error-prone words such as that or to is much higher than the overall error rate suggests. Consequently, one has to be aware of such issues when one relies on automatic annotation, but usually the advantages outweigh the shortcomings. The problem with annotation at the word class level is that it is impossible to search for clause constituents. Thus a search for the conjunction that will yield lots of results of non-subject [that_CL]s as in (2) because those are more frequent than subject [that_CL]s. (2)

Official figures suggest that ACET provided care at home for up to one in four of all those who died of AIDS in the UK last year. (A00 145)

One can of course limit the search to sentence initial that, which delivers sentences such as (3) but fails on sentences like (4). (3)

That the local priests continued to play a significant part in the regulation of local sexual morality from the beginning of the state is shown by accounts of "boy meets girl" from the 1920s. (A07 846)

(4)

Even so, that he was able to expel the powerful figure to whom he may have owed much of his victory in 1016 must say something about the effectiveness of the power base which he had succeeded in creating in England in the intervening years. (HXX 945)

In order to be able to find such cases in the corpora, we need to rely on grammatical parsing. Strictly speaking, PoS tagging is also a kind of parsing, but the term is usually limited to a syntactic analysis of some kind when it is used in corpus linguistics. Parsing is much less reliable than PoS tagging and parsers are less readily available than PoS taggers. Few (and often small) corpora, such as the Penn Treebank (Marcus/Santorini/Marcinkiewicz 1993) or the British component of the International Corpus of English (ICE-GB) are made available in parsed form, due to the manpower required for manual annotation or correction. Every parser requires a so-called parsing scheme – essentially the grammatical model underlying the piece of software. There is much less consensus on what output a parser should produce than there is for PoS-taggers.106

|| 106 Of course there is no universally agreed English tagset for all taggers, but many of the often very fine-grained analyses offered by taggers can be transformed to capture more gener-

64 | Methodological Considerations

For a small part of the study (the extraction of extraposed sentences used for Section 7.2.1), a version of the BNC parsed with a Dependency Parser developed at the University of Zurich (Schneider 2009) was used. The parser mainly used in this study is the Stanford Parser (Klein/Manning 2003), a phrase structure grammar parser which can also deliver dependency output (de Marneffe/MacCartney/Manning 2006) and provides PoS-tagging at the same time. It was trained on the Wall Street Journal portion of the Penn Treebank plus a set of sentences to include clause types underrepresented in the Penn Treebank data. The BNC and the internal academic, fiction, newsmerge, britnews and newspapers corpora – i.e. more than 1.5 billion words – were parsed using Erlangen University’s High Performance Computing facilities,107 then their dependency representations were stored in a relational database to allow for relatively fast queries.108 A web user interface was produced with visualization tools and the option to add results from the corpus directly to a database that could then be used for the present study (with additional information on extraposition, voice, comments, …).109 Only with such a tool at hand did the identification of clausal subjects (particularly in canonical position) become feasible on a sensible scale. Again, one has to bear in mind that the accuracy of parsing is still rather limited,110 so neither precision nor recall are perfect, but in spite of these limitations, the parsed corpora offer so many advantages over any nonparsed corpus in the retrieval of clausal subjects that only due to them was it possible for all the analyses in the present study to be carried out. The problem is that the biggest of all sources of data available for our research, the World Wide Web, is of course not annotated for word class or lemma, let alone syntactically parsed. In addition, none of the currently available || alizations (there is – for example – a converter from the C7 to the C5 tagset; see Garside 1987, 1996, Garside/Smith 1997). Thus even if a tagger was programmed/trained to distinguish between prepositions, certain types of adverbs and certain types of subordinating conjunctions, it is quite easy to lump them together if one decides not to make that distinction in a grammatical description (as, for instance, CamG or Herbst/Schüller 2008 do). 107 On a standard office computer, the parsing of these corpora would have taken more than 12 months. 108 On the server used, the original version of the database took about 3 minutes for a query of clausal subjects in canonical subject position and up to 15 minutes for a query of extraposed clausal subjects, so for queries of extraposition, a regular expression search was often used instead, even though it decreased precision. 109 It is the need for the research in this study that provided the impetus for a successor of the database that is now available as the Erlangen Treebank.info project (Uhrig/Proisl 2011). 110 See Cer et al. (2010) for figures on the parsing accuracy of different parsers.

Data | 65

commercial search engines take sentence boundaries into account nor do any of them offer case-sensitive queries any more111 – two features which at least would have permitted queries of the type “. That” in order to find a sentence-initial [that_CL] (and instances of sentence initial demonstrative that, of course). It is for this reason that, despite its sheer size and the easy availability, Internet data turned out to be of very limited usefulness.

3.2.3 Native speaker interviews As pointed out above, a study such as the present one cannot depend solely on corpus data. But then again, native speaker interviews are time-consuming and expensive to do on a large scale, so a sensible way of combining the two means of information had to be found for the present study. If we assume that we need about 20 native speakers for reliable results112 and that we have to test every item twice for every native speaker informant, as is suggested by Cowart (1997: 7ff and 67ff), testing every item that was tested for the present study (more than 1,150 sentences) would have resulted in over 45,000 individual judgements, which would have been impossible to handle in the context of the present study.113 Most of the example sentences for this study were checked by three or four native speakers of English (see Appendix 27 for details) and only retested if there was serious inconsistency. In addition, all sentence judgement tests took place under the informal, uncontrolled, distorting conditions to which Greenbaum (1988: 87) objects.114 Furthermore, the native speaker informants were usually asked whether they accept or reject a sentence. The five point scale that can be seen in the data listed in the appendix is normally an interpretation of the answer given. Clear yes or no answers were given a 5 or a 1; judgements of the type “Yes, that is ok,

|| 111 See http://www.searchengineshowdown.com/features/byfeature.shtml version March 2008, accessed 17 February 2009. 112 The number is chosen deliberately small. Greenbaum (1988) uses over 40 for his experiments, sometimes even over 100. 113 If we had wanted to verify each corpus example collected for the present study, the number would have to be at least three or four times higher. 114 Sentences were read to informants or, less frequently, submitted to them via email. By some, these messages were answered on paper, by others via email. Some interviews took place in the researcher’s office, some in the subject’s office, a few even in local pubs. Where the sentences were not understandable out of context, the relevant context was retrieved from the corpora. Invented examples were often crafted with the help of one of the informants.

66 | Methodological Considerations

but I would prefer a to-infinitive” or “It may not be wrong but nobody would say it” receive a 4 or a 2 respectively, and “I don’t know” or similar answers a 3.115 Thus the native speaker interviews are of course by no means satisfactory from an empirical point of view, but it should be borne in mind that the corpora, not the interviews, are the primary source of data. The fact that some of the native speaker informants were American and some British can distort the results of the tests, but it turned out that with so few informants, individual differences had much more influence on the judgements than variety, and agreement/disagreement among the informants seemed not to be tied to the country of origin.116 There is a certain wariness towards native speaker intuition in the corpuslinguistics and (partly) the usage-based communities which is no doubt due to the way in which intuition was used in the generative transformational paradigm, which was at its heyday when Labov stated that “linguists cannot continue to produce theory and data at the same time” (Labov 1972: 199). There is much less risk of that in a study such as the present one, where the author is not a native speaker of English, where the native speaker judgements are used merely to complement the corpus data and where at least three native speakers were involved in most of the judgements.

3.3 Acceptability/Grammaticality In a series of publications starting in 1987,117 Sampson has argued that the distinction between grammatical and ungrammatical structures is non-existent, a claim that has triggered a lot of discussion (see for instance Culy 1998 for a rebuttal and Greenbaum 1988: 94f for a general discussion and additional references). The journal Corpus Linguistics and Linguistic Theory even dedicated an entire issue (2007/1) to the question with a target article by Sampson and responses by Pullum, Meurers, Stefanowitsch and others. Since judgement tasks are performed for the present study, a short discussion of the matter is in order.

|| 115 Since the scale is not an interval scale, no mean can be given (see for instance Albert/Marx 2010: 105ff). 116 The number of years spent in a German-speaking country may actually be a more problematic factor, but even that could not be shown due to the small number of informants. See Section 9.1 for a further discussion of issues associated with the native speaker interviews. 117 Sampson’s 1987 article takes a less theoretical approach. It is more concerned with computational models that make a distinction between grammatical and ungrammatical sentences.

Acceptability/Grammaticality | 67

When Sampson claims that a sentence such as the following is a meaningful English utterance despite the fact that the last two instances of the determiner the follow their head noun, he is certainly right: (5)

Norwegians put the article after the noun, in their language they say things like bread the is on table the. (Sampson 2007: 20)

However, the issue is marginal, so if a grammatical model failed to predict sentences such as (5) due to the odd noun phrase structure, it could still be a very good model of English syntax. Stefanowitsch, who supports Sampson’s argument, brings forward language contact phenomena as an argument against the grammatical/ungrammatical distinction (Stefanowitsch 2007: 59f), which is probably a better example. Still, native speakers seem to be perfectly happy to pass judgements on what they think is correct and what is not. Not all of their intuitions may be shared by all other native speakers of the language – and that is why the objection to individuals producing the data cited in their linguistic articles on the go is quite well founded – but if we administer experiments of the type designed by Cowart (1997), then we can safely assume that if there is overall agreement between subjects to reject a sentence, this reflects some sort of unacceptability. Note that the term chosen in the previous sentence was unacceptability, not ungrammaticality. It is unfortunate that neither Sampson nor Stefanowitsch discuss in any detail this distinction presented by Chomsky (1965), who states that “[l]ike acceptability, grammaticalness is, no doubt, a matter of degree [...], but the scales of grammaticalness and acceptability do not coincide” (Chomsky 1965: 11).118 For him, grammaticality is related to competence whereas acceptability is related to performance. Thus he can present grammatical strings of words which are hardly acceptable: (6)

the man who the boy who the students recognized pointed out is a friend of mine (Chomsky 1965: 11)

|| 118 Although Chomsky uses the term grammaticalness instead of grammaticality, we shall nonetheless use grammaticality in the discussion of his view for consistency’s sake. Note however that Chomsky’s choice, he claims, is a deliberate one: I purposely chose a neologism in the hope that it would be understood that the term was to be regarded as a technical term, with exactly the meaning that was given to it, and not assimilated to some term of ordinary discourse with a sense and connotations not to the point in this context. (Chomsky in a discussion; Paikeday 2003: 28)

68 | Methodological Considerations

From that point of view it makes sense to criticise the concept of grammaticality, since it predicts a structure such as (6) that is not usually used in human communication and the usefulness of a grammar that does not model human communication is rather limited. It is probably also correct that we will never be able to predict exactly what is grammatical and what is ungrammatical in the sense of Chomsky’s (1957: 13) statement.119 Thus there is an advantage of usagebased approaches, which primarily try to account for the language as it is used and are less concerned with the question of what bizarre constructions might be possible if speakers had unlimited working memory.120 Thus, for the present study, we shall try to model valency aspects and treat them as rooted “in a general linguistic theory of the occurring and the nonoccurring” (Stefanowitsch 2007: 69), i.e. a relatively usage-based approach. Accordingly, if we want to make the distinction between grammaticality and acceptability, no statements as to the grammaticality of any linguistic construction will be made since the present study does not posit an unobservable competence of any sort. It is, however, useful for our purposes to be able to mark sentences as unacceptable.121 At a less theoretical level, we can easily make a point for parting ‘correct’ from ‘incorrect’ sentences, since from a foreign language learning perspective the distinction is crucial. Foreign language teachers must not leave sentences such as the following one uncorrected, simply because a large number of native speakers probably share the assumption that the sentence is plain wrong:122 (7)

*My father wanted that she goes to the doctor.

|| 119 In 1957, Chomsky claims that grammar is about separating grammatical from ungrammatical sequences. But since he states, as cited above, that grammaticalness is to be regarded as a scale in 1965, it does not appear as if Chomsky himself believes that the exact delimitation of grammatical and ungrammatical structures will ever be possible. 120 In a way, the distinction between grammaticality and acceptability finds a certain parallel in the distinction between System and Norm made by Coseriu (1973: 44) with the Norm being what is common whereas the System contains everything that is possible from the point of view of grammatical rules. 121 Langacker’s choice of wording for one of his starred sentences is worth noting: “In (32) we observe that a finite complement in subject position is unhappy after an auxiliary verb in a question” (Langacker 2008: 430; my emphasis). 122 Since teachers of English as a foreign language also have to take factors such as style level into account and may be influenced by prescriptive rules, acceptability can of course not be equated with what teachers allow.

Acceptability/Grammaticality | 69

There is no reason why linguists should not be allowed to make the same point in their discussion of the verb want and state that (7) is unacceptable, for instance to contrast it with verbs such as wish, as long as this statement is based on sufficient data. Thus the present study will make a distinction between acceptable sentences, unacceptable sentences (marked with an asterisk) and sentences of questionable acceptability (marked with a question mark) where necessary, despite the concerns discussed above.

4 Clausal Subjects in active clauses The aim of the present chapter is to identify restrictions on the formal realisations of subjects occurring with non-copula verbs in active declarative clauses.123 If such restrictions can be found, this will be taken as an indication for the valency complement status of the subject. In a second step, the influence of semantic factors will be minimised by looking at near-synonymous or semantically similar verbs. Finally, a case study on the interdependence between modality and the choice of subject will be carried out to show that other factors than verb valency may be at work in selecting formal realisations of subjects.

4.1 Data analysis 4.1.1 VDE data The Valency Dictionary of English (VDE; Herbst et al. 2004) is used as a starting point for our analysis since it lists permissible clausal subjects for each lexical unit identified. Of the 511 verb entries in the VDE, many are split up into several lexical units or subsenses, so that there is a valency description for 1291 lexical units in total.124 According to the dictionary, only 200125 of them take a clausal subject of at least one of the following forms: [to_INF], [that_CL], [V-ing], [NP_Ving], [wh_CL],126 [wh_to_INF], [for_NP_to_INF]. The following diagram gives an overview of the distribution of these seven forms of subjects across lexical units in the VDE:

|| 123 As mentioned above, this study is limited to English. For a crosslinguistic perspective on clausal subjects see Schmidtke-Bode (2014, Section 6.4.2). 124 This count only comprises full valency descriptions. Idiomatic uses (“Further uses” in the complement block of the VDE) and phrasal verbs are not included since for them, only postverbal complements are specified in the dictionary. 125 The list, however, contains only 199 verbs since no systematic indication as to which clausal subjects are allowed are given for the verb be in the dictionary, which is a sensible choice due to its status as a copula verb. Other verbs often treated as copula were included in the list since the VDE listed restrictions on some. See the discussion at the beginning of Chapter 6 for a brief discussion of what counts as copula verb. 126 The use of the cover-term [wh_CL] in the present study follows Herbst/Schüller (2008). However, many grammarians distinguish between (indirect) interrogatives clauses and nominal relative clauses (CGEL 1056ff) / fused relative clauses (CamG 987ff for the distinction). https://doi.org/10.1515/9783110589801-091

72 | Clausal Subjects in active clauses

VDE Lexical Unit Count 176

155 93

80

39

19

5

Fig. 6: Number of lexical units for the various types of clausal subject in the VDE

It is immediately obvious that [V-ing], [that_CL] and [to_INF] clauses occur with a much higher number of verbs than wh-elements do, with the complex forms [NP_V-ing] and [for_NP_to_INF] in between. In fact there is no single lexical unit in the VDE which accepts wh-elements and no other clausal subject, while there are such items for [V-ing], [that_CL] and [to_INF]. The full list can be found in Appendix 1, but the top of the list (covering the lexical units starting with a and b) will be reproduced here to illustrate what inspired the hypothesis given at the beginning of this study, i.e. that the choice of subject is highly itemspecific.127 Tab. 5: Clausal subjects with verbs in the range a – b in the VDE

[to_INF]

[that_CL]

[V-ing]

account(A)



+

+

act(C)





add(A)





afford(B)



+

agree(E)





[NP_V-ing] [wh_CL] [wh_to_INF]

[for_NP_to_ INF]

+







+

+







+









+

+







+









|| 127 For all tables of this kind in the present study, attested and/or acceptable combinations are indicated by a plus sign; unacceptable combinations are marked by a minus sign; inconclusive evidence is indicated by an o.

Data analysis | 73

[to_INF]

[that_CL]

[V-ing]

[NP_V-ing] [wh_CL] [wh_to_INF]

[for_NP_to_ INF]

allow(A)





+









allow(B)



+

+









amuse(A)

+

+

+

+

+





annoy()

+

+

+

+

+



+

appear(B)

+

+

+

+





+

attract(A)

+

+

+









attract(B)



+

+

+

+





bar(B)



+

+









become()

+



+

+





+

begin()

+



+

+





+

belong(A)





+









bind(B)



+

+









breathe()



+

+

+







breed(C)



+

+

+







bring(B)

+

+

+

+





+

buy()

+

+

+

+







It has to borne in mind, though, that the information on subjects in the VDE is often based on native speaker intuition not complemented by corpus evidence, given that the corpora used for the dictionary were not sufficiently large to allow sensible lexicographic treatment of the relatively rare issue of clausal subjects. Thus, as mentioned above, the list compiled on the basis of the VDE can be nothing more than a starting point for further analyses because it helps to identify candidates for restrictions which then have to be verified with the help of corpora (and possibly native speaker informants).

4.1.2 Corpus evidence In order to verify the VDE data, the parsed corpora were searched for the following patterns in order to find clausal subjects:128

|| 128 csubj = clausal subject; aux = auxiliary; complm = complementizer; %ing: anything that ends in -ing; see de Marneffe/Manning (2008) for a full description of the grammatical model.

74 | Clausal Subjects in active clauses

[to_INF]: [verb] ---csubj-----> [not specified] ---aux-----> to [that_CL]:[verb] ---csubj-----> [not specified] ---complm-----> that [V-ing]: [verb] ---csubj-----> %ing

In order to keep the amount of data manageable, it was necessary to restrict the selection to the three major types of clausal subjects, [to_INF], [that_CL] and [Ving]. Also, being the most frequent types of clausal subject, these seem to be the most versatile as well, so semantic incompatibilities may be found less often than with the rarer types. All results were manually checked and relevant sentences were added to a database. At this stage of analysis, all lexemes that occurred with more than one lexical unit in the list were ignored for two reasons. First, the corpus does not allow querying for specific lexical units, which makes checking for restrictions on one of them difficult if the structure is attested with the other, due to massive noise. Secondly, the division in different lexical units in the VDE follows lexicographic principles and does not necessarily correspond to other linguistic analyses, which again might be different depending on whether one is a “lumper” or a “splitter” (see Goldberg 2006: 45). Thus any analysis relying on differences between lexical units can easily be challenged as long as there is no generally accepted method of identifying lexical units. The table shown above and given in full in Appendix 1 was augmented with the results from the corpus queries to form Appendix 2, which now shows fewer restrictions: Tab. 6: Clausal subjects of the [to_INF], [that_CL] and [V-ing] type with verbs in the range a – b in the VDE, augmented by corpus data

[to_INF]

[that_CL]

[V-ing]

account

o

+

+

act

+

+

+

add

+

+

+

afford

+

+

+

agree

o



+

amuse

+

+

+

annoy

+

+

+

appear

+

+

+

bar



+

+

become

+

+

+

begin

+

+

+

comment

[to_INF] and [that_CL] not attested

Data analysis | 75

[to_INF]

[that_CL]

[V-ing]

comment

belong

o



+

[to_INF] only in one near-quotation from Shakespeare’s Sonnet 23

bind



+

+

[to_INF] and [that_CL] not attested

breathe



+

+

[to_INF] and [that_CL] not attested

breed

+

+

+

bring

+

+

+

buy

+

+

+

All ‘+’-signs from the VDE were kept and only the cells containing ‘-‘-signs in the original table came under closer scrutiny. This means that if either the VDE or the corpus contains a clausal subject, no restriction is given in the list. Thus, as can be seen above, although the corpus search did not yield [that_CL] subjects for breathe, the corresponding cell contains a ‘+’-sign since the VDE allows this formal realisation.129 It cannot be stressed enough that this method introduces a strong bias against restrictions and that many of the ‘acceptable’ examples are extremely marginal, but this approach is deliberate to ensure that the arguments for restrictions in the present chapter do not conflict with any of the sources of data used here and to avoid the impression that the data was prepared in order to support the hypothesis that there are arbitrary restrictions.

4.1.2.1 Restrictions identified Taking a closer look at the table in Appendix 2, we can see that the most striking result is that all but one of the 140 remaining verbs (see above) allow a [V-ing] clausal subject. This one verb is grieve, which is attested both with a [that_CL]130 and a [to_INF] in the corpus: (1)

a.

It grieves me terribly that I was unable to reach out or do anything to save my daughter. (newspapers) b. To see Tullivers so neglected had grieved Thrush Green. (ASE 194)

|| 129 Any verb that showed a restriction after the completion of the corpus search was checked for all three types of clausal subject in the corpus. 130 No [that_CL] in canonical position was found, though, so an extraposed version is used as an example here.

76 | Clausal Subjects in active clauses

However, the verb grieve is not particularly frequent (roughly 3 instances per million words in the BNC) and the patterns exemplified in (1a–b) are just about present in the corpus, so we cannot be sure that this result is not a corpus artefact. A modified version of (1b) with a [V-ing] can be formed: (1)

c.

Seeing Tullivers so neglected had grieved Thrush Green.

Since (1c) is not unacceptable, the corpus may just as well be too small to contain such an example. In any case, even if the exception of grieve did hold, the data tells us that whenever a verb allows for a clausal subject of any form, we can generally expect it to allow a [V-ing]. The reason may be that [V-ing] clauses are more nounlike than [to_INF] and [that_CL] complements, even though every care was taken in the corpus study to only include clearly verbal [V-ing] clauses (i.e. those that showed complementation by an [NP] or similar structures). On the other hand, if a verb allows for a [V-ing] clausal subject, this does not tell us anything about other possible realisations of clausal subjects. For instance, as predicted by the VDE, guide is only attested with a [V-ing] subject but not with a [to_INF] or a [that_CL]: (2)

a.

She once said that having acupuncture had guided her to a “new level” in life, helping her to find love with Martin and giving her the strength to cope with the death of her father last year. (newspapers) b. *She once said that to have acupuncture had guided her to a “new level” in life, helping her to find love with Martin and giving her the strength to cope with the death of her father last year. c. ?That she had acupuncture had guided her to a “new level” in life.131

In the case of matter, where the VDE only predicted a [that_CL] subject (as in (3a)), we can see that it does readily occur with a [V-ing] subject, too, as in (3b): (3)

a. That they’re almost always awful hardly matters. (britnews) b. Packing the right piece matters as much as wearing the right athletic shoes. (newsmerge)

However, no [to_INF] could be found in the corpus, and in this case the result is less likely to be a corpus artefact since matter is much more frequent than grieve discussed above (roughly 50 instances per million words) and since the other

|| 131 While this example is better than (2b), the native speaker informants preferred an [NP] with The fact that.

Data analysis | 77

two realisations both occur at least around ten times in the corpus. Still, of course, as discussed in Section 3.2 above, the figures are not high enough to claim that they suffice to rule out a [to_INF] subject, but we can be more confident that such a use would be marginal than in the case of the [V-ing] for grieve above. Restrictions on the [that_CL] complement also exist, as for instance with introduce, for which again the VDE analysis was confirmed. So we find [V-ing] and [to_INF] subjects as in (4a–b) below, but the [that_CL] is less likely. (4)

a.

Involving the Department of Education and the schools in the process will introduce a wider awareness of road safety in the school. (GXJ 4163) b. He was concerned that to include the large Asian countries in such a scheme would introduce problems that the small nations could find difficult to resolve. (GVK 125) c. ?That they included the large Asian countries in the scheme introduced problems that the small nations found difficult to resolve.

However, once more, the [that_CL] in (4c) cannot be ruled out categorically, although native speakers seem to prefer the sentence with a “The fact that” [NP] subject instead. Note, however, the meaning difference between (4b) and (4c), which is due to the difference in the modality and factual character usually associated with the two structures which will be discussed in more detail in the case study in Section 4.2 below for [to_INF] subjects.132

4.1.2.2 Problematic cases There is a whole set of verbs for which the VDE predicts a [V-ing] and a [that_CL] subject but only the former was found in the corpora. They can be identified from the comment column in the table in Appendix 2. The set contains the verbs belong, bind, entertain, hang, interest, link, persuade and shape.133 These were not counted among the verbs showing restrictions if no other restrictions oc-

|| 132 In order to make (4c) as compatible with a [that_CL] as possible, it was rephrased in the past tense. 133 The existence of such a set where corpus data and native speaker intuition do not coincide can be interpreted in two ways. It could be regarded as an argument against native speaker intuition (if we believe the corpus evidence) or it could be seen as an argument in favour of bigger corpora (if we trust the native speakers). See also the discussion in Section 9.1.

78 | Clausal Subjects in active clauses

curred due to the careful approach taken here, but in fact the corpus data here strengthens our argument for the existence of restrictions. Another highly interesting case from a valency point of view is presented by take as in the following example, where the [that_CL] is found in subject position although take does not usually allow a [that_CL] subject: (5)

That this may not be the case in certain instances does not take much imagination to comprehend. (HP3 929)

The issue will be discussed in more detail in Sections 7.4 and 9.3.2.2.1, so all that needs to be said here is that the [that_CL] in (5) will not be treated as a complement of the verb take but as a complement of the verb comprehend, even though it functions as the grammatical subject in a sentence in which take is the main verb.

4.1.2.3 Summary In total, the combined analysis of the corpora and the VDE data showed that in our list of 140 verbs, 29 show a restriction on at least one clausal subject realisation, even using the most conservative way of counting.134 (A list of examples is given in the conclusion in Section 4.3.) Thus the data does suggest that there are restrictions on the formal realisations of subjects depending on the verb. However, this does not necessarily mean that the restrictions may not be predictable on semantic grounds, so in order to shed more light on that question, a complementary analysis based on native speaker interviews was conducted.

4.1.3 Native speaker data If we want to find out whether restrictions of the kind identified above are arbitrary syntactic restrictions, we will have to exclude semantic (and other) factors as far as possible from our analysis. In order to do so – following the example set by Faulhaber (2011) for postverbal complements – semantically similar verbs

|| 134 The list contains in fact 31 verbs, but furnish was disregarded as the corpus analysis did not yield any evidence for clausal subjects with that verb at all. Furthermore, the ones with dubious examples (e.g. account, establish, pass, ...) were not counted either. Finally remain was not counted for it is often treated as a copula verb and thus rather belongs in Chapter 6.

Data analysis | 79

were identified for which example sentences were crafted135 which then were submitted to native speakers for acceptability judgements.136 All judgements are available in Appendices 3 to 13. The most promising pair of near-synonyms with possible arbitrary restrictions on the form of the subject are convince and persuade. According to the VDE, both are expected to occur with [V-ing] and [that_CL] subjects but not with a [to_INF]. In the parsed corpora, however, we find persuade only with [V-ing] subjects and one instance with a [that_CL] whereas convince seems to allow all three types:137 (6)

a.

In other words, watching the film persuaded me that despite the intricacy of the life-producing process, it is not wrong to think of helping it along in certain ways. (B1J 574) b. That Prince Charles was absolutely committed to her and probably had been since they first met, long before he married Diana, persuaded me that I should see her as soon as possible. (newspapers)

(7)

a.

By itself, changing the party’s name will not convince a sceptical electorate. (newspapers) b. That Iraq never once even nibbled at the goodies has convinced the administration that diplomacy was always doomed, because Mr Hussein meant to hold on to Kuwait no matter what. (ABH 660) c. Just to read the 893 case histories of Pritikin’s first patients reviewed by the Loma Linda University in 1978 would convince anybody of that. (academic)

However, the native-speaker judgements do not sustain the corpus analysis. While acceptability is equally high for non-extraposed [that_CL] in both verbs and similarly for [V-ing], the acceptance rate for non-extraposed [to_INF] is even slightly higher for persuade than for convince. The little difference we can ob|| 135 Despite all the problems associated with such invented examples, there is no alternative given the range of factors that influence the structure of sentences. With corpus examples, it would be impossible to rule out other influences. See for instance Hauser-Suida/Hoppe-Beugel (1972: 65ff) for restrictions imposed by tense in German and the papers in Rohdenburg/Mondorf (2003) for an array of relevant factors. 136 All native speaker data for the present section were collected in actual interviews. Great care was taken to finish the judgement task on one verb before beginning with another in order not to contrast otherwise identical sentences with different verbs directly and thus obtain more ‘natural’ (if that is possible in such an experiment) answers. 137 The example given as (7b) is the only one of a non-extraposed [that_CL] with convince.

80 | Clausal Subjects in active clauses

serve between the two verbs in the native speaker data is related to isolated phenomena which may be indeed due to the semantic differences between the two. One native speaker remarked for instance that convincing can be more easily done by circumstances or facts whereas persuading is prototypically done by people. Such a hypothesis would account for the acceptability differences between the following two pairs: (8)

a. Hearing her voice on the telephone convinced me that she was alive. b. ?Hearing her voice on the telephone persuaded me that she was alive.

(9)

a. ?For the company to promise him stock options may convince him. b. For the company to promise him stock options may persuade him.

In the first pair, hearing her voice on the telephone is a factual element – evidence in a way – which is better suited for convince whereas for the company to promise him stock options in the second pair is an intentional act aimed at changing the referent’s mind and thus better suited for persuade. However, in the first pair, the difference in acceptability may just as well stem from a semantic incompatibility of the postverbal [that_CL] with persuade, given that in the following pair, no acceptability difference could be found: (10) a.

Yes, seeing the merchandise would convince me that you’re telling the truth. b. Yes, seeing the merchandise would persuade me that you’re telling the truth.

We thus cannot reliably identify valency differences for the subject position of convince and persuade with the help of the native speaker data collected here. If we believe the native speaker informants, all three types of clausal subject are acceptable with both verbs. For the other verb groups formed for the present study, no differences in the acceptability with the three major clausal subjects could be identified in the native speaker studies, either: All verbs allowed all three types of clausal subject. Some verbs not included in the VDE were used in this part of the study, too, since the number of semantically similar verbs in the VDE was not deemed large enough to allow for a systematic study of such phenomena. The groups used for the interviews were: – – –

annoy/offend/upset amuse/please confuse/shock/surprise/puzzle

Case study: modality and non-extraposed [to_INF] subjects | 81

However, the judgements on these were often far from unanimous and sometimes contradicted corpus data, so to what extent native speaker data can be regarded as a reliable indicator of actual usage remains questionable. The issue will be taken up again in a critical evaluation of the relationship between corpus data and native speaker data in Section 9.1. The evidence presented by Newmeyer (2003: 167), however, suggests that there are in fact syntactic (rather than semantic) restrictions on the form of the subject with some verbs in that these allow no [that_CL] but an [NP] with the fact that. He specifically cites persuade as one of these verbs, for which we have presented one counter-example above, but also give three others: (11) a.

The fact that / *That the weather was sunny persuaded Mary to do some gardening. (Newmeyer 2003: 167) b. The fact that / *That the WTO riots were violent impressed upon the Italian police the need for extra precautions. (Newmeyer 2003: 167) c. The fact that / *That we had one more applicant tipped the scales in favor of our extending the deadline. (Newmeyer 2003: 167) d. The fact that / *That I failed the exam decided me to study harder in the future. (Newmeyer 2003: 167)

For (11b–d) no counter-evidence was found in the parsed corpora138 and the native speaker informants agreed with Newmeyer’s analysis.139 This kind of restriction then speaks strongly in favour of syntactic restrictions that cannot be explained by semantics alone.

4.2 Case study: modality and non-extraposed [to_INF] subjects As mentioned above, during data collection a high proportion of modal verbs were found in sentences with non-extraposed [to_INF] subjects, so this case study sets out to verify that intuitive observation and present relevant data.

|| 138 There is a [that_CL] subject for impress, but not for the reading in (11b). 139 Of course, Newmeyer himself is a native speaker of English, so whether he uses an asterisk or not can be regarded as a further native speaker judgement.

82 | Clausal Subjects in active clauses

4.2.1 Theoretical background This section will follow the approach taken by the authors of CamG (52), who use the traditional distinction between two types of modality, epistemic and deontic,140 and add a third dimension called dynamic modality.141 For our purposes, though, it is enough to concentrate on epistemic modality and to contrast it with non-epistemic types of modality. In relatively broad terms, epistemic can be “interpreted as showing the status of the speaker’s understanding or knowledge; this clearly includes both his own judgements and the kind of warrant he has for what he says” (Palmer 1986: 51), so basically, epistemic modality is concerned with the likelihood of something being true (from the speaker’s perspective). By contrast, “[d]eontic modality typically has to do with such notions as obligation and permission, or – in combination with negation – prohibition” (CamG 52). Finally, “[d]ynamic modality generally concerns the properties and dispositions of persons, etc., referred to in the clause, especially by the subject” (CamG 52), which “includes the meanings of volition and ability” according to Verplaetse (2003: 154).142 While there is a whole range of devices available to express modality in English, including full verbs (e.g. guess, allow), nouns (e.g. possibility, permission), adjectives (e.g. likely, forbidden) or adverbs (e.g. probably; see Hoye 1997 for a detailed account), the analysis in this section is limited to the most frequent markers of modality, i.e. modal verbs.

|| 140 Palmer (1986: 18) traces the distinction back to Jespersen. 141 Many scholars, however, have introduced further distinctions or decided to use different terminology, so there is root modality with different interpretations of the term (see Nuyts 2001: 25), which in some readings is a cover-term for deontic and dynamic and thus synonymous with agent-oriented modality as opposed to speaker-oriented modality (Bybee 1985, Bybee/Perkins/Pagliuca 1994; see also Krug 2000: 41ff for a short discussion). Quirk et al. (CGEL 219) rely on a gradient between intrinsic and extrinsic modality. While all of these concepts may be helpful or even necessary for a fine-grained analysis of modality in English, the point at stake here can be illustrated with much less-refined basic terminology. 142 While the prototypical elements of meaning such as truth (epistemic) and obligation (deontic) are distinguished by most linguists, there has been considerable debate as to whether volition should be subsumed under deontic or where to include necessity or ability.

Case study: modality and non-extraposed [to_INF] subjects | 83

4.2.2 Presentation of data The parsed corpora were searched for non-extraposed [to_INF] subjects of all the verbs discussed in the previous sections, i.e. verbs that allow at least one type of clausal subject of the type [to_INF], [that_CL] or [V-ing].143 The results were then checked manually and every sentence had to be specifically selected to find its way into the database. The sentences in the database provide the sample used in this chapter.144 In the sample, almost half of the verbs which are used with a [to_INF] subject are preceded by one of eight modal verbs,145 as can be seen in the following table: Tab. 7: Distribution of modal verbs in the sample of sentences with [to_INF] subjects

number of sentences

percentage of sentences146

will

71

6.6 %

would

323

29.9 %

can

25

2.3 %

could

32

3.0 %

may

23

2.1 %

should

7

0.6 %

must

7

0.6 %

might

26

2.4 %

modal verbs total

514

47.5 %

no modal verb total

567

52.4 %

These results are remarkable as they show different values than the distribution of modals over a corpus in general: “Verbs phrases with modals comprise 10– 15 % of all finite verb phrases in all registers” (LGSWE 456). Furthermore, the || 143 The copula verbs be, appear, seem, become and remain were, however, discarded. For a discussion of what counts as a copula verb see the beginning of Chapter 6. 144 If for a given verb only a few sentences with a [to_INF] subject were found, they were all added to the sample. If a verb occurred frequently with [to_INF] subjects, only 8–10 were usually added to the database. 145 The list of modal verbs is based on LGSWE (486) with the omission of shall, which did not occur in the collection of example sentences. 146 Due to rounding, figures do not add up to 100 %.

84 | Clausal Subjects in active clauses

distribution of the modal verbs does not reflect the distribution in a general language corpus (see also LGSWE 486). The following table compares the distribution in the sample with the distribution over the whole BNC. Tab. 8: Distribution of modal verbs in the BNC and in the sample of sentences with [to_INF] subject

number in [to_INF] collection

percentage147

number in BNC

percentage

will

71

13.8 %

328,424

23.8 %

would

323

62.8 %

278,522

20.2 %

can

25

4.9 %

261,778

19.0 %

could

32

6.2 %

159,818

11.6 %

may

23

4.5 %

112,397

8.2 %

should

7

1.4 %

108,970

7.9 %

must

7

1.4 %

69,752

5.1 %

might

26

5.1 %

59,026

4.3 %

The plot of these data shows that would and (to a much lesser extent) might rank higher than in the overall BNC whereas can, should and must rank lower: 70% 60% 50% 40% 30%

with [to-INF] subject

20%

BNC overall

10% 0%

Fig. 7: Plot of the distribution of modal verbs in the BNC and in the sample of sentences with [to_INF] subject

|| 147 Due to rounding, figures do not add up to 100 % in both percentage columns.

Case study: modality and non-extraposed [to_INF] subjects | 85

Since this distribution is averaged over all verbs, it is also important to observe that there are considerable verb-specific differences, as will be seen in the following two charts. Figure 8 charts the collected examples, between 10 and 15 per verb, so it has to be treated with caution due to the small numbers involved. Figure 9 only counts instances of the selected modal verbs when they occur within 1 to 3 words to the left of the verb in the BNC. This means it misses out on many coordinated structures where two verbs “share” one modal verb and on cases in which adverbials between and modal verb and main verb are 3 words or longer: (12)

Hoare also agreed that he had been advised by the Director of Public Prosecutions that charges of seditious libel against individuals attacking Jews as a group would in all probability fail. (CS6 677)

On the other hand, the count erroneously includes cases such as (13) for the verb harm. (13)

‘Who'd want to harm Connon?’ (GUD 696)

Thus Figure 9 also represents tendencies rather than exact figures, but both diagrams should nonetheless be reasonable approximations of the actual situation.

Sentences with [to_INF] subjects say fail indicate

modal

threaten

non-modal

place harm 0%

20%

40%

60%

80%

100%

Fig. 8: Distribution of modal elements among selected verbs occurring with [to_INF] subjects in the data collected for the present chapter

86 | Clausal Subjects in active clauses

BNC frequency count say fail indicate

modal

threaten

non-modal

place harm 0%

20%

40%

60%

80%

100%

Fig. 9: Distribution of modal verbs across all instances of selected verbs in the BNC

4.2.3 Interpretation of the data The most striking feature found in the data is the extraordinarily high frequency of modal verb use (particularly of would), if the [to_INF] occupies the canonical subject position. This might be interpreted as being mainly due to inherent semantic features of [to_INF] complements. Accordingly, the [to_INF] has been linked with potentiality148 by many scholars:149 In general, a to-clause has a meaning that is more hypothetical or potential than the meaning of the corresponding ing-clause (with the same verb). However, the specific meaning difference between a to-clause and an ing-clause depends on the particular controlling verb. (LGSWE 757)150

Thus although we can observe tendencies in the semantics of formal realisations such as [to_INF] or [V-ing], the verb that selects a certain complement

|| 148 Some scholars (Coates 1983, Kleinke 2002) differentiate between potential/possible and hypothetical events. In the following discussion, hypothetical is subsumed under potential. 149 See also, for instance, Quirk et al. (1972: 835f), Bolinger (1977: 73). However, Mair (1990: 36) shows that some of Bolinger’s (1977) asterisks are questionable, and our corpus evidence supports this view. Rohdenburg (1992) discusses the issue historically and contrasts the construction with the corresponding German options. 150 LGSWE’s statement is based on post-verbal complements in this context.

Case study: modality and non-extraposed [to_INF] subjects | 87

often plays a more important role. While Mair claims “that there is no modality inherently associated with either infinitival or gerundial subjects” (Mair 1990: 86), the data presented here suggest the opposite. Mair’s data basis – handcollected examples from the Survey of English Usage – is simply not big enough to single-handedly reject such claims.151 Still, as not even half of the uses in the sample occur with a modal verb, there is clearly no single exclusive meaning “potentiality” associated with [to_INF]s, as can be seen in (14): (14)

To urge Arabs to rise up and kill British troops in Iraq was despicable and turned him into one of the most hated men in British politics. (newspapers)

The quoted tendencies are, however, obvious enough in the present data. They are statistically significant (p < 0.01) and should thus be accounted for by linguistic theory. If we take a closer look at the data we can in fact observe that not all modal verbs are highly frequent; only a certain class of modal verbs is chiefly responsible for this result, namely modal verbs marking epistemic modality, and then again those which are more on the “possible” end than on the “certainty” end of the epistemic scale.152 Thus would moves from second to the first place in a frequency ranking (compared to the BNC), might from eighth to fourth (together with can). The rarest modal auxiliaries in the study are should and must, probably because they serve mainly as markers of deontic modality. In our examples, deontic modality is – not surprisingly – relatively rare and all seven uses of the modal auxiliary must are epistemic, as illustrated in the following sentence: (15)

To be so selfless after hearing she had cancer must have taken tremendous courage. (newspapers)

The modal verb can, which can express possibility as well as ability, also occurs less commonly in our sample of sentences with [to_INF] subjects than in the BNC and most instances in our collection would have to be classified as markers

|| 151 Mair (1990) only found 52 instances of non-extraposed [to_INF], the majority of which presumably occurred with the copula be. 152 Even if one does not classify would as a marker of possibility but of hypothesis (Coates 1983: 5), the findings presented here are still valid given that “WOULD is often little more than a marker of unreal condition” (Coates 1983: 211). This study, though, will follow Palmer (1990: 58) in classifying would as a marker of epistemic modality.

88 | Clausal Subjects in active clauses

of possibility rather than ability, at least not in a strict sense of the term ability.153 (16)

To tackle small problems can create big benefits. (britnews)

While most linguists would not classify such a use as epistemic (root possibility in Coates 1983: 5, dynamic possibility in Palmer 1990: 81ff), its non-factual status is compatible with the meaning of “potentiality” associated with [to_INF]s. We can thus confirm the hypothesis that there is a certain meaning potential in [to_INF]s that expresses potentiality and goes along with modal verbs compatible with this meaning component. It has to be made clear, however, that any strong claims (i.e. that all [to_INF]s necessarily express potentiality) have to be rejected, given the still more than 50 % non-modal uses in the sample. Such a strong claim is also rejected by Duffley: This study has at least shown very clearly what the distinction between the gerund and the infinitive is not. Neither the distinction between particular versus general nor that between reification versus hypothesis/potentiality, nor any of the derivative oppositions in terms of factivity versus nonfactivity, referring versus nonreferring or validated versus validatable, can account for the full range of meanings expressed by these two forms in their use as subject of the sentence. (Duffley 2003: 349)

However, Duffley, who looks at a smaller dataset but in more detail, discusses examples where the [to_INF] “is employed […] for its ability to prospectivize and thereby evoke its events as merely hypothetical” (2003: 339) and finds “a powerful tendency for the to-infinitive to be used in contexts evoking nonreal events (90 percent of its occurrences)” (2003: 349). The fact that hypothetical meaning is neither necessary nor sufficient for the use of a [to_INF] is also backed up by Duffley’s next observation: “However, the -ing form is also used to refer to nonreal events in one-third of its uses” (2003: 349). That the overall, stronger tendency of the [to_INF] to occur with a potentiality meaning stated above is not uniform in character is shown by Figures 8 and 9. The propensity of a verb that is used with a [to_INF] non-extraposed subject to occur with a modal verb appears to be item-specific and unpredictable. Again, we are dealing with tendencies and not rules. The fact that say does not occur with a modal in the sample whereas harm always does does not allow strong claims such as that say never occurs with modals. In fact, a search of the

|| 153 Palmer (1990: 85) suggests subject orientation as a term to overcome the restriction of the term ability to animate creatures.

Conclusion | 89

entire corpus does indeed yield one instance of a sentence with would say which does not figure in the 1082-sentence sample used for this chapter:154 (17)

To be competitive again would say a lot for chemotherapy and the cancer community. (newsmerge)

For some verbs, there seems to be an even more idiosyncratic distribution: The 43 sentences from the whole corpus which contain add with a non-extraposed [to_INF] subject behave much the same way as the overall sample: slightly less than half of the sentences occur with a modal verb, but none of the six instances of add insult to injury do.155 As the numbers are quite small, this tendency – interesting though it might be – is not statistically significant (p > 0.05).

4.3 Conclusion In sum, we have found a substantial number of restrictions on clausal subjects in the corpus data, which are summarized in the following table: Tab. 9: Summary of restrictions on non-extraposed clausal subjects of active verbs

[to_INF]

[that_CL]

[V-ing]

comment

agree

o



+

bar



+

+

[to_INF] and [that_CL] not attested

belong

o



+

[to_INF] only in one near-quotation from Shakespeare's Sonnet 23

bind



+

+

[to_INF] and [that_CL] not attested

breathe



+

+

[to_INF] and [that_CL] not attested

charge





+

|| 154 Due to the high frequency of say in the corpus only a subset of the query results were analysed originally. 155 One might be inclined to treat the (relatively) fixed expression add insult to injury as an idiom and thus separate it from the lexical unit add used in the other examples, but this approach would be misguided as there is no discernable meaning difference between add in add insult to injury and many other uses of add in the corpus. In a construction grammar approach (Goldberg 2006) one could nonetheless treat the combination as a construction in its own right if we assume that it is stored separately due to the frequency with which the items occur together.

90 | Clausal Subjects in active clauses

[to_INF]

[that_CL]

[V-ing]

comment

charm



+

+

no [to_INF] and no [that_CL] attested, [V-ing] only one example

check

+



+

[to_INF] only one example

compare

+



+

compel



+

+

concentrate

+



+

demand

+



+

depend

+



+

entertain



+

+

[to_INF] and [that_CL] not attested; few real [Ving]s but lots of cases with "entertaining" as predicative element

grieve

+

+



no [that_CL] attested non-extraposed, only extraposed

grow





+

borders on copula for some uses (e.g. ‘grow difficult’)

guide





+

hang



+

+

no [to_INF] and no [that_CL] attested

interest



+

+

no [to_INF] and no [that_CL] attested

introduce

+



+

link



+

+

matter



+

+

pay

+



+

persuade



+

+

recall





+

remain

+



+

replace





+

shape



+

+

no [to_INF] and no [that_CL] attested

take

+



+

[that_CL] only in tough movement

wash

+



+

but [for_NP_to_INF] attested

no [to_INF] and no [that_CL] attested

no [to_INF] and only one [that_CL] attested borders on copula; extraposed [that_CL] attested

Conclusion | 91

While the case for a restriction on non-extraposed [V-ing] subjects is weak, [to_INF] and [that_CL] show significant patterns of non-occurrence in, after all, 1.5 billion words of parsed text.156 We have also seen that other factors besides verb valency, such as modality, can play a role in the selection of the type of clausal subject, even though their predictive power is limited. In a small study of semantically similar verbs, some differences in the acceptability of the three major clausal subjects ([V-ing], [to_INF], [that_CL]) were found for persuade and convince, but these are gradual and thus not necessarily ‘bulletproof’ counter-examples against a view that claims that the syntactic form follows from the semantics alone. On the other hand, the range of semantic features associated with the different clausal complements is so wide and shows so much overlap that it is almost impossible to predict any restriction with them. (It is of course always possible to claim that a certain feature is responsible for a restriction in a post-hoc manner.) This is in line with Kleinke’s (2002) observations on the alternation between the two principal non-finite clausal complement in English, [to_INF] and [Ving]. She finds that there is substantial semantic overlap between the two forms and that often the difference is only in one single semantic feature, so that it is often impossible to actually observe a semantic difference (Kleinke 2002: 304). Accordingly, if in many cases there is no or hardly any semantic difference between realisations of clausal subjects, how complex would the semantic component of a grammatical model have to be in order to systematically predict the distribution of [that_CL], [to_INF] and [V-ing] shown above? It would thus be quite difficult to conceive of a model that is based on a finite set of semantic roles (such as the generative model) and that is capable of predicting the restrictions on clausal subjects shown in this chapter. Given that research on postverbal elements has shown convincingly that such restrictions exist,157 it would furthermore not be economical for a grammatical model that has to account for such restrictions anyway to have a second mechanism for subjects that works purely at the semantic level. Thus an item-specific treatment of subjects of non-copula verbs in linguistic theory is very likely called for.

|| 156 As discussed in Section 3.2.2, parsing errors may of course prevent us from finding all instances in the corpus, but not finding a pattern is very likely an indicator of its extreme rarity. 157 See, for instance, Faulhaber (2011).

5 Analysis of subjects in passive clauses If we want to determine whether there are arbitrary restrictions on the form of subjects, a study of passive clauses may be particularly revealing. As many theories derive passive clauses from active clauses or rely on a systematic structural relationship between the two, the role of the subject in the passive voice is directly related to the constituent often called object in the active voice (see for instance CGEL 726f, Aarts/Aarts 1982: 138ff, CamG 246f).158 Given that ample evidence is available that such postverbal constituents are highly verb-specific, it will be investigated in this chapter – whether the same formal restrictions hold for the subjects of passive clauses, – whether there are more restrictions, – whether subjects are less restricted, or – whether we simply find different restrictions than in the active clause. The findings of this chapter are also relevant to the question in what way passives should be modelled: Can they be seen as derived from the corresponding active clause or should they rather be treated as independent? [NP] complements are by far the most frequent and versatile subjects in passive clauses – just as in active clauses – and are thus least relevant for the identification of formal restrictions.159 The focus will thus once again be on the major clausal complements [V-ing], [to_INF] and [that_CL].

5.1 Short theoretical introduction As mentioned above, the traditional view in grammatical description is to regard the subject of the passive as related to or derived from the ‘object’ in the

|| 158 The present chapter is only about English. For a cross-linguistic perspective on passives, see Siewierska (1984). 159 However, not all verbs show postverbal [NP] complements in the same slot as the clausal complements which can become the subject of a corresponding passive clause. Thus, for instance, agree and aim also have passivizable complements in other valency slots. Some verbs such as remark discussed in Section 5.2.2.3 have no pure [NP] complement at all (only prepositional ones), but they allow [NP] subjects in the passive nonetheless (possibly with a stranded preposition). https://doi.org/10.1515/9783110589801-113

94 | Analysis of subjects in passive clauses

corresponding active declarative clause.160 Thus the direct object Mary in the active clause (1a) becomes the subject in the corresponding passive clause (1b): (1)

a. John killed Mary last night. b. Mary was killed last night.

However, not all objects occur as subjects in the passive clause161 and not everything that can occur as a subject in the passive clause is treated as object by all accounts.162 One of the most influential approaches to posit derivation of passives from actives was the generative transformational account put forward by Chomsky (1957: 77f), who made the notion of a ‘passive transformation’ popular163 (see also Stein 1979:11): The idea of representing the English active-passive relation in terms of transformations was not, however, revolutionary. Jespersen, for instance, had spoken of the ‘turning’ and Poutsma of the ‘conversion’ of the verb form from one voice to another; but it was only in the case of transformational theory that the use of transformation was extended, formalised and systematically incorporated into a unified grammatical framework. (Svartvik 1966: 1f)

This derivative view was followed up by many researchers and can also be found in approaches to valency theory. Thus Allerton states that “where two structures are related with real regularity, as in the cases of active-passive” (Allerton 1982: 31), one should recognise a transformational relation between the two structures since having to describe only one structure is more economic. It follows that in such a view of grammar a description of active clause objects is also at the same time a description of passive clause subjects. As mentioned in Section 2.2.2, most German valency grammarians – while not necessarily sub|| 160 That it is the active declarative that is seen as primary is certainly due to the fact that the declarative is the “most common sentence type” (Herbst/Schüller 2008: 149) and that active sentences are more frequent than passive sentences: “Proportionally, passives account for c. 25% of all finite verbs in academic prose [...,] c. 15% [...] in news [...,] c. 2% in conversation” (LGSWE 476). 161 So-called ‘middle verbs’ (CGEL 735f), ‘extent and measure phrases’ (CGEL 735) and ‘benefactive objects’ (Aarts/Aarts 1982: 140) are problematic cases. See also Herbst/Schüller (2008: 169–172) for a discussion of objecthood and passivizability. 162 CamG for instance does not call postverbal [that_CL] complements object, except for those in complex transitive complementation (CamG 1017–1022). Another problem is the so-called prepositional passive discussed in more detail in Section 5.2.2.3. 163 To be precise, the passive is derived from a deep structure that is shared with the active clause, but at least in early transformational work the deep structure had a striking similarity to the structure of the active declarative sentence.

Short theoretical introduction | 95

scribing to a view that derives one structure from the other – rely on the use in the active declarative to describe the complement inventory of the verb. Thus a passive sentence, for instance, can be related to the corresponding active sentence by means of a so-called ‘valency reduction’ (‘Valenzreduktion’, Welke 1988: 69f, Helbig/Heinrich 1978: 11f).164 Again this means that passive clause subjects are complements that occur postverbally in the active clause. However, the authors of VALBU (Schumacher et al. 2004) are aware of the problems that arise in such a derivative view and state that due to the quite different realizations of the semantic roles in the passive sentence compared to the active sentence one could attribute two different complement inventories to active and passive clauses and thus interpret valency as a property of verb forms (Schumacher et al. 2004: 55). For practical lexicographic reasons they discard the separation of active and passive patterns in the dictionary, though. Herbst/Faulhaber reject the primacy of the active clause, too: „Der aktive Aussagesatz wird nicht als primär oder grundlegend gegenüber anderen betrachtet“ [“The active declarative sentence is not regarded as primary to, or more basic than, others.”] (Herbst/Faulhaber 2011: 415). Active and passive are thus treated simply as structural alternatives. Nonetheless, a relationship between the two is acknowledged with regard to the complements that can be used in the active and passive clause: In passive clauses, the complement functioning as subject realizes a predicate complement unit in the corresponding active clause. While this is not to say that any of the two sentence types is primary, the range of subjects in passive clauses depends on the verb’s possible predicate complement units [...]. (Faulhaber 2011: 11)

This is in line with the approach taken by the VDE where, by default, active patterns are given and the ability of individual complements to occur as a subjects in passive clauses is indicated through a system of codes. Lexical functional grammar (LFG), early accounts of which treated passivization as a lexical rule and not an operation on trees (as opposed to the generative transformational account), makes use of a principle of ‘lexical mapping’ (see Bresnan 2001: Chapter 14) to account for the relationship. Thus LFG can in principle account for actives and passives independently from one another.

|| 164 Similarly Sadziński, who states that a reduction of valency slots is characteristic of the passive voice (Sadziński 2006: 971).

96 | Analysis of subjects in passive clauses

The same is true of a valency pattern approach, which does not rely on a complement inventory but specifies patterns of clause structure.165 Such an approach was suggested for English by Herbst (2007), is discussed in more detail by Herbst/Schüller (2008: 137–141) and has been realised in the Erlangen Valency Patternbank166 (Herbst/Uhrig 2009), where active and passive verb patterns are listed separately (including the subject), so that no direct relation between passive clause subjects and active clause postverbal complements is assumed.

5.2 Analysis As our aim in the present chapter is to investigate to what extent the subject in a passive sentence is lexically governed by the main verb, empirical evidence from a variety of sources will be discussed. The VDE is the primary source when it comes to the active clause and the ability of complements that are found postverbally in the active clause to occur as subject in the corresponding passive clause. Furthermore, the VDE was used as the point of departure for the corpus analysis, for which a multi-level process was used: In the first step a list was compiled containing all verbs in the VDE that have at least one clausal complement which can occur as a subject in a passive sentence.167 The list was then used as the basis for queries for clausal subjects of the [V-ing], [to_INF] and [that_CL] type in canonical subject position in the parsed corpora presented in Section 3.2.2. In a third step, evidence that was found to be in contrast to the description in the VDE was submitted to a native speaker informant (NS1) for screening in order to filter out performance errors. Examples discussed below are taken from that list and were verified by another native speaker informant || 165 A similar approach to German syntax is taken by Engel (2004: 104), who distinguishes between Satzmuster (verb plus complements) and Satzbauplan (a Satzmuster with an indication of the optionality of complements and additional semantic information). His patterns are always established on the basis of the active sentence, though (Engel 2004: 104). 166 At present, however, since the Patternbank is based on the VDE, the data is not backed by extensive separate corpus analysis of passives, though this would certainly be desirable. 167 The selection does not include the idiomatic phrasal verbs listed beneath many VDE verb entries as no systematic description of complements in relation to valency slots is given in this part of the dictionary entries. However, a cursory check hinted that a more thorough analysis of, for instance, bring up, bring about, rule out might be worth the effort. The more idiomatic type of what CGEL calls prepositional verbs (see discussion in Section 5.2.2.3) was not analysed in detail as their valency structure in the complement block is usually not complete (only postverbal complements listed under the heading “Further uses”).

Analysis | 97

(NS2).168 Further evidence is supplied by Faulhaber’s (2011) study of restrictions on semantically similar postverbal complements of semantically similar verbs.

5.2.1 Passive clause subjects as subset of active clause complements In the simplest case, the postverbal complements found in the valency slot that corresponds to the semantic role ÆFFECTED (i.e. those often called ‘object’) can occur as subject in a passive clause. Thus for complete, the VDE lists an [NP]169 and a [V-ing] complement in that position and uses subscript P to indicate that both of them can act as passive subjects.

Fig. 10: Complement block for complete (VDE 164)170

The examples are straightforward and no restrictions whatsoever seem to apply on the use in the passive: (2)

a. In addition, they completed the project within budget. b. In addition, the project was completed within budget. (GX7 32)

|| 168 The list was submitted in writing to NS1 and can be found in Appendix 15. NS2 was then interviewed in person on the relevant data. 169 For reasons of consistency, the latest version of the complement labels found in Herbst/Schüller (2008: 118ff) will be used so that [N] in the VDE becomes [NP], [to-INF] becomes [to_INF], etc. 170 The complement block of a VDE entry contains a list of complements, i.e. formal phrase or clause labels, grouped by valency slots (I, II, III in this case). So slot I in this example is the one for the active clause [NP] subject (indicated by subscript A), which is realized as [by_NP] in the passive clause if it is present. Slot II for this example is obligatory (hence “obl”) and contains what is traditionally described as the direct object, which can be realized either by an [NP] or a [V-ing] complement, both of which can become subjects in the passive clause (indicated by subscript P). Slot III offers additional complements that can occur in trivalent clauses (or divalent passive clauses) but cannot occur in subject position at all.

98 | Analysis of subjects in passive clauses

(3)

a.

He added that they had not completed securing the posts before Gary was fatally injured the following day. b. He added that securing the posts had not been completed before Gary was fatally injured the following day. (K2W 260)

Also, no other than the postverbal active clause complements can become passive subjects: (3)

c. *To secure the posts was never completed. d. *That they secured the posts was never completed.

Tables of the following type will be given after each verb discussed in order to summarise the findings: Tab. 10: Summary of the findings for complete171

complete slot

complement

PCUact

SCUpass

II

[NP]

+

+

II

[V-ing]

+

+

II

[to_INF]

-

-

II

[that_CL]

-

-

With admit, the situation is markedly different as only two out of the six complements listed in valency slot II can become the subject of a passive clause:

|| 171 PCU stands for Predicate Complement Unit (i.e. a complement that occurs within the predicate), SCU for Subject Complement Unit (i.e. a complement that occurs in subject position).

Analysis | 99

Fig. 11: Complement block for admit (A) (VDE 15)

Thus, apart from the [NP], only the [that_CL] is perfectly acceptable in the passive: (4)

That some hospitals fail to reach the required state of cleanliness is admitted and may be overly difficult to cure.

The [V-ing] that is acceptable in the active voice as in (5a) below cannot become the subject of a passive clause as predicted by the VDE and demonstrated in (5b), which is a all the more notable because [V-ing] subjects are extremely versatile, as we have seen in Chapter 4 and will see again for passive verbs in the next section. (5)

a. So far no group has admitted carrying out the murder. b. *Carrying out the murder has never been admitted by any group.

Again, complements that cannot occur in the active clause cannot become passive clause subjects either: (6)

*To fail to reach the required state of cleanliness is admitted.

Tab. 11: Summary of the findings for admit (A)

admit slot

complement

PCUact

SCUpass

II

[NP]

+

+

II

[V-ing]

+

-

II

[that_CL]

+

+

II

[wh_CL]

+

-

II

[Q/S]

+

-

100 | Analysis of subjects in passive clauses

admit slot

complement

PCUact

SCUpass

II

[to N/V-ing]

+

-

II

[to_INF]

-

-

The examples of complete and admit represent a large set of verbs for which the possible subjects of the passive clause are a subset (i.e. some or all) of the postverbal complements found in the active clause. Most verbs in the VDE (and possibly most verbs in the language) belong to this category. Of the verbs studied for the present chapter, the following conform to this prototype: announce, avoid, confirm, decide, defend, delay, discover, enjoy, forget, ignore, include, intend, link, mention, remember, show, stop, suspect, tolerate,172 value.173 Thus the evidence presented in this section supports the traditional view that passive clause subjects can be found postverbally in active clauses.

5.2.2 Passive clause subjects not found in the active clause 5.2.2.1 Complement occurs only as passive subject As mentioned above, most theories do not take into consideration the idea that complements may occur as subjects in passive clauses but not as postverbal complements in active clauses. However, the corpus search yielded substantial evidence for such phenomena. The verbs with extra [V-ing] complement form by far the largest group and will be illustrated here with the example of teach:

|| 172 The verb tolerate is not included in the VDE but was shown by Faulhaber (2011: 187) to not allow a postverbal [to_INF] even though semantically similar verbs such as bear and endure do allow it. The restriction seems to hold in the passive as only one counter-example was found which was rated marginal and dispreferred compared to [V-ing] by native speakers: ?He added: “The message has to go out that to win by cheating is ethically unacceptable and will not be tolerated.” (newspapers) It is likely that the only reason it was produced is the coordination between is ethically unacceptable and will not be tolerated. 173 Some of these verbs have complements which are not passivizable according to the VDE but were found as passive subjects in the corpora. A few were supposed to have obligatory extraposition, but were found non-extraposed in the corpora. However, as all complements that can occur as a passive clause subject occur in some active clause pattern, they were subsumed under this category.

Analysis | 101

(7)

a.

Conducting a disciplinary hearing or interviewing potential employees is best taught face-to-face because they rely on human contact. (britnews)

b. #She taught repairing cars.174 Tab. 12: Summary of the findings for teach

teach175 slot

complement

PCUact

SCUpass

II

[that_CL]

+

+

II

[to_INF]

+



II

[V-ing]



+

Other verbs in that group are attack, attempt, decide, ensure, estimate, explain,176 expect, help,177 influence, inform, learn, repeat. There were two good examples of verbs with extra [that_CL] complements as passive subjects, help and support, both of which are illustrated in (8) and (9). (8)

a.

That he was nearly always available as mentor was considerably helped by the fact that unlike many scientists of his intellectual seniority, he spent most of his time on the home ground of his department rather than traveling worldwide to this or that “important” symposium. (academic) b. ?The fact that he spent most of his time at the department helped that he was nearly always available as a mentor.

|| 174 The sentence is acceptable in another reading irrelevant for our purposes, i.e. when repairing cars is understood as an adjunct rather than a complement of teach, or, possibly, when it is treated as a course title. 175 For brevity’s sake, only [that_CL], [to_INF] and [V-ing] complements are listed in the tables since only these were verified with the help of corpora and native speaker informants unless there is a special reason for listing another complement. 176 The VDE entry for explain features an [N(’s) V-ing] complement for active sentences, though. 177 The verb help can of course occur with a postverbal [V-ing] in the active sentence, but only in constructions of the type “cannot help V-ing” (which is treated as subsense  in the VDE).

102 | Analysis of subjects in passive clauses

Tab. 13: Summary of the findings for help

help slot

complement

PCUact

SCUpass

II?

[that_CL]



+

II?178

[V-ing]

–179

+

III

[to_INF]

+

+

(9)

a.

That the secret ingredient to a longer life might be self-esteem is supported by other research confirming that heart disease and cancer are more likely in those suffering from depression. (newspapers) b. ?Other research supports that Miller’s calculations are correct.

Tab. 14: Summary of the findings for support

support slot

complement

PCUact

SCUpass

II

[V-ing]

+

+

II

[that_CL]



+

III

[to_INF]





The case of settle follows the same pattern and is the only case in the VDE where there is a complement that can only occur in the passive:180 (10)

It was settled that very evening that she would produce Streetcar. (VDE 749)

Data from the BNC confirms that the structure presented in (10) is used whereas no postverbal [that_CL] occurs in an active sentence.181 The VDE gives the pat-

|| 178 The [that_CL] and [V-ing] complements were grouped into valency slot number II on semantic grounds.That they seem not to combine with a postverbal [to_INF] may argue in favour of placing them in valency slot III, however. 179 See Note 177. 180 There are actually five more, but they are either more an artefact of the description than a real phenomenon (lose, test, see), of dubious acceptability (pray) or erroneous (match). 181 In the large parsed corpora described in Section 3.2.2 there are seven instances of active settle followed by [that-CL]. However, six of them were rejected by native speakers and one was

Analysis | 103

tern as “[it] be settled + that-CL” (VDE 749), which indicates obligatory extraposition.182 We thus have to add a distinction between extraposed and nonextraposed subjects to the table. Tab. 15: Summary of the findings for settle

settle slot

complement

PCUact

SCUpass canonical

extraposed

II

[wh_CL]

+

+

+

II

[that_CL]





+

II

[V-ing]







II

[to_INF]







Furthermore, read (in the subsense labelled γ in the VDE) is interesting in that it allows a [V-ing], a [to_INF] and a [that_CL] as a passive subject, neither of which occurs in the active:183 (11) a.

Bringing it back is read as an admission that today’s deficits are going to be with us for some time. (newsmerge) b. However, to criticise the Government’s approach to pension planning should not be read as a glorification of its predecessor’s efforts. (newsmerge) c. That the movie seems likely to get at least a sprinkling of Oscars next weekend could be read as not much more than one of Hollywood’s periodic outbursts of patriotism, and of the Academy’s weakness for golden-toned period spectacle. (newspapers)

|| rated at best marginally acceptable: ?“Less talk, if you don’t mind, and more service,” he said at length, just when we had settled that I would go to the ball, Allegra - who had, as usual, a rotten cold - would mind the till and Barbara would take a well-earned break in the Eternal City. (newspapers) 182 In the parsed corpora, there are two apparent counter-examples, but one could also be treated as adjectival and the other is from a non-native context. 183 There is a [wh-CL]P(it) in slot II, though.

104 | Analysis of subjects in passive clauses

Looking at the examples one may be inclined to wonder whether active uses of this particular meaning of read exist, but they do as illustrated by (11d), although the passive is more frequent in the corpus.184 (11) d. Like some Greek-speaking liberal Jews, the early Christians read the prophets as foretelling a universal mission of the Jews to illuminate all peoples. (ADC 34) Tab. 16: Summary of the findings for read (γ)

read (γ) slot

complement

PCUact

SCUpass

II

[NP]

+

+

II

[wh_CL]

+

+

II

[V-ing]



+

II

[that_CL]



+

II

[to_INF]



+

The verb blame deserves separate treatment because it is the only one in our test set185 that allows sentential complements as passive subjects but no sentential complement whatsoever in postverbal position in the active clause according to the VDE – a finding which was confirmed by a corpus search of the BNC and the newspaper corpora. (12) a.

Riding a bike is often blamed for producing thunder thighs. (newsmerge) b. That this has been so severe was blamed on three things. (newspapers)

Why it is that (12a–b) are acceptable remains mysterious and appears to be completely arbitrary.

|| 184 This sense of read is similar to the cases discussed in Section 5.2.2.2 below and might just as well have been discussed there. 185 As pointed out above, only verbs that allowed at least one passivizable sentential postverbal complement were tested systematically. However, every single verb starting with a and b was tested in order to check for a case such as blame.

Analysis | 105

Tab. 17: Summary of the findings for blame

blame slot

complement

PCUact

SCUpass

II

[NP]

+

+

II

[V-ing]



+

II

[that_CL]



+

II

[to_INF]





To sum up, this section has already presented conclusive evidence that complements which do not occur postverbally in the active clause can occur as passive clause subjects, seriously challenging the traditional view outlined in Section 5.1.

5.2.2.2 Possible influence of other valency carriers in the clause Sentences such as (13a) below have attracted the attention of researchers for a long time186 because the two postverbal elements are in a semantic relationship to each other that is similar to that of a copular clause such as (13b), i.e. there is some sort of predication. (13) a. Mr Advani’s enemies consider him dangerous. (VDE 176) b. He is dangerous. In the generative literature, the postverbal elements in (13a) are usually treated in terms of a small clause.187 Similar observations led the authors of the VDE to introduce a formally not specified complement of the type [it + pattern of X], where X can be an [NP], an [ADJ] or an [as_NP] or [as_ADJ] prepositional complement in another valency slot of the verb. The [it + pattern of ADJ] complement is underlined in (13c): (13) c.

Thus he would not consider it appropriate to regard any individual as having a fixed amount of intelligence. (VDE 176)

|| 186 Aarts (1992: 36) traces the observation back to Jespersen, who treats the postverbal elements as a dependent nexus (see Section 2.1.2 for details on his terminology). Chomsky ([1955] 1975) also briefly discusses such sentences. 187 See Aarts (1992) for a detailed account; Giparaitė (2006) offers a more recent generative perspective.

106 | Analysis of subjects in passive clauses

The idea behind such a treatment is that the extraposed [to_INF] in (13c) is somehow governed by the predicative element appropriate (similar to the situation in copular clauses we shall see in Section 6.2). That there is some sort of relationship of that kind can be easily shown if we insert incompatible complements as illustrated below. In (14a), there is a perfectly acceptable copular sentence in which the [that_CL] follows an [NP] headed by fact. However, a [for_NP_to_INF] complement is incompatible with fact, which renders (14b) unacceptable. (14) a. It is a fact that Brexit is a success. b. *It is a fact for Brexit to be a success. The fact that the sentences in (15) show a parallel behaviour to those in (14) even though the constituents in question are both used as postverbal complements indicates that the restriction imposed on the clausal complement by fact holds in such a construction, too. (15) a. They consider it a fact that Brexit is a success. b. *They consider it a fact for Brexit to be a success. Let us now return to the passive of that construction. The [it + pattern of X] complement can also occur there with the same restriction between the two elements: (16) a. It is considered a fact that Brexit is a success. b. *It is considered a fact for Brexit to be a success. Furthermore, a non-extraposed version – though comparatively rare – does occur, too. And, what is particularly interesting, the restriction between fact and the clausal complement still holds: (17) a.

That housing will decline is considered a fact like Isaac Newton’s laws.188 (housingpanic.blogspot.com) b. *For housing to decline is considered a fact like Isaac Newton’s laws.

Thus a model of passive clauses of that sort has to account for restrictions imposed on the subject by elements other than the verb. For sentences such as (17a) above, one could feel reminded of copular clauses such as (14a). However, the analogy only holds for [NP] and [ADJ] complements in the postverbal posi|| 188 The attachment of like Isaac Newton’s laws is ambiguous. It could be treated as a postmodifier to fact but also as an adjunct in the sentence.

Analysis | 107

tion and not for [as X]189 or [to_INF] complements. The latter is particularly interesting in that the restriction identified above still holds even if fact is the predicative noun within a [to_INF] postverbal complement:190 (17) c.

That housing will decline is considered to be a fact like Isaac Newton’s laws. d. *For housing to decline is considered to be a fact like Isaac Newton’s laws.

We can also show that it is still not the verb consider which restricts the form of the passive subject, as [for_NP_to_INF] complements as in (17d) above can occur if we change the predicative noun in the postverbal [to_INF] clause: (17) e.

For him to do that was considered to be a brilliant idea.

In the corpus data analysed, most of the verbs that can occur in such a pattern do not impose many restrictions on the passive clause subject themselves, so we find [V-ing], [to_INF] and [that_CL] alike with many of them. The case of regard is similar to consider in syntactic and semantic structure, but it occurs less commonly with a postverbal [to_INF] and needs an [as_NP] instead of an [NP] complement.191 Again, all three major clausal subjects occur in the passive: (18) a.

To be a single mother in Pakistan is regarded as a disgrace. (newsmerge) b. That the present Queen came so close to topping the poll could be regarded as a substantial achievement. (newsmerge) c. Blowing up the Provos' store of the plastic explosive, Semtex, is regarded as a possibility. (newsmerge)

The case of judge is particularly interesting if we want to compare active clause postverbal complements and passive clause subject. Faulhaber (2011: 184f)

|| 189 The symbol X here stands for a choice of potential elements that can occur as a complement of the particle as, e.g. NP, ADJ or V-ing without claiming that all of them are possible for all verbs. 190 Aarts (1992: 68ff) discusses the relationship between sentences such as (17a) and (17c) within a generative framework. There seem to be competing views as to the question whether the latter is derived from the former by inserting to be or whether there is a process of bedeletion that derives the former from the latter. 191 See also Herbst’s (2009) analysis.

108 | Analysis of subjects in passive clauses

shows that in the active clause judge does not allow a [V-ing] complement in the ÆFFECTED valency slot whereas a [to_INF] is possible.192 Compare: (19) a.

Fertility conditions play a part then, and as hard pruning provokes the greater growth response, we should judge to ease back a little on poor, dry soils compared with more fertile conditions. (CMM 799 cited by Faulhaber 2011: 185) b. *?You should judge doing this. (Faulhaber 2011: 185)

However, in the following sentence, we find a [V-ing] passive clause subject with judge: (19) c.

Breaching Commons security and flour-bombing the Prime Minister will be judged even by many supporters of Fathers 4 Justice to be a protest too far. (newspapers)

At first sight, this may look as if simply fewer restrictions apply in the passive than in the active (as with the verbs discussed in the previous section), but the existence of the [V-ing] in the active as in (19d) below points in another direction: (19) d. In the computer assisted methods (both CAPI as CSAQ) respondents were more positive about data-privacy, and judged answering sensitive questions as less unpleasant. (academic) It appears as if judge does not block the choice of the [V-ing] in the trivalent active pattern and accordingly neither in the divalent (optionally trivalent) passive shown in (19c) above. This is in fact strong evidence for a small clause analysis: If both postverbal elements form a small clause constituent and if the verb only opens up one slot for such a constituent, it may not exert influence on the “subject” of the small clause. While for the active clause, such an analysis is immediately appealing even outside the generative framework, within which it is most prominently discussed, the situation is not as straightforward in the passive clause. For linguists using a surface-oriented approach, the underlined constituent in the following sentence is the subject of the verb consider. (20)

That man was considered a fool. (Aarts 1992: 160)

|| 192 Example (19a) illustrates a structure Faulhaber found but which is not listed in the VDE for this lexical unit.

Analysis | 109

In such a view, an analysis that regards “That man ... a fool” as a small clause is unintuitive and would have to make use of a discontinuous verbless clause constituent. In a generative approach, the phenomenon is easier to account for as Aarts illustrates with the help of the D-Structure (reproduced as 21a) and the S-Structure (reproduced as 21b) of (20): (21) a. [NP e] was considered [SC that man a fool] (Aarts 1992: 160) b. [NP that man]i was considered [SC ti a fool] (Aarts 1992: 160) These structures can be read as follows: The NP that man is the subject of the small clause at D-Structure and is then raised to the subject position of the matrix clause (which is empty at D-Structure) so that only a trace co-indexed with the raised subject remains in the small clause at S-Structure. While this is certainly elegant, such an analysis is not sensible in usage-based approaches, which – for good reasons – doubt the existence of empty/null elements193 and reject the idea that elements in a sentence may be in a different place in an underlying structure. We shall try to reconcile surface-based valency grammar with the evidence presented here in Section 9.3.2. Apart from consider we find assume, declare, describe, feel, find, judge, prove, recognize, regard and take in such passive structures.194 There is a related structure with verbs taking a postverbal [to_INF] complement where the ones just discussed often allow nominal, adjectival or prepositional complements. It appears that the verbs associated with this structure, just like the ones discussed before, exhibit considerably fewer restrictions on the passive clause subject than expected. Let us begin with the discussion of guarantee. According to the VDE guarantee allows for a [to_INF] and a [that_CL] complement in the active clause. Only the [that_CL] can become subject of a passive clause. However, we find all three major clausal complements as passive clause subjects of guarantee in the corpora: || 193 In Goldbergian construction grammar, for instance, “[n]o underlying levels of syntax, nor any phonologically empty elements are posited” Goldberg (2006: 10). Sinclair/Mauranen argue in a similar vein that “now that many linguists respect the actual wordings of corpora, and are committed to describing the text and not some rewritten version of it, notions that there are some words missing or that the text cannot be understood as it stands are no longer tenable” (2006: 150; the statement is about so-called ‘ellipsis’). 194 Make as in “Linking up with peers abroad is made dramatically easier using the Internet.” (newsmerge) is a superficially similar case, but the semantic structure is different. To be able to distinguish such structures, Herbst/Schüller (2008: 139) propose valency constructions, which are pairings of valency patterns and participant structures.

110 | Analysis of subjects in passive clauses

(22) a.

Knocking two small reception rooms into one is virtually guaranteed to create value, as is enlarging a family kitchen so they can eat in it. (newspapers) b. But even to raise the question of what comes after Downing Street is guaranteed to halt conversation in mid-track with any of Blair’s advisers. (newspapers) c. That the believers’ understanding of Christianity is truly what God wills it to be is guaranteed by the Holy Spirit, as, again, Jesus promised explicitly. (newspapers)

While (22c) is the straightforward passive we would have expected, the structure of (22a–b) is very similar to sentences with a semi-auxiliary (CGEL 143). In fact, we could easily replace be guaranteed to with the semi-auxiliary be bound to in both of them without a tangible change in the structure of the sentence. This would, however, suggest that the [V-ing] and the [to_INF] are not selected by guarantee but by the verb following it in the infinitive. Since “the nonfinite construction following these verbs can often be replaced by a that-clause with an indicative verb” (CGEL 1204) as illustrated in (23a–b) for report, the subject and the predicate in the [that_CL] have to be somehow compatible. Since the postverbal construction in (23b) is a predication that might be regarded as a small clause, such an interpretation is indeed plausible. (23) a. The police reported that the traffic was heavy. (CGEL 1204) b. The police reported the traffic to be heavy. (CGEL 1204) c. The traffic was reported to be heavy. (CGEL 1204) Other verbs of that type include say, suppose, think and understand. CGEL includes only be supposed to in the group of semi-auxiliaries but also states that “[t]he boundaries of this category are not clear” (CGEL 143). According to CGEL (1203) no corresponding active version of the type shown in (23a) above can be constructed for say and think.195 While for say196 this is certainly correct, a query of the BNC yields more than 20 counter-examples where think is used in the active voice in such a structure, such as (24a). (24) a.

He knew that she thought him to be a stiff and unimaginative person. (GW2 3246)

|| 195 Similarly, the pattern representations in the VDE depart from the usual (active) format. They read “be said + to-INF” (VDE 720) and “be thought + to-INF” (VDE 869). 196 The fact that say can occur in this particular passive pattern is an idiosyncratic property in itself and has to be represented lexically. (Compare *He is written/told to have left early.)

Analysis | 111

In fact, we can even find an active counterpart for the structure with passive subject [to_INF] for think. (The structure was presented for guarantee in (22b) above.) (24) b. They would support the union if they thought it to be in their interest to do so. (AC2 2297) c. To do so was thought to be in their interest. Thus even though most of the verbs discussed in the second part of this section seem to prefer the passive, we can observe a similarity to the verbs of the consider type covered at the beginning of the chapter in that a small clause analysis is possible and thus once again the postverbal element seems to exert influence on the passive clause subject. However, even though we found considerably fewer restrictions in the passive clause subject than we would have expected, not all verbs discussed in this section seem to occur with all types of subject. Thus for example advertise was only found with a [V-ing], for report there was no [to_INF], and understand only occurred in the pattern with a postverbal complement as discussed here with a [V-ing] subject but allowed the other forms when used without a postverbal complement. It has to be stressed, though, that the corpus evidence is by far not sufficient to base serious arguments on these three cases. Thus the picture that emerges is rather patchy and suggests that, for the verbs studied here, secondary postverbal “valency carriers” are at least as responsible for restrictions on the passive clause subjects as are the verbs themselves. It thus makes only limited sense to produce summarizing tables of the kind shown in the previous sections, as tables would need three dimensions for a given verb (complement, active/passive, and secondary valency carrier) and a lot more corpus evidence would be needed to fill such tables reliably. Given that there were a few verbs for which not all complements were found as passive clause subjects, a more detailed study of such phenomena would be desirable once even larger corpora become available. A tentative proposal for the modelling of such structures is presented in Section 9.3.2.

5.2.2.3 Prepositional passive Furthermore, there is a set of verbs which allow a [V-ing] in the active only with a particle as in (25a), but allow it in the passive without the particle as in (25b), sometimes with the particle ‘stranded’ after the verb.

112 | Analysis of subjects in passive clauses

(25) a.

These two groups agree on rejecting many economic policy prescriptions of the ‘wets’ but disagree on the role of the state. (A6F 1021) b. Officially, lifting the embargo has been agreed in principle, but “now is not the time”. (newspapers) c. Officially, lifting the embargo has been agreed on in principle, but “now is not the time”.

For (25b–c) it would be difficult to argue for a semantic difference. Examples of further verbs showing the same behaviour are aim, allow, arrange, look and remark. The case of look is interesting because there is one sentence that suggests that it can take a [to_INF] as subject in the so-called prepositional passive (CGEL 1164) construction: (26)

I don’t think we are the new race of philistines that some fear, but we do live in a time when to be serious about a non-profit-making venture is looked on as madness – and our consumer society would have us believe that the arts and spirituality have less to communicate than a mobile phone. (newspapers)

The VDE predicts in fact a similar structure but only in an extraposed variant and only for an [as_ADJ] postverbal complement.197 Nonetheless, the pattern for look is comparable to the case of consider discussed above apart from the fact that there is a particle after the verb, so we may hypothesize that madness in (26) could possibly also determine the [to_INF] subject.198 The fact that the particle is found after the verb in the passive and not in front of the subject is one reason for CGEL to treat such verbs and their particles as a single unit called prepositional verb (CGEL 1155ff).199 Herbst/Schüller (2008: 120f) show convincingly how the rejection of prepositional verbs makes the

|| 197 As to the particle, the VDE is inconsistent: In the complement block, passivization is only allowed with on, in the pattern block it is only allowed with upon. 198 Despite a superficial similarity, the particle after the verb is not a particle of the type that is used to form phrasal verbs (or so-called phrasal-prepositional verbs). See CGEL (1150–1167) for a detailed discussion and criteria for the distinction of phrasal verbs from the structure discussed here (“prepositional verbs” in their terminology). 199 While this solution does offer certain elegant analyses, as discussed below, it should be borne in mind that all those verbs which allow an [as X] complement discussed in the previous section would have to be treated as prepositional verbs (type II) in CGEL’s terminology, which is highly problematic because a single unit is less intuitive if it is discontinuous and some of the most important arguments for prepositional verbs (e.g. preposition stranding in the passive) are not relevant to such cases.

Analysis | 113

analysis of active sentences much simpler, and Herbst opposes such an analysis along the same lines for the VDE when he states that “[f]rom a valency standpoint the identification of prepositional verbs is neither theoretically convincing nor is it economical for lexicographic purposes” (VDE xxvi) although he acknowledges that passives are an argument in favour of a prepositional verb analysis. Neither description offers a discussion of how the postverbal particle in (26) above can be accounted for theoretically in terms of valency and participant structure, but Herbst states in the same context that “prepositional phrases can be seen as complex complements consisting of more than one constituent” (VDE xxvi). Then again it would be difficult to argue that the [to_INF] and on in (26) form a discontinuous prepositional phrase since no non-discontinuous version exists with the preposition realized. The case of [that_CL] passive subjects is particularly noteworthy. Let us briefly compare possible active and passive clause patterns as discussed by CGEL (1178). The standard pattern for prepositional verbs in the active is as follows: (27) a.

They agreed on the meeting. (CGEL 1178)

However, “when a prepositional verb is followed by a that-clause or a toinfinitive clause, the preposition disappears, and the prepositional object merges with the direct object of the monotransitive pattern” (CGEL 1178). This explains why (27b) is not possible and has to be expressed as (27c): (27) b. *They agreed on that they would meet. (CGEL 1178) c. They agreed that they would meet. (CGEL 1178) Yet such an analysis is highly unintuitive for the active pattern, as it would claim that the verb in (27c) is a prepositional verb the preposition of which is obligatorily absent (as confirmed by (27b)). However, it is more appealing for the passive: “Yet the preposition omitted before a that-clause can reappear in the corresponding passive [...], even in extraposition” (CGEL 1178). As predicted by their analysis, all four of the following sentences are possible: (27) d. e. f. g.

That they should meet was agreed on. (CGEL 1178) That they should meet was agreed. (CGEL 1178) It was agreed on eventually that they should meet. (CGEL 1178) It was agreed eventually that they should meet. (CGEL 1178)

This behaviour speaks strongly in favour of a prepositional verb analysis. On the other hand, it is difficult to argue for a prepositional verb without preposition in (27c) since it is impossible to know which preposition is omitted and

114 | Analysis of subjects in passive clauses

accordingly which prepositional verb we are dealing with – for agree, both on and upon are acceptable.200 Thus a model where two synonymous prepositional verbs agree on and agree upon exist leads to absurd problems for instance in determining which of the two is used in (27c). The situation is even more complicated in the case of remark. The entry in the VDE places the [that_CL] in another valency slot than the [on_NP] and [upon_NP] complements:

Fig. 12: Complement block for the general sense of remark (VDE 673)

If we follow the CGEL analysis outlined above, we would expect the particle in (28a) to be optional: (28) a.

That the moral imperative was not a sufficient condition has already been remarked upon. (CS7 1331) b. That the moral imperative was not a sufficient condition has already been remarked.201

There is a meaning difference between the two, though. In the latter, the [that_CL] is the complement listed in the VDE and thus corresponds to the following active sentence: (28) c.

They have already remarked that the moral imperative was not a sufficient condition.

|| 200 See also Herbst/Faulhaber (2011: 415): “Es werden keine (oder möglichst wenige) Annahmen über zugrunde liegende syntaktische Strukturen oder ‘fehlende’ bzw. ‘getilgte’ Elemente gemacht.” [“No assumptions (or as few as possible) will be made about underlying syntactic structures or ‘missing’ or ‘deleted’ elements.”] 201 Due to the heavy initial clausal subject, the sentence verges on the border of nonacceptability, but this could be remedied by adding a heavy by-agent phrase, which was omitted to form a structure analogous to (28a).

Analysis | 115

Thus in (28b) the [that_CL] expresses the content of the remark (which corresponds to slot II in the VDE) whereas in the original sentence (28a) the [that_CL] expresses the topic of the remark (which corresponds to slot III in the VDE). Consequently, the corresponding active clause to (28a) is not (28c) but (28d) below: (28) d. They have already remarked upon the fact that the moral imperative was not a sufficient condition. Although this sentence is made up, the structure is sufficiently common in the corpora to base an argument on it. We can observe that the fact is used to allow a combination of the particle upon and the [that_CL]. The passive suggests that the fact is not necessary from a semantic point of view, so if CGEL’s rule quoted above were correct, we would expect upon to disappear in the active and not the fact to appear. Neither would we expect a change in meaning when the particle is omitted in the passive. Thus the explanation offered by CGEL is elegant in the case of agree but not tenable for remark. We have to conclude that a satisfactory analysis of the structures discussed in this section is hard to come by since a complement inventory based valency analysis cannot explain why a preposition that was missing in the active sentence appears in the passive (as with argue), and CGEL’s analysis, which can account for such phenomena, needs to posit underlying structures, which are highly implausible from a usage-based point of view. Furthermore, CGEL’s analysis wrongly predicts the syntactic behaviour of cases of the remark type when there is a semantic distinction lost through conflation. However, a valency pattern-based approach that separates active and passive voice could be used to model this behaviour, as will be suggested in Section 5.3 below.

5.2.3 [to_INF] vs. [V-ing] The availability of [to_INF] and [V-ing] subjects in passive clauses merits a separate discussion for the simple reason that we do find arbitrary restrictions as to their use postverbally in the active clause (see Faulhaber 2011: 170–189)202 and it is highly relevant for the question of the valency status of the subject to see whether such restrictions can be found in passive subjects, too. || 202 However, see also Egan (2008) who says he was surprised “how neat this distribution actually is, how form and function seem to go hand in hand, in an area sometimes said to be characterised by no little degree of chaos and arbitrariness” (Egan 2008: 308).

116 | Analysis of subjects in passive clauses

The first interesting case is allow (in the sense of ‘permit’), which does not permit postverbal [V-ing] complements in the ÆFFECTED valency slot of the active clause:203 (29) a.

The committee is empowered to allow such a request for humanitarian reasons. (VDE 29) b. *The committee is empowered to allow implementing the necessary measures.

And while allow204 does have a prepositional [for N/V-ing] complement, it is not in the ÆFFECTED valency slot, as is illustrated in (29c), and cannot occur as passive subject. (29) c.

Time [ÆFFECTED] should therefore be allowed for visiting some of the major scenic attractions as well as the oldest working whisky distillery in the world, Bushmills. (VDE 30)

However, we do find convincing evidence for [V-ing] in the passive subject position, as illustrated by (29d–e). (29) d. Other instances of racial prejudice can be found in the fact that the church used by the black community is used by the white men as a gambling den in the week although using a white church for such purposes would not be allowed, if even thought of. (HPG 338) e. It says an assault on a person over 18 would be illegal whereas hitting a child is allowed under English law. (newspapers)

|| 203 Faulhaber presents the following example as evidence against this restriction: The weighting screen presents twenty options including on-line help, and saving of changes, while others allow sorting of data, changing of titles, reversing of the XY ranges and so on. (A19 1063; also cited in Faulhaber 2011: 180) The fact that sort, change and reverse do not have a valency slot for of when used as verbs renders a nominal analysis more likely in this case, though. In the BNC, only one example of a clearly verbal [V-ing] is found after allow: This allows one complete message command of four characters to be sent each frame and at the minimum frame rate allows up-dating the channel pulse width settings some forty times every second. (C92 1796) A cursory search in the parsed corpora did not yield any further hits. Thus although we cannot rule out the acceptability of the [V-ing] in postverbal position, its extremely low frequency and its non-indication in the VDE suggest that such a structure would still be marginal. 204 As the present section is concerned with allow in its ‘permit’ sense, the other meanings described in the VDE will be ignored.

Analysis | 117

It appears as if the passive subject [V-ing] in (29e) is related to the postverbal [to_INF] complement in the active clause (29f). (29) f. The law allows people to hit their child. g. *The law allows people hitting their child. Although the [to_INF] complement does sometimes occur as a passive subject, it is dispreferred. There are 4 instances of [to_INF] passive clause subjects in the parsed corpora, and while native speakers did not agree as to their acceptability, they all preferred a [V-ing] over the [to_INF] in the corpus examples such as the following: (29) h. ?Sir Philip said there was no problem with Mr Cameron sometimes meeting donors in his office, but to advertise for funds in return for meetings was not allowed. (newspapers) i. Sir Philip said there was no problem with Mr Cameron sometimes meeting donors in his office, but advertising for funds in return for meetings was not allowed. The following table summarizes the findings for allow: Tab. 18: Summary of the findings for [to_INF] and [V-ing] with allow (A ‘permit’)

allow (‘permit’) slot

complement

PCUact

SCUpass

II

[V-ing]

-?

+

II

[to_INF]

+

-?

We can see that the choice between [to_INF] and [V-ing] is highly voicedependent. It is particularly interesting to compare these results to the nearsynonymous verb permit. Faulhaber (2011: 180) shows that in the active, the [Ving] in (30a) cannot be replaced with a [to_INF] as in (30b).205

|| 205 It has to be noted that permit does allow a [to_INF] complement in postverbal position, but only if it is preceded by an [NP] complement, whereas allow can have a single [to_INF] according to Faulhaber (2011: 181).

118 | Analysis of subjects in passive clauses

(30) a.

Congressional negotiators have imposed limits which may permit building only two more of the bombers. (VDE 585) b. *Congressional negotiators have imposed limits which may permit to build only two more of the bombers.

In the passive voice, however, the [to_INF] readily occurs as extraposed subject in sentences such as (30c) as predicted by the VDE. The VDE also claims that extraposition is obligatory, though, which is doubtful in the light of (30d). (30) c.

It is not permitted to sound a car horn after a certain hour.206 (HWA 3119) d. But to deny customary and reasonable care or to deliberately starve or dehydrate someone because he or she is very old or very ill should not be permitted. (academic)

Example (31) below shows that the [V-ing] can also occur as passive subject, so the active clause postverbal [V-ing] seems to correspond to both of the passive clause subjects [V-ing] and [to_INF]. (31)

Surely photocopying such details should not be permitted. (newspapers)

Tab. 19: Summary of the findings for [to_INF] and [V-ing] with permit

permit slot

complement

PCUact

SCUpass

II

[V-ing]

+

+

II

[to_INF]

207

-

+

Again, the choice between [V-ing] and [to_INF] is different in the active and the passive. Also, we find a difference in the passive between allow and permit which is a strong indicator that the form of passive subjects is determined by the valency of the verb and not by some general principle of semantic compatibility. This finding is backed up by the behaviour of blame, help, support and teach discussed in Section 5.2.2.1.

|| 206 Native speaker informants commented that this sentence is acceptable but restricted to a specific (rather formal) style level. 207 See however Note 205.

Analysis | 119

CamG (1435) presents further arguments for an item-specific (or verb-group specific) view of [V-ing] and [to_INF] as passive clause subjects. They claim that “[p]assives with gerund-participial subjects are uncommon” which – in the light of the evidence discussed in the present chapter – seems to be a slight overstatement. But they present an interesting reason for a restriction on passive [Ving] subjects, which are allowed with the verbs in (32a) but not with the ones in (32b): (32) a.

Taking out a mortgage wasn’t considered/recommended/suggested. (CamG 1435; my emphasis) b. *Painting the house was begun/kept/hated/intended/remembered by Sam. (CamG 1435; my emphasis)

Note that there is a difference in the interpretation of the corresponding actives. In Sam remembered painting the house the understood subject of paint is recovered from the matrix clause: it was Sam who painted the house. In Sam recommended taking out a mortgage, however, the subject of take is not specified syntactically but has to be contextually recovered. It is this type of gerund-participial construction that most readily allows passivization. (CamG 1435)

However, such a generalization seems to hold only partially. There is evidence for [V-ing] complements used as passive clause subjects with verbs of the type where the subject is recovered from the matrix clause, e.g. avoid, complete, enjoy, plan, practice/practise. The following is one of the many examples found in the parsed corpora: (33)

Wearing an open-necked shirt with a suit is best avoided by anyone over 30, unless you are Tim Jefferies, Bryan Ferry or George Clooney on whom a dark suit and a plain shirt look great. (newspapers)

For [to_INF] complements, CamG claims that (except for the cases discussed in Section 5.2.2.2) “infinitivals are restricted to a few catenative verbs (e.g. decide, desire, hope, prefer), and then only in extraposed position” (CamG 1435).208 This statement is in principle in line with our findings apart from the [to_INF] subject we found for permit. The VDE, however, allows [to_INF] passive clause subjects for a lot more verbs.

|| 208 Langacker (2008: 430) also observes a “resistance of infinitival complements to passivization.”

120 | Analysis of subjects in passive clauses

To sum up, the choice between [V-ing] and [to_INF] subjects in the passive is highly verb-specific and is only partially related to the choice of postverbal complements in the active clause.

5.3 Conclusion Given that the basis for the discussion of passive subjects is usually the active declarative clause, the fact that some postverbal elements in the active clause cannot occur as subjects in the corresponding passive clause has been discussed in great detail in the linguistic literature (see for instance CGEL 159–171). However, due to this active-centric perspective, the fact that some passive clause subjects cannot occur as postverbal complements in the corresponding active clause has received much less attention in grammatical description and theory. The present chapter has made available ample evidence to confidently state that such phenomena do exist in English, though. In fact, the number of verbs which exhibit such behaviour is sufficient to suggest that some systematic treatment in the grammar is called for to account for it (and that it is thus not enough to dismiss the problem as ‘exceptions’). So, to come back to the questions asked at the beginning of this chapter, it is not a matter of whether there are more, fewer, the same or simply different restrictions. We find all four phenomena, depending on the verb chosen. The logical conclusion for any model of grammar must be to give up the derivation of the passive from the corresponding active clause, although it is of course sensible to posit a regular relationship between active and passive clauses since it holds for a majority of passive clauses. A valency model that describes a complement inventory (as is done in the complement block of the VDE) could be made to account for such phenomena with minor modifications, i.e. not only possible subjecthood in active and passive clauses would have to be specified for the complements but also whether they can occur postverbally in active and passive clauses. However, a model that is based on valency patterns (as mentioned above) has considerable advantages over the valency complement view. Not only does it allow a treatment of active and passive as structural alternatives without claiming the primacy of one over the other, it also can cover the prepositional passive issue discussed in Section 5.2.2.3, for which a complement-based valen-

Conclusion | 121

cy approach has to fail.209 However, a purely formal pattern-based approach cannot account for the case of remark discussed above in Section 5.2.2.3, as we find two [that_CL] complements in different semantic roles. Thus a pairing of form and participant roles (whether drawn from a general or a verb-specific set is irrelevant here) in the sense of the valency constructions introduced by Herbst/Schüller (2008: 139) is called for to model such syntactic behaviour. In addition, the sentences of the guarantee and consider type discussed in Section 5.2.2.2 suggest that item-specific restrictions are not only based on the verb but possibly also on a predicative element in such structures. See Section 9.3.2 for the question of how to model such relationships in a valency approach. We can thus conclude that the range of permissible passive clause subjects is highly item-specific and has to be specified as a lexical property of the verb, preferably in terms of patterns that are independent from active clause patterns and include information on participant roles.

|| 209 The VDE provides plenty of evidence for such a view as there are numerous patterns which are listed as only occurring in the passive. Examples include be drawn + against N, be drawn + to-INF, be employed V-ing, be fastened + by N to N, be fed + from N, be formed + of N, be found + ADV, be heard + to-INF, be quoted + at N, be seen + to-INF, be set + to-INF, be set + with N, be spoken + for, be struck + dumb (+ by/with N), be sworn + into N.

6 Analysis of subjects in copular clauses The treatment of copula verbs in a separate chapter is necessary because they do not show the same valency restrictions as non-copula verbs. Be – the principal copula in the English language – allows subjects (and also postverbal complements, see CGEL 1171ff) of any form. The sentences in (1) illustrate different formal realisations of the subject: (1)

a.

She is a busybody who likes talking and has cultivated her own style. (A06 912) b. To substantially increase sale and to produce reasonable profits is a considerable achievement. (K4S 1434) c. Singing along isn’t a big success. (HH0 3997) d. That the number of refugees is increasing so dramatically is a major cause for concern. (K41 828)

If such sentences are described simply as divalent uses of the verb be, which requires a subject and a predicative element (noun phrase or adjective phrase in these examples), then be cannot account for restrictions of the kind found in (2b–d): (2)

a.

Pakistan’s support for US operations is considered crucial because it is adjacent to Afghanistan. (newspapers) b. *That Pakistan supports US operations was adjacent to Afghanistan. c. *To support US operations would be adjacent to Afghanistan. d. *Supporting US operations is adjacent to Afghanistan.

Given that be allows all relevant subject realisations, as seen in (1b–d), the fact that the subjects in (2b–d) are incompatible with the predicate cannot be a property of the verb. It must be the adjective adjacent that does not allow a clausal subject, very much in the same way as it does not allow the nominal subject in (2e). (2)

e.

*Pakistan’s support for US operations is adjacent to Afghanistan.

The reason why all these subjects are unacceptable is a semantic incompatibility between the subject and adjacent, which has traditionally been treated under the heading of selectional restrictions in generative transformational gram-

https://doi.org/10.1515/9783110589801-143

124 | Analysis of subjects in copular clauses

mar (Chomsky 1965: 94ff, 114ff)210 or in terms of semantic valency (Helbig 1992: 7–9). What this chapter sets out to verify is whether syntactic restrictions on the subject depending on the predicative element also apply.

6.1 What counts as a copular clause? In the introduction to this chapter, the expression copula(r) was used without further definition, as if the term were uncontroversial. As far as be is concerned, there is no problem with that; everyone agrees that be is “the principal copula” (CGEL: 1171).211 But while Quirk et al. include a whole range of verbs in that category (e.g. appear, look, seem, remain, become, get, etc.), Huddleston/Pullum insist on restricting the term to the verb be. They propose “the term ‘complex intransitive’ for the more general construction, partly to bring out the parallel between [complex intransitive and complex transitive constructions], partly because complex-intransitive verbs other than be are not mere syntactic copulas but do express semantic predication” (CamG 218). Due to their largely similar syntactic behaviour, this chapter will follow the practice of CGEL in calling such verbs and clauses copula(r) in a broad sense where possible and make distinctions on a lexical basis where necessary.

6.2 Theoretical treatment of copula verbs and copular clauses In a valency model, if the adjective determines the form of the subject212 it usually follows that it must be a valency carrier.213 One could of course argue that the subject requires the adjective, but requiring actual lexical words instead of more abstract classes ([NP], [that_CL], [to_INF]) is less appealing for a model of syntax. In Eisenberg’s (2006: 34) terms, predicative elements in German copular clauses govern nominal subjects categorically, i.e. if there is a member of the

|| 210 See also Matthews (1993: 162ff) for discussion. 211 We shall use the term copular clause to refer to clauses in which a copula verb functions as the main verb. 212 Generative grammar models such behaviour with the help of a mechanism called ccommand (see Eroms 2000: 63f for discussion). 213 Dik (1983: 128f) proposes a similar structure. He suggests a process of copula support, which introduces a copula verb and thus “accounts for the occurrence of be in cases in which a non-verbal predicate such as clever is used as the main predicate of some predication” (Dik 1983: 129).

Theoretical treatment of copula verbs and copular clauses | 125

category predicative element, a nominal subject is automatically permitted. For clausal subjects, he assumes lexical governing of the conjunction dass (“that”), i.e. if an adjective governs the conjunction, this is a lexical property of the adjective. The property of predicative elements to select subjects is by no means a recent discovery,214 and while it has been taken as a fact for over thirty years in some lines of research in German linguistics,215 there are many open questions as to the exact nature of the relationship between the different constituents/ dependents and head(s). One option would be to continue seeing the verb as primary valency carrier and the adjective as a secondary valency carrier. This approach is advocated by Helbig (1992: 111) at the level of syntactic valency. Thus in (3)

My teacher is happy with my progress.

Helbig would see My teacher and happy as immediate dependents of is, and with my progress as a dependent of happy: is

My teacher

happy

with my progress Fig. 13: Syntactic structure of (3) in Helbig’s model (adapted from Helbig 1992: 111)

At the semantic level, however, he sees the predicative element as the valency carrier:216

|| 214 Mair (1990) uses the terms adjectival and nominal predicate. 215 See Herbst (1983: 19f) for references. Zifonun/Hoffmann/Strecker also follow this route in their treatment of clausal subjects in German copular clauses: “Hier selegiert der nicht-verbale Teil, das Adjektiv oder das Substantiv, das Subjekt” (Zifonun/Hoffmann/Strecker 1997: 1451). [“Here the non-verbal part – the adjective or the noun – selects the subject.”] 216 Helbig attributes no semantic content to the copula and treats the predicative element as representative of the whole predicate.

126 | Analysis of subjects in copular clauses (is) happy

My teacher

with my progress

Fig. 14: Semantic structure of (3) in Helbig’s model (adapted from Helbig 1992: 111)

It follows that, in Helbig’s view, the adjective is not responsible for the form of the subject at the syntactic level, so any formal restriction on the subject related to the predicative adjective must be purely due to semantic restrictions. It has to be noted, though, that Helbig does not address the problem of clausal subjects. Eroms (2000) sees the subject as dependent on the finite verb (or the category INFL), basically for reasons of concord. For the verb sein (‘be’), he maintains a structural valency (Eroms 2000: 139) of two and states at the same time that, logically, sentences of the following type have to be analysed as containing a monovalent predicate: (4)

He is ambitious

“Unabhängig von logischen Überlegungen” (‘independently of logical considerations’), Welke (1988: 159f) attributes no valency to the copula purely for semantic reasons and thus rejects Helbig/Schenkel’s ([1969] 1973)217 position, who see the copula as valency carrier and opt for hierarchical valency relations (basically as advocated in Helbig 1992 cited above). Welke’s stemma is similar to Helbig’s semantic level; apart from more brackets to indicate the irrelevance of grammatical elements, the most notable difference is that he puts the particle with next to the adjective and not next to my progress, which is in a way contrary to the concept of valency presented in Section 2.2.2. (is) happy (with)

(My) teacher

(my) progress

Fig. 15: Representation of (3) in Welke’s model

|| 217 Welke cites the 1982 edition, which, just like the 1973 edition used for the present study, is an identical reprint of the original 1969 edition.

Theoretical treatment of copula verbs and copular clauses | 127

Another option might be to regard the copula verb as the actual valency carrier and have it inherit certain valency requirements/restrictions from the adjective as a secondary valency carrier. One could also postulate a direct valency relation between adjective and subject, but in this case the subject may be required by two valency carriers (the verb and the adjective). Concord may be an argument in favour of a verb-centered analysis (as with Eroms), but not all scholars agree on this. (5)

The man who marries young is happy. (Napoli 1989: 9; her emphasis)

Napoli, who works with a semantic approach rooted in Government and Binding Theory, maintains that in (5), “happy alone, not is happy, is the predicate. […] Is is present in [(5)] purely to satisfy needs of the syntax and does not contribute to the semantic interpretation in the same way semantically full lexical items do” (Napoli 1989: 9; her italics). Such a semantic interpretation in fact relegates the copula to an inferior status, a position shared by the computational model used for the parsed corpora (see Chapter 3), the Stanford typed dependencies representation (de Marneffe/Manning 2008): happy

nsubj teacher

poss My

cop is

prep_with progress poss my

Fig. 16: Stanford typed dependencies representation of (3)

Here, is is a dependent of happy via a cop (=copula) relation. CamG (2002: 217) makes a distinction between the semantic and the syntactic level that parallels that made by Helbig discussed above: From a semantic point of view a PC [=predicative complement] tends to be more like a predicator than an ordinary complement such as a subject or object. [...] syntactically, a PC is a complement, but semantically it characteristically has a predicative function. (CamG 217)

(6)

a. Ed seemed quite competent. (CamG 217) b. She considered Ed quite competent. (CamG 217)

128 | Analysis of subjects in copular clauses

c.

Ed was considered quite competent. (CamG 217)

On the basis of sentences (6a–c) they “speak of Ed not as the subject of the PC but as its predicand” (CamG 217; their emphasis) in order to capture the generalization that quite competent is predicated of Ed in all three sentences although Ed is not a subject in (6b) but an object.218 In this analysis, Ed is the subject of the verb seemed or of the predicate as a whole. In a footnote, Huddleston/Pullum (CamG 217, fn3) also mention what they call predicative elements as adjuncts, which is an issue that has been widely discussed in the relevant literature. The examples (7a) and (8) are instances of the structure in question.219 (7) (8)

a.

Peter died a millionaire. (Aarts/Aarts 1982: 129) He wrote most of his poetry drunk. (CamG 217)

While the semantic relationship between a millionaire and Peter or drunk and He is identical to that of quite competent and Ed in (6a–c) above, it could hardly be justified to treat died and wrote as copula verbs. From a valency point of view, this structure raises the interesting question whether the predicative element in (7a) should be treated as a valency carrier in the same way as it is in (7b) in some of the approaches mentioned above.220 (7)

b. Peter is a millionaire.

As the verbs died and wrote would definitely be analysed as valency carriers in a valency approach, treating a millionaire as valency carrier in (7a) in analogy to (7b) would result in two valency carriers both requiring the subject Peter. In addition, the optionality of the predicative element also argues against its status

|| 218 In a small clause-analysis, one could in fact argue for the status of Ed as a subject in the postverbal small clause of (6b). See Section 5.2.2.2 for a brief discussion of small clauses. 219 Cases such as “He saw him dead.” (GLV 189) could be regarded as related problems in a small clause-analysis. In both cases the optional adjective adds a new predication to the clause without an explicit copula. 220 It is problems such as these where Argument Structure Constructions as posited by Goldberg (1995) are an interesting alternative since they permit structures of the type shown in (7a) without having to include the pattern in the lexicon simply because the construction (in the sense of the sequence of verb arguments) carries meaning of its own and combines (“fuses”) with the valency structure of the verb if the two are compatible.

Data Analysis | 129

as a valency carrier.221 Nonetheless selectional restrictions between the subject and the predicative element can be observed: (7)

c.

*Peter died a tree.

The matter is further complicated by the lack of such structures with non-NP subjects, so it is impossible to make a statement about syntactic valency. Still, it appears reasonable to treat neither selectional restrictions nor predication as sufficient criteria for postulating a valency relation.

6.3 Data Analysis To investigate the influence of predicative elements in copular clauses on the form of subjects in some more detail, two systematic studies and two case studies will be carried out, one for adjectives and one for nouns. The systematic studies cover extraposed subjects for the simple reason that they are the unmarked and by far more frequent case (see figures in Sections 7.3.4 and 7.3.6). Furthermore, the study of extraposed clausal subjects can be performed in a systematic way because for them a considerable amount of data is available preanalysed in the form of (native speaker assessment based) Herbst (1983) as well as the (corpus-based) VDE (Herbst et al. 2004) and its spin-off, the Erlangen Valency Patternbank (Herbst/Uhrig 2009). Thus, particularly with the help of the latter database, it is possible to easily identify “suspicious” cases among the 544 adjectives and 274 nouns treated. Since for non-extraposed clausal subjects no such resources were available, they are discussed in case-studies that cover a much smaller set of lexical items, which of course reduces the chance of finding relevant restrictions. All studies are based on corpus data which was augmented by native speaker interviews. The systematic studies draw on data from a manual inspection of instances from corpora of roughly 780 million words222 found with queries of the following type in Bonito (Rychlý 2007) and BNCweb (Lehmann/Schneider/Hoffmann 2000, Hoffmann et al. 2008):223 “[Ii]t” []{1,6} “typical” “not”? “to” [pos=“V.*”] within

|| 221 There are, however, obligatory elements with a similar semantic structure. See Note 23 on page 15 for an example from CamG that is treated as object there. 222 Besides the BNC, the academic, popmerge and newsmerge corpora presented in Section 3.2.2.1 were used for this research. 223 For the CQP syntax of BNCweb, the has to be replaced with a plain s.

130 | Analysis of subjects in copular clauses

This query retrieves all sentences in which It or it is followed, with a maximum of 6 words intervening, by the noun or adjective in question followed an optional not and then by to and a verb (or, alternatively, by that for [that_CL] (without not)).224 The case studies rely on the parsed corpora of more than 1.5 billion words described in Section 3.2.2.3.

6.3.1 Adjectives and extraposed clausal subjects Our first systematic study covers the two most important extraposed subjects, [to_INF] and [that_CL], both of which are highly frequent with adjectives. Of the 544 adjectives listed in the VDE/the Patternbank, 220 allow the [it] + [to_INF] pattern225 and 142 the [it] + [that_CL] pattern. As many allow neither and many both, a preliminary count results in 38 which allow a [that_CL] but no [to_INF], and 119 which allow a [to_INF] but no [that_CL]. Both groups will be dealt with separately. A systematic treatment of extraposed [V-ing] subjects was not carried out since, in the light of the discussion by CGEL (see Section 7.1.1), it was considered to be marginal. However, given that the Patternbank lists only 13 adjectives which can occur in such a pattern, we can be quite sure that restrictions apply there, too, and probably on a greater scale. For instance, there is the specific behaviour of worth, which takes a [V-ing] extraposed subject but no [to_INF] or [that_CL].226 In the VDE, a total of 249 adjectives occur with either a [to_INF] or a [that_CL] extraposed subject but not with a [V-ing] in the same position. Checking all 249 adjectives in the corpora and then with the native speakers would

|| 224 It is obvious that [that_CL] complements without that were not found with this method of querying. Even the parsed corpora did not allow such queries at the time the data for the present chapter was gathered. See however the Erlangen Treebank.info project (Uhrig/Proisl 2011) that allows queries with negated elements due to a newly developed piece of software (see Proisl/Uhrig 2012 for details). 225 3 adjectives allow for the pattern in 2 lexical units each, thus the total lexical unit count is 223. 226 The exact theoretical status of this construction remains unclear, though. Since we can have It is worth ensuring that kitchen staff are not tempted to leave the dirty cutlery soaking until morning. (A0C 964) but not *Ensuring that everything is locked up is worth., one might be tempted to analyse the construction as an impersonal construction and not as an instance of extraposition (see discussion in Section 7.1.2). However, given that we can have a tough movementlike structure with worth (see Section 7.4), an extraposition analysis is probably to be preferred.

Data Analysis | 131

have resulted in at least 1,500 extra native speaker test items,227 which would not have been possible – and would probably have been quite unnecessary – in the context of this study.

6.3.1.1 [that_CL] but no [to_INF] Appendix 17 lists the 37 relevant adjectives identified based on the Erlangen Valency Patternbank.228 A first verification using the Patternbank shows that of these 37, 6 (awkward, credible, incongruous, typical, unusual, urgent) allow a [for_NP_to_INF] or [for_NP] + [to_INF] pattern. We can exclude unusual due to the problematic analysis in the VDE. There, the example of the + [to_INF] pattern given would possibly fit the [it] + [to_INF] pattern (i.e. extraposed subject) better: (9)

Wasn’t it unusual to have been so aware of the nuances of class division at that age? (VDE 910)

A corpus search shows that of the remaining five cases all allow a [to_INF] without the [for_NP] component, too: (10) a.

As we get older and perhaps more worldly, it is sometimes awkward to admit that we ever believed such nonsense, and even more difficult to admit that not everything we imbibed was bad or wrong. (AE8 198) b. Having spent years attacking her as the author of awful policies, it is not credible to claim now that her departure hardly matters. (EDT 304) c. It would be incongruous to see her as an influence on later writers who may never have heard of her. (AN4 3156) d. It is rather typical to have slight variations in mythological stories – usually related to family relationships – appear in ancient stories. (fiction) e. It suddenly became rather urgent to find out how much Mrs Sweet knew. (GUF 1307)

Thus 31 examples are left as candidates for restrictions. Corpus evidence of these occurring with [to_INF] clauses along with some invented examples and || 227 The figures assume that we would have to test about 1.5 example sentences per adjective with four native speaker informants. 228 One of the 38 examples mentioned above, heartbroken, is due to a lexicographic error in the VDE and is thus ignored in this chapter.

132 | Analysis of subjects in copular clauses

the respective native speaker judgements can be found in Appendix 18. Not surprisingly, acceptable structures were found in the corpora that were not listed in the Patternbank, but the proportion of rejections (usually not by all but by some informants) is considerable, so many of them seem to border on the verge of unacceptability. This can be taken as an indicator that – given that extraposed subjects with adjectives are fairly common – the VDE (and consequently the Patternbank) has a relatively broad coverage of the established uses of the construction. The corpus examples that were accepted by all native speaker informants seem to be straightforward enough, such as a the following example: (11)

‘It seems particularly unfortunate to find within the elegant and modern buildings of Feltham, so carefully landscaped, all the defects of poor regimes’ (Home Office 1989). (CRT 41)

Some others were found to favour a certain syntactic environment, so for instance obvious was only fully accepted by all informants when it occurred with seem as a copula verb (with which it seems to occur more frequently in this construction), although most native speakers also accepted sentences with be: (12) a.

So when Steve came back from the office it seemed obvious to ask him to wash the dishes and scrape the remains of our vegetarian lasagne off the baking tray. (newsmerge) b. It is fairly obvious to say there are two classes of buildings to be studied — those which still stand, and those which have been destroyed, fallen down, or otherwise disappeared. (B1P 555)

Furthermore, plain is only attested followed by to see + [that_CL] as in (13) below, so an analysis that treats to see as an additional complement of plain and the following [that_CL] as the extraposed subject may be a viable alternative, in particular as no non-extraposed [to_INF] is attested. (13)

It is horrifyingly plain to see that the Mondays229 are in a tragic state, and the excellent photographs highlight this tragedy. (CK4 3249)

A complex adjective analysis similar to the one briefly mentioned in Section 7.4 on tough movement could also be considered, but due to the transparency of the construction, an analysis in terms of lexical specification (Herbst/Uhrig 2010:

|| 229 The sentence is about a band called Happy Mondays.

Data Analysis | 133

131) may be preferable (see Section 9.3.2 for a brief discussion of this structure with plain). Then there are cases on which the native speakers did not agree at all, such as positive or the invented example for characteristic below. To make sure our conclusions do not rely on problematic cases, even these will be treated as acceptable in our analysis. (14) a.

?It’s positive to win, but for every winner there are countless losers, and who wants to think about those? (academic) b. ?It would be characteristic to see him do that. (invented)230

Despite all the acceptable and problematic data, it is still possible to identify adjectives which do not allow an extraposed [to_INF] subject at all, such as apparent, certain, conspicuous, indisputable, manifest and notable. One might be inclined to state that these form two relatively homogenous semantic groups, one with adjectives that have to do with noticing (apparent, conspicuous, manifest, notable)231 and one with certainty (certain, indisputable) and that it can be predicted on the grounds of their semantics alone that they occur with facts only, and accordingly with [that_CL] complements only since facts are usually expressed by [that_CL] complements whereas [to_INF] clauses express potentiality.232 While the whole argument may work prototypically (and the semantic groups are in fact quite neat and noteworthy!), such a hypothesis must be rejected here on the grounds that we can find adjectives which fit into these groups just as neatly as the others do but do not show the same syntactic restriction. For the ‘noticing’-group we have seen above that obvious does in fact allow [to_INF] complements,233 and since it is a very good near-synonym for apparent or manifest, we can observe that the semantic prediction becomes

|| 230 Note that characteristic allows a [to_INF] if preceded by an [of_NP] as in It was characteristic of her creative and theoretical writings to establish original links between various intellectual movements of the transitional period she lived in. (GTD 1647) 231 If we decided not to treat plain to see as an instance of extraposed [to_INF] subject but as a complex adjective or as an adjective with an obligatory [to_INF] complement, we could add it to this group, too. 232 See the discussion of [to_INF] in relation to modality in Section 4.2. 233 As noted above, obvious does have a preference for seem (and possibly other markers of epistemic modality such as perhaps) when it occurs with an extraposed [to_INF] subject, but since even a search of the 1.5 billion words parsed corpora for a nominal subject it with the copula seem and a [to_INF] complement did not yield relevant results for the queried adjectives (apparent, manifest and certain), these seem not to occur in the construction even under the same modality conditions.

134 | Analysis of subjects in copular clauses

untenable. The situation is similar for the ‘certainty’-group, where correct, sure and true can be cited as counter-examples. While sure, semantically a good match for certain, allows neither [that_CL] nor [to_INF] extraposed subjects, correct and true come close to the meaning of indisputable in the following examples of extraposed [to_INF] clauses: (15) a.

For this reason it may not be correct to relate thermodynamic quantities obtained from one cooling curve to those obtained by another as has been pointed out by Gee (1966). (H0U 821) b. It is true to say that regular rail travellers in Britain know their own country better and in greater detail than those who travel by any other form of transport. (VDE 884)

Thus the prototypically correct semantic explanation can at best predict tendencies but fails to account for the exact facts.

6.3.1.2 [to_INF] but no [that_CL] The picture Appendix 20 gives for adjectives followed by an extraposed [that_CL] that can only occur with an extraposed [to_INF] according to the VDE/Patternbank is even more colourful (i.e. varied) than the one given for the previous section.234 Not only is there a smaller proportion of sentences that all native speaker informants accept, it also becomes very obvious that the native speakers have different personal preferences (which will be briefly discussed in Section 9.1). Both observations are indicators of the problematic character of the data collected from the corpus. Again it appears as if the coverage provided by the VDE includes the vast majority of the adjectives established in that construction and that most of the ones we deal with here are less central. There are still a few adjectives of which we can state, with varying degrees of confidence, that they seem to allow an extraposed [that_CL] subject despite not being listed in the VDE/Patternbank with it. We shall start with these clearly acceptable cases, then discuss problematic instances and finally present evidence for arbitrary restrictions. The adjectives false, handy, harsh, hurtful, improper and inadequate seem to be perfectly all right with a [that_CL] extraposed subject, as in the following example:

|| 234 Appendix 19 gives a list of all candidate adjectives from the VDE.

Data Analysis | 135

(16)

It was hardly the tour operators’ fault that the Jamaica test was abandoned and it may seem harsh that they have to compensate those who missed out on cricket watching. (newsmerge)

The cases of accurate, excessive, foolhardy, irrational and negligent are less clear since for each one of them, only one example (given in (17)) was accepted by all native speaker informants alike, either because only one was found in the corpora, as for excessive, foolhardy and negligent, or because other examples were rejected. (17) a.

Nor is it necessarily accurate that doctors’ lax prescription habits are to blame, for this is to underestimate the staggering adaptability of bacteria allowing them to survive in any environment. (newsmerge) b. It does seem excessive that there are six dissertations under way on Apollonius Rhodius’ Argonautica and none on Ennodius, but perhaps there I betray a prejudice. (academic) c. From that point on Strauss lived in semi-disgrace and, in retrospect, it seems both courageous and foolhardy that he should have attempted an opera on a pacifist subject to a libretto by an exiled anti-fascist Jew in 1938. (newsmerge) d. A few simple facts paint the real picture: It may seem irrational that both the cost of labor and net wages could improve under the new system, but that in fact seems to be the case. (academic) e. One of the first acts of the newly free eastern European countries was to commission their own national museums, so it seems almost criminally negligent that Scotland’s ruling elite preferred to spend most of the past 50 years squabbling over narrow political definitions of nationhood and the minutiae of devolution. (newsmerge)

Some adjectives were not attested in the corpora and were tested with invented examples, but the responses were mixed. All instances are listed in Appendix 20 as well, so it will suffice to give one such example: (18)

?It is constitutional that people are allowed to own guns in the US. (invented)

As in the previous section, some adjectives seem to favour certain linguistic environments. Thus proper (which in the construction we are dealing with often occurs in the combination right and proper) seems to prefer a putative [that_CL] as extraposed subject: In the BNC, all 9 sentences found with the query it {be/V} proper that contained should in the extraposed [that_CL] subject and 8 out of the

136 | Analysis of subjects in copular clauses

10 instances found with it {be/V} right and proper that did so, too. All examples tested with improper also had a putative should in the [that_CL]. The examples tested with proper and an indicative verb instead were generally accepted by the native speaker informants but not unanimously. Herbst (1983: 146) gives a semantic explanation of this preference, i.e. that what the [that_CL] expresses in putative uses is rather a notion or idea than a fact as in indicative examples. Herbst takes this semantic difference as an argument in favour of a distinction between two types of [that_CL], one of which can be replaced with an [if_CL] in conditional contexts, but drops this distinction in later works such as the VDE.235 No systematic study was carried out on the acceptability or preference of indicative vs. putative extraposed [that_CL] subjects, but there seem to be further adjectives which favour this form of complement, for instance prudent. Both proper and prudent also regularly occur with a subjunctive verb in the [that_CL] in the corpora. The evidence for counter-productive consists only of two examples, but these are too interesting to ignore: (19) a.

Is it not illogical and counter-productive that the British electoral system denies or impairs their voting rights? (AJ6 849) b. Is it not counter-productive that benefit officers have been told to push people on to invalidity benefit to bring down the claimant count, which passes for Britain’s unemployment total, only to complain about it and start looking for ways to scrap it? (newsmerge)

One native speaker informant rejected (19b) and accepted (19a) only due to the coordination with illogical. What is most remarkable, though, is that both examples (from entirely unrelated sources) are in the form of a negated interrogative clause – a rather striking coincidence, if it is one.236 Even the large parsed corpora did not yield any further instances of counter-productive (or counterpro-

|| 235 The distinction turns out to be problematic in the light of large corpora since putative should and [that_CL] complements that can be replaced with [if_CL] complements do not necessarily coincide. For instance, likely, which is listed as not being able to occur with an [if_CL] in conditional contexts (Herbst 1983: 147) does occur several times with putative should in the [that_CL] in the parsed corpora. 236 Since the inversion at the beginning of interrogatives triggers obligatory extraposition, one might be led to hypothesize that [that_CL] subjects may occur with counter-productive in canonical position and are only forced to extraposed position by the interrogative pattern, but no single instance of a preverbal [that_CL] subject was found in the parsed corpora. (The spelling variant counterproductive was checked, too.)

Data Analysis | 137

ductive) with an extraposed [that_CL] subject, so it was impossible to determine whether these two sentences follow a rare but regular pattern. For a majority of the corpus examples tested there was no agreement by the informants. Two such examples will be given for illustration, but plenty can be found in Appendix 20. (20) a.

?The first one, I had a relationship with a guy who was really nice, but we were both very difficult and disturbed people at the time and it wasn’t suitable that we should have a baby. (newsmerge) b. ?But Mr Luis Miyares, the director of the Fondo de Bienes Culturales, which oversees art sales, says bluntly: We do not want them to succeed. In our society, health-care is free, education is free, funerals are free, electricity, telephone and water are cheap. And if the artists receive all these benefits, it is correct that they pay us part of their income. (newsmerge)

The second example seems to involve a non-native or translation context and is noteworthy because correct is represented in the VDE with two lexical units, one of which (correct (A) in the sense of ‘true’) is listed with extraposed [to_INF] and [that_CL] whereas the second one (correct (B) in the sense of ‘acceptable, in line with established conventions’) is listed only with a [to_INF]. The corpus examples that were accepted by all native speaker informants were all instances of lexical unit (A). Example (20b) was the only instance of lexical unit (B) and was rejected by two out of four native speaker informants. A large number of adjectives were identified for which no [that_CL] complements in extraposed subject position are attested and for which no single native speaker came up with a sensible invented example. These are bold, careless, civil, clumsy, complacent, early, easy, economic, exact, expensive, extreme, fruitful, fruitless, hopeless, inept, negative, radical, safe, sentimental, smart and ungrateful. In this list no semantic groups of the kind identified in the last section can be made out. The adjective negative provides an interesting example to support our conclusion that there are quite arbitrary restrictions at work, since positive shows a different pattern. The latter only allows a [that_CL] according to the VDE, the former only a [to_INF], which is marked as rare:237 (21) a.

And it’s very positive that this area continues to rise. (VDE 599)

|| 237 A query of the parsed corpora confirmed the rarity but more than 10 good examples also seem to confirm the acceptability of the construction.

138 | Analysis of subjects in copular clauses

(22) a.

I think it’s offensive and negative to talk about somebody having a black mark against their name or having a blackened character. (VDE 549)

For positive with a [to_INF], we saw above that the only corpus example – (14a) reproduced below as (21b) – was not accepted by all speakers. (21) b. ?It’s positive to win, but for every winner there are countless losers, and who wants to think about those? (academic) There are, however, more than ten good examples in the larger parsed corpora,238 so we shall continue to treat this construction as acceptable. Since for negative with a [that_CL] the corpora used for this chapter did not contain any example, the search was widened to the parsed corpora. One single instance was retrieved – by a non-native speaker239 of English:240 (22) b. ?It was also negative that we didn’t play so well in the second half because there were too many changes. (newspapers) We can thus conclude that arbitrary restrictions on the use of the [that_CL] extraposed subject with adjectives seem to apply and that they occur with more adjectives than the restrictions on the [to_INF] extraposed subject clauses.

6.3.2 Nouns and extraposed clausal subjects As will be pointed out in Chapter 7 extraposed subjects are predominantly [to_INF] and [that_CL] complements. These will thus be the primary interest in this section and due to the much smaller number of relevant nouns (compared to the number of adjectives), they can be studied together in one section. Extraposed [V-ing] clause complements will be dealt with separately due to their much lower frequency and more limited acceptability.

|| 238 There is a considerable number of [for_NP_to_INF] extraposed subjects, too. 239 In this example it was former England football manager Sven-Göran Eriksson. 240 We cannot even rule out the possibility that the example contains a typing error since there seems to be the more established alternative to express the same meaning with a nominalised form of negative (contrasted beautifully with adjectival positive in the following example): It’s positive that they’re trying to understand what the coaches and players are saying but it’s a negative that they aren’t taking more action,“ said Millward. (newspapers)

Data Analysis | 139

6.3.2.1 [to_INF] vs [that_CL] The Erlangen Valency Patternbank identifies 40 lexical units which occur in one or both of the patterns [it] ... noun ... that_CL and [it] ... noun ... to_INF. This includes three pairs of lexical units belonging to the same lexeme (business, disgrace, job), the relationship of which will be discussed at the end of this section. Table 20, derived from the Patternbank data, claims that there are only 6 (blessing, blow, crime, luck, policy, will) out of the 40 lexical units241 that allow both patterns. Tab. 20: VDE/Patternbank data for extraposed [to_INF] and [that_CL] subjects; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses

extraposed [to_INF]

extraposed [that_CL]



+

advantage

+



bet (A)



+

blessing (A)

+

+

blow (B)

+

+

business

+



case (C)



+

crime

+

+

custom

+



delight

+



disgrace (A)

+



disgrace (B)



+

duty (A)

+



error

+



fault



+

fun

+



goal (A)

+



habit

+



idea

+



illusion

+



achievement (A)

|| 241 The table contains 39 relevant lexical units since business (A) and (B) were grouped together (see Note 242) and since joke only occurs with a [V-ing].

140 | Analysis of subjects in copular clauses

extraposed [to_INF]

extraposed [that_CL]

importance



+

instinct

+



interest (B)

+



job (D)

+



job (special use 1)



+

joke (special use 1)





judg(e)ment



+

luck

+

+

mercy (α)



+

mistake

+



need



+

opinion



+

policy

+

+

practice (B)

+



purpose

+



responsibility



+

role

+



task

+



tendency

+



will

+

+

The table suggests that there are strong valency restrictions with the majority of the nouns, but since only limited corpus data was available at the time the VDE (which provides the data used in the Patternbank) was created, these results had to be double-checked with the help of modern corpora. The results of the corpus search are given in the following table: Tab. 21: VDE/Patternbank data augmented with corpus results; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses

extraposed [to_INF]

extraposed [that_CL]

achievement (A)

+

+

advantage

+

+

bet (A)

+

+

Data Analysis | 141

extraposed [to_INF]

extraposed [that_CL]

blessing (A)

+

+

blow (B)

+

+

business

+



case (C)

+

+

crime

+

+

custom

+

+

delight

+

+

disgrace (A)

+

disgrace (B)

+

duty (A)

+

+

error

+

+

fault

+

+

fun

+

+

goal (A)

+

+

habit

+



idea

+

+

illusion

+

+

importance

+

+

instinct

+



interest (B)

+

+

job (D)

+



job (special use 1)



+

joke (special use 1)





judg(e)ment

+

+

luck

+

+

mercy (α)

+

+

mistake

+

+

need

+

+

opinion

+

+

policy

+

+

practice (B)

+

+

purpose

+

+

responsibility

+

+

role

+



142 | Analysis of subjects in copular clauses

extraposed [to_INF]

extraposed [that_CL]

task

+



tendency

+



will

+

+

The situation has changed dramatically as there are very few ‘suspicious’ cases left, although it has to be borne in mind that Table 20 is by no means ‘wrong’. It merely covers the frequent established uses whereas Table 21 also includes cases which border on the verge of unacceptability. Often, only one example was found in the corpus, so in order to exclude ‘errors’ (see discussion in Section 3.3 above), native speaker interviews were conducted, on the basis of the corpus evidence. The following table reports the results: Tab. 22: Data from VDE/Patternbank, corpora and native speaker interviews combined; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses

extraposed [to_INF]

extraposed [that_CL]

achievement (A)

+

+

advantage

+

+

bet (A)

+

+

blessing (A)

+

+

blow (B)

+

+

business

+

+

case (C)

o

+

crime

+

+

custom

+

+

delight

+

o

disgrace (A)

+

disgrace (B)

Comment

only with premodifier

lexical units (A) and (B) combined

see separate discussion +

duty (A)

+

o

error

+

+

fault

+

+

fun

+

+

goal (A)

+

+

see separate discussion

only one example from dubious source for [that_CL]

Data Analysis | 143

extraposed [to_INF]

extraposed [that_CL]

habit

+



idea

+

+

illusion

+

+

importance

+

+

instinct

+



interest (B)

+

+

job (D)

+



job (special use 1)



+

joke (special use 1)





judg(e)ment

+

+

luck

+

+

mercy (α)

o

+

mistake

+

+

need

+

+

opinion

+

+

policy

+

+

practice (B)

+

+

purpose

+

o

responsibility

+

+

role

+



task

+



tendency

+



will

+

+

Comment

see discussion in Section 6.3.4.1 all cases “in someone’s interest”

only occurs with V-ing

no agreement among informants

As expected, some of the corpus evidence was of limited acceptability, although there were no instances on which all native speakers agreed in their verdicts on all instances of a noun to reject the examples. The evidence is given in Appendices 21 and 22. Sentences (23a–d) exemplify four such cases. (23) a.

?Its message was simple: it was an Islamic duty that Abu Zeid be killed. (newsmerge) b. ?It is a great delight that the trade has stopped, even if the reasons have nothing to do with animl [sic!] welfare. (newsmerge) c. ?War has been my life, and although my loyalties lean toward its continuation, it would be a mercy to see it end. (fanfiction)

144 | Analysis of subjects in copular clauses

d. ?“It is our purpose that the dialogue leads to establishment of trust, friendship and co-operation between the two countries,“ India’s foreign ministry secretary, Salman Haider, said. (newsmerge) Such use may be non-native or a translation as is likely in the first example or it may be due to differences in national varieties as in the last example. To be on the safe side, no argument will be based on any of these dubious cases. The same policy also leads to items being classified as acceptable in a certain construction even if the evidence is relatively doubtful. Thus business242 is classified as acceptable with an extraposed [that_CL] despite the complete absence of corpus evidence because the native speaker informants accepted (24): (24)

It is none of your business that I plagiarized my PhD thesis. (invented)

At this point, a brief discussion of problems related to the distinction between lexemes and lexical units in the tables used in this chapter is in order. The nouns disgrace and job are listed separately with their two respective lexical units since both accept different patterns (see Note 242 for a discussion of the two lexical units of business). This would make a strong case in favour of the view that valency is a property of the lexical unit, but one has to be careful using the VDE’s classification for linguistic description given that it was developed primarily with the lexicographic application in mind: The sense distinctions given […] do not aim to provide a description of word meaning comparable to that of a general dictionary. Establishing senses according to their valency patterns in some cases results in a rather different identification of senses than in conventional dictionaries. (VDE xxxviii)

In the case of disgrace, however, the VDE classification mirrors that of other dictionaries (Longman Dictionary of Contemporary English, 5th edition [LDOCE5], Oxford Advanced Learner’s Dictionary, 7th edition [OALD7]), with the two senses as defined in LDOCE5: 1 [U] the loss of other people’s respect because you have done something they strongly disapprove of [...]

|| 242 The VDE identifies two lexical units in the case of business. As both occur in the [it] + toINF and in the [it] + V-ing pattern, they were treated as one item for the purpose of this chapter. One might argue that example (24) only represents the lexical unit business (B), which would be a further hint at arbitrary valency restrictions, but on the basis of the limited evidence such a conclusion would have to remain speculative in character.

Data Analysis | 145

2 be a disgrace used to say that something or someone is so bad or unacceptable that the people involved with them should feel ashamed

Nonetheless any argument based on the different syntactic behaviour of the two lexical units is dangerous as their semantics seem to show a certain overlap: (25) a. Her behaviour has brought disgrace on her family. (OALD7, sense 1) b. That sort of behaviour is a disgrace to the legal profession. (OALD7, sense 2) The meaning of disgrace in (25b) could also be regarded as a more familiar use of the word used in (25a).243 However, the examples from the VDE are in line with the meaning difference between the two senses, in that the first one takes a more neutral, outside perspective whereas the second expresses a subjective element of annoyance: (25) c.

Certainly it was no disgrace to lose to Ghana, an accomplished young team. (VDE 237, sense A[=1]) d. It’s a disgrace that they want to deport an innocent woman like me, when murderers are walking free from the justice system. (VDE 237, sense B[=2])

The fact that it’s no disgrace seems to occur more readily with a [to_INF] but less so with a [that_CL]244 confirms that sense B/2 might be associated with some sort of additional pragmatic meaning of annoyance that cannot be expressed in the same way under negation. To what extent this justifies different lexical units or may just be treated as different meanings associated with different patterns is debatable. So is in fact the question whether different senses of such words can be postulated at all. Wiegand states on the issue: “Es gibt keine Wörter, die polysem sind, sondern nur Wörter, die wir als polysem analysieren” (Wiegand in a discussion at the Eighth International Symposium on Lexicography, May 2–4, 1996, at the University of Copenhagen; confirmed by personal communication) [“There are no polysemous words, there are only words that we analyse as polysemous.”]. One should thus be aware that the distinction of different lexical units is often arbitrary at least to a certain extent.

|| 243 The fact that shame could be used as a synonym for both senses (and German Schande as well) is another indicator of their semantic closeness. 244 In the parsed corpora, the [to_INF] is more than 10 times more frequent with no disgrace than the [that_CL].

146 | Analysis of subjects in copular clauses

In the case of job, the semantic difference between the two lexical units is clearer and will be illustrated with the relevant examples from the VDE: (26) a.

They are the policymakers. It’s their job to see that the tools get used wisely. (VDE 448) b. It was a good job a friend was with me to raise the alarm. (= it was fortunate that) (VDE 448)245

Research in the parsed corpora confirms the distribution, i.e. neither instances of the first meaning with a [that_CL] nor instances of the second meaning with a [to_INF] were found. What to make of this result is very much theorydependent. If one regards lexical units as primary, one can posit a valency restriction on both lexical units. However, one could also argue that there is one item job that can occur in both constructions and acquires a slightly different meaning if it does. Whatever the theoretical treatment, though, the fact that job shows such a behaviour is another indicator of the highly item-specific and storage-related character of English subject selection. So to sum up, we can observe that there are five246 nouns left which show restrictions of the kind we were looking for in the revised version of our table: habit, instinct,247 role, task, and tendency. It is worth noting that only restrictions on the [that_CL] and not on the [to_INF] could be identified reliably.

6.3.2.2 [V-ing] There is only a small number of nouns which frequently occur with an extraposed [V-ing] clause complement. The Erlangen Valency Patternbank lists business, fun, job and joke in the pattern [it] ... noun ... V-ing. All of these are perfectly established usage and can be illustrated with examples from the BNC: (27) a.

More often the problem is to limit the list of recipients, for it can be a costly business sending out fragile pottery or expensive jewellery. (EX6 1180)

|| 245 The American native speaker interviewed on the issue claimed that this meaning of a good job must be British, and searches in COCA (Davies 2008-) and the BNC as well as the Longman Dictionary of Contemporary English, 5th edition confirm this hunch. 246 Plus, possibly, job and disgrace, depending on whether one is willing to accept the division into lexical units. The verb joke is treated in the section on [V-ing]. 247 But note that instinct was accepted with a [that_CL] non-extraposed subject by native speakers in spite of absence of corpus evidence in the case study in Section 6.3.4 below.

Data Analysis | 147

b. It was no fun loving a person like Memet, especially when he wouldn’t commit himself to her in a civilized and law-abiding manner. (G1D 1614) c. As you know it is a 24 hours job looking after a child/young adult with special needs. (K97 18252) d. It’s no joke being left with a young child to bring up on your own. (H9D 2061) No restrictions as to their acceptability seem to apply, contrary to CGEL’s view that “[e]xtraposition of -ing clauses is in fact uncommon outside informal speech” (CGEL 1393). Due to the indeterminacy between extraposed and rightdislocated structures discussed in detail in Section 7.1.1, no such distinction will be upheld in the present chapter. This is also reflected in the query used for the research in the corpora in that optional commas and dashes are accounted for: “[Ii]t” [word!=”[,\.;!?]”{1,6} “habit” “[,-]”? “not”? [pos=”V.G”] within 248 All nouns that occur at least in one of the two patterns [it] ... noun ... that_CL and [it] ... noun ... to_INF were queried and examples found in the corpus were submitted to native speaker informants. The results of the interviews are listed in Appendix 23. As with the phenomena discussed in the chapters before, there are some items that seem to allow the structure in question but are not listed with it in the VDE/Patternbank, such as blessing, blow, habit or responsibility: (28) a.

But it seems that it is a mixed blessing having the most successful racing driver of all time as a boss. (newsmerge) b. It was a big blow having to send on a substitute after about ten minutes. (newsmerge) c. It’s a great British television habit, reading people in exactly the way they ask themselves to be read, flattening British society into something whimsical and apolitical and untrue. (newsmerge) d. It is a colossal responsibility being the midwife. (newsmerge)

There are also cases where there was no agreement among informants, such as mistake or fault, and acceptability varied across the actual instances: (29) a.

?I should have realized it was a mistake getting involved with Compeyson. (FPU 1655)249

|| 248 The query translates as follows: It or it followed by up to six ‘words’ that must not be punctuation marks, followed by the relevant noun, followed by an optional comma or dash, followed by an optional not, followed by an -ing verb form, all within the same sentence.

148 | Analysis of subjects in copular clauses

b. ?*Oh it’s my fault leaving it there. (FMF 353)250 Then there is role, which occurs only once but was accepted by all informants, or illusion, which does not occur at all but was still accepted by both informants interviewed on it: (30) a.

It’s a very demanding role being Prince of Wales, but it’s an equally more demanding role being King. (newsmerge) b. It’s an illusion believing that you can fly. (invented)

The case of idea is interesting in that we can illustrate the full acceptability scale here: (31) a. *It was my idea asking you. (AT4 2237) b. ?It was a good idea taking your own dressing gown though wasn’t it? (KE3 271) c. It seemed a good idea - making a film of Muhammad Ali’s comeback fight in the heart of Africa. And so it was. (newsmerge) All native speakers rejected (31a), no agreement could be reached on (31b) and everyone liked (31c), which would, however, have to be classified as dislocated rather than extraposed if we were to make that distinction (see Section 7.1.1 for why the present study does not make the distinction). Much of the judgement data collected shows that testing actual data of extraposed or dislocated [V-ing] clauses is problematic (see Section 7.1.1 for details). The most interesting cases for our purposes are again the ones that do not usually allow the structure. The following quite heterogeneous group of nouns seems to show a restriction of that kind: duty, goal, instinct, judg(e)ment, need, opinion, tendency and will. Thus we can say that some nouns show restrictions on extraposed (in a broad sense, including dislocated) [V-ing] subjects. This is interesting also for the item-specific restrictions on extraposition discussed in more detail in Chap-

|| 249 Even though there are enough hits in the corpora to suggest that mistake does occur in the [it]...noun...V-ing pattern, there was no example among those submitted for judgement that all native speakers agreed to accept. A general preference for a [to_INF] in the position could be observed among the native speaker informants. 250 Three out of four native speaker informants rejected this sentence and corrected it to read my fault for leaving, whereas three out of three accepted It was my own fault, getting pregnant again, a personal matter, not something you should take to the union. (academic)

Data Analysis | 149

ter 7, since, for instance, duty can occur with a [V-ing] in canonical subject position: (32)

Attracting the sympathy of erstwhile opponents is the proper duty of a political party. (newsmerge)

6.3.3 Case study: adjectives and non-extraposed clausal subjects As mentioned at the beginning of this chapter, there is a lack of pre-analysed data for non-extraposed clausal subjects with adjectives, so in order not to miss relevant cases by relying on the analyses of extraposed subjects available, the first case study looks at all adjectives beginning with the letter h that are listed in the Erlangen Valency Patternbank (Herbst/Uhrig 2009). This rather harsh restriction had to be made due to the sheer number of adjectives in the VDE (544). Also the native speaker checks in this section were done with the help of only two informants. The parsed corpora were searched for the three principal clausal subjects in English, [that_CL], [to_INF] and [V-ing], using the following queries:251 – [to_INF]: [adjective] ---csubj-----> [not specified] ---aux-----> to – [that_CL]: [adjective] ---csubj-----> [not specified] ---complm-----> that – [V-ing]: [adjective] ---csubj-----> %ing All results were manually checked and only actual examples of the construction in question were added to a database. Where the number of results was sufficient to suggest that the construction is acceptable, no further checks took place. Combinations which occurred only once were checked by the native speaker informants to verify their acceptability. In the same interviews, judgements on invented examples exemplifying structures not found in the corpus were also elicited. Table 23 presents the results. Due to the small scale data available, all results presented in this section have to be interpreted as preliminary and in need of verification by a larger scale study.252

|| 251 csubj = clausal subject; aux = auxiliary; complm = complementizer; %ing: anything that ends in -ing; see de Marneffe/Manning (2008) for a full description of the grammatical model. 252 Appendix 24 lists all native speaker judgements.

150 | Analysis of subjects in copular clauses

Tab. 23: Results of corpus research and native speaker interviews on clausal subjects in canonical position of copular clauses with adjectives starting with h as predicative element

half-hearted

[to_INF]

[V-ing]





[that_CL] Comment –

handy

+

+

+

happy







only example of [to_INF] rejected

hard

+

+

+

[that_CL] only with “hard to [verb]”, see discussion on tough movement

hard-pressed







harmful

+

+



harsh

+

+

+

hateful







hazy







healthy



+

o

heartbroken







heartless

+

+

o

heavy



o



helpful

+

+

o

heroic

+

+

+

hesitant







hilarious

+

+

+

honest

o

+

+

honourable

+

+



hopeful

o

o



hopeless

+

+



horrendous

+

+

+

horrible

+

+

+

horrific

+

+

o

hospitable







hostile

+





hot



+

o

human

+

+

o

problematic examples of [V-ing] (no agreement on one; not the same lexical unit, highly colloquial for the other)

[that_CL] only one example “too horrible to”, but NS accepted alternatives without “too”

only one example of [to_INF] and [V-ing], the latter rejected

Data Analysis | 151

[to_INF]

[V-ing]

[that_CL] Comment

humane

+

+

o

hungry







hurtful

+

+

+

hypocritical

+

+

o

It is not surprising that some adjectives – such as hesitant or heartbroken – obligatorily require an animate (usually human) subject and thus allow no clausal subjects at all. Various others – for instance hilarious or horrendous – allow all three realizations of clausal subjects.253 However, we also find adjectives that permit some but not all clausal subjects. The adjective healthy, for instance, is only attested with a [V-ing] subject as in (33a), and the [to_INF] in (33b) was rejected and corrected to [V-ing] in (33c) in the native speaker interviews. The [that_CL] in (33d) was accepted, however. (33) a.

Many non-veggies think giving up red meat is healthy but by going vegetarian you risk losing out on valuable nutrients like iron. (newspapers) b. ?To eat lots of junk food does not seem healthy. (invented) c. Eating lots of junk food does not seem healthy. (invented) d. That he ate lots of junk food was not healthy. (invented)

While these examples make it appear as if the [to_INF] were dubious with healthy, we find ample evidence in the corpora that healthy does allow a toinfinitive complement in extraposed position: (33) e. f.

It is not healthy to have such a consensus on Europe across the three main parties. (AHN 2098) So all those of us who have been taught it is healthy to eat potato skins had better think again. (BN4 1663)

Thus, healthy is not incompatible with [to_INF]s as such. In addition, if we test the non-extraposed counterparts of (33e) and (33f) we obtain the perfectly acceptable (33g) and (33h).

|| 253 Many adjectives in the list are not attested with [that_CL]s in the corpus but were accepted by native speaker informants, which may be related to the strong propensity of [that_CL]s to occur in extraposed position.

152 | Analysis of subjects in copular clauses

(33) g. To have such a consensus on Europe across the three main parties is not healthy. h. So all those of us who have been taught that to eat potato skins is healthy had better think again. We cannot even argue that the oddness of (33b) is due to the verb eat in the [to_INF] given that it is eat in (33h), too. The fact that the [to_INF] is the subject of a subordinate clause in (33h) cannot account for the difference in acceptability, either, given that the structure of the acceptable (33g) is parallel to the problematic (33b) with the exception of the copula and the resulting do-supported negation. This result is puzzling in that it shows that there are other restrictions at work than just a simple selection of subject category (in this case [to_INF]). Furthermore, the adjective honourable does not allow a [that_CL] subject, but is perfectly acceptable with a [to_INF] (34a) or a [V-ing] (34b). (34) a.

To redouble humanitarian efforts during Ramadan would also be both honourable and politically effective. (newspapers) b. Although it is a short stride from national pride to xenophobia, celebrating a sporting victory can be honourable enough when the performance is as stylish as it was now. (newsmerge)

Although this looks like an arbitrary restriction, there might be a possible semantic explanation: As mentioned above,254 it is sometimes claimed that [that_CL] complements usually express facts, whereas non-finite forms such as [V-ing] or [to_INF] can also be used to express (actual and potential) activities without factual character. However, Faulhaber (2011: 170–189) shows convincingly how many such semantic accounts contradict each other and fail to predict the actual realisations of complement clauses in postverbal position.255 Since quite often such explanations can only be applied in a post-hoc manner it is questionable to what extent they can predict the choice of complement here. Let us thus examine harmful, which shows the same pattern: (35) a.

Using fitness equipment incorrectly can be harmful to your body. (HJB 203) b. To awaken a child for food is very unnecessary and harmful. (academic)

|| 254 See also 4.2 for a discussion of potentiality with [to_INF] complements. 255 See also Klotz (2007) for a smaller scale study of postverbal clausal complements on the basis of the VDE.

Data Analysis | 153

c.

(36)

*That he created a sterile environment through excessive cleanliness was harmful to his immune system. (invented) That he only left her 3,000 pounds was not particularly helpful, either. (invented)

Interestingly, although not all native speaker informants agreed on it, (36) appears to be less awkward than (35c), but it would be difficult to decide what in the semantics of helpful makes it allow factual subjects and what in the semantics of harmful makes it disallow them, so it looks as if there are at least some arbitrary restrictions in the selection of [that_CL] subjects. The same pattern was also found for hopeful and hopeless. To sum up, we can say that we were able to identify some likely candidates for restrictions on clausal subjects of adjectives in canonical position even with the limited data available and that we can treat this as a further pointer to the existence of item-specific restrictions on English subjects.

6.3.4 Case study: nouns and non-extraposed clausal subjects For the second case study, all nouns starting with the letter i were selected from the Erlangen Valency Patternbank (Herbst/Uhrig 2009) and the parsed corpora were searched in a similar way to the first case study. The modified queries can be represented as follows: – [to_INF]: [noun] ---csubj-----> [not specified] ---aux-----> to – [that_CL]: [noun] ---csubj-----> [not specified] ---complm-----> that – [V-ing]: [noun] ---csubj-----> %ing Again, corpus examples of patterns that occurred only once or were otherwise dubious were verified with the help of two native speaker informants who were also asked to judge invented examples of patterns that did not occur in the corpus. Since some of the items in question are of very low frequency, since a considerable portion of sentences had to be invented and since only two native speakers were involved in this rather small scale case study, the results gathered in this subsection must be treated with great caution. The following table presents the results of the combined corpus and native speaker evidence:256

|| 256 All native speaker verdicts can be found in Appendix 25.

154 | Analysis of subjects in copular clauses

Tab. 24: Results of corpus research and native speaker interviews on clausal subjects in canonical position of copular clauses with nouns starting with i as predicative element

[to_INF]

[V-ing]

[that_CL] Comment

idea

+

+

+

[that_CL] only once

ignorance

+

+

+

[V-ing] only once

illusion

+

+

+

[that_CL] only once

image



o



But lots of [V-ing] with inversion; see separate discussion in Section 6.3.4.2

immunity







impact



o



one example of [V-ing] by Belgian painter; rejected

importance







see separate discussion in Section 6.3.4.1

of importance

+

+

+

[that_CL] only once; almost always “of [adjective] importance”; see separate discussion in Section 6.3.4.1

impression

o

+

o

impulse

o

+

o

incentive

+

+

o

inclination

+

+

o

[to_INF] only once

increase

+

+

+

plus lots of [V-ing] with inversion

influence

o

+

+

[to_INF] only once in a very specific construction

insight



+

+

[that_CL] only once

insistence







[V-ing] only with inversion

inspiration

+

+

+

instinct

+

+

+

instrument

+

+



insurance

o

+

+

interest

o

+

+

interval







interview







introduction

+

+

+

invitation

+

+

o

[to_INF] 'borderline', 'marginal'; lots of [V-ing] with inversion

plus some [V-ing] with inversion

Again, it is not surprising that some nouns do not allow clausal subjects at all. It appears that immunity, insistence, interval and interview are such cases, proba-

Data Analysis | 155

bly also impact since the following corpus example was not only rejected by both native speaker informants but is also from a non-native context: (37)

?Seeing his work was a lasting impact. (newspapers)

A variety of nouns seem to allow the full range of possible clausal subjects, at least under the relatively lax criteria applied in this section. These include idea, ignorance, illusion, influence, inspiration, instinct and possibly introduction. For instance example (38a) of instinct with a [that_CL] was crafted on the basis of a corpus example with a [V-ing] given as (38b): (38) a. But that he immediately went after them was pure instinct. (invented) b. What Harry didn’t realize was that going after Snape was instinct, too. (fiction) It would take a larger study with more native speaker informants to reliably determine if the [that_CL] that was accepted on the basis of an invented example here is actually fully acceptable, in particular in the light of the corpus evidence discussed in Section 6.3.2.1 which led us to claim that there is a restriction with instinct that forbids extraposed [that_CL] subjects. Again, we also find nouns that do not allow all realisations alike. Thus, for instance, instrument is attested with several examples of [V-ing] subjects in the corpora (usually as blunt instrument) but with no [that_CL] and no [to_INF]. While it was possible to craft borderline-acceptable examples with a [to_INF], the [that_CL] in (39b) was rejected by both native speakers. (39) a.

It will point to one of the conclusions of the governmentcommissioned Deloitte and Touche report which suggests that giving a set budget to a newly opened hospital is a “blunt instrument”. (newsmerge) b. *That they gave a set budget to a newly opened hospital was a “blunt instrument”. (invented)

For insight, we also may have identified a restriction on the availability on the [to_INF], but again there is not enough data for us to make a claim other than that there is a need for further investigation. There were one [that_CL] and one [V-ing] example found in the corpus, and the [V-ing] example given in (40b) had to be modified (reproduced as (40c)) in order to be accepted by the native speaker informants. (40) a.

That “the country” in whose interests policy is designed is to be understood in class terms is, of course, no recent insight. (academic)

156 | Analysis of subjects in copular clauses

b. ?Seeing him in action at his GEC HQ off London’s Park Lane was a rare insight. (newspapers) c. Seeing him in action at his GEC HQ off London’s Park Lane was a rare insight into his everyday working life. (modified version of previous example) The invented [to_INF] example was rejected: d. *To realize her own issues with him was an interesting insight. (invented) Since it is often difficult to tell why exactly an example is rejected by native speaker informants, this again is of course not enough to rule out the acceptability of a [to_INF] subject with insight altogether. Finally, the case of influence shows very nicely that the syntactic environment has an influence on the acceptability that is hard to describe. The corpora provide sufficient evidence for the acceptability of [V-ing] as a subject of influence (as in (41a)), no evidence for [that_CL] subjects (but (41b) was just about accepted) and one example of a [to_INF] given as (41c) below, which was accepted, too: (41) a.

Selling The Big Issue is a stabilising influence in Goddard’s life. (newspapers) b. That he sold The Big Issue was a stabilising influence in Goddard’s life. (invented) c. T S Eliot once remarked of Henry James, “There will always be a few intelligent people to understand James, and to be understood by a few intelligent people is all the influence a man requires. (newsmerge)

It appears as if in all the example sentences it would be much less likely to have influence on its own or just with a premodifier – the postmodifier in the first two and the complementation around influence in the last appear to play a decisive role in the acceptability of the sentences, which would be difficult to model in valency theory since they do not appear to fill a valency slot of influence in the traditional sense. We thus managed to identify ‘suspicious’ cases for restrictions on subjects of some nouns discussed in the present section, but as for the first case study, we have to admit that the limited availability of data does not allow us to make any strong claims, even though the general tendency towards item-specificity observed in this study so far seems to be confirmed further.

Data Analysis | 157

6.3.4.1 Special case 1: of importance During the data analysis for the present case study, it turned out that while importance on its own does not allow any form of clausal subject, the phrase of importance allows all of them:257 (42) a.

*To encourage succinct writing about science is a great importance and these winners will make a real contribution to science communication. (invented) b. To encourage succinct writing about science is of great importance and these winners will make a real contribution to science communication. (newspapers) c. Yet finding out what happens in higher education is of utmost importance in understanding the patterns of gender inequality that exist. (FA6 114) d. That the eggs should be fresh, free-range and from a well-fed chook is of paramount importance. (newsmerge)

One could argue that of importance functions in a similar way to the adjective important in these sentences and that the construction should not be treated as a case of noun valency at all. Syntactically, of course, as most instances of the construction contain an adjective modifying importance,258 and as adverbial modifiers of the whole structure are not acceptable, we still have to treat the structure as a prepositional/particle phrase with a noun phrase complement: (43)

*To encourage succinct writing about science is very of importance.

At the semantic level, however, the premodifying adjectives fulfil the function of expressing degree in a similar way to degree adverbs (see CamG 720–725), often, as in the examples above, – but not always (see (44) below) – expressing a high degree of importance.

|| 257 The possible complements shown here make a good case for a large corpus, given that Kaltenböck in his study based on the 1-million-word ICE-GB did not find any instance of nonextraposed [to_INF] subjects with a PP predicate (Kaltenböck 2004: 153). 258 In the BNC, the sequence of importance occurs 410 times, the sequence of [adjective] importance 1753 times. In the data analysed for the present chapter, the numbers were skewed towards the latter construction even further. This is probably due to the fact that heavier postverbal elements boost the acceptability of sentences with non-extraposed clausal subjects.

158 | Analysis of subjects in copular clauses

(44)

Thus the premium need not necessarily be connected with the product that carried the premium; the idea is to stimulate purchases of the product — selling the premium is of secondary importance. (K94 1117)

While it is of course possible to follow Kaltenböck (2004), who treats such items as PP predicates, their peculiarities make any treatment in a traditional grammatical approach quite problematic and are a further indicator that linguists should regard meaningful units of whatever size as basic in the description of language very much in the way Sinclair (2004: 24ff) or construction grammar259 suggest. In Section 9.3.2 a model to deal with such cases in a valency framework will be discussed.

6.3.4.2 Special case 2: image Of all nouns examined for this chapter, image is the only one to frequently show the particular pattern exemplified in (45a). (45) a.

Hovering in front of her was the image of a large eagle. (fiction)

While the structure looks like a copular clause with a [V-ing] subject and a predicative element headed by image superficially, there is an alternative interpretation. For instance, the subject-related question seems to indicate that the postverbal element might be the subject since the alternative is infelicitous: b. ?What was the image of a large eagle? – Hovering in front of her. c. What was hovering in front of her? – The image of a large eagle. That in fact the postverbal element should be treated as subject in such structures can be observed in (46), where the verb is in the plural and thus shows concord with images and not with the [V-ing] clause preceding it. (As pointed out in Section 7.1.3.2, clausal subjects are usually treated as singular by concord.) (46)

Running through McClaren’s mind were images of England’s recent rugby prowess, as well as the determination paraded by his own play-

|| 259 While Croft/Cruse (2004: 265ff) are right to point out that construction grammar (they use lower case to refer to the whole set of construction grammars) is far from uniform, this statement is in line with any such approach, in particular with Goldberg (1995, 2006) and Fillmore/Kay/O’Connor (1988).

Conclusion | 159

ers since the first-half nadir against Andorra in Barcelona last season. (newspapers) We thus can safely assume that the given structure is an instance of some kind of inversion. Some sentences might be treated as instances of locative inversion, but this is clearly not true of the following one: (47)

Adding to the doubts is M Chirac's long-standing image as a political weathervane and an impulsive leader. (newsmerge)

CamG suggests the neutral term subject-dependent inversion, thus also acknowledging that “the subject occurs in postposed position” (CamG 1385). If we compare the inverted structures with their canonical counterparts, an interesting question arises as to what the verb in the sentence is: (48)

The image of a large eagle was hovering in front of her.

In (48), was hovering would be treated as verb by CGEL, or as VHCact by Herbst/Schüller (2008), which could be taken to mean in turn that the original sentence (45) contains a discontinuous verb in the progressive aspect. CamG calls the preposed element a VP (1386) even though the copula is located outside this ‘VP’.260 To sum up we can state that image attracts inverted structures in a way no other item in our test set does, which is a further indicator of the highly idiosyncratic behaviour of words that can be modelled best in theories which feature a substantial storage component.

6.4 Conclusion Overall, the analysis of subjects in copular clauses has yielded convincing evidence that highly item-specific restrictions are at work there, too. The present section will first give an overview of the results collected in the data analysis and then discuss theoretical implications and issues of modelling.

|| 260 CamG (1385f) shows that even though the inverted structures show progressive aspect morphology, they do not necessarily have a progressive meaning and thus not all such inverted sentences correspond to canonical progressive sentences.

160 | Analysis of subjects in copular clauses

6.4.1 Summary of results For extraposed subjects of adjectives, it was shown that there are the itemspecific restrictions summed up in the following table: Tab. 25: Summary of restrictions on extraposed subjects in copular clauses with adjectives as predicative elements

[to_INF]

[that_CL]

apparent



+

bold

+



careless

+



certain



+

civil

+



clumsy

+



complacent

+



conspicuous



+

early

+



easy

+



economic

+



exact

+



expensive

+



extreme

+



fruitful

+



fruitless

+



hopeless

+



indisputable



+

inept

+



manifest



+

notable



+

radical

+



safe

+



sentimental

+



smart

+



ungrateful

+



Conclusion | 161

It was also shown that similar restrictions exist for nouns, which are summed up here: Tab. 26: Summary of restrictions on extraposed subjects in copular clauses with nouns as predicative elements; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses

[to_INF]

[that_CL]

[V-ing]

comment

disgrace (A)

+



+

distinction of lexical units difficult for [V-ing]

disgrace (B)



+

+

distinction of lexical units difficult for [V-ing]

duty

+

o



goal

+

+



habit

+



+

instinct

+

o



invented [that_CL] example was accepted in canonical position

job

+



+

whether [V-ing] examples are same lexical unit is questionable

job (special use 1)



+



joke (special use 1)





+

judg(e)ment

+

+



need

+

+



opinion

+

+



role

+



+

task

+



+

tendency

+





will

+

+



only one example of [V-ing]

Even though the restrictions on the [to_INF] only hold if the distinction between different lexical units is kept,261 the columns for [that_CL] and [V-ing] indicate that there is item-specific variation. Thus both systematic studies on extraposed subjects in copular clauses come to similar conclusions based on a substantial test-bed of data. Even though the limited data available for the two case studies does not allow to

|| 261 See Section 6.3.2.1 above for the discussion of job and disgrace.

162 | Analysis of subjects in copular clauses

make strong claims about subjects in canonical position, the results do point in the same direction as the systematic studies. For adjectives starting with h and for nouns starting with i the most interesting cases are repeated here: Tab. 27: Summary of restrictions on clausal subjects in canonical position in copular clauses with adjectives as predicative element

[to_INF]

[V-ing]

[that_CL]

harmful

+

+



healthy



+

o

heartless

+

+

o

helpful

+

+

o

honest

o

+

+

honourable

+

+



hopeful

o

o



hopeless

+

+



horrific

+

+

o

hostile

+





hot



+

o

hungry







hypocritical

+

+

o

Tab. 28: Summary of possible restrictions on clausal subjects in canonical position in copular clauses with nouns as predicative element

[to_INF]

[V-ing]

[that_CL]

influence

o

+

+

insight



+

+

instrument

+

+



The fact that Table 27 contains more o-cells than the tables before is an indicator of the problems that arose in the case studies due to the limited availability of data. It is nonetheless likely that many of the restrictions identified in Table 27 hold, since the interpretation of data was done very carefully. For the nouns listed in Table 28, there are only two possible restrictions, so based on the pre-

Conclusion | 163

sent data we can neither confirm nor reject any hypotheses solely based on the case study of nouns. Overall, the results obtained in the present chapter can only be interpreted as strong evidence for a treatment that regards subjects of copular clauses as syntactically governed by the predicative element.

6.4.2 Theoretical implications The results of the data analysis are particularly relevant for the modelling of copular clauses in valency theory. While in the two previous chapters, we basically found the position of valency theory in general confirmed, the situation is more complex for copular clauses. If we recall the discussion of theoretical positions in Section 6.2, there is no general agreement as to the exact syntactic structure of such clauses. The model we propose here is in principle a modified version of Herbst/Schüller’s (2008) model. Their model, which makes use of valency and constituent structure (see Section 2.2.2), assigns copular and non-copular clauses identical syntactic analyses: (49) a. She killed a teacher. b. She is a teacher. Both examples can thus be described as follows in their notation (see Herbst/ Schüller 2008: 173–178): sentence: declarative-‘statement’ (49a)

She

killed

a teacher.

SCU: NP

PHU: VHCact:2

PCU: NPÆFFECTED

Fig. 17: Analysis of a non-copular clause in Herbst/Schüller’s (2008) model

sentence: declarative-‘statement’ (49b)

She

is

a teacher.

SCU: NP

PHU: VHCact:2

PCU: NPPREDICATIVE

Fig. 18: Analysis of a copular clause in Herbst/Schüller’s (2008) model

164 | Analysis of subjects in copular clauses

Thus the only difference between (49a) and (49b) is at the semantic level, where the postverbal complement takes different semantic roles depending on the verb. This analysis is similar to Helbig’s position discussed in Section 6.2, but the evidence collected in the present chapter challenges such a view since we found that predicative elements impose valency restrictions on the subject.262 For extraposed subjects, the relationship is less controversial, so Herbst (1983) or Herbst et al. (2004) would regard the underlined [that_CL] in the following structure as governed by the adjective clear: (50) a.

It is clear that the views of the unions were ignored. (ACH 400)

Whether and in what way this analysis could be combined with the view that it and the [that_CL] form one constituent together as proposed in Herbst/Schüller (2008)263 remains unclear. Here we shall instead propose a uniform treatment that analyses both extraposed and non-extraposed subjects as governed by the predicative element as a valency carrier. This does not imply any change in the constituent structure part of the clause. A first approximation may look as follows: (50) b. That it represented a threat was clear enough. (ANU 1264) clause

subject

predicate

constituent structure predicator predicative element

That it represented a threat was clear enough. valency structure Fig. 19: Possible structure for non-extraposed copular clauses

The valency structure in this particular example only consists of a link from clear to the [that_CL];264 the copula verb does not even figure in the valency structure. The reason for such an analysis is – as already mentioned the theoret-

|| 262 For adjectives we have reasonable evidence, for nouns the same behaviour is to be expected. 263 See the discussion in Chapter 7. 264 One may also posit a valency relationship to enough, but an analysis as adjunct to clear is preferred here.

Conclusion | 165

ical introduction – that no formal valency restrictions whatsoever are imposed by the copula be. Similar to the argument that in sentences with meteorological verbs the preverbal it is not a complement of the verb and just supplied because of the necessity for English clauses to have a subject, we can claim that the copula is not a complement of the valency carrier (the predicative element) in such clauses and that it is simply supplied because of the necessity for English clauses to have a verb.265 There are three disadvantages to such an analysis. The first and minor one is that the parallel treatment of copular and non-copular clauses with the verb as the highest-ranking element is lost. The second and possibly more serious problem is that there are sentences of the type given in (51a), which are only reasonable in a context where further information is known, for instance as an answer to (51b). (51) a. Yes, I am. b. Are you sure you want to talk about it? (A7J 1679) If we regard sure as the valency carrier in (51b), what is the valency carrier in (51a)? One may of course argue for an elliptical treatment in which the valency carrier is omitted because it is known, but assuming an underlying structure to be able to explain an actual structure is not very appealing for a surfaceoriented model with a strong focus on usage. Furthermore, there are similar structures discussed by Herbst (VDE xxxi): (52) a. Does Hannah know? (VDE xxxi) b. Where is he now? Does Hannah know? (VDE xxxii) Again, (52a) is only possible in a context such as the one given in (52b) and this is reflected in the model:266 Hence complements such as the wh- and that-clauses in (17) c. But you know how folks are here in Zennor. d. Did you know they were here? can be described as contextually optional complements to indicate that they are only optional if their referent can be identified from the context. (VDE xxxii; their emphasis)

One may thus argue that it is desirable to treat cases such as (51a) and (52a) in the same way in a syntactic model and, since the complement status of the || 265 Word grammar (Hudson 2010: 293f) also regards the subject as the subject of the predicative element. 266 See also the discussion of optionality in Section 2.2.2.

166 | Analysis of subjects in copular clauses

‘missing’ element in (52a) is hard to deny, the ‘missing’ element in (51a) should be treated as a complement as well. Furthermore, different copula verbs show different behaviour in this respect, so the item-specific (i.e. valency) character can hardly be doubted:267 (53) a. She is very nice. b. She really is. (54) a. She seems very nice. b. *She really seems. The third and most convincing argument against the simple structure presented in Figure 19 is the quantitative valency of be. One can argue that the copula must be the primary valency carrier in the clause since it is divalent268 and thus blocks the occurrence of additional complements such as John in the following example. (55) a.

*That it represented a threat was John clear enough.

Semantically, there is absolutely no reason to reject (55a), and a parallel structure is in fact possible in German in which the underlined element corresponds semantically to the [to_NP] in (55b), which is usually analysed as a complement of clear (VDE 138). (55) b. That it represented a threat was clear enough to John. (56)

„Nur dass es ziemlich zur Sache geht“, war Gao klar, der zuvor für eine Universitätsmannschaft im Fußballtor stand. (taz, 24 March 2007)

One may of course argue that the valency of clear simply does not allow such an [NP] complement in English. However, the fact that no English adjective has this valency slot while verbs often allow a postverbal [NP] after the verb (even with an adjective following as in the case of so-called ‘complex transitive’ structures) speaks for a model in which the copula exerts the restriction, even though we have seen at the beginning of the chapter that no formal restrictions whatso-

|| 267 Since the regular form of (54b) is She really does., one may argue that is in (53b) is a proform for is in (53a), but since the pro-form is only takes up instances of the verb is, such a treatment seems unnecessarily complicated and could only be justified with the help of metatheoretical considerations such as symmetry. 268 For exceptional cases of trivalent use see the entry of be in the VDE.

Conclusion | 167

ever are enforced by be, i.e. that qualitative valency does not seem to restrict the occurrence of complements. We can thus modify our model to account for these valency restrictions and add the additional valency relations:269 clause

subject

predicate

constituent structure predicator predicative element

That it represented a threat was clear enough. valency structure Fig. 20: Modified structure for non-extraposed copular clauses

Such a treatment with multiple incoming edges also allows for an elegant treatment of ‘complex transitive’ structures (an issue raised in Chapter 5, in particular in Section 5.2.2.2) and of so-called tough movement structures to be discussed in Chapter 7. Section 9.3.2.2 will show how these structures can be modelled according to the same principles in a valency model and thus present more evidence for the solution suggested here.

|| 269 Theoretically, a structure in which the subject is only the child of one element might be possible, too, but in that case one would either have to posit an inheritance relation where be inherits the restrictions from clear and imposes them onto the subject (which requires a more complex model and thus has to be rejected, other things being the same) or one could have a complex head of the clause, consisting of the verb and the predicative element, within which we would have to posit valency relations as well.

7 Extraposition Extraposition is a term used for structures such as that in (1a), where an element (usually a subordinate clause) is found towards the end of its superordinate clause and it occurs in the position the extraposed element occupies in canonical structures such as (1b).270 (1)

a. It was stupid to barge in here like this. (JYC 4090) b. And to treat him like the prodigal son was stupid. (CBN 2328)

The hypothesis presented in Section 1.3, i.e. that subject extraposition might be lexically determined, may appear surprising at first sight, given its traditional treatment in terms of a (“usually optional”) rule, as also proposed for the generative framework by Rosenbaum (1967: 6). However, the table found in Appendix 26, the top of which is reproduced here, seems to suggest otherwise. It gives the distribution of extraposed and non-extraposed [to_INF] subjects in the VDE for all verbs that allow this form of subject in the active voice: Tab. 29: Extraposition and non-extraposition of [to_INF] subjects with active verbs according to the VDE; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses

extraposed

non-extraposed

amuse(A)

+

+

annoy()

+

+

appear(B)

+

+

attract(A)

+

+

become()

+



begin()

+



bring(B)

+



buy()

+



call(F)

+



cause()

+



|| 270 Extraposition is not limited to subjects and there are further uses of the term in generative grammar, but only subject extraposition is covered in the present study. Again, the study is limited to English. For a recent cross-linguistic survey of such structures see Schmidtke-Bode (2014, Section 5.5.). https://doi.org/10.1515/9783110589801-189

170 | Extraposition

extraposed

non-extraposed

come(F)

+

+

commit(A)

+



compare(B)

+



concentrate(A)

+



confirm()

+



cost(A)

+

+

cost(B)

+



count(B)

+



decide(alpha)

+



delay(B)

+



demand(A)

+



discourage()

+



do(alpha)



+

ensure()

+



excite()

+



fall(E)



+

feel(C)

+



fill()

+



form()

+



free()

+



Even if only some of these restrictions hold, we will have to treat extraposition as an item-specific phenomenon and list its occurrence in the lexicon of a valency model. There is a second purpose to the present chapter: The last of the hypotheses presented at the beginning of this book (see Section 1.3) claims that the term subject covers an array of properties that happen to coincide in one constituent in canonical clauses but less so in non-canonical clauses. We would thus expect the subject properties discussed in Section 2.3.2 to perhaps cause problems when applied to sentences showing subject extraposition and will discuss this issue in detail in Section 7.1.3.

7.1 Theoretical treatment This section is devoted to the treatment of extraposition in various approaches to syntax. Such a discussion is necessary because the models differ greatly from

Theoretical treatment | 171

one another in their analysis, in what counts as extraposition and what does not, and because the theoretical implications drawn from the analysis of course depend on our theoretical treatment of the phenomenon. In most recent grammars of English, the term extraposition is used for a “[p]ostponement which involves the replacement of the postponed element by a substitute form” (CGEL: 1391), as in the following example: (2)

a.

It was clear that she had been worrying. (HP7.535)

The use of the word replacement in the above citation seems to suggest that Quirk et al. see sentence (2a) above as derived from sentence (2b) below, which is backed by their use of the formulation “is moved to the end of the sentence” (CGEL 1391).271 (2)

b. That she had been worrying was clear.

As pointed out by Kaltenböck (2004) from a functional perspective treating the non-extraposed form as basic may be misleading, since it is much less frequent than the extraposed form “and therefore has to be taken as the statistically marked construction of the two” (Kaltenböck 2004: 2). Even though the definition given by Quirk et al. is relatively general, only it is mentioned as the substitute form in the subsequent paragraphs. We shall follow this practice in the present study even though there are other researchers who have argued for a wider range of possible anticipatory elements. For instance, Mindt (2007: 105), who carried out an extensive study of adjectives followed by a [that_CL], prefers the term non-intentional subject in order to cover a range of problematic cases.272 But as the pronoun it accounts for 96.9 % of her non-intentional subjects and as the other examples are not straightforward cases of extraposition, the distinction is of limited use here. Similarly, for Jespersen the term extraposition has a wider scope than for most grammarians nowadays, as it includes instances of dislocation (see post-

|| 271 Aarts uses a similar terminology in that he identifies two stages in the process of extraposition: “In the first stage the content clause is displaced to the end of the matrix clause, after which anticipatory it [...] is slotted into the matrix clause Subject position” (Aarts 2011: 183). He goes on to state, however: “I present extraposition in this way only for expository purposes. There is no suggestion that the two stages are processes which take place in the mind” (Aarts 2011: 183). 272 Non-intentional subjects cover cases such as “The Bible is quite clear that these evil spirits (and the things that they do) are dangerous.” (Mindt 2007: 105) which do not contribute relevant insights to the present chapter.

172 | Extraposition

poned and anticipated identification in CGEL 1310 and the discussion in Section 7.1.1 below) as well as cases where the extraposed element is represented by a pronoun other than it in the “sentence proper” (Jespersen 1933: 95). Extraposition in the narrower sense in which it is most often used today is treated under the heading of preparatory it: But it has also another very important function, namely as “preparatory it” to represent a whole group of words which it would not be convenient to put in the place required by the ordinary rules of word-order without causing ambiguity or obscurity. The group itself (an infinitive with its complements, a clause, etc.) then comes afterwards in “extraposition.” (Jespersen 1933: 154)

Jespersen’s reason for extraposition, i.e. to avoid “ambiguity or obscurity”, is noteworthy and at least partially compatible with processing accounts presented in Section 7.3.1 below. In the context of noun phrase extraposition illustrated in (3), no indication as to what the actual cause of a preference of this structure over the canonical structure might be is given.273 (3)

It is strange the number of mistakes he always makes. (Jespersen 1933: 155)

Jespersen treats [to_INF] subjects as subjects irrespective of whether they are extraposed or not. No indication as to the factors influencing the choice of either construction is given. The same applies to the case of [for_NP_to_INF] (Jespersen 1933: 343f). Thus Jespersen’s preparatory it corresponds more closely to extraposition as we understand it in the present study than his broader concept of the term extraposition, which is, however, relevant for the interpretation of Erdmann’s results cited below.

|| 273 Besides the length and structural complexity, concord may play a role in this particular example as the canonical structure would require a singular verb near a phrase that expresses plurality while being in the singular. This can indeed lead to confusion. (See, for instance, The number of inquiries are higher than in any other month since the branch opened in November 1991. [K55 4087] as opposed to the much more frequent structure with singular verb The number of injuries is actually going down in proportion to the increased number of skiers. [AHC 1117].)

Theoretical treatment | 173

7.1.1 Extraposition vs. dislocation 7.1.1.1 The problem Some grammatical models, such as CamG, make a distinction between extraposition as in (4a) and (right)-dislocation as in (4b): (4)

a. It is no joke coming home at night to an empty house. (newsmerge) b. It is no joke, trying to build a new National Theatre. (newsmerge)274

Superficially, both sentences look very similar with it as subject, no joke as predicative element and a [V-ing] clause following. The only reason for which these are treated as instances of different constructions is the comma between joke and the [V-ing] clause in (4b). According to CamG the comma275 in writing represents an audible prosodic boundary: In speech the two constructions are strikingly distinct prosodically. Unlike the extraposed constituent, the dislocated constituent almost invariably constitutes a distinct intonational phrase and is separated from the nucleus of the clause by an intonational boundary. (CamG 1414)

The spoken character is important since dislocation “is often found in oral personal narratives” (CamG 1408). CGEL claims that the [V-ing] is “resistant to bearing the main information focus” so that the main focus is “on the first element of the predicate” (CGEL 1393): (5)

It’s |FÙN being a HÓSTess| (CGEL 1393)

However, Mair, whose corpus contains roughly 50 % spoken material, notes that “neither of these two prosodic criteria proves very helpful”276 (Mair 1988: 56) for the distinction between extraposed and dislocated [V-ing] clauses since “right-dislocation structures are not always separated from the rest of the clause by a distinct intonational break” (Mair 1988: 56). || 274 (4b) is difficult to assess without context. It is the first sentence of a commentary in the Irish Independent and it becomes clear that it has to be analysed as an instance of rightdislocation if we look at the entire paragraph: It is no joke, trying to build a new National Theatre. Apart from the brutal logistics, one is at the mercy of events, dear boy, events, in this case the unexpected offer of a waterfront site owned by the Dublin Docklands Development Authority. (newsmerge) 275 CamG (1414) points out that a dash also commonly fulfils the same function. 276 Mair does of course not refer to CamG in his discussion but CamG’s position appears to be basically that of Postal’s (1974) study.

174 | Extraposition

CamG (1413f) identifies two further properties in which the two constructions differ. The first is the status of it, which they claim is a non-referential dummy in (4a) and an ordinary referential pronoun in (4b). If we accept this line of argument, then it should be possible to replace it in (4b) with another pronoun such as this – similar to the full range of personal pronouns that is possible with right-dislocated [NP]s: (4)

c.

?This is no joke, trying to build a new National Theatre.

However, (4c) is only of limited acceptability, even in spoken language, so the referential character of the pronoun in dislocated structures is at least doubtful, as long as we treat (4b) as a clear case of dislocation, which CamG would have to do due to the intonational boundary represented by the comma after joke. The second difference between the two constructions according to CamG is “in the information status of the right-dislocated and extraposed constituents. [... T]he right-dislocated phrase is required to be discourse-old, whereas the extraposed constituent may be discourse-new” (CamG 1414). The property is illustrated with the help of the limited acceptability of a dislocated noun phrase in which the indefinite article is an indicator of the discourse-new status: (6)

#

It’s really interesting, a book I’m reading. (CamG 1414)

In the case of [V-ing] clauses, the discourse-new status is not available as readily from their form and it may be argued that often a clause by its very nature contains both old and new information, although possibly less so for typical subjectless [V-ing] clauses. Kaltenböck (2004) also distinguishes the two constructions based on their communicative function. CGEL, on the other hand, does not draw the distinction between extraposition and dislocation (CGEL 1392f) for [V-ing] clauses. Instead, Quirk et al. classify both variants as instances of extraposition and state that “[e]xtraposition of -ing clauses is in fact uncommon outside informal speech” (CGEL 1393),277 “that the participial clause has just as much affinity with a noun-phrase tag278 (cf 18.59; eg: He’s a friend of mine, that man) as with a genuine extraposed subject” (CGEL 1393), and that the structure could be punctuated with or without a comma.

|| 277 Since CamG (1188f) does not mention any correlation to register or medium for extraposition, this might only apply to the structures CamG would analyse as instances of dislocation. 278 Right-dislocated [NP]s are called noun-phrase tag by CGEL.

Theoretical treatment | 175

Furthermore, Mair notes that – contrary to what CGEL (1393) claims – extraposed [V-ing] clauses are not rare or marginal (Mair 1988: 54) and that, while they show a general affinity towards spoken informal language, their acceptability in non-informal language is highly item-specific (Mair 1988: 59). The most important claim made by Mair is that he discards the distinction between extraposition and right-dislocation (and in many cases non-finite adverbial clauses) altogether and opts for a solution that treats them as instances of an indeterminate construction in the sense of Matthews (1981: 17–21).

7.1.1.2 Evidence Mair’s position is supported by CamG’s observation that “with [extraposed] gerund-participials speaker judgements vary” (CamG 1189). It is also confirmed by the corpus research and native speaker interviews carried out for the analysis presented in Section 6.3.2.2, in which the distinction between extraposed and dislocated [V-ing] clauses proved impossible to maintain: The native speaker informants not only disagreed as to the acceptability of corpus examples, they also disagreed strikingly about the necessity of an intonational boundary before the [V-ing] clause, i.e. in their classification of the element as dislocated or extraposed in CamG’s terms. For instance, the three native speakers of British English interviewed on the issue of (7) below all accepted it, but one only on the condition that the intonational boundary between fun and playing be present, one only on the condition that it be absent, and one accepted both variants. (7)

They think it is just harmless fun, playing on the ice-covered reservoir. (newsmerge)

On the other hand, there are clear-cut instances of dislocation such as (8a), for which all native speakers agree on the necessity of an intonational boundary between bet and producing, and there are cases of extraposition, where all informants accepted the sentence without the intonational boundary, such as (8b). (8)

a.

You know, I can’t blame John Travolta. It would have seemed a safe bet, producing and starring in a major science fiction epic with all the high-tech hullabaloo that was afforded to The Phantom Menace, The

176 | Extraposition

Matrix, and all those other major science fiction epics. (academic; slightly modified)279 b. Rugby Football League spokesman Dave Callaghan said: ‘It is a blow losing a sponsor of the quality of Stones Bitter but work has already started on finding somebody who will have a long-term commitment to the exciting principle of Super League.’ (newsmerge) Not having to distinguish between the two structures is probably the only viable alternative if one does not want to introduce artefacts into the analysis. So far, we have only looked at [V-ing] clauses, simply because they occur in both extraposition and in right-dislocation, if one distinguishes the two. One may argue, however, that [NP]s also show both variants, though extraposition to a lesser extent: “NPs generally cannot be extraposed, but there is a very limited range of types that can: [...] The NPs concerned all have the form ‘the ... N + relative clause’” (CamG 1407f). The following example is an instance of the relevant structure: (9)

It’s extraordinary the amount of beer he puts away. (CamG 1407)

Accordingly, example (10a), in which the [NP] takes a different form, would not count as an instance of extraposition, whereas (10b) clearly shows extraposition: (10) a.

It’s stressful this managing lark, you know, I’m starting to think I was a bit hard on Wood sometimes. (HP5.248) b. It’s stressful to manage the team, you know, I’m starting to think I was a bit hard on Wood sometimes.280

Again, CamG would probably make a distinction here and classify (10a) as a case of dislocation and not of extraposition, since the underlined element in (10a) is analysed as “a sort of ‘afterthought’” CamG (1411). In both examples, the underlined element corresponds to the sentence-initial it and can replace it to form the canonical sentence. An analysis of (10a) as an instance of extraposition

|| 279 The original reads “would seem” instead of “would have seemed”. One native speaker informant objected to the aspect and as it makes no difference to the argument, the example was modified accordingly. 280 Example (10b) is slightly infelicitous and might be rephrased either using would as a modality marker (It would be stressful to manage the team) or with a [V-ing] in extraposed/dislocated position (It’s stressful(,) managing the team). The argument is still valid, though.

Theoretical treatment | 177

is thus not excluded. By analogy, and following Jespersen, whose extraposition has a wider scope, one could also treat (11) in the same way. (11)

He’s a funny man, Dumbledore. (HP1.219)

The reason why this approach is not advocated in the present study is that the pronoun he in (11) is not as semantically neutral as it is in (10b) with the extraposed [to_INF]. For postposed [NP] complements, the status of it is more difficult to decide. Thus if we extend (12a) below with an afterthought to form (12b), the term extraposition as defined above might be less suitable due to the referring character of it, which could be said to refer to the real world object just as well as to the postponed [NP]: (12) a.

It shows us nothing more or less than the deepest, most desperate desire of our hearts. (HP1.157) b. It shows us nothing more or less than the deepest, most desperate desire of our hearts, this mirror.

We thus can conclude that for [NP]s, much speaks for dislocation in the majority of cases and that extraposition is rare. But again, the distinction is difficult to draw.281 Finally, [that_CL] and [to_INF] clauses “are rarely right-dislocated, though not wholly excluded from the construction” (CamG 1414). Due to their very low frequency (especially in the mostly written corpora used for the data analysis in this study), dislocated [that_CL] and [to_INF] clauses will not be discussed in the present chapter, even if it were possible to distinguish them from extraposed ones at all.

7.1.1.3 Conclusion for the present study As one of our hypotheses in this study is that subject extraposition is itemspecific, we would expect not to find it with all lexical items alike. No hypothesis can be made at this point as to whether dislocation, as treated by CamG, is a generally applicable mechanism or might be regarded as item-specific, so treating both right-dislocation and ‘genuine’ extraposition as instances of extraposition – as CGEL does – may result in additional noise in the data. It is for this

|| 281 CGEL also used the term extraposition for clausal elements and states that dislocation (“postponed and anticipated identification” in their terminology [CGEL 1310]) is “restricted to informal spoken English, where it is very common” (CGEL 1310).

178 | Extraposition

reason that it would of course be desirable to follow CamG in keeping extraposition and dislocation distinct in the present study where possible, but, as noted above, the distinction cannot be made for [V-ing] clauses and is basically irrelevant for [that_CL] and [to_INF] clauses. Thus we are faced with the same dilemma as Mair and have to come to the same conclusions: The distinction between right dislocation of constituents and extraposition of clausal subjects is real enough in many areas of English syntax. Operating with this distinction in the present case, however, is both facile and misleading – facile because, in disregarding the potential for overlap with adverbial -ing-clauses, it obscures the true complexity of the empirical data, and misleading because in some instances the analyst will be forced to make a distinction where none exists in the data. (Mair 1988: 59)

7.1.2 Obligatory extraposition There are clauses which show the same structure as the extraposed sentences illustrated above but which do not have a non-extraposed counterpart: (13) a. It appears that he is now employing Occlumency against you. (HP6.61) b. *That he is now employing Occlumency against you appears. Typical verbs that occur in such structures are seem, appear, chance, happen and turn out (CamG 960). It is obvious that a definition that relies on the replacement of an element by it cannot be used to describe (13a) unless one wants to argue that there is some sort of underlying representation in the form of (13b) which is altered by a replacement rule. Since the phenomenon is clearly itemspecific, one could take this behaviour as evidence for treating extraposition as a valency phenomenon. CGEL (1392) takes a similar approach and treats such cases as instances of obligatory extraposition. However, Huddleston/Pullum (CamG 67, 241f, 960ff) exclude sentences such as (13a) from their treatment of extraposition due to the unacceptability of (13b). Similarly, Halliday states that if there is extraposition, “there will be a marked variant with the clause Subject at the beginning” (Halliday 1994: 98), which would imply that (13a) does not show extraposition. In terms of valency theory, the case of (13) represents an interesting problem that will be discussed in detail in the remainder of this section. We can hypothesize that appear needs two further elements to occur in a clause in order for the sentence to be grammatical. These could be it and the extraposed ele-

Theoretical treatment | 179

ment, but we could also add another element (e.g. a predicative adjective) to (13b) in order to form the relatively straightforward282 (13c): (13) c.

That he is now employing Occlumency against you appears obvious.

Our hypothesis is, however, contradicted by the fact that we cannot rephrase (13d) below in terms of (13e) even though another element is present. (13) d. It appears to me that he is now employing Occlumency against you. e. *That he is now employing Occlumency against you appears to me. So since there are two complements in (13e) quantitative valency cannot be the reason for its unacceptability and consequently we cannot claim that quantitative valency must be the reason for the unacceptability of (13b) above. Thus we have to find other ways of modelling such behaviour. Matthews283 gives the following description of the structure of extraposed sentences such as (2a): (2)

a.

It was clear that she had been worrying. (HP7.533)

[T]he predicator clear takes a single complement, the clause [that she had been worrying]. This could, in principle, supply the subject [...]. But the more usual pattern is with the clause removed outside the subject-predicate relation [...]; this leaves no subject and it is supplied. (Matthews 1981: 276)

He makes an important point that is less pronounced in the approaches mentioned so far: the anticipatory it does not fill a valency slot of the predicator in any traditional model.284 If we adopt a more usage-based, surface-oriented perspective, Matthews’ argument would be more difficult to sustain. If we accept that it does not fill a valency slot, appear still only has one complement in (13a), which seems to be enough to satisfy its valency requirements. Thus, if only one complement is required by the valency carrier and if this complement is in canonical subject position as in (13b), we would expect || 282 (13c) is slightly infelicitous due to the long subject in canonical position, but this effect could be reduced significantly by adding a conjoined clause such as and we have to think very carefully what we should do about it. 283 Although Matthews does not identify himself as a valency grammarian, much of his model is compatible with valency theory; see Section 2.2.2.3. 284 Herbst et al. (2004) seem to adopt the same line of argument as they indicate obligatory and optional extraposition as a property of the complement in the complement block of verb entries in the VDE; in adjective entries they use square brackets around it in order to indicate its non-complement status.

180 | Extraposition

the sentence to be grammatical. As it is not, there are two alternative ways of modelling this fact. First, one could reject the position that it does not fill a valency slot and maintain the opposite. While this is certainly possible, it leads to additional problems in that it does not have an identifiable referent and likely does not fill an argument slot of the verb. The second option is to posit obligatory extraposition in the case of appear in (13a). If we treat extraposition as a property of the complement itself, as in the VDE,285 we would have to argue that the obligatorily extraposed [that_CL] in (13a) is not the same complement as the optionally extraposed [that_CL] in (13c), since the latter does not show such a restriction. Such a treatment can be observed in the relevant entry for appear in the VDE:

Fig. 21: Complement block of appear (VDE 40)

According to the VDE model, the [that_CL] in (13c) is not governed by the verb appear but by the predicative adjective obvious. Given that extraposition is more natural for such sentences, the VDE only lists an extraposed variant as [it + pattern of II], which means that an element can occur as extraposed subject of the sentence if the [ADJ] complement listed in valency slot II has a valency slot for this element (see also the discussion in the previous chapter). There is ample corpus evidence to show that the non-extraposed variant exists as well, as in (14) below, so the form of the complement as [it + pattern of II] may be misleading and one could rephrase it as [pattern of II]A(it) in the VDE style.

|| 285 Such an interpretation is suggested by the notation used but may of course be no more than a descriptive device.

Theoretical treatment | 181

(14)

That the arbitrageur can lend money at the riskless rate of interest appears sensible [...]. (FSA 246)

For the obligatorily extraposed variant shown in (13), the VDE’s analysis is particularly interesting since it treats the [that_CL] as one complement that is obligatorily extraposed and represents two arguments of the verb.286 The reason behind this is the parallel we can observe between the following two sentences: (15) a. He appears to fancy me. (adapted from the VDE 40) b. It appears that he fancies me. The verb appear in (15a) takes two complements (he and the [to_INF] clause), which are analysed as representing one argument each in the VDE.287 We can then state that the semantic relationship between he, appear and fancy is the same in both sentences in (15). In (15b), appear adds a meaning component of epistemic modality to what is expressed in the [that_CL] and one could indeed argue that it does exactly that in (15a), which corresponds roughly to (15c): (15) c.

He apparently fancies me.

In that respect the case of appear is similar to certain adjectives288 in that one could see the same relationships between (15a–c) as between (16a–c).289

|| 286 For seem, no such analysis is offered. 287 In this respect, the VDE differs from Jespersen’s and the generative analysis mentioned in the following footnote in that the two complements would have to be analysed as representing only one argument in these approaches. 288 Jespersen offers a similar analysis and proposes a split (notional) subject in the following examples: He seems to be all right. (1933: 343) They would be certain to miss him. (1933: 343) She is not likely to be up so early. (1933: 343) Erdmann (1987: 57ff), for instance, follows Jespersen in this respect (and indeed even seems to transfer Jespersen’s notional concept to the syntactic level). In the generative framework, such structures are treated as examples of so-called raising, since the subjects in (15a) and (16a) are analysed as being ‘raised’ from an inferior position in the subordinate clause to the subject position in the main clause. Accordingly, appear is called a raising verb and certain a raising adjective. See for instance Haegeman (1994: 319f). 289 The situation becomes slightly more complicated in the case of the adjectives through the introduction of will in (16b) and (16c) to express the futurity inherent in (16a). For another similar phenomenon see the discussion of examples (23a–c) in Section 5.2.2.2.

182 | Extraposition

(16) a. He is certain to turn up. b. It is certain that he will turn up. c. He will certainly turn up. If we regard appear purely as a marker of epistemic modality which happens to be in the form of a verb, we might have to regard he in (15a) as a complement of fancy and not necessarily of appear; appear would instead have everything else in the clause as one (discontinuous) complement, similarly to the situation in (15b). The analysis of appear as a modality marker can also be backed up by looking at the copula be, which cannot occur in this construction on its own as in (17a) but can as soon as we add a modal verb to express the epistemic modality as in (17b). (17c) illustrates that the same obligatory extraposition takes place here as in the case of appear and seem. (17) a. *It is that she no longer trusts you. b. It may be that she no longer trusts you. (CGEL 1392) c. *That she no longer trusts you may be. Thus one might be inclined to regard seem and appear as copular verbs with a ‘built-in’ modal verb. While the fact that modal verbs can still occur with the two (illustrated in (18)) argues against an analysis of the two as actual modal verbs, given that as a rule modal verbs cannot be combined (see (19)), we maintain the analysis since some of the verbs classified as semi-auxiliaries by CGEL (143–146) do in fact also combine with modal verbs and we would of course not want to claim that appear or seem are prototypical modals syntactically. (18)

At first sight it may appear that all the children are doing the same thing but within each group allowance is made for ability and the children follow their own carefully planned programme at their own pace — this is updated regularly. (APJ 174)

(19)

*I tell you what we might should do. (Kortmann 2006: 609)290

There is another reason given by CamG to show that instances of it followed by appear or seem should not be treated as cases of extraposition:

|| 290 As the present study tries to model Standard English (see Section 3.1 for the problems associated with the term), (19) is asterisked even though it is acceptable in certain dialects of the southern United States (see Nagle 2003) and Scottish and Tyneside English (Kortmann 2006: 609).

Theoretical treatment | 183

(20) a. It seemed that he was trying to hide his true identity. (CamG 961) b. It was later confirmed that he was trying to hide his true identity. (CamG 961) c. *It seemed and was later confirmed that he was trying to hide his true identity. (CamG 961) Example [(20b)] is a clear case of the extraposition construction: if [(20a)] also belonged to this construction it should be possible to coordinate the contrasting parts seemed and was later confirmed. However, [(20c)] shows that this can’t be done, suggesting that the content clause does not have the same function in the two cases. (CamG 961)

There are two flaws in the argument presented here. First, the sentence is much more acceptable if we use the [that_CL] in a coordinated structure with a repeated it as in (20) d. It seemed and it was later confirmed that he was trying to hide his true identity. There is no reason to arbitrarily claim that only the contrasting elements should be coordinated and we can see in (21b) that we should still expect unacceptability in (20d) if the that-clause really had different functions in (20a) and (20b), regardless of the presence of the repeated pronoun:291 (21) a. *She is and loves a teacher. b. ??She is and she loves a teacher. The second problem is the assertion that one should be able to coordinate the contrasting elements if both sentences were instances of the same construction, i.e. extraposition. The examples in (22) definitely show extraposition: (22) a. It is shocking that such a thing could happen. b. It is disgraceful that such a thing could happen. c. It shocked me that such a thing could happen. If we want to combine (22a) and (22b) by means of coordination, there are three theoretically possible ways, given in (23a–c):

|| 291 One could argue that the repeated it might be necessary due to it having two different functions in (20a–b), but the fact that both partial structures can share the same [that_CL] in (20c) should be sufficient to refute CamG’s argument because the [that_CL] still seems to have the same function given that the coordination is acceptable.

184 | Extraposition

(23) a. It is shocking and disgraceful that such a thing could happen. b. *?It is shocking and is disgraceful that such a thing could happen.292 c. It is shocking and it is disgraceful that such a thing could happen. The coordinated adjectives in (23a) and the coordinated matrix clauses in (23c) lead to acceptable sentences whereas the coordinated predicates in (23b) are at least clumsy. But as (23a) is the one that coordinates the contrasting parts, the findings could still be seen in line with CamG’s analysis. If we look at (23d), however, we can observe that the coordination of the contrasting elements does not lead to an acceptable sentence, even though both original sentences show extraposition: (23) d. *It shocked me and is disgraceful that such a thing could happen. Again, the version with repeated it is much more acceptable: (23) e.

It shocked me and it is disgraceful that such a thing could happen.

So it appears as if the restriction CamG observed is not due to extraposition but follows a more general principle that forbids coordination of verb phrases – or at least of differently structured verb phrases with an extraposed subject. However, there are (rare but acceptable) examples of such coordination, so the exact nature of the restriction remains mysterious: (24)

It is pretty straightforward and follows Johnny’s philosophy that a race driver is “a simple animal, and doesn’t have to be a rocket scientist.” (www.ericjacobsen.org)293

The last argument presented by CamG against the obligatory extraposition analysis is the case of as if-clauses. CamG (962) points out that all [that_CL] complements in what they call the impersonal construction with appear or seem allow an as if-clause as an alternative. Thus (13a) can be modified to yield (13f) without loss of acceptability.

|| 292 In the non-extraposed variant such coordination of VPs is also problematic: *?That such a thing could happen is shocking and is disgraceful. 293 This example may also be used to argue that the extraposed subject is not VP-internal. Or it may not be accepted as a valid example because – if we insist on the VP-internal status of the extraposed subject – it is not an instance of basic coordination in the sense of CamG (238).

Theoretical treatment | 185

(13) a. f.

It appears that he is now employing Occlumency against you. (HP6.61) It appears as if he is now employing Occlumency against you.

This is not true of optionally extraposed sentences such as (25a): (25) a. It surprised some that he did not retire to Cambridge. (A68 30) b. *It surprised some as if he did not retire to Cambridge. Thus, given that as if-clauses never occur as subjects in canonical position, an extraposed subject analysis may appear infelicitous, but the VDE opts in favour of this solution. As it would generally be desirable to treat (13a–b) as parallel, (13a) should not be treated as an instance of extraposition, CamG claims,294 but of course the VDE also treats them in parallel as instances of obligatory extraposition. However, the question in the VDE approach is where to stop: CGEL, while it does not agree with CamG on appear or seem, seems to be in line with its analysis when it comes to further superficially similar constructions: If- and when-clauses behave very much like extraposed subjects in sentences like: It would be a pity if we missed the show. [cf It is a pity that we missed the show] It’ll be a great day when you win the sweepstake. It is doubtful in each case, however, whether the clause could act as subject, although it could act as initial adverbial clause: If we missed the show, it would be a pity (cf 10.9ff). On balance, therefore, these appear to be adverbials rather than extraposed subjects. (CGEL 1392; their italics)295

In the VDE, the [if_CL] is analysed as an extraposed subject, albeit it is only listed for the verb surprise as in the following example:296 (26)

It wouldn’t surprise me if she quit. (VDE 834)

|| 294 CamG backs up the analysis with what they treat as another instance of the same impersonal construction as in It looked as if he was trying to hide his true identity. (CamG 962), which does not allow that instead of as if and is thus even more clearly not an instance of extraposition. 295 For German, Oppenrieder comes to a similar conclusion and states that such expressions have a ‘taste’ of subject or object respectively when extraposed. When they are found at the beginning of the sentence, they require the addition of a ‘real’ subject or object (Oppenrieder 2006: 907). 296 The same construction is, however, generally subsumed under the heading of [wh_CL] in the VDE as with matter: It doesn’t really matter if you are wrong. (VDE 526; see also VDE xvii)

186 | Extraposition

The when-clause is subsumed under the [wh_CL] category and as such is also listed as extraposed subject. The extraposed subject analysis in the VDE even extends to items such as [about_NP] as for matter: (27)

It doesn’t matter about the bed. (VDE 526)

In such cases it becomes clear that from a semantic point of view an analysis in terms of an impersonal construction with [about_NP] as an additional complement may be preferable over the extraposed subject analysis. There is another class of sentences which look as if they are instances of obligatory extraposition due to the fact that no canonical variant is readily available: (28) a.

It emerged that across the very varied fleets the cost of asbestos removal would be £25,000 per vehicle, an investment producing no financial return. (A11 974) b. It emerged that not only was there no reliable evidence of guilt, but, on the contrary, there was considerable proof of innocence. (A1G 104)

However, the reason behind their apparent inability to be ‘transformed’ into a canonical variant is the complex structure of the [that_CL] complement, which is related to weight and processing constraints (see Section 7.3.1 below). That emerge does allow a [that_CL] in canonical subject position is illustrated in (28c):297 (28) c.

That this is the correct interpretation of the role of words in trusts emerges more clearly in the next section. (B2P 210)

We shall thus limit the use of the term obligatory extraposition to those cases in which the valency carrier generally disallows the type of element (e.g. [that_CL]) in canonical subject position in a given pattern (e.g. appear without predicative element). Given that there is no one simple solution, the remainder of this chapter will follow CGEL in using the term obligatory extraposition despite the at least partially valid arguments against such a treatment brought forward by CamG.

|| 297 One could also argue that the fact that additional elements are present postverbally may influence the ability of a non-extraposed [that_CL] subject to occur, similarly to the case of (13c) discussed above. However, the complement status of the elements here is debatable as opposed to (13c), where obvious is definitely a complement of appear.

Theoretical treatment | 187

To sum up, in the comparatively static models used for the analysis in this chapter, there is no convincing and consistent model to capture all generalizations one would want to make. So the reference grammars only capture either the parallel between (29a) and (29b) (CGEL) or the one between (29b) and (29c) (CamG). (29) a.

It is perfectly clear that Charles and Diana are less and less comfortable in each other’s company. (CEN 3385) b. It appears that Charles and Diana are less and less comfortable in each other’s company. c. It appears as if Charles and Diana are less and less comfortable in each other’s company.

The valency analysis offered by the VDE captures both parallels but this comes with the negative effect that complements that can never occur in canonical subject position, such as [as if_CL] complements, are labelled as subjects only due to an analogy to other complements that can occur in subject position, so the generalization over extraposed and non-extraposed subjects is weaker in this model. A model that could capture all these parallels would be more attractive than CGEL, CamG or the valency approach. Currently, construction grammar (see for instance Goldberg 1995, 2006) is the most likely candidate for such a model as the strong focus on storage that most flavours of construction grammar advocate would probably lead to a model which stores all variants seen above and which allows emergent generalizations that are not mutually exclusive. The organization of the stored language data could be imagined in the form of a network model as presented by Bybee (2007: 323–326). Such a model will be discussed in Section 9.3. However, it is still a problem to model the scope and limitations of generalizations (see Herbst 2011 for a critique).

7.1.3 What is the subject? As we have seen above, most researchers do not treat the anticipatory it as a dependent of the valency carrier. However, in the grammatical structure, items that are not required by the verb can still be subjects — as suggested for meteor-

188 | Extraposition

ological verbs (“It’s raining.”)298 in most approaches. The present section reviews the different positions taken and the arguments for and against the respective analyses.

7.1.3.1 Survey of different positions Quirk et al. (CGEL 1391) state explicitly that there are two subjects in a sentence that shows extraposition: the postponed subject (which is the notional subject) and the anticipatory subject (it).299 This view is challenged by Huddleston/Pullum (CamG 239ff), who introduce a uniqueness constraint on subjects, i.e. there cannot be more than one subject in a clause.300 They check both elements, it and the extraposed subject,301 against the list of subject properties they established and conclude that it is the sole subject. We shall discuss the criteria in detail below. In Halliday’s (1994: 97) analysis, it and the extraposed subject together form the subject of the clause; in the accompanying text, the embedded clause at the end of the clause is “functioning as Subject [...] with an anticipatory it occurring in the normal Subject position” (Halliday 1994: 98). The precise nature of the relationship between the two elements remains unclear. Similarly, Herbst/Schüller (2008: 124) treat it and the postponed subject as one single discontinuous element. Thus we can see that almost every conceivable position on this matter has been argued for in the analysis of English. In German linguistics, there is no consensus of any sort, either. Engel (2004: 137) uses the term Korrelat (‘correlate’) for es (‘it’), as it correlates with an embedded clause. He makes use of obligatory extraposition and seems to regard

|| 298 Such meteorological verbs will not be discussed in detail here, but see for instance the discussion in Emons (1974: 118–124) and the references there, or Fischer (1997: 97), who analyses It in It rains. as part of the verb (!) and can thus claim that the verb is zero-valent. 299 The VDE seems to follow that position, which can be inferred from their explanation of the notational convention to indicate obligatory extraposition of a complement: “when occurring as subject, extraposition with a dummy subject it is obligatory” (VDE xii). 300 See Section 2.3.2 for details. 301 The use of the term extraposed subject for an element that is explicitly not treated as subject may appear illogical, but the authors defend their terminology as follows: These terms are intended to capture the fact that they are semantically like the subject of their basic counterpart, but they are not to be interpreted as kinds of subject. The subject is a syntactic function, and these elements are no more subjects than a former president is a president or than an imitation diamond is a diamond. (CamG 243)

Theoretical treatment | 189

the extraposed subject as the subject of the clause.302 Eisenberg (2006: 178) treats the correlate es and the extraposed element together as subject of the clause. The internal structure of the subject is similar to an attributive construction, where es is head (Eisenberg 2006: 330).

7.1.3.2 Criteria for subjecthood revisited The criteria for subjecthood were discussed in Section 2.3.2 above. In this section, we will see to what extent they apply to anticipatory it and extraposed subjects. A substantial portion of the discussion is based on the extensive treatment by Huddleston/Pullum (CamG 241–243), who are in favour of an approach that treats only it as the subject in extraposed sentences. A critical evaluation of their arguments is given where necessary. 1.

Category

Prototypically, subjects are noun phrases. Although both it and the extraposed subject can usually occur as only subject, it would be better suited as it is a noun phrase and thus conforms to the prototype: (30)

That riots and local socialism were both found in the most acutely stressed parts of the country is no accident. (AS6 878)

(31)

It is a tool or aid to a banker making a lending decision. (B1W 1648)

As discussed in Section 7.1.2, matters become more complicated if we treat when-clauses as in (32a) as extraposed subjects since they cannot be subjects in the canonical subject position (see (32b)). An analysis in terms of two subjects (see above) may thus be less appropriate. (32) a. It was a relief when six o’clock arrived [...]. (HP7.421) b. *When six o’clock arrived was a relief. If we treat it and the extraposed subject as one constituent, it is not obvious what category this constituent should belong to. If we want to follow Eisenberg (2006; see above) in calling it the head of the construction, our complex element could be treated as an NP and would thus conform to the prototype.303 || 302 A similar position seems to be taken by Marx-Moyse (1983), who speaks of “vorausweisendes es” (‘cataphoric it’) and calls the extraposed subject “Subjektsatz” (‘subject clause’). 303 It has to be made clear that such considerations are more relevant to rather traditional approaches to grammar; the fact that we can identify chunks or discontinuous elements that

190 | Extraposition

2.

Position

It occupies the default subject position whereas the extraposed subject (hence the name) does not. If our analysis involves a discontinuous constituent, there is no single position that could be determined.304 3.

Case

As neither clausal subjects nor it are marked for nominative/accusative case, this criterion is not applicable. 4.

Agreement

The relevant paragraph in CamG reads as follows: In extraposition the verb agrees with it. Clauses generally have the default 3rd person singular feature, but a clause-coordination can have a plural interpretation, as reflected in the are of To promise you’ll do something and to actually do it are two quite different things. There can be no extraposition counterpart precisely because are does not agree with it. (CamG 242)

On the other hand, many coordinated clauses occur with a verb in the singular:305 (33) a.

That he bought a new car and that he has a shiny yacht annoys me like hell. b. *That he bought a new car and that he has a shiny yacht annoy me like hell.

|| do not readily lend themselves to a simple categorization is common and accepted in usagebased approaches such as Construction Grammar. 304 Note that the term position is often used differently in the generative model, where it refers to a place in the underlying deep structure which can but need not correspond to the position the element occupies on the surface (see Section 2.2.1 for more details). 305 Even variations of CamG’s example can be shown to allow both singular and plural agreement: (1) To promise you’ll do something and to actually do it is not the same. (2) To promise you’ll do something and to actually do it are not the same. (3) To promise you’ll do something and to actually do it is not the same thing. (4) To promise you’ll do something and to actually do it are not the same thing. (5) To promise you’ll do something and to actually do it are not the same things. By comparing the last two examples we can exclude that agreement is determined by the postverbal [NP] complement. Thus coordinated clausal subjects can receive singular or plural interpretation basically depending on whether the speaker chooses to treat them as one unit or separately.

Theoretical treatment | 191

If we compare (33a) to its counterpart that shows extraposition (34c), there is no way to tell what the verb agrees with, so the criterion is difficult to apply.306 (35) c.

It annoys me like hell that he bought a new car and that he has a shiny yacht.

CGEL’s test for agreement between subject and subject complement (if the latter is a noun phrase) cannot be applied satisfyingly either. However, since CamG is right in that no coordinated clauses with a plural interpretation can occur as an extraposed subject, we can still count agreement in favour of it as the subject. 5.

Inversion

The post-operator position is taken by it in sentences that show subject extraposition: (36) a. It was obvious that he had only just got home [...]. (HP5.373) b. Was it obvious that he had only just got home? However, one may want to question the applicability of the criterion to clausal subjects if we follow CamG’s test: This property provides an indirect way of identifying the subject in structures not illustrating inversion: you can be confirmed as subject of You know the candidates because it is you that comes to occupy the post-auxiliary position when we convert this declarative to the corresponding closed interrogative Do you know the candidates? (CamG 237)

Thus if we take sentence (37a) and transform it into the corresponding closed interrogative in (37b), we find that again it takes the post-operator position and that the [that_CL] is impossible in the same position, as illustrated by (37c), i.e. that extraposition is obligatory in inverted structures:307 (37) a.

That such traditions continued in the early 17th century not only in Italy but also in France should hardly surprise us[.] (J1A 473)

|| 306 Eisenberg (2006: 289f) only gives German examples comparable to (33a) although plural agreement with dass-clauses should be possible in German as well. Thus he concludes that contrary to nominal (“nominative”) elements, clauses are not marked for number, even when coordinated, which is at least questionable. 307 Newmeyer shows “that if we lengthen the constituents following the sentential subject, acceptability improves considerably” (Newmeyer 2003: 157) and cites the following example in support: “Why did [that Mary liked old records] lead Bill into a three-hour discursus on how wonderful the 1960s were?” (Newmeyer 2003: 157)

192 | Extraposition

b. Should it surprise us that such traditions continued in the early 17th century not only in Italy but also in France? c. *Should that such traditions continued in the early 17th century not only in Italy but also in France surprise us? However, we would of course not want to claim that it is the subject of (37a), thus the indirect test is of rather limited usefulness with clausal subjects. Nonetheless, since most researchers would agree that (37b) is an extraposed structure and since it takes the post-operator position, we can count this criterion in favour the subject status of it instead of the extraposed element. 6.

Open interrogatives

According to CamG (243) “[i]n extraposition neither it nor the extraposed clause can be questioned, so this test is not applicable.” This can be doubted since the native speakers interviewed on this issue do not agree with this statement as they would accept such questions both for copular and for non-copular clauses:308 (38) a.

It is likely that the school will be closed unless the culprit behind these attacks is caught. (HP2.191) b. What is likely? c. #It. d. That the school will be closed until the culprit behind these attacks is caught.

(39) a.

[I]t follows logically that somebody at the Ministry might have ordered the attacks [...]. (HP5.134f) b. What follows logically? c. #It. d. That somebody at the Ministry might have ordered the attacks.

The form of questions (38b) and (39b) suggests that the element which is the focus of the question is a subject. The acceptable answers to these questions, (38d) and (39d), tell us that the questioned subject is the [that_CL].

|| 308 CamG claims later that the question for the extraposed subject needs to contain the dummy pronoun it (“*What does it invariably annoy him?” [CamG 962]), but no reason is given why the question should have this form. If we accept their premise, questioning the extraposed subject is indeed impossible.

Theoretical treatment | 193

In some cases the answer has a slightly different form than the element in the original sentence: (40) a. b. c. d. e.

It is a pity it broke [...]. (HP5.749) What is a pity? #It. #It broke. That it broke.

This is hardly surprising since the extraposed clause without that cannot occur in the canonical subject position either: (41) a. *It broke is a pity. b. That it broke is a pity. The fact that the answer to (40b) is (40e) with that and not the bare clause (40d), which is present in the original sentence (40a), seems to imply that there is some kind of relationship between that-clauses and subordinate clauses without that. This issue and the obligatoriness/optionality of that in [that_CL] complements are discussed in Section 7.2.2 below. 7.

Tags

Huddleston/Pullum (CamG 243) observe that it is repeated in interrogative tags and take this as another argument in favour of the subject status of it: (42) a.

It annoyed you that he left early and that he took all the booze with him, didn’t it?

Their point is not precise, though, as the criterion is not the repetition of the subject in the tag, but the occurrence of an element that agrees with the subject. As we have seen above in the section on agreement, all clauses with extraposition show singular agreement, so it is in fact impossible to decide whether the it in the tag represents the anticipatory it or the extraposed subject. Even in the case of the coordinated [that_CL]s in (42a) we cannot tell as they also take it as pronoun in the tag if they occur in canonical position, since – as we have seen above – coordinated clauses with a plural interpretation cannot occur in extraposition. (42) b. That he left early and that he took all the booze with him annoyed you a lot, didn’t it? c. *That he left early and that he took all the booze with him annoyed you a lot, didn’t they?

194 | Extraposition

8.

Coordination

With regard to the occurrence with a coordinated VP, Huddleston/Pullum (CamG 243) take a rather one-sided view by just testing it with a relatively uncommon structure which is rather difficult to process (reproduced as (43)) in order to claim that it must be subject: (43)

It was obvious to everyone that he loved her and had been from the very beginning. (CamG 243)309

They ignore the fact that it is also possible (and probably more natural) to have an extraposed subject with coordinated VPs as in (24) (repeated below), where the extraposed subject appears not to be part of the VP:310 (24)

It is pretty straightforward and follows Johnny’s philosophy that a race driver is “a simple animal, and doesn’t have to be a rocket scientist.” (www.ericjacobsen.org)311

Thus, the criterion of coordination is not as clearly in favour of the “it is subject” analysis as Huddleston/Pullum make it appear. Coordination probably offers the best argument to back up the “discontinuous subject” option, as can be seen in (24), where the VPs is pretty straightforward and follows Johnny’s philosophy are coordinated and there is only one discontinuous subject present.312 9.

Obligatoriness

As we have seen above, Huddleston/Pullum (GamG 238f/243) check for obligatoriness by comparing a sentence to its “maximal finite reduction” if the necessary information is recoverable from the context. So if we want to know what the subject in (44a) is, we can rephrase it in the form of a question as (44b) and take a look at the corresponding answer (44d). (44) a. It is probable that this will have much to offer. (A04 723) b. Is it probable that this will have much to offer?

|| 309 Some native speaker informants rejected this example. 310 We also commonly find extraposed subjects with coordinated matrix clauses: It is compelling and it is noteworthy that he was still a staff analyst at the Pentagon when he gave this speech. (rosecoveredglasses.blogspot.com) 311 See however Note 293 for a short discussion of the status of the VP in this example. 312 The example of coordinated matrix clauses given in Note 310 above is however problematic for a discontinuous subject analysis since there are two instances of it and only one extraposed subject.

Theoretical treatment | 195

c. #That this will have much to offer is.313 d. It is. The problem with this criterion is that it does not require the subject (of the original sentence in question) in the maximal finite reduction, it requires a subject that is usually a pronoun replacing the original subject. So although Huddleston/Pullum claim that only it and not the [that_CL] fulfils the criterion of obligatoriness, it is impossible to argue in favour of their analysis on the evidence of the incongruity of (44c) and the acceptability of (44d) as the same situation arises with non-extraposed sentences, too: (45) a.

That he was the most inspiring and best loved of all Hogwarts headmasters cannot be in question. (HP7.24) b. That he was the most inspiring and best loved of all Hogwarts headmasters cannot be in question, right?314 c. #That he was the most inspiring and best loved of all Hogwarts headmasters cannot.315 d. It cannot.

As it would not make sense to argue that it is the subject of (45a) just because it is the subject of (45d), the whole line of argument collapses. Thus the obligatoriness criterion as Huddleston/Pullum test it does not yield conclusive evidence. If we understand obligatoriness in the sense that, leaving position aside, an element has to occur in a sentence, we find that the extraposed subject is at least as obligatory as it in most canonical sentences: (46) a.

It is therefore advisable that you complete your shopping well before nightfall. (HP3.148) b. That you complete your shopping well before nightfall is therefore advisable. c. #It is therefore advisable.

While (46b) is marked due to the heavy initial subject, it is still acceptable. (46c) is only acceptable if we treat it as a personal pronoun (referring to an element/entity that is recoverable from the context) and not as anticipatory it.

|| 313 The hash symbol indicates that this sentence is unacceptable as minimal response to (44b). It may be perfectly acceptable in other contexts. 314 This form of question has been chosen due to the fact that no non-extraposed clausal subject is possible in closed interrogatives. 315 The statement of fn. 313 applies here, too.

196 | Extraposition

The situation is different if we have obligatory extraposition, as in sentences with subject-auxiliary inversion:316 (47) a.

Sir ... is it important to know all this about Voldemort’s past? (HP6.203) b. ?Sir ... is to know all this about Voldemort’s past important? c. #Sir ... is it important?317

For (47c), the same restriction on the interpretation of it applies as for (46c). With regard to the restricted acceptability of (47b), we have to conclude that both it (in the sense it is used in (47a)) and the extraposed subject are obligatory in (47a). 10. Uniqueness In their discussion of the subject in extraposition constructions, Huddleston/Pullum (CamG 241) take the uniqueness property as their starting point: There can only be one subject in a clause, so we have to decide between it and the extraposed subject. The solution proposed by Quirk et al. (see Section 7.1.3.1 above), who treat both elements as subjects, is not even considered and neither is a discontinuous subject analysis, which would definitely meet the criterion. The evidence to back up their view is provided by agreement and tags in coordination (CamG 239) for nominal subjects, but as we have seen both are not reliable for clausal subjects. Thus, in the light of the evidence presented in this chapter, it would be premature to argue that there can only be one subject in a clause, particularly as it is so difficult to decide which element it is. 11. Reflexive pronouns The fact that the form of the reflexive pronoun is determined by the subject will not help us to decide whether it or the extraposed [that_CL] in (48a) is subject because we can observe in (48b) that a [that_CL] also triggers the form itself. (48) a.

But it revealed itself that there was a natural progression. (www.variety.com) b. That she was destined to march to a different drummer revealed itself early. (www.apsu.edu)

|| 316 Sentences of the It seems that... type behave in the same way but since not all scholars would analyse these as instances of extraposition, an interrogative example with inversion was chosen here because there is less disagreement (see Section 7.1.2 above for discussion). 317 This is of course acceptable as a sentence with referential it, but not as a rephrasing of (47a).

Theoretical treatment | 197

12. Passivization According to CGEL’s passivization property, the direct or indirect object of an active clause becomes the subject of a corresponding passive clause. If this condition is accepted to hold for clausal subjects, it follows that the [that_CL] that is object in (49a) is subject in (49b). (49) a.

They had agreed that it was far too dangerous to try and communicate with Mr. Weasley while he walked in and out of the Ministry, because he was always surrounded by other Ministry workers. (HP7.188) b. It had been agreed that it was far too dangerous to try and communicate with Mr. Weasley while he walked in and out of the Ministry, because he was always surrounded by other Ministry workers.

It has to be stressed, though, that this property only applies to the model of Quirk et al., who regard both elements as subjects anyway. Huddleston/Pullum are careful to emphasize that (a) they do not regard the [that_CL] in (49a) as object (CamG 1017f), (b) they do not regard passivization as necessary or sufficient criterion for objecthood (CamG 1018) and (c) they accept two types of passives (extraposed and non-extraposed) related to sentences with what they call content clauses as internal complements318 (CamG 1434). The second half of the passivization property is only applicable to clausal subjects to a limited extent, as there is a need for reformulation of [that_CL] and [to_INF] clauses if they are to become by-phrases.319 If we tolerate that reformulation for our test, we see that in (50b) not the it from (50a) but the extraposed subject has been expressed in the by-phrase. (50) a.

[…] it struck Harry that Dumbledore might not have been quite truthful. (HP1.157) b. […] Harry was struck by the fact/idea that Dumbledore might not have been quite truthful. c. #[…] Harry was struck by it.

This is particularly interesting as Huddleston/Pullum (CamG 68) also mention the relationship between subjects in active clauses and by-phrases in passive

|| 318 “Content clause as internal complement” is prototypically a clausal object in CGEL’s terminology. 319 The term by-agent phrase as chosen by CGEL (725) is not particularly felicitous for our purposes as it is problematic to attribute agentivity to clausal subjects. Following Herbst/ Schüller (2008: 118), the neutral term by-phrase will be used instead.

198 | Extraposition

clauses, so there is an inconsistency between this correspondence and their claim that it is the (only) subject in sentences such as (50a). This test is also an argument against a discontinuous subject analysis since the two parts cannot occur together in a by-phrase: (50) d. *[…] Harry was struck by it the idea that Dumbledore might not have been quite truthful. 13. Theme/topic “The subject is typically the theme (or topic) of the clause” (CGEL 726). The adverb typically suggests that there are exceptions, and indeed the extraposed subject cannot be said to be the theme of the clause if we follow the definition cited in Section 2.3.2; only non-extraposed clausal subjects act as theme. Quirk et al. acknowledge this in a note to their paragraph on extraposition: The it used in extraposition is called ‘anticipatory it’ because of its pronominal correspondence to a later item. But informationally, this it is similar in effect to prop it (cf 6.17), as in ‘It started to rain’ which likewise enables us to end the clause at a focal point. (CGEL 1392)

The question arises, however, whether in such sentences it alone can satisfactorily be treated as the theme despite its apparent lack of semantic content.320 Here, Kaltenböck’s analysis, in which he treats the anticipatory it as sitting on a gradient between prop it and referring it is interesting in that for him, all three types have reference, the type of which he labels as follows: – “’ambience’ (+ possible adverbial restriction)” for prop it, – “state-of-affairs (clause)” for anticipatory it, and – “’single’ entity/referent (NP)” for referring it (Kaltenböck 2004: 47) So for some sentences, we can definitely see that it fulfils the theme/topic criterion better than the extraposed subject does. 14. Given information In the case of extraposition, the anticipatory it, as it is semantically practically empty, could be regarded as an element that is given by default, similarly to existential there to be discussed in Section 8.3. As for the extraposed subjects,

|| 320 CGEL (349) states “that the pronoun is not quite void of meaning, since it arguably has cataphoric reference (forward coreference) to a clause [...] in the later part of the same sentence.”

Theoretical treatment | 199

the general tendency of focus and new information to coincide (CGEL 1361) comes into play, but since it is a clause it is likely to contain both given and new information. As we can see from the following short excerpt, the [that_CL] presents information that is more new than given.321 (51)

Harry kept to his room, with his new owl for company. He had decided to call her Hedwig, a name he had found in A History of Magic. His school books were very interesting. He lay on his bed reading late into the night, Hedwig swooping in and out of the open window as she pleased. It was lucky that Aunt Petunia didn’t come in to vacuum anymore, because Hedwig kept bringing back dead mice. (HP1.67)

Information structure will be discussed in more detail in Section 7.3.2. 15. Agentivity As we have seen above, prototypical agentivity cannot apply to clauses, but if we apply Dowty’s (1991) proto-roles (see Section 2.3.2 for a brief discussion), we can say that the extraposed [that_CL] in (52) is more proto-agent-like than protopatient-like and the anticipatory it – since it fills no argument slot and is thus assigned no semantic role – is not relevant to agentivity. (52)

It struck Harry as ominous that Hagrid thought a biting book would come in useful, but he put Hagrid’s card up next to Ron’s and Hermione’s, grinning more broadly than ever. (HP3.16)

7.1.3.3 Summary The following table presents the findings from Section 7.1.3.2. It conveys a much messier picture than the equivalent table in CamG (241) and thus needs further discussion.

|| 321 Huddleston/Pullum (CamG 68) distinguish between discourse-old and addressee-old information, but in the quoted example there is no difference between the two as the addressee (=the reader) will usually have read the book (=the discourse) so far and as the author does not know what is known to the reader other than what was in the discourse (or what is shared knowledge of the world in general, of course).

200 | Extraposition

Tab. 30: Summary of the criteria for subjecthood with extraposition

Criterion

anticipatory it

postponed subject

discontinuous subject

category

+



?

position

+



(+)

case

N/A

N/A

N/A

agreement

?

?

(+)

inversion

+



(+)

open interrogatives



+



tags

+

+

(+)

coordination

(+)

(+)

+

obligatoriness

(+)

+

?

uniqueness

?

?

+

reflexive pronouns +

+

+

passivization



+



theme/topic

?





given information

?



?

agentivity



(+)

?

Notes

conflicting evidence

rare cases for it; question whether extraposed subject is VP-internal depends on interpretation of the other criteria

only in some models; not prototypically

In the light of such evidence it must be clear that any decision as to which element(s) should be treated as subject has to remain arbitrary to some extent. It is a matter of the researcher’s focus to which criterion he or she attributes most weight. A purely numerical approach (counting the pluses and minuses in the respective columns) is not feasible because some criteria may be subdivided or merged. Therefore the list of criteria is not definite and needs discussion in order to allow for a decision in favour of one of the approaches presented in Section 7.1.3.1:

Theoretical treatment | 201

1.

Both elements are subjects.

The main problem with an analysis that allows for two subjects in a clause is that the syntactic model would have to be adapted accordingly, with a subject slot behind the verb. However, the uniqueness constraint seems to be accepted (sometimes only implicitly) by many researchers.322 2.

Both elements form a discontinuous subject.

The major drawback of the analysis of the two elements as one discontinuous subject is that the formal status of such an element is unclear. An interpretation in which it is the head of the construction, as proposed by Eisenberg for German (see Section 7.1.3.1 above), is semantically counter-intuitive. Criteria such as position or inversion are difficult to apply due to the discontinuity. The main advantages of a discontinuous subject approach are its uniqueness and the possible coordination of VPs. But as long as the internal relations between the different elements are not satisfyingly described and the disappearance of it in non-extraposed constructions and answers to open interrogatives has not been modelled, this proposal remains tentative. Even if we were to acknowledge a discontinuous subject as in the model of Herbst/Schüller (2008), it is not clear what form such an element should take as complement of a verb (particularly if we want to generalize over the extraposed and the non-extraposed variants of the complement). 3.

The postponed subject is the only subject.

The main arguments for treating the postponed subject as the subject of the whole sentence are of a semantic nature. It is relatively uncontroversial that it realizes an argument of the verb in question, that it is the element of which something is predicated, etc. In addition, the fact that the same element does without doubt occupy the subject position in the non-extraposed variant makes a strong case for its subject status. The strictly syntactic criteria, the most obvious of which is position, can, however, be regarded as arguments against such an analysis.

|| 322 This of course does not take coordination into account, which has to be considered independently. One may, however, follow Huddleston/Pullum (CamG 239), who only allow coordination at the formal level and thus evade the problem. See, however, the discussion of problematic cases starting on page 51 above.

202 | Extraposition

4.

Anticipatory it is the only subject.

As we have seen above, the case CamG makes for anticipatory it as the subject of sentences which show extraposition is not as straightforward as claimed there. Open interrogatives, passivization and agentivity are criteria against its subject status, and some of the other arguments brought forward by CamG cannot stand up to closer scrutiny. Nevertheless, at the level of syntax, anticipatory it makes the most convincing subject out of all candidates. If we separate the levels of syntactic structure and of verb valency as rigorously as suggested by Herbst/Roe (1996),323 at least some of the criteria we have in our list above become questionable: Agentivity, for instance, describes a semantic role that is dependent on a particular verb, first of all at the semantic level. Only in a second step can this term be used for the description of the corresponding verb complement at the syntactic level.324 In a third step, which is a considerable abstraction, it is observed that the verb complement assigned the agent(ive) role often occupies the subject position in active declarative clauses.325 Thus while the criterion of agentivity may help in the identification of the subject element in a clause, we have to bear in mind that agentivity is essentially verb-related and is thus better suited to identifying an element in relation to the verb than at the level of syntactic structure.326 Passivization does not provide a viable syntactic argument, either. If it were a property of a syntactic category such as subject, there would be no need to indicate it in a valency dictionary such as Herbst et al. (2004). The reason why we need this information in a valency description of a verb is that it is verbspecific, or, to be more precise, complement-specific. Thus we can only say that a certain complement of a certain verb can occupy a postverbal position in active sentences and a preverbal (=subject) position in passive sentences and that some other complement may or may not occur as subject in active or as byphrase in passive sentences.

|| 323 See Section 2.2.2 for a brief summary of their position. 324 This is of course a simplified account as the match between semantics and syntax is imperfect. See Helbig (1992), Herbst/Klotz (2002), Faulhaber (2011) for detailed accounts of the problems that arise in this context. 325 It is often claimed that, as a general tendency, the most agent-like element in the active clause becomes its subject (see for instance Goldberg 2006: 184ff or Herbst 2007: 27). See also the discussion in Section 2.3.2. 326 It is also possible to assume semantic roles on both levels, as Herbst/Schüller do, whose model distinguishes between verb-specific participant roles and more general clausal roles, “which arise from sentence structure” (Herbst/Schüller 2008: 160).

Theoretical treatment | 203

A similar line of argument can be followed in the discussion of open interrogatives: What is questioned is an element at the level of verb valency, because the question contains the valency carrier and asks for a dependent element. It is thus not surprising that anticipatory it, which has no complement status, cannot be questioned. In fact, the situation is the same as for prop it in (53): (53) (54)

It’s still raining in there. (HP7.200) It is essential that you understand this! (HP6.477)

Both its cannot sensibly be questioned precisely because they are not complements of the verb. In both cases, the it is just there for structural reasons, in order to occupy the subject position of the clause. The main difference between the two semantically empty forms is that in (53) the it is needed as a subject because the verb does not license a complement whereas in (54), the complement licensed by the valency carrier is actually present in the sentence, only in a different position. But still, the subject position is not taken, so we need some sort of element there, which is it.327 This requirement for subjects in English declarative clauses is a purely syntactic constraint, and if we accept this explanation it automatically follows that we have to treat it as subject and thus position (in canonical and in inverted clauses) as the most important criterion for the distinction of the different subject candidates. This is in line with the position of Matthews (1981) presented in Section 7.1.3.1. If we treat it as subject, the question arises as to what the status of the postponed subject is. Here, a treatment comparable to that of other postverbal complements will be proposed. If we were to describe the relationship between it and the postponed subject, we would probably argue for an appositive relation more than for a head-dependent relation as the one proposed by Eisenberg (see Section 7.1.3.1). At the semantic/communicative level, anticipatory it works as a cataphoric pronoun which indicates that the element that would be expected in subject position actually will occur later in the sentence. It is, however, questionable whether it makes any sense at all to posit a syntactic relationship between the two, other than that one is subject because the other is not.

|| 327 The reasons why the complement does not take the subject position are discussed in detail in Section 7.3 below.

204 | Extraposition

7.2 Syntactic properties 7.2.1 Extraposed to which position? While we have discussed the position of anticipatory it in detail above, we have largely neglected the position of the postponed subject. While Quirk et al. (as cited in Section 7.1.3.1) only vaguely state that it is moved to the end of the sentence, Huddleston/Pullum are more precise in locating it “at the end of the matrix clause, in what we are calling extraposed subject position” (CamG 1403) and in stating in a footnote that adjuncts may follow it, “provided they too are relatively heavy” (CamG 1403). In order to verify and refine this analysis, 1,000 sentences from a partially parsed version of the BNC (Schneider 2009) were extracted that had it as subject and that later in the same sentence. This method of looking for that as a word form was necessary because [that_CL]s were not part of the parsing scheme used. Accordingly, more than half of the sentences (557) did not show extraposition with a [that_CL]328 and 13 sentences were unclear, erroneous or contained more than one extraposed [that_CL] which would have resulted in multiple classification. Of the remaining 430 sentences, the vast majority (312) contained a [that_CL] as the only post-verbal element. 65 contained another element between the valency carrier (verb/noun/adjective) and the [that_CL].329 The most important result is that no non-clausal elements came after the [that_CL] in any of the sentences and that no complements (not even clausal complements) of the valency carrier occurred in that position either. There were 42 instances where the [that_CL] was not the last element in the sentence, though, as it was followed either by a subordinate clause or by a coordinated main clause. In 11 cases, it was impossible to decide whether the following clause was part of the [that_CL] or the superordinate clause. The key facts are summarised in the following table:

|| 328 The largest group within the “scrap” were 92 cleft sentences. 329 In 24 of the 65 sentences the intervening elements(s) were complements of the valency carrier, in 32 cases there were adjuncts (most of them relatively short such as however, therefore, also, too, moreover), 9 are indeterminate between complement and adjunct (for a thorough discussion of complement status see Somers 1984).

Syntactic properties | 205

Tab. 31: Position of extraposed [that_CL] subject

that-clause is

percentage330

only element

73 %

final element preceded by sth. else

15 %

followed by another clause

10 %

(ambiguous cases)

3%

Our sample of extraposed [that_CL] subjects thus seems to confirm CamG’s position as to where the extraposed subject occurs. We can add two more observations, which seem to apply to [that_CL] complements only: 1. No non-clausal adjuncts were found to follow the extraposed [that_CL], whereas both clausal and non-clausal adjuncts preceded it. 2. No sentences were found in which the extraposed subject [that_CL] was both preceded and followed by another element.331 Due to the nature of our sample we cannot say whether these are tendencies for postverbal [that_CL]s in general, or whether this behaviour is specific to extraposed subject [that_CL]s. [to_INF]s seem to be all right with a combination of both elements, though. Compare: (55) a.

It was so great of him to let me sleep because I know he was tired too. (mindaandsam.blogspot.com) b. ?It was so great of him that he let me sleep because I know he was tired too.332

|| 330 Due to rounding, figures do not add up to 100 %. 331 Further research in the BNC yielded one example where the extraposition occurred in an adverbial subordinate clause at the beginning of a sentence so that the extraposed [that_CL], which is preceded by a [to_NP], is followed by the matrix clause: Under Section 36 of the Criminal Justice Act 1988, where it appears to the Attorney General that the sentencing of a person in the Crown Court has been unduly lenient he may refer the case to the Court of Appeal, with leave, for the sentence to be reviewed. (EEC 319) 332 Even if we accept (55b) as a spoken/borderline case, it is less acceptable than (55a).

206 | Extraposition

7.2.2 [that_CL] with and without that As we have seen above, content clauses333 in extraposed subject position can occur with or without introductory that. While CGEL speaks of “deletable that” (CGEL 1183), Erdmann (1987: 55) follows Jespersen in rejecting such a deletion since both variants have always existed independently of one another and neither is based on the other. The present study will not rely on the concept of deletion but will treat that as optional in such complements. The major reference grammars treat the optionality of that as the default case, i.e. both variants are usually allowed. Nonetheless, according to Greenbaum (1988: 95) the bare variant is marked compared to the variant with that. Kaltenböck (2004: 76–88) gives a very detailed account according to the text types in the ICE-GB corpus and according to matrix clause predicates. While the latter do not seem to cause large differences in proportion between the two (see 2004: 81), the text type seems to play a role with the highest proportion of that in instructional and academic writing and the lowest proportion in private dialogue and reportage (Kaltenböck 2004: 78). CGEL (1049f) and CamG (951ff) present grammatical restrictions on the use of the bare variant which are summarised below, insofar as they are relevant to subjects. 1.

‘Gap’ is subject

The only construction in which that is obligatorily omitted is described by CGEL as sentences with a “pushdown wh-element” (CGEL 1050) and by CamG as “unbounded dependency construction” (CamG 953) in which the “subject is realised by a gap” (CamG 953). Both grammars do not specifically mention extraposed subjects in this context, but the unacceptability of (56b) as opposed to (56a) indicates that the rule applies to them as well. (56) a.

And who does it appear ___ will go home Thursday night? (realitynewsonline.com) b. *And who does it appear that ___ will go home Thursday night?

2.

Pre-verbal position

All content clauses in pre-verbal position require the presence of that, and both reference grammars agree that this is in order to avoid an interpretation of the

|| 333 In this section we shall borrow CamG’s term content clause in order to avoid the clumsy wording ‘[that_CL] with and without that’.

Syntactic properties | 207

subordinate clause as the main clause. Thus for non-extraposed subjects, that can never be optional. (57) a.

That he was the most inspiring and best loved of all Hogwarts headmasters cannot be in question. (HP7.24) b. *He was the most inspiring and best loved of all Hogwarts headmasters cannot be in question.

3.

Clarification of adjunct position

If there is an adjunct right before the content clause as in (58b) or if the content clause starts with an adjunct as in (58a), that cannot be omitted, as illustrated in (58c): (58) a.

It worries me that, at the moment, the public feel let down by the court system. (news.bbc.co.uk) b. It worries me at the moment that the public feel let down by the court system. c. *It worries me at the moment the public feel let down by the court system.

According to CGEL (1050), this is because without that it would be impossible to determine whether the adjunct belongs to the matrix clause or the subordinate clause. However, the restriction seems to hold even for cases in which an alternative interpretation is not possible. If we move the adjunct last time in (59a) to the final position in the matrix clause, we get (59b), which is problematic. So (59c) should only allow for one sensible interpretation, but it still remains dubious, although (59d) is perfectly all right.334 (59) a.

It’s a pity that last time I missed it because I got up a little late. (library.thinkquest.org) b. ?It’s a pity last time that I missed it because I got up a little late. c. ?It’s a pity last time I missed it because I got up a little late. d. It’s a pity I missed it because I got up a little late.

Thus, though CGEL correctly describes the facts, we have to disagree with their argument.

|| 334 Not all native speakers seem to have that restriction as one informant accepted (59b–c) as well.

208 | Extraposition

4.

Coordination

CGEL also mentions that that must be retained “to prevent a coordinated thatclause from being misinterpreted as a coordinated main clause” (CGEL 1050). This is not obligatory, though, as pointed out in CGEL (1043) and illustrated in (60), which is ambiguous but acceptable in both readings. (60)

5.

I must admit that I didn’t even know he was on trial, but it appears that he was and he just got away with it. (healthcareblog.com) Clause or long phrase intervening

CGEL also shows that “a clause or a long phrase” (CGEL 1050) intervening between the verb and the content clause leads to the retention of that. There is a certain overlap between this factor and the clarification of the adjunct position as outlined above, since said long phrase may well be an adjunct, as in the following example: (61)

It appeared to me in the Town Hall meeting last night that some neighbors are concerned the plan is a way to provide additional parking for the Nationals and encourage fans to park in the neighborhoods. (tommywells.org)

But we find that the phenomenon has indeed to be treated as a separate factor since non-adjunct elements may just as well trigger the obligatory presence of that if they are heavy enough: (62) a.

So it seems pretty obvious to me he doesn’t want someone like his mother. (associatedcontent.com) b. So it seems pretty obvious to me that he doesn’t want someone like his mother.

(63) a.

So it seems pretty obvious to anybody who has followed his misfortune that he doesn’t want someone like his mother.335 b. ?So it seems pretty obvious to anybody who has followed his misfortune he doesn’t want someone like his mother.

|| 335 This example was constructed in order exclude other possible factors of influence when compared to (62a–b), and to make the point more obvious. It was formed on the example of the following original sentence: It must be obvious to anybody who has followed his misfortune that his guilty plea was a desperate attempt to obtain some prospect of an end to his nightmare. (larvatusprodeo.wordpress.com)

Factors influencing extraposition | 209

The content clause may be preferred in canonical subject position, though, if the element following the verb becomes too heavy compared to the content clause itself (cf. Section 7.3.1 below).

7.3 Factors influencing extraposition So far we have only dealt with questions of analysis, but we have ignored why extraposition occurs in the first place. As there is, at least theoretically, a canonical alternative to most sentences which show extraposition, we need to account for this deviation from the default pattern. Various scholars have suggested a wide range of factors that influence whether extraposition occurs or not, which will be discussed in turn in the following subsections.

7.3.1 Weight/Processing336 The tendency of longer constituents to follow shorter ones is a well-established fact in linguistics, and Behaghel (1909) is usually quoted as the first to observe it (see also Wasow 1997: 348; Wasow 2002). Thus, for instance, it is not possible to transform the following sentence into a non-extraposed version: (64) a.

It was plain that whatever ‘everyone’ was saying, she was not going to believe it until Dumbledore told her it was true. (HP1.14) b. *That whatever “everyone” was saying, she was not going to believe it until Dumbledore told her it was true was plain.

However, less ‘heavy’ [that_CL] subjects can occur in canonical position with plain: (64) c.

Mr Jafferjee said: “That this was a significant failure within the Mental Health Care regime is plain. (newspapers)

Most authors refer to such elements using the concept of weight. Wasow (2002: 16ff) gives a brief overview of the different definitions of weight that have been proposed, some of which can only sensibly be applied to an analysis within the

|| 336 CamG (1405) treats weight and processing issues under the heading of pragmatics, which is at least debatable since measurements of weight are usually some sort of syntactic measurement; processing as a cognitive phenomenon also would require a very broad definition of pragmatics in order to be subsumed under it.

210 | Extraposition

grammatical framework for which they were developed. The simplest measure (and the one that is easiest to apply) is the number of words an element consists of. Hawkins (1994: 74f) claims that counting words is sufficient for the practical analysis of data, although his exact (theoretical) model uses a more elaborate but highly theory-dependent measure of counting nodes in a constituent structure. Wasow (2002: 32ff) challenges Hawkins’ simplification and concludes “that a single measure of weight may subsume what appear to be effects of both length and grammatical complexity” (Wasow 2002: 40), which is probably related to the fact that early research on the matter337 focused exclusively on grammatical complexity, thus treating subordinate clauses as heavier than noun phrases of the same length. The reason behind this analysis was the observation that in some cases shorter complex elements (e.g. a noun phrase with a short relative clause) lead to unacceptability if not in end-position whereas longer not-so-complex elements (e.g. a noun phrase with a long prepositional phrase as postmodifier) in the same position do not (Chomsky [1955] 1975: 477). In generative transformational grammar, ease of processing was proposed as an explanation of the observation that complexity seems to take precedence over length, which is still in line with today’s positions (cf. for instance CamG 1405f).338 We can see from examples with multiple embedding that complexity indeed seems to play a role: (65) a.

*That that he was angry was so obvious embarrassed her. (CamG 1406) b. It embarrassed her that it was so obvious that he was angry. (CamG 1406)

In such an example, processing constraints will lead to restrictions on the canonical form. The reason is not, however, the complexity in the sense of the number of nodes as such, which is much higher in the non-extraposed subject of (66):339

|| 337 Ross (1967: 56) for instance treats an NP as heavy if it dominates S, i.e. if there is an embedded clause. 338 Szmrecsányi on the other hand presents evidence from experiments which suggests that “researchers can feel safe in using the measure that is most economically to conduct, word counts” (Szmrecsányi 2004: 1031). 339 The number of nodes is of course highly dependent on the syntactic model used, but it is very unlikely that any model would lead to a higher number of nodes in (65a).

Factors influencing extraposition | 211

(66)

That the other tobacco diseases which of late years have begun to trouble the farmers in Rhodesia are due to an impoverished soil is suggested by the appearance of eelworm in this crop. (popmerge)

CamG proposes that “[t]he sequence of two identical subordinators in [(65a)] makes this basic version unacceptably difficult to process” (CamG 1406). This is correct insofar as multiple embedding with a double that is not usually possible, no matter where the element is placed:340 (67) a.

?When Johnson quit on Wednesday, the McCain headquarters issued a statement saying that that he had been selected in the first place raised “serious questions about Barack Obama’s judgment.”

The sentence becomes fully acceptable when there is the fact that instead of the second that as in (67b), though: (67) b. When Johnson quit on Wednesday, the McCain headquarters issued a statement saying that the fact that he had been selected in the first place raised “serious questions about Barack Obama’s judgment.” (nytimes.com) However, the same alteration does not lead to an acceptable sentence in the case of (65a): (65) c.

*?That the fact that he was angry was so obvious embarrassed her.

Although (65c) is more acceptable than (65a) it is still far from being perfectly acceptable. Thus the double subordinator that, which was proposed as the reason for the unacceptability of (65a) cannot be the sole factor at work. The difference in acceptability between (67b) and (65c) confirms the position that subjects in canonical position have to be less complex than extraposed subjects and postverbal complements can be. It is not possible, though, to give a hard and fast limit on the complexity of the subject. Karlsson’s (2007a) model of “constraints on multiple initial embedding of clauses” correctly predicts that (65a) is not acceptable, but neither this paper nor his paper on multiple centerembedding of clauses (Karlsson 2007b) can account for the unacceptability of (65c), which contains a combination of initial and central embedding.341 One

|| 340 See also, for instance, Rohdenburg (2003) for a discussion of horror aequi. 341 Initial embedding means that the subordinate clause occurs before any other element of the superordinate clause; central embedding applies if some material of the superordinate

212 | Extraposition

possible reason might be that too many “pending” clauses, i.e. clauses the predicate of which have not been read yet, lead to difficulties in processing and thus to unacceptability. In the case of (65a) and (65c), we have three pending clauses – the main clause, the first that-clause and the second that-clause – and all three predicates immediately follow one another. So unclear boundaries at the end of the subordinate clauses may also be responsible for the unacceptability. Generally, multiple embedding seems to be much less problematic if nonfinite clauses are involved: (68) a.

The deep trouble for Mr Gantt is that many black Democrats agree, however reluctantly, that the voters of North Carolina are simply not ready to vote for a black man, and that to nominate one would merely allow Mr Helms to extend his 24-year lease on the state’s Senate seat. (newsmerge) b. To try to outlaw something that’s widely seen as acceptable behaviour brings the law itself into disrepute. (britnews)

The second example also shows that length and complexity do not necessarily trigger extraposition if other factors, such as information packaging, favour the non-extraposed variant.342

7.3.2 Information packaging In English (and many other languages) the focus is usually at or towards the end of the sentence. So one function of extraposition with anticipatory it is that it “enables us to end the clause at a focal point” (CGEL 1392). Usually, focal information is also the new piece of information in a sentence. And, since often new information needs more words than given information, which can be represented through pronouns and other pro-forms, the locus of new and focal information often coincides with the syntactically heavier constituents discussed in the section before.343

|| clause is to the left and to the right of the subordinate clause; accordingly final embedding refers to subordinate clauses in final position within the superordinate. 342 CamG also notes that while “[i]n cases where both orderings are permissible, the weight factor may well result in the extraposed version being preferred” the “acceptability of the basic version certainly does not require that the content clause be fairly short” (CamG 1405). 343 See also the discussion of criteria 13 and 14 in Section 7.1.3.2

Factors influencing extraposition | 213

For CamG, extraposition of subjects is the unmarked case when the subject is a [that_CL], a [wh_CL] or a [to_INF].344 Thus, two “pragmatic constraints relating to familiarity status” (CamG 1404) of the non-extraposed variant apply. The first is that non-extraposed subjects are always treated as background knowledge (CamG 1404f). This does not mean that it has to be discourse-old, “all that is necessary is that the speaker be able to treat the information as familiar to the addressee” (CamG 1404), which is in line with Miller’s (2001) analysis. The following example, which is from the article on St. Basil the Great in the Catholic Encyclopedia of 1907, is an extreme case of a long non-extraposed subject: (69)

That he was in communion with the Western bishops and that he wrote repeatedly to Rome asking that steps be taken to assist the Eastern Church in her struggle with schismatics and heretics is undoubted; but the disappointing result of his appeals drew from him certain words which require explanation. (popmerge)

It is hard to tell whether the writer would really believe the initial coordinated [that_CL] complements to be shared background knowledge, although the use of undoubted points in that direction, but according to CamG (1405), discoursenew information that “is not being asserted, but rather is being treated as background knowledge” can occur in this position, too. No clear criteria as to what counts as an assertion are given. For Miller it is not necessary for the information to be discourse old as long as it is “directly inferrable” (2001: 683), which is of course also questionable in example (69). The second constraint is that “[v]ery often a non-extraposed content clause represents presupposed information” (CamG 1405) as in the following example: (70)

That the skin survives these daily torments is a remarkable tribute to its toughness. (CamG 1404)

The problem is that presupposition is neither a necessary nor a sufficient criterion345 for non-extraposition, so CamG’s ‘constraint’ is more of a tendency.

|| 344 Similarly CGEL: “But it is worth emphasizing that for clausal subjects (though cf 18.34 [which restricts V-ing extraposition]) the postponed position is more usual than the canonical position before the verb (cf 10.26)” (CGEL 1392). 345 The optional extraposition noted for factive predicates (Kiparsky/Kiparsky 1970: 145; CamG 1405) shows that presupposition cannot be tied to non-extraposed clausal subjects, for instance.

214 | Extraposition

Finally, there is a further information-structural tendency discussed elsewhere in CamG: Given that [V-ing] clauses usually occur in canonical subject position whereas [to_INF] clauses usually occur in extraposed position, CamG claims that “[t]o a significant extent, then, the difference between infinitival and gerund-participial is a matter of information packaging” (CamG 1254). Mair argues that such a difference between the two can even be maintained if both are extraposed: The gerundial construction, because of its affinity to noun-phrase tags and free -ingclauses, makes it easier to “split” the focus of emphasis in an utterance. By extraposing the infinitival subject clause the sentence (37a)To stay in an American hotel can be extremely expensive. can be reordered to yield: (37b) It can be extremely expensive to stay in an American hotel. Separating an extraposed infinitival subject clause from its matrix by a distinct intonational break is unusual, so that sentences such as (37b) often have one informational focus only (“hotel” in the unmarked case). An extraposed gerundial subject, by contrast, is very easy to separate from its matrix. A pause between the matrix and the extraposed subject, and heavy stress on both “expensive” and “hotel” would not be unusual at all in: (37c) It can be extremely expensive – staying in an American hotel. (Mair 1988: 61)

However, Mair also acknowledges that such cases are comparatively rare in his data. The most comprehensive study on the information structure of extraposition is Kaltenböck’s (2004) functional account in which he distinguishes “two basic types of it-extraposition: Given Complement Extraposition and NewComplement Extraposition” (2004: 203). While the latter is more frequent (71.5 % in Kaltenböck’s collection), the proportion varies considerably with text type. See Kaltenböck (2004: ch. 5.1) for the exact details. To sum up, we can state that there are no hard and fast rules as to how information structure influences the choice of the canonical or extraposed variants, but information structure is one of several factors playing a role in the selection of one option over the other.

7.3.3 Factivity A further factor relevant to extraposition is described by Kiparsky/Kiparsky (1970) in their article “Fact”, where they argue that predicates can be divided into two groups, so-called factive and non-factive ones. For our purposes, this distinction is relevant since “[f]or the verbs in the factive group, extraposition is

Factors influencing extraposition | 215

optional, whereas it is obligatory for the verbs in the non-factive group” (Kiparsky/Kiparsky 1970: 145).346 For sentential subjects they list the following verbal predicates as non-factive: seems, appears, happens, chances, turns out. We can confirm that chances and happens never occur with a clausal subject in canonical position in the parsed corpora, but the situation is more difficult for the other three. The case of appears (and the largely parallel seems) has been discussed in detail in Section 7.1.2 where it became obvious that non-extraposed variants are possible with these verbs if certain other elements are present in the clause (and possibly determine the range of permissible subjects). For turns out, a similar example could be found: (71)

As G K Chesterton said of Gnosticism, ‘That Jones shall worship the god within him turns out ultimately to mean that Jones shall worship Jones.’ (newsmerge)

Nonetheless the observation that sentences of the following type do not have a non-extraposed variant is correct: (72)

It turns out that there are generally not enough dislocations originally present in most crystals to account for the very extensive slip which can take place in a ductile material. (CEG 456)

However, the matter becomes more complicated if modality and [to_INF] complements come into play.347 Whether the canonical form is allowed seems to also depend on the interpretation of the subject as a fact, as can be seen in the following examples. In (73a) the extraposed subject does not receive a factive interpretation due to the modal verb would: (73) a.

It would be so great of him to let me sleep because I know he is really tired, too.

If we change the sentence to past tense, however, the extraposed subject does receive a factive interpretation:

|| 346 Kiparsky/Kiparsky also list adjectives as non-factive predicates, but since the obligatory extraposition explicitly applies only to verbal non-factive predicates, they can be largely ignored here. See Herbst (1983: 221ff) for a discussion of adjective valency and factivity. 347 It was noted by Kiparsky/Kiparsky that for some speakers “many of the syntactic and semantic distinctions we bring up do not exist at all” (Kiparsky/Kiparsky 1970: 147), so generalizations over multiple users of the language may prove difficult.

216 | Extraposition

(73) b. It was so great of him to let me sleep because I know he was really tired, too. (mindaandsam.blogspot.com) If we look at the same pair in the (less likely) non-extraposed variant, the first one works and the subject does not receive a factive interpretation: (73) c.

To let me sleep would be so great of him because I know he is really tired, too.

In the second version, however, there seems to be a clash between the subject and the past tense verb, at least for some native speakers:348 (73) d. ?To let me sleep was so great of him because I know he was really tired, too. Our hypothesis here is that the [to_INF] receives a non-factual interpretation if it is at the beginning of the clause, which then results in the clash that renders (73d) less acceptable if a predicate that requires a factual subject follows. The hypothesis is backed by the fact that we can replace the [to_INF] with a [V-ing] to form the perfectly acceptable (73e): (73) e.

Letting me sleep was so great of him because I know he was really tired, too.

This is in line with the results of the case study on non-extraposed [to_INF] subjects and modality in Section 4.2, where we were able to observe a tendency of such subjects to be associated with epistemic modality. However, to state that the semantics of a (non-extraposed) [to_INF] is non-factive and of a [V-ing] is factive is not right, as noted by Berndt, who rejects a general factive interpretation of [V-ing] elements since he gives examples which “illustrate the possibility of putting one and the same gerundial construction to various uses and demonstrate, in our opinion, the complete untenableness of attempts to ascribe some kind of ‘typical’ meaning to a particular construction type of gerundive nominals” (Berndt 1991: 49).349 Thus the observation made here must again be regarded as a tendency but not a hard and fast rule.

|| 348 One native speaker informant rejected both (73c) and (73d). 349 Mair argues, in a similar vein, that “the alleged ‘factivity’ of the gerund construction may be nothing more than an incidental result of the fact that the gerund is more noun-like than both infinitives and that-clauses and hence less likely to be extraposed. And, as is well known, a subject in topic position tends to be presupposed as given in the unmarked case” (Mair 1988: 60). However, it was argued here that the [to_INF] in canonical position is not treated as factive

Factors influencing extraposition | 217

We thus have to regard factivity as one of several factors involved in the decision as to which complement can be extraposed or not – only the obligatory extraposition with the five factive predicates listed above seems to hold generally.

7.3.4 Register It has been suggested that there is a register distinction guiding the uses of extraposition and non-extraposition. For instance, Collins, on the basis of a limited dataset of Australian English, finds that “[t]he pressure to extrapose would seem to be at its strongest at the speech end of the speech - writing continuum and at the informality end of the formality - informality continuum” (1994: 14f). However, given that – as Kaltenböck (2004: 23) points out – only 7 instances of non-extraposition were found in the spoken data, all of which were in the Hansard transcripts, the predictive power of Collins’ figures is limited. Kaltenböck himself used a larger sample, i.e. the entire British component of the International Corpus of English (ICE-GB), to obtain more reliable but less clear-cut results as summarized by the following table: Tab. 32: Frequency of canonical and extraposed variants in spoken and written texts (250 each) from Kaltenböck (2004: 132)

canonical

extraposed

n

%

n

%

Spoken

79

9.8 %

730

90.2 %

Written

138

12.4 %

971

87.6 %

Total

217

11.3 %

1701

88.7 %

In fact, it could be argued that Collins made his statement on a wrong assumption. He observes: “65 % of non-extraposed sentences occurred in the written data, and 35 % in the spoken” (1994: 14). This is in line with Kaltenböck’s figures, where roughly 64 % of non-extraposed sentences occurred in the written texts. However, as the data by Kaltenböck shows, the number of clausal sub-

|| and thus has to be extraposed if it should be interpreted as factive, which is not in line with Mair’s statement.

218 | Extraposition

jects in total is much lower in the spoken data than in the written data, so having a higher proportion from the written mode almost follows automatically and does not necessarily mean that the spoken data favours extraposition. Still, as Kaltenböck’s data suggest, there seems to be a stronger preference for extraposition in the spoken data, a fact that he explains by “the increased processing effort associated with clausal subjects” (Kaltenböck 2004: 133). It is important to note, however, that both Collins and Kaltenböck include obligatory extraposition (see discussion in Section 7.1.2), so that for a substantial portion of their extraposed sentences there is no non-extraposed variant available, so the numbers cannot necessarily be read as proportions in which the speaker opted for one or the other of two alternatives. Also, if the proportion of obligatory extraposition differs between the two subcorpora, this may skew the results even further. One more mode-related feature addressed by Kaltenböck is the overall proportion of extraposed [that_CL] and [to_INF] subjects. While in Erdmann’s (1987) study, which was based on a written corpus, [to_INF] extraposed subjects were the most frequent ones, they only come second after the [that_CL] subjects in Kaltenböck’s (2004) study, which is based on 50 % spoken material. Kaltenböck gives the following reason: A closer look at the to-infinitives in our corpus reveals a striking imbalance of distribution between the spoken and the written mode: unlike extraposed that-clauses, which are fairly evenly distributed over both modes with only a slight preference for written texts [crossreference omitted], to-infinitives occur much more frequently in writing (405 vs. 260 instances; = 60.9% vs. 39.1%); hence Erdmann’s unusually high number of to-infinitives. (Kaltenböck 2004: 92)

As for the formal vs. informal divide, Kaltenböck (2004: 133–135) also gives very detailed numbers from ICE-GB that confirm Collins’ tendencies cited above as he observes “a gradual increase of non-extraposition from the spoken or more informal categories to the more formal written ones” (Kaltenböck 2004: 133). Zhang (2015) claims that academic writing is particular likely to contain extraposed subjects, but the differences she presents are not all statistically significant and the more detailed account given by Kaltenböck does not support her claim.

7.3.5 Type of complement The type of the complement itself is another factor responsible for the choice of an extraposed over a non-extraposed variant. Erdmann (1987) shows how the

Factors influencing extraposition | 219

different realisations of subjects have different probabilities of occurring in extraposed position; the overall results of his corpus study are summarised in the following table:350 Tab. 33: Proportion of extraposition by complement type, based on Erdmann (1987: 41); figures not exact

clause type

canonical

extraposed

[that_CL]

10 %

90 %

[to_INF]

12.5 %

87.5 %

[for_ N_to_INF]

9.5 %

90.5 %

[V-ing]

50 %

50 %

It looks as if only [V-ing] shows a significantly different pattern from the other three types distinguished by Erdmann. His proportion for extraposed [V-ing] clauses is probably too high, though: Mair (1988), who worked on 445,000 words of spoken and 430,000 words of written English from the Survey of English Usage, found that in spoken language 45 % of [V-ing] clauses were extraposed and in written language 30 %. The reason for the discrepancy may be that Erdmann follows Jespersen’s broad definition of extraposition and thus includes instances of dislocation without anticipatory it as well (see Erdmann 1987: 34). A larger corpus-based study, viz. that of Kaltenböck (2004), gives a detailed account based on ICE-GB, summarized in the following table: Tab. 34: Frequency of non-extraposition compared with subject it-extraposition; figures from Kaltenböck (2004: 152), who also includes other types of complements

non-extraposition

it-extraposition

n

%

n

%

[that_CL]

12

1.6 %

731

98.4 %

[to_INF]

34

4.9 %

665

95.1 %

|| 350 For better readability Erdmann’s factors (“...9 times more frequent than...”) were converted into percentages and rounded to the nearest 0.5 %, which leads to the undesired effect of apparently higher precision than actually backed by the data.

220 | Extraposition

non-extraposition

it-extraposition

n

%

n

%

2

1.8 %

107

98.2 %

[V-ing]

128

72.7 %

48

27.3 %

Total

176

10.2 %

1551

89.8 %

[for_NP_to_INF]

Furthermore, as shown above, there are complements that never occur as subjects in canonical position but do occur in a structure that looks like obligatory extraposition. These include when-clauses as well as [if_CL], [as_if_CL], [as_though_CL] and [like_CL] complements. It is likely that there is a cline between extraposition and impersonal constructions and that all these are somewhere on that cline between the extreme positions.

7.3.6 Type of valency carrier If we include the type of valency carrier in our table of clausal subjects, the picture becomes more complex:351 Tab. 35: Proportion of extraposition by complement type and valency carrier, based on Erdmann (1987: 41, 92, 94, 101–103); figures not exact

clause type

type of valency carrier

[to_INF]

verb adjective

[for_NP_to_INF]

extraposed

28.5 %

71.5 %

6%

94 %

noun

18 %

82 %

verb

50 %

50 %

adjective [V-ing]

canonical

4%

96 %

noun

18 %

82 %

verb

91 %

9%

adjective

29 %

71 %

noun

44 %

56 %

|| 351 Erdmann makes no statement about [that_CL] complements.

Factors influencing extraposition | 221

Erdmann’s figures are sufficient to show that the form of the complement and the type of valency carrier both play a role in determining whether extraposition takes place or not. The findings are generally in line with the reference grammars’ statements as to the preference of [to_INF] subjects to be extraposed. The picture for [V-ing] is in line with Mair’s statement that “[o]ne of the factors most likely to prevent extraposition is the presence of a verbal (rather than a nominal or adjectival) matrix predicate” (Mair 1988: 55). The more detailed analysis by Kaltenböck presented below relies on a larger corpus basis. By and large it seems to confirm Erdmann’s figures. The data on [for_NP_to_INF] is too sparse to base any claim on it. However, it does not seem to confirm Mair’s position on [V-ing] with a nominal predicate, which seems to be quite rare, too. Tab. 36: Proportion of extraposition by complement type and valency carrier, based on Kaltenböck (2004: 152f)

clause type

[(that)_CL]

[to_INF]

[for_NP_to_INF]

[V-ing]

type of valency carrier

canonical

extraposed

n

%

n

%

verb

3

0,9 %

343

99,1 %

adjective

6

2,0 %

293

98,0 %

noun

2

2,4 %

81

97,6 %

verb

16

21,1 %

60

78,9 %

adjective

11

2,1 %

512

97,9 %

noun

7

8,5 %

75

91,5 %

verb

0

0,0 %

5

100,0 %

adjective

0

0,0 %

86

100,0 %

noun

1

6,3 %

15

93,8 %

verb

55

93,2 %

4

6,8 %

adjective

31

43,7 %

40

56,3 %

noun

35

89,7 %

4

10,3 %

7.3.7 Valency carrier Since the VDE lists extraposition properties as optional or obligatory, it was possible to compile a list of all complements that are listed as obligatorily extraposed in the active clause with their respective valency carrier:

222 | Extraposition

Tab. 37: Obligatory extraposition in active clauses in the VDE; for lexemes with multiple lexical units, the lexical unit in the VDE/Patternbank is given in parentheses

valency carrier

complement

comment

amuse(A)

[wh-CL]A:it

see discussion below

appear(B)

[(that)-CL]A:it

sort of copula; discussed in Section 7.1.2

concern(A)

[that-CL]A:it

counter-example

do(alpha)

[to-INF]A:it

counter-example

fall(E)

[to-INF]A:it

follow(B)

[that-CL]A:it

obligatoriness of extraposition depends on pattern (in VDE)

grieve(α)

[to-INF]A:it

counter-example

grieve(α)

[that-CL]A:it

happen(B)

[that-CL]A:it

hit(E)

[that-CL]A:it

only dubious counter-examples

look(C)

[as if-CL]A:it

complement can never be in canonical subject position

look(C)

[as though-CL]A:it

complement can never be in canonical subject position

look(C)

[like-CL]A:it

complement can never be in canonical subject position

matter()

[about N]A:it

occur(A)

[that-CL]A:it

occur(B)

[that-CL]A:it

obligatoriness of extraposition depends on pattern

owe(B)

[to-INF]A:it

possibly not a subject; counter-example

pay(α)

[to-INF]A:it

entry messy -> influence of pattern/other complements

remain()

[it + pattern of II]A:it

problematic counter-examples (only in coordination)

remain()

[to-INF]A:it

non-extraposed sentence only with tough movement

seem()

[Q/S]A:it

complement can never be in canonical subject position

seem()

[that-CL]A:it

sort of copula; discussed in Section 7.1.2

seem()

[as if-CL]A:it

complement can never be in canonical subject position

seem()

[as though-CL]A:it

complement can never be in canonical subject position

seem()

[like-CL]A:it

complement can never be in canonical subject position

seem()

[so/not/otherwise]A:it complement can never be in canonical subject position

sound(B)

[to-INF]A:it

problem with pattern of X

sound(B)

[as if-CL]A:it

complement can never be in canonical subject position

sound(B)

[as though-CL]A:it

complement can never be in canonical subject position

sound(B)

[like-CL]A:it

complement can never be in canonical subject position

surprise(B) [if-CL]A:it

complement can never be in canonical subject position

Factors influencing extraposition | 223

valency carrier

complement

comment

take(γ)

[to-INF]A:it

counter-example

tend(B)

[that-CL]A:it

counter-example

touch(B)

[to-INF]A:it

only one highly dubious counter-example

wash(C)

[to-INF]A:it

counter-example

The table was then verified with the help of the parsed corpora presented in Section 3.2.2 in order to find possible counter-examples. So, for instance, the [that_CL] with concern or the [to_INF] with wash can occur in the canonical position, even if they do so relatively rarely: (74)

That United were denied the chance to take the title in front of their supporters tonight will not have concerned Ferguson after Villa’s 1-0 defeat. (newsmerge)

(75)

A former chief risk officer at one of Britain’s biggest banks says that Northern Rock’s operating model was very risky: “To say that nobody could have envisaged what happened doesn’t wash at all.” (newspapers)

Some counter-examples are dubious, others are problematic in that the verb only occurs in coordination so that the structure may be influenced by the other valency carrier in the coordinated structure: (76) a.

To champion the sovereignty of weakness was - and remains - a profound challenge to the human agenda. (newsmerge) b. To take the country into not one but two recessions in which millions lost their jobs was unforgivable and remains so. (newspapers)

So for remain with a [to_INF] subject, it would be impossible to argue either for or against obligatory extraposition on the basis of the data. The complements that can never be in canonical subject position were already discussed above and need not be discussed here. For amuse, the [wh_CL] class has to be divided up into when-clauses, which fall into the aforementioned category, and others which – while they are not attested in the corpus – may still be acceptable:

224 | Extraposition

(77)

Where the King lost his crown amused me.352

There are, however, a few good examples for which no counter-examples were found. Compare for instance: (78) a.

It fell upon defeated Germany to take full responsibility for the incoming Germans. (VDE 299) b. ?To take full responsibility for the incoming Germans fell upon defeated Germany.

(79) a.

It grieved me that I had added to the sorrows she already had to bear. (VDE 361) b. ?That I had added to the sorrows she already had to bear grieved me.

As mentioned above, matter is the only verb in the VDE that seems to allow an extraposed [about_NP] subject which cannot occur in non-extraposed position: (80)

It doesn’t matter about the bed. (VDE 526)

This is, however, a borderline case in that the extraposed subject status is more than questionable (i.e. one may rather want to analyse the structure as an impersonal construction) and thus our discussion should not rely on such examples, even though there is the occasional prepositional subject as mentioned in Section 2.3.2. There are also verbs that seem to resists extraposition of their clausal subject altogether. The VDE gives for instance convince and persuade and the tests carried out with native speakers and reported in Appendices 3 and 4 confirm this finding.353 One can only speculate as to the reason for this restriction, but it may be due to a danger of confusion with postverbal clausal elements in the case of a divalent pattern and due to too complex intervening elements in the case of a trivalent pattern. Looking at the VDE, we can observe that the majority of verbs actually only allows for non-extraposed clausal subjects in the active clause: Less than 30 % of all [that_CL] active clause subjects are listed with

|| 352 This example was made up and accepted by two out of three native speaker informants. See Appendix 8. 353 There is, however, one example of an interrogative (which leads to obligatory extraposition) with convince in the parsed corpora where the extraposed element is an [if_CL]: But would it convince Tube users that we were fully paid-up members of the decadent west if we wore our work passes on our jackets, or carried a copy of Pride and Prejudice, or played the new Foo Fighters album very loudly on our iPods? (britnews)

Factors influencing extraposition | 225

optional or obligatory extraposition; this does of course not say anything about token frequency, which may just as well be the opposite, given that frequent verbs such as seem show obligatory extraposition. If we return to the table presented at the beginning of this chapter, we can observe that many of the restrictions on extraposition identified there are perfectly valid. So, to pick out two random verbs, commit seems to resist extraposition of the [to_INF]: (81) a.

To blame the lexicographers for recording common usage commits the irrational and primitive error of shooting the messenger. (newsmerge) b. ?It commits a terrible error to blame the lexicographers for recording common usage.

Similarly concentrate: (82) a.

The problem is that there are 75 ridings a year and to give a wider berth would concentrate our low flying over other parts of the countryside to the detriment of the inhabitants there. (HHV 16197) b. ?It would concentrate our low flying over other parts of the countryside to give a wider berth.

Given the abundant evidence from the VDE on this issue, further systematic studies are not necessary to claim that extraposition in the active clause is highly dependent on the valency carrier.354 In the passive voice the situation is different. For [that_CL] subjects, Erdmann remarks that passive verb forms are substantially more frequent than active ones in the it-matrix clause and can be regarded as the default for extraposed structures (Erdmann 1987: 70). For [to_INF] subjects, on the other hand, he finds that passive verb forms are not frequent in the matrix clause (Erdmann 1987: 89). We need not carry out a systematic here study to show that item-specific restrictions on extraposition are at work in the passive, too, since the data presented in Section 5 indicates that permit allows a non-extraposed [to_INF] in the passive whereas for other verbs, we can confirm CamG’s findings that “infinitivals are restricted to a few catenative verbs (e.g. decide, desire, hope, prefer), and then only in extraposed position” (CamG 1435).

|| 354 Although we have only looked at verbs here, data presented in Section 6.3 such as the difference between extraposed and non-extraposed [V-ing] with duty seems to suggest that the situation might be similar for other valency carriers.

226 | Extraposition

7.3.8 Other complements/valency pattern Some of the restrictions listed in the table in the previous section are not restrictions imposed by the valency carrier as such but restrictions that exist only when the valency carrier is used in a certain pattern.355 Thus for instance occur needs extraposition of the [that_CL] in the monovalent pattern but not in the divalent pattern: (83) a.

It frequently occurs that goods have to be returned to suppliers for various reasons, e.g. damaged or defective goods. (EA9 1728) b. ?That goods have to be returned to suppliers for various reasons, e.g. damaged or defective goods occurs frequently. c. It did not occur to him that he might also have been frightened. (CEC 1375) d. That it could help the client to better manage future projects did not occur to them. (newsmerge)

According to the VDE, pay and follow show the same behaviour, thus sentences (84a–c) are acceptable whereas (84d) is not.356 (84) a.

It follows from this interpretation that the difference between religion and morality is reduced. (VDE 322) b. That the difference between religion and morality is reduced follows from this interpretation. c. […] it follows that the BBC must be for the government in this crisis. (VDE 322) d. *That the BBC must be for the government in this crisis follows.

There are more cases in the VDE which are not listed in the table above357 but which can be found with similar restrictions, such as harm with a [to_INF] subject: (85) a. It shouldn’t harm to point this out. (VDE 375) b. ?To point this out shouldn’t harm.

|| 355 In some such cases one could also argue for different lexical units of the verbs in the different patterns and accordingly for different valency carriers. 356 This is in line with Erdmann (1987: 40). 357 The reason for these items not being listed in the table is that the VDE only differentiates between them at the pattern level but not at the complement level in the complement block and that the table was created on the basis of the complement block.

Factors influencing extraposition | 227

c. Yet to let him off would harm the rule of law. (newspapers) d. Yet it would harm the rule of law to let him off. Furthermore, there are cases where extraposition is not possible if certain other complements are present. According to the VDE, attract allows extraposition of [that_CL] and [to_INF] subjects in the divalent pattern but not in the trivalent ones:

Fig. 22: Complement block for attract from the VDE (58)

Corpus research seems to confirm this treatment since for instance the nonextraposed [to_INF] occurs in both divalent and trivalent patterns as in (86a–b), but the extraposed variant was only found in the divalent pattern as in (86c). (86) a.

But to do so could attract severe penalties such as deduction of points and, more seriously, a ban from European competitions. (newspapers) b. Labour sources dismissed the claim, pointing out that to take such drastic action would only create martyrs and might even attract more MPs to the rebel cause. (newspapers) c. As I thought it might attract suspicion to walk the same way again, I looked at the shops in Tottenham Lane instead. (fiction)

Similar but even more complex is the case of frighten, where extraposition is possible in one trivalent pattern but not so much in others:

228 | Extraposition

Fig. 23: Complement block for frighten from the VDE (334)

The native speakers interviewed on the topic confirmed the VDE’s analysis: (87) a. That he had escaped from prison frightened me to death. b. It frightened me to death that he had escaped from prison. c. That he had escaped from prison frightened me into staying in our flat. d. ?It frightened me into staying in our flat that he had escaped from prison.

7.4 Case study: tough movement In some sentences with extraposition, it is not unproblematic to identify the extraposed subject, as there are two elements which could replace it in canonical subject position. Thus based on (88a) both to follow this route (as in (88b)) and this route (as in the corpus example (88c)) can occur in canonical subject position: (88) a. It is tough to follow this route. b. To follow this route is tough. c. In bad weather even this route is tough to follow. (CMD 254) The [NP] this route in (88c) can be analysed as a complement of follow since it fills a valency slot of that verb as in (88a) and (88b). Such an analysis was proposed for instance by Lees (1960: 216–218), who showed that the verb’s selection restrictions apply to the subject of sentences such as (88c). It was taken up in the generative transformational literature (e.g. Rosenbaum 1967, Postal 1971), where the phenomenon received its name ‘tough movement’ because it was analysed to involve a movement transformation and occurs chiefly with adjec-

Case study: tough movement | 229

tives such as tough as in the example given above.358 As Oehrle noted, “[o]n the classic analysis, all three sentences share a common underlying structure” (Oehrle 1979: 583) with different rules that operate on the structure underlying (88b) in order to yield the structures presented in (88a) and (88c). However, the classic analysis is not accepted by all researchers in the generative tradition. Oehrle (1979: 585f), for instance, claims that tough to follow in (88c) above must be a constituent but tough to follow this route in (88a) is not a constituent, so he goes on to propose two distinct lexical structures for adjectives such as tough, one responsible for (88a–b) and the other for (88c) (Oehrle 1979: 591). CGEL discusses examples such as (88c) without recourse to derivative mechanisms simply by positing that “[t]here is an analogous construction in which the adjective is complement359 to an infinitive clause acting as (extraposed) subject” (CGEL 1229). The authors do, however, state that “the subject of the sentence is identified with the unexpressed object of the infinitive clause” (CGEL 1229), so the subject’s relationship to the verb is expressed. They also note that generally “there is no semantic implication between” (CGEL 1229) sentences (89a–b) below although with some adjectives, such an entailment relation exists, as in (90a) which implies (90b). (89) a. The bread was hard to bake. (CGEL 1229) b. The bread was hard. (CGEL 1229) (90) a. Jenny is nice to know. (CGEL 1229) b. Jenny is nice. (CGEL 1229) CamG treats the phenomenon in the larger context of what they call hollow clauses, i.e. non-finite clauses “in which a non-subject NP is missing” (CamG 65). This category includes, besides the [to_INF] type we have seen so far and some others that are of no relevance here, [V-ing] clauses as in the following example: (91)

This idea is worth giving some thought to. (CamG 65)

|| 358 Tough movement is a shorthand term for what in the generative theory is a process of object-to-subject raising. The term raising basically covers all cases in which an argument that is semantically part of a subordinate clause is realised as a constituent in the superordinate clause. Depending on the version of the framework, there is a transformational relationship between the non-raised deep structure and a raised surface structure (standard theory, minimalism) or not (GB). Osborne (2007) uses the term rising to model such structures in a dependency framework. 359 CGEL’s term complement corresponds to predicative element in the present study.

230 | Extraposition

However, since these “are licensed as complement to the adjectives worth and worthwhile, and to the preposition for with a purpose sense” (CamG 1246) only, they need not be discussed further here and are just another indicator of the item-specific character of complementation with regard to subjects and related constructions. CamG’s analysis is noteworthy insofar as it rejects a raising analysis for sentences such as (88a) above. The authors provide syntactic tests, such as the nonacceptability of existentials and clefting with such structures, which, they claim on the basis of their acceptability with other raising structures such as the one with seem, should be allowed if the subject really is a raised element from the subordinate [to_INF] clause (CamG 1247f).360 They also present the following pair of sentences and claim that if raising were at work, the acceptability of (92a) would (wrongly) predict the acceptability of (92b). (92) a.

It has been a pleasure to listen to someone with so much enthusiasm. (CamG 1248) b. ?Someone with so much enthusiasm has been a pleasure to listen to __. (CamG 1248)

The issue is to some extent theory-internal and there may actually be other factors at work (see for instance the constraints on information structure discussed in Section 7.3.2) preventing the non-extraposed variant in this particular example, so whether such structures are treated as instances of raising or not is only of theoretical interest within models that allow such a mechanism at all. In principle, the phenomenon of tough movement is not limited to noun phrases, as can be illustrated with the help of the following examples in which a [that_CL] can be ‘raised’ from the subordinate clause to canonical subject position in the matrix clause: (93) a.

It was quite difficult to believe that there could be any animation in them. (G1L:2500) b. To believe that there could be any animation in them was quite difficult. c. That there could be any animation in them was quite difficult to believe.361

|| 360 CamG also claims that there is a semantic difference between the ‘tough-moved’ structure in (88c) and the extraposed (88a) in that in the former, tough to follow “denotes a property that is predicated of” (CamG 1248) this route whereas this is not the case in the latter. 361 One native speaker informant called this use “archaic”.

Case study: tough movement | 231

Given that Kaltenböck (2004: 93) observes that in such examples “it is not entirely clear whether the to-infinitive […] is best to be taken as part of the complement clause or rather as part of the matrix clause”, we shall take a closer look at clausal subjects in these structures and thus try to determine the valency relationships in such sentences in the remainder of this section. The evidence collected for the present study contains both material for and against an analysis in which the verb of the [to_INF] clause and not the predicative adjective is seen as generally governing the subject in structures of the type shown in (93c). As pointed out in the literature (CamG 1246, 1250; Mair 1990: 59), tough movement structures also occur with nouns (to a lesser extent) and verbs (very rarely). The theoretical treatment, however, often ignores verbs, as Mair confirms:362 Except for the odd speculative footnote (e.g. Chomsky 1981: 319, on take), the verbal matrix predicates illustrated above [take, cost, require, need, be for, be up to] do not figure in the extensive literature on Tough Movement [...]. (Mair 1990: 60)

For verbs, Mair only found instances of tough movement with take in his data from the Survey of English Usage (SEU), but predicted the structure to occur with a few other ones. With the help of our parsed corpus we can show that his example of a sentence with require, which is reproduced as (94a), is a strikingly accurate proxy of actual usage as in the corpus example given as (94b) below. (94) a.

The questionnaire should not require more than ten minutes to complete. (Mair 1990: 60) b. Between the two phases subjects answered two brief questionnaires which together required approximately five minutes to complete. (HPM 1051)

If we look at the case of take presented in Section 4.1.2.2 above, we noted there that generally take cannot occur with a [that_CL] subject, but it does in the presence of a postverbal [to_INF] clause whose verb takes a [that_CL] postverbal complement in finite active clauses. The relevant examples are reproduced here: (95) a.

That this may not be the case in certain instances does not take much imagination to comprehend. (HP3 929) b. ?That this may not be the case in certain instances does not take much imagination.

|| 362 For a list of valency carriers that typically occur in such structures see CamG (1246, 1250).

232 | Extraposition

c.

It does not take much imagination to comprehend that this may not be the case in certain instances.

This behaviour is strong evidence for relating (95a) and (95c) by some sort of rule. The case of remain is more difficult to decide. We can still relate the tough movement construction in (96a) to the extraposed construction in (96b), where it becomes clear that the whether-clause depends on see, but the nonextraposed version in (96c) is not possible. (96) a.

Whether such modest dieting is enough remains to be seen. (ABG 2096) b. It remains to be seen whether such modest dieting is enough. c. *To be seen whether such modest dieting is enough remains.

Furthermore, there are variants of the construction without a postverbal [to_INF],363 such as (97a–b): (97) a.

Whether an electric car can succeed in its present form remains as big a question as ever in my mind after a test drive in central London. (newsmerge; double after removed) b. Whether the credentials of Nasa and its staff can be salvaged by the repair of the Hubble remains in doubt. (newsmerge)

These do not look like tough movement structures due to the missing postverbal [to_INF], but remain on its own does not seem to have a valency slot for whether-clauses in subject position. However, since that remain is sometimes treated as a copula verb,364 the form of the subject found in such sentences may be determined by the postverbal element in such sentences. The case of hard also provides inconclusive evidence. The [that_CL] in canonical subject position was only found with postverbal [to_INF] in what looks like a tough movement structure:365

|| 363 The pattern with remain + to be seen is by far the most frequent one; it accounts for roughly 98 % of all occurrences of a whether-clause subject with remain in our parsed corpora; the others are perfectly acceptable, though. 364 See CGEL (1172). 365 In extraposed position, a [that_CL] seems to be acceptable with hard as in the following example: It did seem a little hard that, just when things were beginning to go really well, the even tenor of their lives should be upset by the arrival of someone else — someone who, although a friend, was none the less a stranger. (BMU 2572) One native speaker commented that all examples of this kind constitute a wrong use of hard and should be constructed with harsh instead.

Case study: tough movement | 233

(98) a.

That Leapor is concerned with land use in ‘The Month of August’ is hard to dispute; her observations, however, are oblique. (AN4 3003)

This counts as evidence in favour of an analysis in which the [that_CL] fills a valency slot of the verb of the postverbal [to_INF] clause, in this case dispute. The situation for infinitival subjects with hard is quite complex. Generally, hard allows a [to_INF]366 complement as a canonical subject, as in the following example: (98) b. To write about the events up there was hard, sometimes a real struggle, but it had to be. (ECG 1468) Then there are (infrequent) cases in the corpus where one could analyse the subject [to_INF] as a dependent of the postverbal [to_INF], i.e. as instances of tough movement: (98) c.

To go out on penalties is hard to accept. (newspapers)

The vast majority of these sentences, however, contain a verb in the postverbal [to_INF] clause that does not usually allow a [to_INF] clause as a postverbal complement. By far the most frequent verb in this pattern is take as in (98d), so one might argue that hard to take is an idiomatic expression that should not be split into smaller parts,367 but (98e–g) cannot be accounted for by such an analysis:368 (98) d. To lose like that after playing so well is hard to take. e. To be passed out in 2C would be hard to explain, but fortunately Lauria scraped up a return to 2H. (newspapers) f. To have lost that much and still have light winds and to have them for the foreseeable future is pretty hard to deal with. (newspapers) g. For this to happen within the past 12 months is hard to believe. (newspapers) While (98e–g) look like instances of tough movement superficially, we could not transform any of them into the ‘non-tough-moved’ variant without changing the

|| 366 It also allows a [for_NP_to_INF], which can be subsumed under the [to_INF] category here. 367 Since hard to take commutes with alternatives such as difficult to take and hard to bear, such a treatment would be questionable, though. 368 See however Chomsky ([1981] 1993: 312), who considers analysing easy-to-please in John is easy to please as a complex adjective.

234 | Extraposition

clause type of the complement that functions as subject: For explain, we would have to use a [that_CL] or a [V-ing], for deal with a [V-ing] and for believe a [that_CL]. As we have seen in (98a) above, a [that_CL] is in fact possible in tough movement with hard, and the following example shows that [V-ing] clauses can occur there, too: (98) h. Earlier the expert admitted that stopping a horse was hard to detect. (newspapers) Thus there is no syntactic reason that would require a change of form of the element that is moved to the subject position from a supposed ‘underlying’ or ‘basic’ sentence such as the extraposed version. Accordingly, we must deduce that the subject [to_INF] in (98e–g) is not a valency complement of the verb in the postverbal [to_INF], but it does of course realise a participant role of that verb. We thus have to conclude that while a tough movement analysis has its advantages for the analysis of some sentences (such as the one with take cited above), it cannot be extended to all sentences of a similar pattern369 and the open question in this respect is how to determine where the generalization stops. Thus we can agree with Mair in that a gradient analysis is probably best: Tough-Movement structures and SVC patterns of the type ‘She was pretty/to look at’ (cf. ‘*It was pretty to look at her’) should be regarded as opposite ends of a gradient rather than two discrete structural categories. (Part of the reason for the stalemate in generative grammar between the proponents of a ‘movement analysis’ and those of an ‘object deletion analysis’ is that in their attempts at rigid formalisation researchers fail to allow for precisely this indeterminacy in much of their data.) (Mair 1990: 63)

However, our ‘gradient’ has at least three poles: the two mentioned by Mair and a third structure exemplified in (98d–f) above. We can also agree with Mair as to his analysis of the problems with ‘rigid formalisation’ or simply the failure to allow for indeterminacy: If we have to decide for every given sentence whether it is an instance of structure A or structure B, large overlap between structures – as illustrated above with hard, for instance, – will result in relatively arbitrary groupings which do not capture all the generalizations one might want to make.

|| 369 There is another similar pattern mentioned for instance in CamG in the same context: “The document is now ready [for you to sign __].” (CamG 1247) The difference is that the adjectives that occur in this pattern do not allow the basic variant, so one cannot regard these cases as instances of tough movement: “*It is now ready [for you to sign the document].” (CamG 1247)

Conclusion | 235

Again, this shows that any model in which every element has to belong to exactly one category is too inflexible to account satisfactorily for all sorts of generalisations. A possible treatment of such structures in valency grammar will be proposed in Section 9.3.2.2; a more flexible model that takes into consideration cognitive aspects and may also be able to account for the evidence presented here is briefly discussed in Section 9.3.3.

7.5 Conclusion We can conclude this chapter with a confirmation of the original hypothesis presented in Section 1.3: Whether subject extraposition is impossible, optional or obligatory is indeed a property of the valency carrier.370 In fact, we have to go even beyond the original hypothesis and state that whether subject extraposition is impossible, optional or obligatory can also depend on a combination of the valency carrier with other complements in the clause. However, as we have seen, valency is not the only factor contributing to the choice for or against extraposition. Besides some general tendencies as to the occurrence of extraposition depending on the type of complement and on the type of valency carrier, there are processing constraints and functional factors such as information packaging at work. Furthermore, semantic factors (e.g. factivity) play a role, too. From a theoretical point of view, these results create interesting problems in modelling: As outlined in Chapter 2, the term valency is used to describe itemspecific (or subclass-specific – see Engel 2004: 89) information on the number and form of the complements of a valency carrier. It is thus not immediately clear whether we would want to describe optional or obligatory extraposition as a valency property: Depending on the theoretical position taken as to the status of it and the extraposed subject (see Section 7.1 for discussion), extraposition may well be beyond the traditional notion of valency. In a purely complement inventory-based approach to valency, it would in fact be difficult to model the influence of the other complements in the sentence, since one would have to posit for instance one [that_CL] complement that can be extraposed and a second one that cannot. The choice between these two would still be problematic. A much simpler and more accurate – though more verbose – way of modelling such behaviour of lexical items is the use of valency patterns of the type used in

|| 370 As mentioned above, there is strong evidence for verbs, some evidence for nouns and less evidence for adjectives.

236 | Extraposition

the Erlangen Valency Patternbank (Herbst/Uhrig 2009), where both extraposed and non-extraposed patterns are listed where they occur. Such a model will be outlined in Section 9.3.2.1. Whatever the exact theoretical model might be for extraposed sentences (e.g. whether it allows for obligatory extraposition or not), we can be certain that there are item-specific restrictions at the level of the valency carrier and at the level of the pattern, and any model has to be able to account for these. The present chapter also provided considerable evidence for the hypothesis that subject properties do not coincide in one constituent in non-canonical clauses. The discussion in Section 7.1.3 showed how difficult it is to identify one element as the subject in extraposed structures compared to prototypical canonical instances such as noun phrases in pre-verbal position. Furthermore, the analysis of so-called tough movement structures and related problems has shown that there is a very complex interplay of constructions which cannot sensibly be accounted for in a static model of grammar. Together with the array of factors responsible for the choice of extraposed or canonical structures, such constructions can only lead us to reject any model that uses independent syntactic, semantic and pragmatic components (as often found in early generative literature). Instead, the evidence points in the direction of a model that allows for storage and generalizations (or ‘rules’?) side by side for similar phenomena, possibly based on approaches such as Bybee’s network model (see for instance Bybee 2007, 2010) with emergent structures. A rough sketch of such a model will be discussed in Section 9.3.

8 Existentials The present chapter deals with structures such as the following: (1)

This time, however, there was a change. (A0F 9)

(2)

There’s a gym installed at my villa. (JYD 1892)

These so-called existential sentences are interesting and relevant for a study of subjects for at least three reasons: First, it is not self-evident what should be analysed as the subject in such sentences. Secondly, there are sentences which look as if there is a predication involving an element one could call subject and some sort of predicate after the main verb, as in (2), so the role of that predication and its ‘subject’ needs discussing. Finally, in the light of their departure from the canonical patterns of declarative sentences, it may also be necessary to ask to what extent a category such as subject actually makes sense in the description and theoretical modelling of existential structures. In order to approach these questions systematically, we shall first discuss the boundaries of the construction and give a survey of different structures and their properties before we match them against the properties for subjecthood. The final section deals with models of the syntactic structure of existentials.

8.1 Phenomena covered 8.1.1 Limitation to there-clauses Examples (1) and (2) are prototypical instances of a type of clause that is said to serve to assert the existence of entities or “to bring the existence of an entire proposition [...] to the attention of the hearer” (CGEL 1403). They represent what the term existential371 is commonly used for, i.e. “the type introduced by unstressed there, accompanied by the simple present or past of be” (CGEL 1403). There is, however, less agreement as to the boundaries of the construction. Example (3), which is categorised as an existential sentence in CGEL, is treated as a presentational clause in CamG due to the fact that it has “some other verb than be” (CamG 1390).

|| 371 The term was originally proposed by Jespersen (1924: 155). https://doi.org/10.1515/9783110589801-257

238 | Existentials

(3)

In 1983, there appeared a new study of the issue by Dr Robert Eisenman, Chairman of the Department of Religious Studies at the University of California, Long Beach. (EDY 1078)

On the other hand, (4) and (5) are treated as existentials by CGEL (1402f) even though they do not contain there, but not by CamG.372 (4)

I have a car blocking my way. (CGEL 1402)

(5)

One finds a similar conflict in other cultures, other mythologies, other cosmologies. (EDY 1429)

From a functionalist perspective, Gast/Haas (2011) treat there-existentials and so-called ‘locative inversion’ as in (6) as formal sub-types of presentational/presentative structures, i.e. structures whose function it is to introduce novel participants (marked by subscript NOV below). (6)

Until the end of the war so very few folk had beards, and then only short ones nicely trimmed, but into the room came [a young man with a black fuzz of over eight inches]NOV. (Gast/Haas 2011: 129)

Due to their slightly different functions (see Gast/Haas 2011: 130) and their quite different structures, locative inversion will not be covered in this chapter. Finally, Kjellmer calls the it in (7) “existential it” (Kjellmer 2001: 328) and relates it to existential there but comes to the conclusion that it appears to be part of a relatively fixed expression “when it comes/came time” (Kjellmer 2001: 334). (7)

Here in Canada we are badly isolated from regular contact with soccer, so when it comes time for the World Cup, the cable sports network dredges up some living bodies to do the commentary. (J1G 415)

In this study, following CGEL, presentational sentences will be treated as a subtype of the existential construction and not, as in CamG, as a different, though related construction.373 However, following CamG, the term existential will be

|| 372 While sentences such as (4) and (5) are similar to existentials as defined above in their information structure and thus may merit to be classified as the same construction on such grounds, they are relatively straightforward in terms of syntactic structure and valency, with I and one as SCU, have and find as VHC and the postverbal NPs as PCU. (Terminology follows Herbst/Schüller 2008.) 373 Despite the proclaimed distinction between the two types of clauses, CamG treats them in the same section, entitled “Existential and presentational clauses” (CamG 1390ff).

Phenomena covered | 239

restricted to the there-type of clause, not only due to its overwhelming predominance but also because it is grammatically distinct from the other types presented by Kjellmer and CGEL. The term existential there will be used to refer to the preverbal there in such structures.374

8.1.2 Existentials vs locative there-clauses Furthermore, existentials have to be kept apart from a superficially similar locative structure presented in (8): (8)

Aye, where’s my card! I didn’t get, oh there’s my card! (KCU 3104–5)

Although Herbst/Schüller claim that in existentials “in some cases an element of the locative meaning of the particle use can also be made out” (Herbst/ Schüller 2008: 72), as in (9a–b) below, there are reasons that argue against their interpretation of these two sentences as existentials.375 (9)

a.

There, hesitating on the threshold, was the most beautiful girl he had ever seen in his life. (Herbst/Schüller 2008: 72) b. “There’s Swallow,” said Dempsey. (Herbst/Schüller 2008: 72)

Both sentences might rather be read as inverted structures with a space adjunct at the beginning, i.e. not as existentials. That is, the first word is the locative

|| 374 Erdmann (1976: 76) lists a range of alternative terms that have been proposed for the existential there such as “pro-nominal there”, “anticipatory there”, “formal there”, “preparatory there”, “empty there”, “expletive there”. 375 See however Freeze (1992), whose article on “Existentials and other Locatives” claims that his “unified theory of locative predications, based on an extensive cross-linguistic corpus, will result in the surprising conclusion (inter alia) that forms like English existential there are locative” (Freeze 1992: 554). However, this is only possible in his account due to heavy use of structure-altering mechanisms in the D-structure of generative grammar. Collins (1992) offers a slightly different perspective: “It should be remembered, however, that every assertion of existence has a locative dimension, whether explicit or implicit. In the case of existential sentences the assertion of existence is a function of the entire construction, not simply the there (and it is of course also a function of certain there-less sentences)” (Collins 1992: 421).

240 | Existentials

particle376 there and not the form there used in existentials. For (9a), such a reading is in fact obligatory if we follow the reference grammars’ analysis that existential there “is always unstressed” (CamG 1391; similarly CGEL 1405),377 although such a distinction is of course less than straightforward in the written language. Besides the locative reading, example (9b) could theoretically be read as an existential sentence with unstressed there,378 but it would be difficult to maintain that the locative meaning of the there is still present in such a use, given that it is possible to add a space adjunct as in (9c), including the locative particle there, or even a locative expression with a meaning contradicting the meaning of the locative particle there as in (9d). (9)

c. “There’s Swallow over there”, said Dempsey. d. “And there’s Swallow right here next to me”, said Dempsey.379

|| 376 While the terminology differs between grammars, the analysis remains constant. The term particle follows Herbst/Schüller (2008); other terms include adverb (CGEL) and intransitive preposition (CamG). 377 A slightly different position is taken by the Cambridge English Pronouncing Dictionary (18th edition): “The weak forms occur only when there is used existentially as in ‘there is’, ‘there are’, ‘there was’, ‘there won’t be’, etc. The strong form /ðeəʳ/ [US-symbol] /ðer/ is also used in such expressions, and is the normal pronunciation for there as a place adverbial, e.g. ‘there it is’” (EPD18: 491; their emphasis). However, in the treatment of weak forms, it is made clear that strong forms are usually only used “when the word is being quoted […], when it is being contrasted […] and when it is at the end of a sentence” (EPD18: 580). 378 If we take a look at the larger context of (9b) in David Lodge’s novel Small World, we can see that the sentence is definitely locative: “Which one is Professor Swallow?” Persse enquired, looking round the room. “He’s here somewhere.” Dempsey rather unwillingly scanned the sherry drinkers in search of Philip Swallow. […] Then, to his extreme annoyance, a tall, slim, distinguished-looking man of middle age, with a rather dashing silver-grey beard, and a good deal of wavy hair of the same hue around the back and sides of his head, but not much on top, darted forward to greet the girl, blocking Persse’s view of her. “There’s Swallow,” said Dempsey. “What?” said Persse, coming slowly out of his trance. “Swallow is the man chatting up that rather dishy girl who just came in, the one in the black dress, or should I say half out of it? (Lodge 1984: 8) 379 A locative reading is not wholly excluded here. If there is pronounced as full form and if there is an intonational boundary between Swallow and right, one could understand the sentence as containing an instance of dislocation. (See the discussion in Section 7.1.1.)

Survey of structures | 241

Examples (10a–b) show that this type of structure also exists in naturally occurring language and thus support the idea that existential there carries “none of the locative meaning of the place-adjunct there” (CGEL 1405).380 (10) a. There was no one there but the family. (A0D 1909) b. There is no prison here. (A03 849) Breivik/Martínez Insua find that in fact both types of there “were already distinguished in Old English” (Breivik/Martínez Insua 2008: 352) and cite evidence from Scandinavian languages to support their claim that a grammaticalization process had already begun prior to the Old English period.381 The process of semantic reanalysis described by them is highly plausible and fully compatible with later research, e.g. Gast/Haas (2011: 149f) and with other approaches to similar phenomena (see for instance Dauses 2000, in particular Chapter 2).

8.2 Survey of structures Both major reference grammars (CGEL and CamG) divide up existential clauses based on the main verb (be or other verb) and based on the displaced subject382 (only [NP] or [NP] with some extension). Be is by far the most common main verb in existential clauses and also offers a wider range of possible complementation patterns than the other verbs that occur in the construction. In the BNC, a search for “there_EX0 _VB?”383 (existential there followed by a form of the verb be) yields 199,652 results whereas a search for “there_EX0 _VV?”384 (existential

|| 380 Hannay (1985: 23–28) gives an overview of the diverging views on the relationship between existential there and locative there in the (predominantly generative) literature. 381 To what extent the existential construction has come to dominate over competing constructions in the evolution of the English language is illustrated convincingly using evidence from the Old Bailey Corpus (Breivik/Martínez Insua 2008: 354). 382 For lack of a better term displaced subject (CamG 67) is used for the postverbal constituent(s) No statement as to the subjecthood of the item(s) is made by this choice. 383 The CLAWS5 tagset distinguishes between the existential and adverbial uses of there. Even though it has to be borne in mind that this is a particularly error-prone distinction for a computer program, the frequency difference between be and the other verbs is sufficiently great for us to assume that the classification errors are negligible for our purposes. 384 This search pattern also finds instances of verbs followed by to be, such as seems to be, appear to be, proved to be, continues to be, needs to be which might also be counted for the be category. If we discard these cases, only 2,511 hits for verbs other than be remain.

242 | Existentials

there followed by a verb that is not be, do, or have) only returns 3,915 hits.385 Existential clauses with be and only an [NP] as displaced subject are called bare existentials (CGEL 1406, CamG 1393); if there is more than an [NP] after the verb be, they can be called extended existentials (CamG 1393). As pointed out above, if the main verb is not be, the term presentational is often used. We shall follow this threefold distinction, which is also used by Aarts (2011), although it would of course also be possible to distinguish between bare and non-bare presentational clauses.

8.2.1 Bare existentials A minimal example of a bare existential is given as (11): (11) a.

There is no cure. (A01 7)

Example (12) below contains a more complex noun phrase with a postmodification and two extra elements, thus and inevitably, but is still a bare existential because there is no further complementation after the verb other than an [NP]. (12)

Thus there is inevitably a link between democracy and equality. (EVP 594)

We shall see in the next section, however, that the distinction is not always straightforward and that ambiguity between the structures can arise due to the structural identity of postmodifications in the [NP] and extensions.

8.2.2 Extended existentials Non-bare existentials are treated differently by CGEL and CamG. CamG uses the term extended existentials and draws a distinction between various types of

|| 385 These numbers can only be taken as rough indicators since there are also 21,844 hits for existential there followed by a modal verb (which is followed by be in the vast majority of cases), 10,962 instances of a form of have (most of which are followed by been, some by to be) and 418 instances of do (about 280 of which are followed by seem [and about 220 of these are in turn followed by to be]). Furthermore, subject-operator inversion is not reflected in the queries, and semi-auxiliaries (there is to be, there is going to be) may also slightly distort the results. Nevertheless, with the results being almost two orders of magnitude apart, there can be no doubt as to the overwhelmingly higher frequency of be in existential clauses.

Survey of structures | 243

extensions that is not based on a single criterion; it rather groups together items on the basis of their meaning (locative and temporal extensions), their function (predicative extensions) or their form (e.g. infinitival extensions) (all CamG 1394). CGEL classifies according to the canonical clause type corresponding to the respective existential clause and offers a separate treatment of “[e]xistential sentences with relative and infinitive clauses” (CGEL 1406). Given that CGEL wants to “relate basic clauses to existential forms by means of a general rule” (CGEL 1403) and speaks of an “original clause” (CGEL 1403) (=basic clause), a derivative interpretation seems to be suggested. The following table presents the classification and examples in CGEL: Tab. 38: Basic clauses and their existential counterparts; constructed on the basis of CGEL (1404)

clause type

basic clause

existential clause

SVC

Something must be wrong.

There must be something wrong.

SVA

Was anyone in the vicinity?

Was there anyone in the vicinity?

SV

No one was waiting

There was no one waiting.

SVO

Plenty of people are getting promotion.

There are plenty of people getting promotion.

SVOC

Two bulldozers have been knocking There have been two bulldozers knockthe place flat. ing the place flat.

SVOA

A girl is putting the kettle on.

There’s a girl putting the kettle on.

SVOC

Something is causing my friend distress.

There’s something causing my friend distress.

CGEL posits two restrictions on the basic clause in order to allow a “conversion” into an existential with be: the subject generally has to be indefinite and the verb phrase must contain some form of the verb be (CGEL 1403). Breivik (1983), however, opposes the indefiniteness constraint using counter-examples of the following type: (13)

There are the facilities here, in the experimental psychology faculty. (FR3 1676)

Collins takes a similar position in his study of existentials, where he shows that (in his data) 4.5 % of postverbal [NP]s are definite (Collins 1992: 422) for which he offers the following convincing explanation:

244 | Existentials

The restrictions on the post-verbal NP are more profitably interpreted from a pragmatic than a grammatical perspective. What is crucial about the post-verbal NP is the discourse salience of the entity or entities referred to, the newness of the information it expresses (and definite NPs may be used to newly identify a referent in certain contexts). (Collins 1992: 422)

8.2.2.1 Relative clause extensions No restrictions on definiteness apply to existentials with relative clauses. They are formed on the following pattern: (14)

there + be + noun phrase + relative clause (CGEL 1406)

The pattern is exemplified in (15a) (relative pronoun is subject) and (16a) (relative pronoun is direct object). CGEL also relates this type of existential to a basic clause pattern which is illustrated (in a reduced form) in (15b) and (16b) respectively. (15) a.

[H]e’d only known there was something that kept him empty inside, stopped him from being a proper person. (F99 2301) b. Something kept him empty inside.

(16) a.

All these moves will help, but there is something that we as individuals can do — if we want to. (K9H 135) b. We as individuals can do something.

The pattern cited in (14) is more than surprising, given that in CGEL’s syntactic model adnominal relative clauses of the type shown above are usually treated as postmodifications within the [NP] and should thus not be listed separately from the [NP]. If we thus reduce the pattern to the more standard there + be + NP, it becomes indistinguishable from the pattern of bare existentials, and indeed one may ask whether there are syntactic reasons to separate the two. We shall discuss this issue in more detail in Section 8.2.2.4 below.

8.2.2.2 Infinitival extensions The extended existential with infinitive in CGEL “is problematic to the extent that it cannot be directly related to the basic clause types” (CGEL 1407). CGEL relates the structure presented in (17a) below to the “stiffly formal” (CGEL 1407) relative clause construction in (17b) but does not posit a derivative relationship.

Survey of structures | 245

(17) a. You think there’s something to talk about? (GV8 1928) b. (?)You think there’s something about which to talk?386 This relationship is less obvious in cases such as (18a), where the form nowhere cannot be postmodified by a relative clause (see (18b)). (18) a. It was hot and sticky and there was nowhere to sit. (FBL 1068) b. *It was hot and sticky and there was nowhere where to sit. There are two further issues in (19a) which suggest that there is a problem with the relationship between the [to_INF] and relatives. (19) a.

There are glorious walks to be found in the Sierra and although there is no proper beach to speak of in Deya, there is a small cove within walking distance and the town Puerto Pollensa — with a long sweeping unspoilt beach — is just a short drive away. (BPF 1787)

The first existential clause seems to prevent a [to_INF] relative clause as in (19b), but even if we transform it into a finite relative clause as in (19c), there remain considerable semantic differences. (19) b. *There are glorious walks which to be found in the Sierra. c. There are glorious walks which are to be / can be found in the Sierra. In fact, the basic clause in (19e) probably captures the meaning of the original clause better than the relative construction and better than (19d), which is an indicator of the modal component inherent in many [to_INF]s (as discussed in Section 4.2 above).387 (19) d. Glorious walks are to be found in the Sierra. e. Glorious walks can be found in the Sierra. However, the active would usually be preferred if there are no further constituents after Sierra, probably for information structural reasons:

|| 386 Not all native speakers seem to accept such a construction. This even holds for CGEL’s own example “At last there was something about which to write home.” (CGEL 1407), which is of questionable acceptability for some speakers. 387 Such a view is in line with an analysis within “a functional-semantic framework, where infinitivals are relatable to general modal concepts ([There are many questions to be asked.] being interpreted as a communicative variant of Many questions should be asked)” (Collins 1992: 424).

246 | Existentials

(19) f.

You can find glorious walks in the Sierra.

The second existential in (19a) also changes substantially in meaning when the [to_INF] is converted to a relative clause as in (19g). (19) g. ?There is no proper beach of which to speak in Deya. The reason here is the status of negative marker + X + to speak of as some sort of relatively fixed, idiomatic grammatical expression – illustrated in (20) below – which would probably best be described in terms of construction grammar (see for instance Fillmore/Kay/O’Connor 1988 and Kay/Fillmore 1999 for a treatment of similar though more complex constructions). (20)

She was disappointed, even aggrieved, that the remains would not be brought home for a funeral. Since there were no remains to speak of. (FRC 8–9)

It is thus problematic to treat infinitival extensions as related to a more basic version or simply as a variant of a relative clause extension and we should probably consider the construction as structurally sufficiently different to be treated in its own right. CamG also points out that the postverbal [NP] can receive different interpretations in relation to the verb of the [to_INF] so that in “[(21a)], a few replies is understood as subject of the infinitival, while in [(21b)] one letter is understood as object of the infinitival” (CamG 1394). (21) a. There are still a few replies to come. (CamG 1394; their emphasis) b. There’s one letter (for you) to sign. (CamG 1394; their emphasis)

8.2.2.3 Participial extensions In CGEL’s taxonomy of existentials, it seems as if [to_INF]s are the only nonfinite clauses that occur. This is due to the claim that most existentials are related to corresponding basic clauses as in Table 38 above. Given the correspondence, we would expect the basic sentence form of (22a) below to be (22b) which is of course not acceptable with stative verbs such as belong.388 The nonprogressive (22c) is a more sensible basic clause counterpart, but it would vio-

|| 388 While there are claims that the progressive is on the rise with stative verbs, Leech et al. (2009: 130) were unable to confirm this trend in a corpus study.

Survey of structures | 247

late the restriction that basic clauses need a form of be in their VP in order to allow conversion into existentials.389 (22) a. There is plenty of furniture belonging to the house. (HH8 962) b. *Plenty of furniture is belonging to the house. c. Plenty of furniture belongs to the house. Similarly, CamG (1394) treats sentences such as (23) as participial extensions: (23)

There was a note put through her letterbox. (ABW 2149)

Law (1999: 184) calls this structure the passive existential construction. It appears that a debate in generative grammar revolves around the question of how these should be modelled. The fact that the construction merits a distinct name of its own seems to stem more from the apparent difficulty to agree on a derivation in the framework of generative grammar than from interesting linguistic behaviour of such structures. Again, we can observe a structural ambiguity in some such sentence that will be discussed in the following section.

8.2.2.4 Extented existentials vs. bare existentials with postmodified or complemented [NP] As mentioned above, many extended existentials are superficially indistinguishable from a bare existential with a postmodification or a valency complement.390 We shall review examples for infinitival, participial and relative clause extensions as discussed in the previous sections here. For [to_INF] extensions, there is a superficially similar construction where the [NP] of a bare existential contains a valency complement in the form of a [to_INF] clause as in (24a). The fact that example (24b) is completely nonsensical can be seen as an indicator of the different status of the [to_INF] clause, since the postverbal [NP] plans cannot be interpreted as the subject of the verb publish (and since it cannot be interpreted as the ‘object’ either, since that position is taken by the text of the lecture).

|| 389 See CGEL (1403) and the discussion of restrictions on the basic clause at the beginning of 8.2.2 above. One might also argue that in fact (22a) cannot be paraphrased at all in terms of a basic clause simply because it is a bare existential clause. We shall come back to this question in the next sub-section. 390 It has to be noted that not all grammars make a distinction between the different structures, so LGSWE treats all of them uniformly as complex NPs (LGSWE 946–951).

248 | Existentials

(24) a. There are also plans to publish the text of the lecture. (GXE 466) b. *Plans publish the text of the lecture. However, there are cases where ambiguity arises. The [to_INF] clause in (25a) below could be analysed in a similar vein to the relative clause examples mentioned above but it could also be regarded as a complement of the noun policy, given that the VDE lists [to_INF] as a possible complement. (25) a.

There is no planning policy to ensure that they are not concentrated in one area which may lead to enormous pressures on local support services. (FTY 357)

Rough paraphrases of the two interpretations, which are very close to each other semantically, are given in (25b–c). (25) b. No planning policy ensures that XYZ. c. No planning policy to ensure that XYZ exists. Thus, while the two constructions are syntactically distinct in a large number of cases, we cannot assign every case to either one or the other category. For participial structures following the head noun, CamG (1394f), makes a distinction between bare existentials with an [NP] containing a participial postmodifier and extended existentials where the displaced subject [NP] is followed by a participial extension (as in (26a)). (26) a.

There are some people going to be disadvantaged by the new tax system. (CamG 1395) b. *Those people going to be disadvantaged by the new tax system will have to be compensated in some way. (CamG 1395)391

In their reasoning, the unacceptability of (26b) indicates that the participial clause in (26a) is not part of the noun phrase since we would expect such a noun phrase to be acceptable independently of the existential construction. As we can see in (22d), this is true of the noun phrase in (22a) above, so (22a) may alternatively be treated as a bare existential. (22) d. Plenty of furniture belonging to the house had been stolen the night before.

|| 391 Hannay claims that such elements can in fact occur in what he calls “with absolute constructions” (1985: 87): With some people going to be disadvantaged by the new tax system, I think that the government will have a hard time.

Survey of structures | 249

While CamG’s reasoning appears to be valid for the examples given above, indicating that there are two distinct superficially identical constructions, we also find instances which show conflicting evidence: (27) a. There was a war raging inside her. (based on H9V 1455) b. A war was raging inside her. c. A war raging inside her made it impossible for her to think clearly. The reformulation of (27a) in terms of a basic variant in (27b) is perfectly acceptable, which may be counted in favour of an extension analysis, but the acceptability of (27c) could be seen as an argument in favour of the one-[NP]with-postmodifier analysis. No indication is given in CamG of how to analyse a sentence in case of conflicting evidence, and we will again have to accept the indeterminacy of sentences as to whether they contain an extension or not. For the past participials, we can also find relatively clear cases of existentials with a postmodification (and, in this case, an adverbial extension) as in (28), and of extended existentials with a past participial extension, as in (23) above (repeated below), as also mentioned by CamG (1395). (28)

There were [some letters written by her grandmother] in the safe. (CamG 1394)

(23)

There was a note put through her letterbox. (ABW 2149)

We can actually show that the two participial clauses do not commute by combining them into one sentence: (29)

There was a note written by her grandmother put through her letterbox.

Nonetheless, since the extension in (23) can also be used as a postmodification as in (30), we have the same structural ambiguity as with the other extensions: (30)

A note put through her letterbox informed her of her grandmother’s death.

When it comes to relative clauses, CamG is much more careful as to whether it is possible to make a distinction between extension and postmodification at all. The authors do present evidence in favour of such a distinction, though: (31) a. There are [people that have an IQ far greater than that]. (CamG 1396) b. There was [one man] that kept interrupting. (CamG 1396)

250 | Existentials

In (31a) the relative clause is naturally taken as a modifier of people: the sentence asserts the existence of people with an IQ far greater than ‘that’. Example (31b), however, might be construed as the existential counterpart of One man kept interrupting. Note that the latter is a paraphrase of (31b), whereas (31a) cannot be paraphrased as People have an IQ far greater than that. (CamG 1396; numbers changed)

They also add another argument: One piece of supporting evidence that can be adduced is the possibility of having a relative clause after a proper name. Thus in answer to the question Who might be able to help? one might reply: Well, there’s John you could try. This cannot have an analysis like [(31a)], for John you could try is not a possible NP. (CamG 1396)

Collins draws the same distinction: Existential sentences may occur with a relative clause as extension. Some such cases, where the relative clause serves clearly as a restrictive modifier within NP structure, can be readily handled as bare existentials. (Collins 1992: 423)

The fact that both can even occur together is a further indicator of their different status: (32)

I was not unhappy to be sitting opposite an attractive lady from Coll and there was only one person I knew of who lived on Coll. (GXA 940)

(33)

‘There were youngsters who couldn’t read who had to learn from tapes and constant repetition,’ says Mr Chance, ‘and they never faltered. (K4V 2686)

Even though many sentences with a similar structure can be interpreted as two co-ordinated relative clauses, such a reading is excluded in (33), where the extension (who...repetition) is predicated of the entire [NP] including the relative clause postmodification (who...read) and not just of the head youngsters. On the other hand, there are also [NP]s in non-existential structures that contain two such relative clauses in the same hierarchical relationship: (34)

A lady I know who saw his mother’s funeral said that Yeats wouldn’t go into the church. (ADM 1288)

We can thus conclude that there are good reasons to treat existentials with displaced subjects that are postmodified or have a valency complement as structurally different from existentials that contain an extension of the same form. Syntactically, however, it is often impossible to assign an actual example to one of the two constructions. The syntactic difference seems to coincide with a se-

Survey of structures | 251

mantic difference in that – as mentioned above – bare existentials assert the existence of an entity and the postmodification is mainly used for the identification of the entity (“restrictive modifier” in the citation by Collins above) whereas extended existentials contain a predication which (usually) asserts new information. If one accepts this semantic difference as a criterion for syntactic structure, one can classify more data as either one or the other, but borderline cases, such as (25a) above, will still remain. Furthermore, to sum up the evidence presented in all of Section 8.2.2, we can say that due to the restrictions found on the ‘basic’ variant CGEL’s derivative approach for extended existentials is highly problematic and that it may thus be more appropriate to treat extended existentials independently of some supposedly underlying structure.

8.2.3 Presentational existentials As mentioned above, existentials in which the verb is not be are often treated separately. The use of the terms presentational in CamG (1402) or presentative (CGEL 1408, Hannay 1985: 9) is an indicator of the fact that they serve “to bring something to the discoursal stage deserving our attention” (CGEL 1408). CGEL claims that they are “a rather less common, more literary type of existential clause” (CGEL 1408) whereas CamG (1402) does not note such a preference. The most notable difference compared to ‘ordinary’ existentials is that there are restrictions on the verbs that can occur in such constructions, as can be seen in (35): (35)

*There killed a man his family.

Most authors agree that generally the verbs have to be intransitive (see for instance CGEL 1408, CamG 1402),392 but example (36) illustrates that intransitivity is not a sufficient criterion: (36)

*?There sneezed a man at the funeral.

|| 392 CGEL states that “exceptions are idiomatic or dubious” (CGEL 1408) whereas CamG notes that “transitives cannot be wholly excluded, as illustrated in the attested example There seized him a fear that perhaps after all it was all true” (CamG 1402 fn.17). Hannay (1985) gives two further examples, one of them biblical: “And there followed him a certain man... (Mark, 14.151) [...] There entered the hall a procession of dancing girls” (Hannay 1985: 9).

252 | Existentials

This is why usually some kind of semantic groups of verbs are given. CamG states that a “high proportion have to do with being in a position or coming into view” (CamG 1402), CGEL requires a “fairly general presentative meaning: verbs of motion (arrive, enter, pass, come, etc), of inception (emerge, spring up, etc), and of stance (live, remain, stand, lie, etc)”393 (CGEL 1408). Martínez Insua compiled a list consisting of “verbs of being and occurrence -e.g. come, exist, hang, happen-, temporal verbs -e.g. arise, burst, emerge, loom-, verbs of continuation develop, remain, linger, persist-, and verbs of motion -e.g. arrive, enter, come, pass” (Martínez Insua 2002: 134). The obvious incongruity of the three statements (to which “high proportion”, “etc” or “e.g.” add) suggests that no exhaustive list can be given. Corpus research in the BNC confirms that the verbs given in CamG (1402) and the ones cited above account for the bulk of the relevant sentences in the corpus (one verb cited in none of the accounts is given in (37) for reference), albeit with huge variation in frequency.394 (37)

On either side of this scene there crouches a human watcher and the air is filled with small birds — thirty-one of them altogether. (BLX 1674)

The choice of verb in such structures is thus relatively difficult to model, since a purely storage-based model is at a loss if no exhaustive list of items can be given and a semantic model fails if the verbs that occur are difficult to group in a semantically straightforward fashion. The question will be taken up again in the discussion of potential models in Section 8.4.4.

8.3 What is the subject? Generally, there appears to be less controversy in the literature as to what should be analysed as the subject of an existential clause than there is in the case of extraposition (see Section 7.1.3). Thus CGEL classifies there as subject based on the following three criteria:

|| 393 CGEL also shows that a wider range of verbs become available when there is an initial space adjunct in the sentence. Compare: Into the room (there) had staggered a total stranger. ?There had staggered into the room a total stranger. ?*There had staggered a total stranger into the room. (CGEL 1410) 394 Only one hit for loom was found in this construction in the BNC, for instance.

What is the subject? | 253

(i) It often determines concord [...] (ii) It can act as subject in yes-no and tag questions [...] (iii) It can act as subject in infinitive and -ing clauses: I don’t want there to be any misunderstanding. He was disappointed at there being so little to do. There having been trouble over this in the past, I want to treat the matter cautiously. (CGEL 1405)395

However, one can also find the term “pseudo-subject” (CGEL 756), which suggests that existential there is not a ‘real subject’ – an interpretation that seems to be backed by the treatment of “the existential subject there (cf 18.44ff), as distinct from the ‘true subject’ following the verb” (CGEL 897). Such a wording seems to suggest an underlying level of representation similar to Chomsky’s deep structure, on which the postverbal element (which is explicitly not the subject) has its ‘true’ function of subject.396 This may of course be related to the treatment of many types of existential clauses in terms of basic clauses as discussed above. The fact that the term “notional subject” (CGEL 1405) is used in quotation marks may suggest that the level at which that element is the ‘true subject’ may not be a level of syntactic but rather of semantic representation. Anderson draws a distinction between the syntactic subject (there – based on its “syntactic ‘behaviour’” [Anderson 1997: 22]) and the morphosyntactic subject (the postverbal [NP] – based on agreement). Breivik/Martínez Insua (2008: 351f) call existential there a subject-position holder and a particle and use notional subject for the postverbal [NP]. CamG (241ff) also agrees that there has to be the subject on the basis of the criteria for subjecthood presented in Section 2.3.2 (and discussed for extraposed sentences in Section 7.1.3.2). The present section will re-examine these criteria with respect to their applicability to there and the postverbal element(s) in existential sentences. Again, the discussion closely follows CamG’s (241–243) treatment. 1.

Category

As existential there is not the same as the particle there and as it occurs in a position in which particles do not usually occur, it is likely to belong to a differ-

|| 395 The details for criteria (i) and (ii) will be discussed below and have thus been omitted here. 396 It has to be noted, however, that both quotes are from parts of the grammar that are not directly related to existentials and may thus not reflect a conscious decision for or against a certain way of modelling existentials.

254 | Existentials

ent word class.397 The canonical subject position in front of the verb makes a pronoun analysis (similar to it in extraposed sentences) plausible, and this is indeed what makes CamG classify existential there as a “dummy pronoun” (CamG 1390):398 As a pronoun there can be best regarded as a peripheral member of the personal pronoun category. It has the distinctive personal pronoun property of being able to occur as subject in an interrogative tag, as in [(38)]. And it is comparable with the other dummy pronoun, it, a core member of the category. (CamG 427)

(38)

There is an obvious solution to this problem, isn’t there? (CamG 427)

While the arguments brought forward by CamG are certainly valid, their conclusion is nonetheless problematic. The fact that existential there “can fill the subject position in interrogative tags” (CamG 1391) may be taken as an argument in favour of a pronoun analysis because it allows us to capture a generalization. Given that existential there is marked neither for person nor for gender nor for case, a personal pronoun analysis is however doubtful. The ability to occur in canonical subject position is shared by indefinite pronouns such as something (CGEL’s terminology; a compound determinative in CamG’s terminology), but these cannot occur in tag questions. Accordingly, the generalization is only of limited use given that it creates exceptions in another place of the grammatical system. And, if we want to be precise about the application of the criteria used to determine the word class of there, we have to exclude any criteria related to subjecthood in order to avoid circularity in the present chapter. Thus it would be equally valid if not preferable not to assign any word class at all to existential there, a strategy implicitly adopted by CGEL and explicitly by Herbst/Schüller: Since there in such uses does not share enough properties with members of any other word class occurring in similar positions, it would not make sense to classify there in such uses as belonging to any other word class either. (Herbst/Schüller 2008: 72)

|| 397 CamG points out that “[h]istorically, dummy there derives from the locative there [... but] has been bleached of its locative meaning and reanalysed as a pronoun” (CamG 1391). See the discussion in Section 8.1.2 for details. 398 Similarly Aarts, who calls existential there a pronoun and describes it “as a meaningless word that fills the obligatory Subject slot” (2010: 48).

What is the subject? | 255

This argument will be followed here and thus existential there will not be assigned a word class in the present study and treated as a word of unique function instead. However, since we use CamG’s criterion of category here, we can note that in CamG’s classification, both there and understandable complications in (39) fulfil the criterion. (39)

There were understandable complications. (EFN 663)

Matters become more complicated in the case of (40): (40)

There was a boy sitting in the car. (K8V 1640)

If we decide that a boy is the displaced subject, we have a perfect NP in postverbal position; if we decide that a boy sitting in the car is the displaced subject, we still have to decide whether to treat it as an NP with a postmodifier or as some sort of small clause (and whether the latter is a sensible category for subjects). It thus has to be concluded that category is not particularly helpful for the establishment of subjecthood in the case of existential clauses, owing to the difficulty of reliably assigning a category to the items in question. 2.

Position

The canonical subject position in existential clauses is occupied by there. 3.

Case

As we have seen above, there does not inflect for case, but personal pronouns can occur in postverbal position in existential clauses. If they do, they are usually in objective case as in (41), as also noted by CamG (241). A corpus search seems to suggest that subjective case can only occur with locative there as in (42). (41)

There was me and Charlie and Maurice and Jack left and we went when they closed. (A73 731)

(42)

At the end she turned off to the right and there was I with nothing to look at. (FAP 1970)

Thus case argues against the subject status of the postverbal [NP] in the main clause of existentials.

256 | Existentials

4.

Agreement

Typically, the postverbal [NP] in English existentials seems to determine concord: (43) a.

On this floor there is one bathroom, but there are twenty of us. (HTG 168) b. There are only two bedrooms at Tom’s house. (HHC 236)

However, we find singular verbs with plural postverbal [NP]s, too, even though these appear to be much rarer: (43) c.

‘There’s only two bedrooms,’ said the woman. (ABX 1027)

Accordingly, we cannot decide which element the verb agrees with in all cases, but the postverbal [NP] seems to determine concord in the majority of cases. The issue is rather complex and will thus be treated separately in more detail in Section 8.4.2 below. 5.

Inversion

Existential there occupies the subject position in sentences with subjectauxiliary inversion: (44)

Are there any reported rapes between staff and students? (KRL 1291)

Hannay (1985: 10) points out that existentials with verbs other than be do not allow subject-auxiliary inversion: (45)

?Did there live in that castle anyone important? (Hannay 1985: 10)

He takes this behaviour as an indicator of the syntactically distinct status of what he calls there-presentatives as opposed to existentials. However, the BNC contains one acceptable counter-example: (46)

Did there emerge in the twentieth century a distinctive network of Asian capital, embracing Chinese, Indian and Japanese capitalists, distinct from the European network which had dominated South East Asia from the middle of the nineteenth century? (HJ0 2015)

Thus even for presentational existentials the inversion criterion speaks in favour of there as the subject.

What is the subject? | 257

6.

Open interrogatives

CamG observes that “the post-verbal NP can be questioned and then clearly behaves like a non-subject” (CamG 243) as illustrated by (47a) and the (almost) corresponding question in (47b). (47) a. There are several options open to us. (CamG 241) b. How many options are there available to us? (CamG 243) However, example (48b) below illustrates that we can question the postverbal [NP] of (48a), which is relatively uncontroversially the subject occurring in inversion after a fronted locative expression, in a similar way as in (47b), so saying that the postverbal [NP] behaves like a non-subject is misleading. (48) a. At the far end were hundreds of skinned carcasses. (ABC 873) b. How many skinned carcasses were at the far end? In fact, if we recall the criterion, it is about the presence of subject-auxiliary inversion if the subject is not questioned, so if CamG claims that there is such inversion in (47b), this is only possible if it is presumed that there is the subject in the first place. So in order to properly apply the test, it would be helpful to use examples with auxiliaries, so that inversion can be reliably identified. Thus if we take an existential like (49a), we have to ask which of the two questions given in (49b–c) is the preferred option. (49) a. There will be 300 people at the concert. b. How many people will there be at the concert? c. How many people will be there at the concert? Corpus evidence in the BNC for this type of question is limited, but there are four instances of the type will there be (one of which is given as (50)) and none of the type will be there, so this speaks in favour of the former type. (50)

HOW MANY DISABLED ELDERLY WILL THERE BE IN THE FUTURE? (ECE 816)

All three native speakers interviewed on the issue agree that (49b) is the preferred form, too. The underlined item in (49c) is best treated as an instance of locative there.399

|| 399 Such a treatment is backed by the fact that one informant commented that (49c) is only acceptable without at the concert.

258 | Existentials

So far, we have concentrated on questions starting with how many, but the typical questions for subjects are usually of the who/what pronominal type. The corpus evidence in the BNC for who with the modal verb will consists of one example:400 (51)

Sir, — When the local fox hunts are just a memory who will be there to control the ever-increasing fox population? (B03 1278)

Since a locative reading of there cannot be ruled out in this example, it is of very limited help, so three native speakers were interviewed on the following set: (52) a. There will be no one to remember us. b. Who will there be to remember us? c. Who will be there to remember us? While they all accept both (52b) and (52c), there is a tendency to prefer the latter. However, their pronunciation of there in (52c) as a strong form suggests that it is the locative adverb and not the existential there, which would be in line with its postverbal position. In sum, the position in open interrogatives thus also speaks in favour of there as the subject in existential clauses. 7.

Tags

As we have seen above, the element repeated in tag questions is there: (53) 8.

There’s something you want to know, isn’t there? (CEH 3448) Coordination

Despite the acceptability of (54) – which is an indicator of the subjecthood of there – CamG is right in observing that VP-coordination401 is not common with there. (54)

There are several options open to us and have been from the beginning. (CamG 243)

|| 400 The queries used were “who will be there” and “who will there be”. 401 VP here is used in CamG’s sense. A coordination of VPs in the sense of CGEL does of course exist but is not helpful in determining subjecthood: Pearce dismissed the programme’s central theme with a pat circular argument: ‘There is, and will be, no socialist revolutionary government to suppress, therefore it will not be suppressed.’ (ARD 569)

What is the subject? | 259

9.

Obligatoriness

The maximal finite reduction of existentials contains there as a subject, which also speaks in favour of its subjecthood: (55) a. b. c. d.

There were horses in the field when we came. (KCH 4081) Were there horses in the field when you came? #Horses were. There were.

10. Uniqueness As discussed in Sections 2.3.2 and 7.1.3.2, only CamG insists on the uniqueness of the subject. However, contrary to the case of extraposition, researchers generally seem to agree that there is the subject in existential clauses, generating no problems for the uniqueness constraint.402 11. Reflexive pronouns The criterion of reflexive pronouns does not apply as no reflexive pronoun can occur in an existential clause in one of the relevant functions. 12. Passivization Existentials such as (56a) do not have a corresponding passive, which is hardly surprising given that one of their characteristics is the copula or intransitive verb: (56) a. There must be a lot of documentation. (BND 1574) b. *A lot of documentation must be been (by there). c. *There must be been (by a lot of documentation). Accordingly, no element can occur as a by-phrase. However, there are infinitival existential structures after certain verbs where there occupies the position of a so-called raised object (a generative term also used in CamG 1391): (57)

So you’d expect there to be a big change in the social impacts of unemployment. (KRG 1448)

|| 402 If we were to follow Anderson’s distinction between syntactic and morphosyntactic subject mentioned above, uniqueness would not be met, of course.

260 | Existentials

These structures can have passive counterparts403 and it is the ‘object’ (in traditional terms) that moves to preverbal position.404 (58)

There is expected to be a £10 million write-down on its land bank. (AHJ 199)

Accordingly, the criterion of passivization speaks in favour of there as the subject in (58) since it moved from object to subject position in CGEL’s terms.405 There are some cases where existentials appear to contain passive morphology, such as (59a) and (60a), without following to be: (59) a.

In addition, there are attributed to lenders voting rights on shares charged to them (except where they are controlled by the chargor). (BP5 615)

(60) a.

Less tangibly, there was noted in the interviews the tendency for the specialist team members to be somewhat more articulate in the responses they made. (ALN 222)

In the case of the former, however, a reading in which the main clause is not passive is also possible, as illustrated in (59b). If we adopt such a reading, we may just as well interpret (59a) as showing heavy unit shift of the [NP] subject of

|| 403 Not all verbs that allow an active with a postverbal existential structure also allow the passive, though: I never wanted there to be a war between us, Ingrid. (J19 2845) *There was never wanted to be a war between us, Ingrid. Conversely, with the verb say, “only the passive is possible” (CGEL 162) to express this meaning (this is, however, a property of this particular reading of say and has been briefly discussed in Section 5.2.2.2): Between men and women, too, there is said to be a barrier. (A05 1141) *He said there to be a barrier between men and women. 404 In sentences such as (57), the status of the postverbal elements is in fact a matter of debate in linguistic theory. Thus CGEL (1217) states that expect in such structures should be analysed semantically as taking one proposition as its argument, i.e. there to be a big change in the social impacts of unemployment, but takes the fact that the first NP can become the subject in the corresponding passive clause as a criterion to treat it as object of expect on the syntactic level. See also the discussion in Matthews (2007b: 9ff). 405 Note however that agreement remains as described above, with the [NP] in the postverbal to be-clause determining concord: But in the first 10 months of 1989 alone over 7,000 have arrived, and there are expected to be 10,000 by the end of the year. (A8P 66)

What is the subject? | 261

the small postverbal clause (or, in terms of the more traditional analysis, of the [NP], leaving the participial extension in front). (59) b. In addition, there are voting rights attributed to lenders. (60) b. ?There was the tendency to do XYZ noted in the interviews. Why (60a) does not allow such a rephrasing (at least no fully acceptable sentence similar to the pattern given in (60b) could be constructed) remains unclear. A possible reason might be the slightly different grammatical structure, as attribute in (59a) has a valency slot for the [to_NP] whereas in the interviews in (60a) is an adjunct. The more likely interpretation is, however, that there are restrictions on information structure at work.406 Whatever the reason may be, it would probably be misguided to analyse such structures as passive existentials. 13. Theme/topic See following paragraph. 14. Given information As already remarked in Section 2.3.2, CGEL does not always make a clear distinction between theme, topic and given, so these will be treated under the same heading here. According to CGEL, existentials provide “some kind of dummy theme which will enable the originator to indicate the ‘new’ status of a whole clause, including its subject” (CGEL 1402). Furthermore the restriction on many existentials for the postverbal [NP] to be indefinite (see above) is directly related to the fact that it is discourse-new, as is also noted by Dowty, who remarks “that existential constructions in many languages have a grammatical form that removes the NP from normal grammatical subject status, possibly displacing it with a dummy NP or locative (Clark 1978), thereby signifying that its referent is NOT connected to previous discourse in the way that subject status would otherwise indicate” (Dowty 1991: 564). The existential there can thus been regarded as theme (and possibly given) by default. 15. Agentivity Existential there is never agent. In bare existentials, no agent role is expressed, so the criterion does not apply. In extended existential clauses such as (61) one might argue that the postverbal [NP] can be an agent.

|| 406 One might also suspect that length is responsible but *There was a tendency noted in the interviews to do XYZ. was also judged unacceptable.

262 | Existentials

(61)

There’s Paul McCartney doing a kids broadcast and a record about the BEETLES [sic!] on this weird Australian label which is Australians doing impersonations of a Liverpool accent. (K3J 515)

However, given the analysis of existentials provided here, it would probably make more sense to state that the [NP] in question is agent only in the postverbal predication that comes after be and not in the main clause, as the verb expressing the action of which the [NP] can be agent is part of that predication. Further proof that the agentivity of the [NP] is dependent on the verb in the postverbal predication is (62) below, which does not contain an agent because the verb of the postverbal predication is in the passive. (62)

In 1991, there were 701,000 pupils being educated in Roman Catholic schools of whom only 11.5 per cent. were non-Catholics. (FBY 89–90)

Finally, Gast/Haas (2011: 145) suggest a further kind of evidence, viz. the availability of existential there to subject-to-object raising (in the generative sense), as in the following example (see also the paragraph on passivization above): (63)

On neurophysiological grounds one would expect there to be considerable differences between different individuals with regard to the most effective channels of learning; therefore, one must regard statements such as that quoted with a certain amount of scepticism. (B3D 50)

So, to sum up, while there are issues with some of the criteria, most notably agreement, the majority is clearly in favour of an analysis of existential there as a subject. Or, in Hudson’s words, “[t]he grammatical subjecthood of THERE is as close to being a fact as anything in grammar” (Hudson 1999: 200).

Theoretical treatment | 263

8.4 Theoretical treatment This section concludes Chapter 8 with a discussion of the theoretical treatment of existentials in grammatical models. Note that some approaches on which the present study relies heavily do not feature in this chapter. As the reference grammars’ analyses have been discussed in the descriptive part above, they need not be repeated here and since Herbst/Schüller (2008) do not offer an indepth analysis of existentials, their model will only serve as the background of the discussion of a possible treatment in valency theory in in Section 8.4.4. 8.4.1 Postverbal [NP] in extended existentials as subject of a small clause? As we have seen above, extended existentials can usually be rephrased so that the postverbal [NP] occurs as the subject of the modified version (or basic clause in CGEL’s terms). (64) a. There was no force involved. (JY5 2706) b. No force was involved. While nobody would contest the status of the underlined element as the subject of the clause in (64b), we have seen above that usually there is treated as the subject in (64a). Nonetheless, since the logico-semantic relationship between no force and involved is the same in both versions, one may want to treat no force as the subject of involved in both. This, however, would speak in favour of an analysis of no force involved as a sort of clausal constituent in the sentence, i.e. a small clause in the terminology of generative grammar, which would in turn mean that the structure of an extended existential may not contain a postverbal [NP].407 On the other hand, as mentioned above, some grammars (e.g. LGSWE) treat the extension as part of the postverbal [NP], which keeps the structure of existentials uniform but obscures the relationship between postverbal [NP] and the extension and treats structures as [NP] that cannot occur as [NP]s in other structures. Neither CGEL nor CamG make statements about the syntactic status of the extension in extended existential clauses. In the following sub-sections, we shall review some arguments brought forward by Williams (1984), who is in favour of a single [NP] analysis, in order to arrive at a balanced decision as to how extended existentials could be modelled.

|| 407 The problem is in fact very similar to the question of verbs of the consider type, where the postverbal element(s) can be regarded as one item or as two items, depending on the criteria used. See Section 5.2.2.2 for a discussion.

264 | Existentials

8.4.1.1 Heavy [NP] shift On the basis of the evidence given in (65a–b), Williams (1984) shows that heavy [NP] shift can occur in small clauses but cannot occur in existentials as illustrated by the pair (65c–d). (65) a.

I consider several of George’s recent acquaintances sick. 1984: 135) b. I consider sick several of George’s recent acquaintances. 1984: 135) c. There are several of George’s recent acquaintances sick. 1984: 135) d. *There are sick several of George’s recent acquaintances. 1984: 135)

(Williams (Williams (Williams (Williams

Consequently, he maintains, the postverbal element(s) cannot form a small clause and that thus the underlined element is not a subject. However, as we have seen above, there are cases which can be analysed as containing heavy [NP] shift:408 (66)

There should be produced to the tenant not only a copy of the policy and/or a summary of the insured risks but also evidence of the payment of the last premium. (J77 853)

It is true, though, that the subject position in small clauses of the postverbal consider type are much less restricted formally than the postverbal elements in existentials are, i.e. we can get [V-ing] and extraposed [to_INF] and [that_CL] subjects there (see Section 5.2.2.2 for discussion). Thus an analysis that treats both structures as parallel may not be fully convincing.

8.4.1.2 Simplicity Williams also claims “that the bare NP structure can generate all of the strings” (Williams 1984: 132) we can find in existentials. However, as we have seen above, there are postverbal elements in existentials that do not occur as [NP]s. The fact that not the entire postverbal ‘string’ of words found in (26a) can be used as an [NP] in (26b) contradicts Williams’ argument.409

|| 408 See also the discussion of example (59a) above. 409 In the generative terminology, this would mean that the rule generating NPs would have to over-generate in order to create the postverbal string in (26a).

Theoretical treatment | 265

(26) a.

There are [some people] going to be disadvantaged by the new tax system. (CamG 1395) b. *Those people going to be disadvantaged by the new tax system will have to be compensated in some way. (CamG 1395)

8.4.1.3 So-called ‘predicate restriction’410 Small clauses can consist of two [NP]s, i.e. a subject and a predicative element, as in (67a) whereas, according to Williams (1984), no two [NP]s can occur postverbally in existentials as in (67b): (67) a. I consider a friend of mine an imposter. (Williams 1984: 132) b. *There was a friend of mine an imposter. (Williams 1984: 132) However, Hannay shows that [NP]s can also exist as predicates, not only in the form of a bare [NP] (such as ‘president’) as in (68) but also as a fully-fledged [NP] as in (69), which would speak in favour of an analysis of the underlined constituents as subjects in the small clause. (68)

There’s a friend of mine captain of the local football team. (Hannay 1985: 166)

(69)

There’s one of the players a very good friend of mine. (Hannay 1985: 166)

Since a friend of mine captain of the local football team in (68) cannot sensibly be analysed as an [NP], the argument speaks in favour of a small clause analysis with the underlined elements as subjects. The predicate restriction is often also discussed on the basis of adjectives. Compare: (70) a. There was a first year student sick yesterday. (Hannay 1985: 22) b. *There is a first year student intelligent. (Hannay 1985: 22) Hannay411 rejects the widespread interpretation of (70a–b) that “draws a distinction between property-assigning adjectival predicates and state-descriptive

|| 410 See Hannay (1985: 21–23) for details. 411 Hannay’s model is rooted in functional grammar, so he models the restrictions identified above as a restriction on the focus (“Focality Condition” [Hannay 1985: 160]) and posits a requirement of a “non-identifying environment” (Hannay 1985: 160–163).

266 | Existentials

ones” (Hannay 1985: 146) and presents counter-evidence such as (71), in which the italicized part is property-assigning but still acceptable. (71)

If you ask me there’s nobody here really clever. (Hannay 1985: 166)

Thus there seems to be no generally applicable restriction on adjectives that would speak against a small clause analysis even though a tendency against property-assigning adjectives can be observed.

8.4.1.4 [NP]s which resist conversion into a clause On the other hand, Williams shows that the underlined part in (72a) can occur as an [NP] in other contexts such as (72b) but cannot be turned into a clause as in (72c). (72) a. There is a man with a green coat. (Williams 1984: 133) b. The man with a green coat is here. (Williams 1984: 133) c. *The man is with a green coat. (Williams 1984: 133) Such a restriction – the nature of which is unclear as it only occurs with a certain sense of with – is a point in favour of an [NP] analysis. One may be inclined to argue that the evidence presented in (72a–c) speaks in favour of a bare existential analysis, but at least semantically, Williams is right when he claims that (72a) “does not have the flavor of a purely existential sentence” (Williams 1984: 133). Nonetheless we have seen that the arguments against the [NP] analysis are convincing enough to reject it. This would leave the small clause analysis as the preferred alternative in the generative framework. However, in the light of the problems with a small clause analysis pointed out above, Hannay (1985: ch. 3+4) argues against the small clause as modelled in generative grammar and calls his postverbal element in extended existentials an “embedded predication” (Hannay 1985: 73), which might be useful in order to avoid the generative connotations of the term small clause in a nongenerative approach and to terminologically reflect the differences compared to small clauses of the consider-type. We shall thus follow Hannay in calling the postverbal structure in extended existentials a predication rather than a small clause.

Theoretical treatment | 267

8.4.2 Issues of agreement As outlined above, the typical case appears to be that the postverbal [NP] determines concord since concord usually changes when the number of the postverbal [NP] changes. However, in word grammar (Hudson 1999; see also Hudson 1990, 2010), Hudson – as cited above – treats there as the subject of the existential clause. He states first that “if [there] is a pronoun, it must have a number” (Hudson 1999: 200) and then goes on to explain that in word grammar the best way to model this is in parallel to the [NP] analysis in which “the determiner is a pronoun which [...] has the common noun as its complement” (Hudson 1999: 200). Given that in his model, pronoun and complement have the same number and the same referent, he suggests treating “the delayed phrase in each case as the complement of the pronoun that replaces it” (Hudson 1999: 201). Figure 24 shows “how the number agreement is explained by the extra dependency (labeled ‘c’ for complement) shown beneath the words” (Hudson 1999: 201). (73)

There is a fly in my soup. (Hudson 1999: 201)

Fig. 24: Relationships in example (73) according to word grammar (Hudson 1999: 201)

While Hudson’s analysis offers a consistent model of the structure and a neat explanation of certain agreement phenomena with coordinated structures, it does not account for non-concord and it still has the disadvantage of having to posit agreement between there and the postverbal [NP] and then there and the verb, which seems psychologically less plausible than a direct agreement of the verb with the postverbal [NP]. Similarly, CamG argues that “the dummy pronoun there does not have inherent person-number properties but inherits them from the NP that it ‘displac-

268 | Existentials

es’ as subject” (CamG 242). To support this claim, CamG compares the situation to relative clauses in which the verb agrees with the antecedent and not with the relative pronoun that is the actual subject of the relative clause. Examples (74a– b) are used to illustrate the point: (74) a.

There tends to be a single pre-eminent factor in the breakup of a marriage. (CamG 242) b. There tend to be several contributing factors in the breakup of a marriage. (CamG 242)

Ultimately, the choice between the verb-forms tends and tend is determined by the personnumber of the underlined NP, but that NP cannot be the subject of tend, for it is not located in the tend clause, but in the be clause. (CamG 242)

Such an inheritance relationship between there and the displaced subject helps to explain away the ‘odd’ concord and is in fact needed in catenative structures. For CamG, this includes structures with modal verbs and related constructions (see CamG 1209–1220 for an in-depth discussion of advantages and drawbacks of such an analysis). CGEL and Herbst/Schüller (2008) prefer what CamG (1213) calls a “verb group” analysis, with CGEL applying the term marginal auxiliaries “to all four categories (marginal modals, modal idioms, semi-auxiliaries and catenative verbs)” (CGEL 236), which means that agreement may be describable without there inheriting the person-number from the postverbal [NP]. This is the analysis adopted in the present chapter since it is much simpler in its grammatical structure and keeps there and the verb in the same clause as the postverbal elements, which is particularly preferable if we want to model that there + verb combine to form a complex element which acts in a way as valency carrier, as advocated below.412 In such an analysis, we can maintain the (more intuitive) assertion that the postverbal [NP] determines concord. Let us now turn to those cases where concord is not determined by the number of the postverbal [NP]. CGEL (1405) states that in informal contexts there may trigger singular agreement on the verb even if the postverbal [NP] is in the plural as in (43c) reproduced here: (43) c.

‘There’s only two bedrooms,’ said the woman. (ABX 1027)

|| 412 It has to be stressed that the present chapter is intended to present a model that is wellsuited for existentials and is thus written from the comfortable position of not having to say anything about the suitability and implications of the selected model for other aspects of the description of English grammar. Accordingly, no statement will be made as to a general preference of the ‘verb group’ analysis over the ‘catenative’ analysis.

Theoretical treatment | 269

This would suggest that – with regard to concord – in informal contexts, there acts as the subject whereas the postverbal [NP] acts as the subject in other contexts. CamG states that the agreement in (43c) is only possible because “the copula is cliticised to the subject” (CamG 242).413 A corpus search of there is in the BNC seems to confirm this as an overall tendency, although there are a few cases which seem to contradict it. Most are from transcribed speech though, so the quality of the transcription (or the recording in the first place) may be partly responsible. Others, even though they are written language, do not reflect the style one would usually expect in writing. Interestingly, the tendency seems to apply to there has been vs there’s been as well, which may be an indicator that – as CGEL claims – the style level is responsible for the difference and not whether “the copula is pronounced as a full independent word” (CamG 242), which might only be a secondary effect of the style level. (75) is an example of the singular agreement with there’s been + plural noun taken from an interview. (75)

Yes there’s been many cases where lesbian and gay couples have left a gay bar, have given each other a goodnight kiss at a bus stop or at a tube station or in the street and they’ve been arrested under public decency laws and dragged through the courts and fined up to £200. (KRT 1759)

Martínez Insua/Palacio Martínez (2003) analysed a 1-million-word subcorpus (50 % spoken) of the BNC and found that it contained 2410 there-existential clauses, 487 ‘minimal’ and 1923 extended ones.414 Of these, 233 showed ‘nonconcord’ (2003: 276), 64 were minimal and 159 extended. This means that in 13.1 % of minimal clauses and in 8.3 % of the extended clauses, the verb does not show concord with the postverbal [NP].415 Breivik/Martínez Insua (2008: 361) even go so far as to claim that “there1 + singular be (notably the contracted form

|| 413 Breivik/Martínez Insua (2008: 358) find that in the subcorpus of the BNC they used more than 80 % of the instances showing non-concord contain the form there’s, but there is and there was do occur as well. 414 Their classification of existentials as minimal is applied to instances in which no element whatsoever follows the [NP]; no distinction between postmodification and extension is made. 415 Martínez Insua/Palacio Martínez actually claim that “the frequency of non-concord is markedly higher among TCs with an extension […]. Minimal existential TCs […] as opposed to those containing any kind of extension after the PVNP […] display lower rates of non-concord” (2003: 276). Apparently, they were misled by the higher overall numbers. It is true that extended existentials are more common in the group of non-concord showing existentials, but this is only due to the fact that they are more common overall.

270 | Existentials

there1’s) has acquired the status of a fixed pragmatic formula.” They do not, however, explain how this can be reconciled with the occurrence of the contracted form of has discussed above. Their analysis is, in fact, very close to what Chomsky suggests in a footnote: As is well known, agreement with the associate is sometimes overridden, as for example in there’s three books on the table, there’s a dog and a cat in the room (vs. *a dog and a cat is in the room). The phenomenon, however, seems superficial: thus, it does not carry over to *is there three books ..., *there isn’t any books ..., and so on. The form there’s seems a frozen option, not relevant here. (Chomsky 1995: 384)

Thus, to return to the original problem here, the question as to which element the verb agrees with cannot be answered conclusively since agreement is dependent on the style level.416 A model of existentials ideally should be able to account for such conflicting evidence. In non-informal style, the simplest (and thus possibly preferable) interpretation is, however, that the postverbal [NP] determines concord.

8.4.3 Further studies There are further studies on the structure of existentials that did not fit the treatment in the previous chapter but add new aspects that are worth discussing. || 416 In a more comprehensive study, Riordan (2007) found that a whole array of factors determines the occurrence of concord in existentials. His own summary of his distributional analyses will serve here to show that the situation is more complex than can be modelled in the present chapter: We saw an effect of type of determiner on nonconcord, with postcopular NPs containing cardinal numerals and non-count quantificational nouns significantly associated with greater levels of nonconcord. There was an effect of the length of the predicate on concord, such that shorter predicates were more associated with nonconcord, contradicting a previous finding in the literature. There were strong effects of social and discourse factors on concord: increasing age and formality of discourse context were associated with lower levels of nonconcord. (Riordan 2007: 261) Since Riordan’s study is based on a spoken corpus recorded in an academic setting, some variables are not independent (e.g. higher age correlates with certain social roles and thus possibly with a higher degree of formality). Consequently, he is careful not to make strong claims about ‘English’, but even if the study may need revalidation based on other data, the determiners of agreement he identifies appear convincing. For a variationist study based on a UK dialect see Rupp (2005).

Theoretical treatment | 271

Collins (1992) adopts a relatively traditional treatment of existentials in general but proposes a markedly different analysis for some of the existentials that have a relative clause or participial extension. These he calls cleft existentials for various syntactic, semantic and pragmatic reasons. Semantically, he claims that “[w]hereas standard existentials predicate the existence of an entity or entities, cleft existentials are concerned with the description of complex events” (Collins 1992: 426). In this respect, the division appears to correspond to that between bare and extended existentials presented above. However, for Collins there are no locative extensions since the existence of a locative element makes the structure a standard existential. While this may be right as a tendency, it still remains doubtful whether all existentials with a locative extension can be seen predicating the existence of something as can be observed in the following examples: (76)

There’s no gun in this house. (GW3 3228)

(77)

That’s that’s the hundred metres that’s the length of his string. There’s his kite on the end of that. (FMJ 472–3)

While (76) conforms to the prototype, (77) – while possibly not the description of a complex event – is less straightforwardly conceived as an assertion of existence but rather as a predication as to the position of the kite. The latter type is comparatively rare, though. In generative grammar, there are a number of ways to model the structure of existentials. As discussed above in Section 8.4.1, Williams (1984) argues in favour of an analysis that treats the underlined words in (78) as an [NP] whereas Lasnik (1995) prefers to analyse them as constituting a small clause, i.e. with a knock as subject of the underlined structure. (78)

There was a knock on the door and Scales came in. (GW3 1994)

Some of the arguments brought forward for and against the respective analysis are discussed in Section 8.4.1 above, others are of a theory-internal kind (i.e. what principle of generative grammar prevents which sort of derivation, etc). Law’s (1999) introduction gives a general idea of the debate: Chomsky (1986, 1991) suggested that at some (abstract) level of representation [(79b)] the associate DP [=the postverbal NP] ends up in the preverbal subject position by a process of expletive replacement, where it would agree with the finite copula verb, just like it does in [(79a)]: (Law 1999: 183)

272 | Existentials

(79) a. There are/*is many books on the table. (Law 1999: 183) b. There + many booksi are ti on the table. (Law 1999: 183) c. Many books are/*is on the table. (Law 1999: 183) To draw a close parallelism between [(79a)] and [(79c)], however, one may have to assume that the expletive there in [(79b)] deletes at some point in the derivation (Chomsky 1986) [...]. (Law 1999: 183)

From a non-transformational point of view it would be highly unintuitive to assume that there is an existential there at some underlying level of representation in (79c) which is then deleted in the derivation; other transformationalists (e.g. Williams 1984) model the relationship between (79a) and (79c) as the result of a process of “there-insertion” (which is also the title of Williams’ paper). Neither of the two treatments is compatible with any sort of usage-based approach. Finally, Erdmann’s (1976) study of existentials is based on a rather idiosyncratic theoretical model of relations which is too complex to discuss in great detail in the context of the present chapter. Basically, he assumes underlying representations of sentences which are based on relations of different types between two elements and appear to be similar to X-bar theory. The underlying representation of (80) is reproduced in Figure 25. (80)

There is someone on the stairs. (Erdmann 1976: 178)

Fig. 25: Graphical representation of (80) in Erdmann’s relational model (adapted from Erdmann 1976: 185)

As we can see, his approach departs sharply from other approaches in that the two nominal elements someone and the stairs appear in a relation and the preposition has a relation to that relation. There is definitely not the subject at the underlying level given that it does not figure in the underlying representation.

Theoretical treatment | 273

Thus heavy use of derivational mechanisms is needed in Erdmann’s model in order to arrive at surface structures, so his theory is also incompatible with usage-based approaches and a (surface-oriented) valency grammar as proposed by Herbst/Schüller (2008).

8.4.4 Proposed structure Given the conflicting evidence and competing treatments outlined above it is not surprising that existentials are problematic to model in a valency framework, too. In terms of the valency structure, bare existentials resemble the so-called obligatory extraposition with verbs such as appear and seem discussed in Section 7.1.2. Thus in (11a), we have one element (there as subject) in front of the verb and one after it. (11) a.

There is no cure. (A01 7)

Both are obligatory, so (11a) cannot be rephrased as (11b). (11) b. #No cure is.417 Thus we have a typical divalent use of be. As discussed in more detail in Section 6.2, the valency of be is generally problematic in that one may argue that postverbal complement instead of the verb is responsible of valency restrictions, even though the divalent syntactic structure of such sentences seems to be determined by be.418 If we assume that be is the primary valency carrier in (11a), we can state that both there and the postverbal [NP] are complements of the verb as they are obligatory and formally specified. However, while the postverbal element also likely fills an argument slot, it is more difficult to claim argumenthood for there given its non-realization in the semantically similar (11c).419 || 417 This sentence is of course a grammatical and acceptable sentence of English in certain communicative contexts. Thus in a short exchange of the type Can you offer a cure that is 100% reliable? – No cure is. it is perfectly natural. However, (11b) is no acceptable alternative to (11a) in the context in which it was used. (The example is taken from a leaflet on AIDS.) 418 The unpleasant side-effect of an analysis in which the postverbal element is a valency carrier would be that we would have to posit that the nouns which occur in postverbal position of existential clauses have a valency slot for there. This is of course neither desirable from a theoretical point of view nor psychologically plausible. 419 See Helbig (1992: 9f) and Herbst/Klotz (2002) for a discussion of mismatches between arguments and complements in valency theory.

274 | Existentials

(11) c.

No cure exists.

We thus have to conclude that bare existentials cannot be modelled satisfactorily in a traditional valency approach without devising specialised mechanisms or exceptions to rules that are needed elsewhere in the model. For the discussion of extended existentials, we are faced with two questions. The first is whether the postverbal [NP] and the extension form one constituent or two and the other is what type of constituent the entire postverbal element (in the case of a positive answer to the first question) or the extension (in case of a negative answer to the first question) represents. The treatment of extended existentials that will be advocated here analyses the extension (underlined in (81)) as a predicate to the postverbal [NP], irrespective of the formal realisation of the extension. (81)

When we reached the start after a nervous descent there were no fewer than seven climbers ahead of us. (ECH 283)

Whether one would want to analyse the whole postverbal part as one constituent is a matter of personal preference to some extent. The fact that it contains a predication would speak in favour of it, agreement would speak against it since clauses are usually treated as singular (see discussion on extraposed clauses in Section 7.1.3.2). With regard to presentational existentials, Williams argues that the underlined words in (82) cannot form a clause “since arrive and the other verbs that appear in these TISs [= there-insertion sentences] are not verbs that take clausal arguments” (Williams 1984: 146). (82)

There arrived some people very sick. (Williams 1984: 146)

However, as we have seen above, only a relatively limited set of verbs can occur in such presentational existential structures, although it is not certain whether a comprehensive list can ever be compiled. Since general semantic groups or rules appear to be insufficient to predict their occurrence, it would of course be desirable to state the possibility to occur in an existential clause as a lexical property of the verb, similar to other valency information, but if no predictions can be made reliably, a valency model does not appear appropriate, either. A more promising model might be an argument structure construction model as presented by Goldberg (1995), where verbs that are not incompatible can be used in a certain argument structure pattern which is associated with a certain || In the generative framework, existential there is not assigned a theta role either (see Chomsky [1981] 1993: 35).

Theoretical treatment | 275

meaning, but no storage of the potential of a verb to occur in a certain argument structure construction is necessarily assumed. It would require a much largerscale study of presentational existentials to verify to what extent a network model as discussed in Section 9.3.3 might be able to predict the verbs that can be found in such structures based on different types of input data. Still, in order to model existentials in a valency approach, there would be a need to posit larger and internally complex valency carriers (cf. also the categories of lexically and contextually specified elements in the Erlangen Valency Patternbank [Herbst/Uhrig 2009]) within a valency pattern approach, where agreement is modelled via a corresponding sentence type (see Section 9.3.2). It will be argued here that there combines with a verb (typically with a form of be) to form a larger unit with one valency slot that allows for a noun phrase or a predication, while one can still identify, conjugate and possibly replace the verb in the complex valency carrier. However, this is already far removed from conceived valency theory since the verb is not treated as the primary valency carrier in the sentence, so the only similarity to valency is that the combination of there + be opens up a slot for another complement and restricts the form of that complement. This is in principle in line with the idea of multi-word valency carriers mentioned by Herbst/Uhrig (2010: 131). Thus we have to conclude that any description of existentials that relies on a strictly verb-centric valency model is unsatisfactory, as the idiomatic character, the unique syntactic structure and the distinct pragmatic function420 of the construction will not be represented adequately in such a model. There is an important difference between existential there and the dummy subject it in extraposed sentences: it can occur in subject position whenever the slot is not occupied, be it due to extraposition or due to the fact that a verb such as rain does not open up any valency slots. And it is a word that can usually occur in subject position when it is used with referential function. Existential there, however, is more difficult to model as its occurrence is limited to a small set of closely related grammatical constructions, so it is impossible to assess its meaning or function independently of this family of existential constructions. Thus, any treatment in terms of general syntactic rules or valency will need to account for the specific properties of the construction using the tools offered by the model and thus make its description more complex than it probably is. We have to agree with Lakoff (1987), who takes the peculiarities of existentials as an

|| 420 It is outside the scope of the present study to describe the pragmatic functions of and restrictions on existentials in detail. See CamG (1396ff) for an in-depth discussion of such factors.

276 | Existentials

argument in favour of a construction grammar approach in the sense of Fillmore/Kay/O’Connor (1988) and against an autonomous treatment in the syntax: Of course, grammatical construction theory is not the only contemporary theory that permits the direct pairing of syntactic and semantic information. However, it is the only theory I am aware of that permits the pairing of complex syntactic configurations with the appropriate pragmatics – in this case, conveyed illocutionary force. Grammatical construction theory permits such pragmatic factors to enter directly into the composition of sentences. Generative theories with an autonomous syntax cannot do this. (Lakoff 1987: 481)

We thus simply have to accept that constructions without an easy-to-model internal syntactic structure can be learned and used independently of supposed general rules in the language,421 which is probably due to the fact that units of meaning422 are often larger and more complex than traditional morphemes – an insight that has gained substantial ground in linguistic theory over the past 20 years (see for instance Sinclair 2004, Goldberg 1995). As to the question of the usefulness of the concept of subject, we have to conclude that existentials are among the structures in the English language for which the concept is of limited use since many but not all the properties typically associated with it are found in the existential there (e.g. no concord, no aboutness). Also, cross-linguistically, this seems to be true of such structures as observed by Gast/Haas: It is well-known that presentationals are often characterized by a non-canonical distribution of subject-properties over the (pro)nominal constituents of a sentence […]. Moreover, many presentational constructions contain an ‘expletive subject’, i.e. a (mostly pronominal) element which does not play an obvious role in the argument structure or interpretation of a sentence while still exhibiting some (or all) subject properties [references omitted].” (Gast/Haas 2011: 128).

It even appears sensible to posit that, in the light of constructional approaches and network models, the concept subject may do more harm than good since it does not help in any way to describe this family of structures that are most likely “learned pairings of form and function” (Goldberg 1995: 5) but forces on them

|| 421 This is in line with Breivik/Martínez Insua’s position mentioned above, who argue “that there1 + be functions as a presentative formula today, and that the use of non-concord […] is the result of a long process of grammaticalization and subjectification” (Breivik/Martínez Insua 2008: 351). 422 It follows automatically that meaning is taken in a broad sense in such a statement so that it includes not only lexical meaning but also functions such as focusing and the like.

Theoretical treatment | 277

an analysis in terms of categories that are smaller than psychologically plausible and thus simply not suitable for the description and modelling of their particular properties.

9 Conclusions and perspectives for syntactic models The present study has provided a review of the theoretical treatment of various structures as well as a critical evaluation of these models based on large quantities of data from a range of sources. The point of the present chapter is not only to summarize the findings and problems discussed above, but also to offer an integrated perspective of a possible theoretical treatment of these phenomena in a valency framework and beyond.

9.1 Data for linguistic analysis The data-intensive approach taken in this study led to a whole array of problems of data analysis and data interpretation. One set of issues is related to the corpora used and the other one to the native speaker interviews. Compared to many other studies on syntax, the corpus basis of the present study is quite substantial. It does not only rely on the BNC with its 100 million words of running text but also on a range of additional corpora presented in Section 3.2.2, which – together with the BNC – amount to more than 1.5 billion words of text. Due to their opportunistic sampling method, they cannot be regarded as representative of ‘the English language’ to the extent to which the BNC can, but representativeness is less of an issue in a study of the nature of a syntactic phenomenon. Whether the corpora contain a higher or lower proportion of clausal subjects than the population (i.e. ‘English’) does is of marginal importance for the present study.423 However, there are a few issues that are associated with such huge and less carefully sampled corpora that have to be addressed. The main problem is noise, i.e. problematic instances of data one would not want to include in a carefully designed corpus. These comprise (but are not limited to) the following: – imperfect pre-processing (it turned out that headlines were often merged with the following sentence, for instance) – non-native usage (often cited in newspaper articles)

|| 423 We can thus agree with Schierholz’ (2001: 96) argument that representativeness of the sample is not the primary concern in the choice of corpus as long as we do not make claims about the frequency distribution in the population based on the non-representative sample. https://doi.org/10.1515/9783110589801-299

280 | Conclusions and perspectives for syntactic models

– –

archaic language (some of the texts were about 100 years old) poetry (which was explicitly excluded from corpora such as the Brown corpus due to its often highly deviant syntax and creative use of language).

While the first point may only lower precision and recall of search queries, the second and third (and to some extent the fourth) are much more difficult to detect, although detection is usually possible through careful checks of the linguistic context. The problem of how to deal with such instances remains, though. Should we discard words (supposedly) uttered by non-native speakers of English but published in British newspapers? Should we ignore evidence if it turns out to be over 50 years old? If so, since the corpora apart from the BNC do not contain metadata and we can thus only check the status of individual instances by trying to find the source, should we only discard evidence that appears to be dubious, i.e. contradicts the theory? All decisions of that sort are problematic, so in order not to tamper with the data, no filter of any such kind was applied. The respective instances were, however, marked with a comment. Furthermore, there are technical limitations that have to be mentioned in relation to the corpus evidence. One is that in the parsed corpora the reliability of the stochastic parser used is far from perfect. The figures published by Cer et al. (2010) suggest that we can expect about 85 % correct attachment of labelled dependencies with the Stanford Parser that was used here, but these figures might be overly optimistic given the varied nature of corpus text used in the present study. In an evaluation using a different corpus, Proisl (personal communication) found that labelled attachment with the model used here drops to around 75 %.424 We thus have to expect that, for instance, we have missed out on [to_INF] subjects that were present in the corpora but were analysed as [to_INF] adjuncts of purpose instead. It was also impossible to automatically retrieve [that_CL] subjects without that, but the analysis of extraposed [that_CL] subjects is based on sufficient data, so that the overall results should not have been influenced in a meaning-changing way by this limitation. Let us briefly put the size of the corpora into perspective: For the BNC, Aston/Burnard claim that it “corresponds to roughly 10 years of linguistic experience of the average speaker in terms of quantity – though not, of course, in

|| 424 A certain error rate is inherent given that there are syntactic ambiguities where even the human annotator cannot decide which structure is ‘correct’, such as so-called PP attachment, i.e. structural ambiguity between postmodifier and adjunct in cases such as He saw the spy with the binoculars. As a consequence the figures often appear lower than they are because human annotators could never score 100 % on this kind of task, either.

Data for linguistic analysis | 281

quality” (Aston/Burnard 1998: 28). Their estimate of how much language people are exposed to seems to be rather conservative, though. Dąbrowska assumes “an eight-hour ‘language day’ – that is to say, that most people spend about eight hours a day engaged in some kind of linguistic activity (talking, reading, watching television, listening to the radio, browsing the Internet, etc.)” (Dąbrowska 2004: 19), which means that the BNC would correspond to roughly four years of linguistic experience of the speaker.425 Given that most native speakers interviewed for the present study were language professionals, Dąbrowska’s estimate may actually be closer to the reality of their linguistic exposure. If we accept that estimate, our corpora would correspond to 60 years of linguistic experience. Thus if we find a structure only once in the corpora (e.g. a [that_CL] non-extraposed subject with a given verb), there is only a 50 % chance that a 30-year-old speaker has come across this structure even only once in his entire life so far.426 Given that many of the structures we studied are syntactically complex and are thus probably over-represented in our almost exclusively written corpora and given that items that occur once in a corpus are often overrepresented statistically, the chance is probably much lower. For most structures the corpus evidence was sufficiently large to identify actual restrictions and to enable us to base serious arguments upon them. Since the data collection was carried out against the hypotheses (i.e. even dubious evidence that contradicted the hypotheses was accepted as valid), the results should of course not be read in a way that suggests that every construction with a plus-sign in the tables presented above is equally acceptable – the opposite is the case. But since much of the relevant data is made available in the appendices, the reader can of course derive his or her own judgements from the primary data. Generally, the data available for the active clause was sufficient; for the passives there was less data and there were huge differences between the different lexical units studied. For non-extraposed subjects in copular clauses, the availability of data was most severely restricted and the native speaker informants’ verdicts were indispensable. However, there are also problems with the native speaker data that need mentioning. For purely practical reasons, most items could only be tested with

|| 425 Calculated on the basis of a rate of 150 words per minute (see Dąbrowska 2004: 19 for details). 426 Using the figures quoted by Aston/Burnard, the corpora would correspond to 150 years of exposure and the chance of a 30-year-old native speaker having heard the construction once would only be 20 %.

282 | Conclusions and perspectives for syntactic models

two, three or four native speaker informants, which was considered sufficient because the main idea was to verify corpus data and exclude obvious errors. It turned out that ‘obvious’ errors were not so obvious after all. We can illustrate problems of that kind when we take a closer look at the evidence used for the systematic study in Section 6.3.1.2 on adjectives that occur with an extraposed [to_INF] according to the VDE but not with a [that_CL]: Since the VDE provides very good coverage of almost all relevant adjectives that allow an extraposed [that_CL], the vast majority of the adjectives that had to be checked because they were not listed with a [that_CL] in the VDE disprefer the [that_CL] to some extent (at least compared to the [to_INF]). Nonetheless, in the native speaker interviews, NS1 and NS4 fully accepted 146 and 179 items respectively and categorically rejected 13 and 6. NS2 and NS3 fully accepted 80 and 84 and rejected 91 and 93. All native speakers were interviewed independently of one another. The figures are charted in the following diagram:

200 150 rejected (1)

100

fully acceptable (5) 50 0 NS1

NS2

NS3

NS4

Fig. 26: Acceptance and rejection of test items for adjectives followed by a [that_CL] (see Appendix 19)

While the picture seems to suggest that NS2 and NS3 virtually share intuitions, this is by no means so. They exactly agree in their verdict for the full acceptance in 39 instances, for the rejection in 46 instances and they do not fully agree in 127 instances.427 In the light of such contradictory evidence it becomes obvious || 427 This sounds slightly worse than it is since sometimes one speaker may have fully accepted a sentence which the other would have accepted but dispreferred over another structure. Furthermore, of the sentences tested, about 50 were made up (often with the help of NS4), so not all of the disagreement is on corpus examples.

Data for linguistic analysis | 283

that a binary acceptable/unacceptable distinction is hard to maintain and that it is important to additionally take frequency data from corpora into account. In the case of the adjectives with extraposed subject that is comparatively easy to do due to the usually high overall frequency of the construction. In the corpora used above for research of the construction (about 780 million words), no instance of illegal with an extraposed [that_CL] subject was found, so three native speakers were asked for their verdict on the following example: (1)

It was illegal that she was fired just because she was pregnant. (invented)

It was accepted by two out of three informants, but if we compare this structure to the infinitive variants ([to_INF] plus [for_NP_to_INF] and [of_NP_to_INF]), which occur roughly 2,000 times in the parsed corpora, it becomes immediately obvious that whatever the native speaker informants might say when they are presented with it, this construction is practically never used (or, to stay with the model given above, it is likely that a speaker never hears this structure in his or her entire life).428 In such cases, where the difference between the realisations in the corpus is sufficiently large, even Stefanowitsch’s method for accounting for negative evidence (Stefanowitsch 2006) would be usable in principle if it were possible to determine the frequency of the construction as such (i.e. extraposed subject with adjective). Unfortunately, it is currently impossible to determine this frequency for two reasons: First, due to the large number of instances a manual classification is out of the question. Secondly, the phenomenon resists automatic analysis given that no distinction between the referential pronoun it and a post-adjectival complement on the one hand and dummy it with an extraposed subject on the other hand can be made reliably.429 In the case of rarer data, the situation is much more difficult. If we find one example of hilarious with a non-extraposed [that_CL] and five or six430 with a non-extraposed [to_INF] in the parsed corpora, we cannot tell from the corpus data whether a [that_CL] is as acceptable as the [to_INF], so we have to revert to native speaker judgements if we want to make a statement about hilarious at

|| 428 The situation is similar for a [that_CL] subject with persuade, where only one example was found and native speaker informants accepted the construction. Newmeyer’s (2003) conflicting results presented at the end of 4.1.3 might thus be explained by the fact that one instance in our corpora may correspond to less than one instance in the lifetime of a native speaker. 429 Similar difficulties arise due to the ambiguity of the forms that and to. 430 One of the sentences has different readings and might or might not be counted.

284 | Conclusions and perspectives for syntactic models

all.431 But since we saw with illegal that native speaker judgements need not necessarily reflect actual usage, it is hard to decide whether it is preferable to make a statement about acceptability or not if the corpus evidence is too thin. The whole question of what data is adequate is related to the question of what it is we would like to describe in a grammatical model. If a grammatical description aims at being a model of the psychological reality of speakers, then – given the huge disagreement among native speaker informants – it may be problematic to base such a description on a corpus, since a corpus contains average usage over a whole range of individuals with individual grammars. Thus one may be inclined to follow the route taken by the Kiparskys, who were faced with a similar problem of non-agreement among native speakers: There are speakers for whom many of the syntactic and semantic distinctions we bring up do not exist at all. Professor Archibald Hill has kindly informed us that for him factive and non-factive predicates behave in most respects alike and that even the word fact in his speech has lost its literal meaning and can head clauses for which no presupposition of truth is made. We have chosen to describe a rather restrictive type of speech (that of C.K.) because it yields more insight into the syntactic-semantic problems with which we are concerned. (Kiparsky/Kiparsky 1970: 147)

While, at first sight, the idea of modelling the language of one person in order to achieve a psychologically adequate model has a certain appeal (and is related to Chomsky’s ideal speaker-listener), the general applicability of such a model is so severely restricted that it may not be able to predict how language is actually used. Since we know that (a) native speakers tend to change their minds432 and (b) that native speaker judgements do not necessarily reflect their language use, any model that relies exclusively on native speaker judgement is of course bound to be much less accurate in describing language usage433 than a model based on actual language use found in corpus evidence. In sum we can state that if we want to model language use, it is helpful to rely more on corpus data than on native speaker interviews. If the former is not available in sufficient quantities, it may become indispensable to rely on the latter, but the examples cited above indicate that such a step will reduce the quality of the results in many cases since there is a substantial divergence between what native speakers accept and what they produce. || 431 Just accepting one instance in a large corpus as proof of existence is what Schierholz calls “naive Corpuslinguistik” (2005: 13). 432 See, for instance, Quirk/Greenbaum (1970: 48f). 433 However, it has to be stressed that Chomsky’s aim is not to describe language use but to describe the internal grammar of a human being.

The concept of subject | 285

For the present study, after reviewing thousands of corpus examples and eliciting roughly 3,500 judgements from native speaker informants, we can be confident that most of the restrictions presented in this study will remain valid even when verified with larger corpora and more native speakers. The policy of accepting counter-examples even on a dubious empirical basis makes the descriptive part of the present study weaker but the argument for arbitrary restrictions stronger, since it is likely that in actual usage there are many more restrictions for most native speakers than claimed in this study.434

9.2 The concept of subject One of the hypotheses stated at the beginning of this study was that the notion of ‘subject’ may not be a sensible concept in grammatical theory and description. Chapters 7 and 8 on extraposition and existentials – and particularly the detailed discussions of what should be analysed as the subject in such structures – have shown that some of the defining criteria run into problems in noncanonical structures. Following up on these discussions it will be argued here that these problems are due to the fact that subject incorporates a variety of properties that only coincide in the same constituent in canonical structures.

9.2.1 Problems of diverging subject properties Let us recall that traditionally subjects are prototypically defined in terms of category, aboutness, agentivity, agreement, givenness and position, for instance.435 But only the subject in the first of the following set of sentences matches all these criteria: (2)

She is sleeping now. (G0X 3077)

(3)

My sister is a secretary and I am studying biology. (A6V 2174)

The subject in a copular clause, as in (3) above, cannot be called agentive in the sense of the original definition quoted in Section 2.3.2 as “the case of the typi-

|| 434 It follows that for a purely descriptive enterprise, such as a lexicographic project, relying exclusively on the vast amount of corpus data available would have been preferable. 435 See Section 2.3.2 for a discussion and a detailed list of relevant criteria. The properties mentioned here were chosen in order to illustrate how they can diverge.

286 | Conclusions and perspectives for syntactic models

cally animate perceived instigator of the action identified by the verb” (Fillmore 1968: 24) since there is no action. With clausal subjects, as in the following example, the subject is not animate and thus less typically agentive, although concepts such as Dowty’s (1991) proto-roles or Herbst/Schüller’s (2008: 158ff) clausal roles can remedy that issue.436 (4)

That it should suddenly be privatised and handed over to a Cambodian-Japanese joint venture has shocked many Cambodians. (newspapers)

Accordingly, a clausal subject in a copular clause is neither animate nor the instigator of an action: (5)

That she did not do so herself was beside the point. (GU9 907)

Also, a clausal subject is not the prototypical noun phrase either, so the criterion of category is not fully met. Thus the fact that we have used the term subject for clausal subjects throughout the present study in analogy to the prototypical nominal subjects at least deserves reflection despite it being in line with the major reference grammars and most grammatical models. In clauses showing extraposition, as we have seen above, there are two elements that might be regarded as subject in principle: (6)

‘It’s true that they never consulted me. (CDB 58)

There are some criteria that speak in favour of it as a subject in (6), others speak in favour of the [that_CL]. Specifically, in favour of it we can list category, agreement (to some extent), position, possibly givenness.437 On the other hand, aboutness has to be counted in favour of the [that_CL] and so does agentivity if we regard it as a matter of degree or in terms of proto-roles/clausal roles in noncopular clauses such as the following: (7)

It amazes me that there are still some people who equate morality with Christianity. (K55 8936)

For existential clauses, the situation is equally difficult. Thus in the following example, the verb seems to agree with the postverbal [NP] and the sentence is about it. On the other hand, it is not usually given nor is it in preverbal position.

|| 436 A similar agentive interpretation of some subjects that are not typically agents in German is noted by Reis (1982: 182). 437 See discussion in Section 7.1.3.2.

The concept of subject | 287

(8)

There are many good reasons. (A67 1662)

Similar issues in the identification of subjects occur for instance in so-called cleft or pseudo-cleft sentences ((9) and (10) respectively): (9)

It is this typological point that I would like to consider now, bearing in mind the question of historical divergence and convergence that I referred to above. (FAD 72)

(10)

‘What you’re hearing is the sound of sacred cows dying.’ (CSJ 247)

The motivation behind the use of such structures438 is usually that they allow to assign topic or focus status (and/or given or new status) to elements that would not have the same status in the canonical clause (or at least not to the same extent). Halliday’s analysis of cleft sentences illustrates the difficulty in assigning one element subject status: (11)

Pensioner Cecil Burns thought he had broken the slot machine; but it was not the machine he had broken – it was the bank. (Halliday 1994: 98)

Halliday analyses it + he had broken as the subject in its clause in example (11), which is another indicator of how inadequate the notion of ‘subject’ is for such constructions. Thus we can observe that subjecthood is not a unified notion in noncanonical structures of English if subject is understood in the sense as defined in Section 2.3.2, particularly if we include the semantic and pragmatic properties. The problem arises from the fact that instead of treating such noncanonical constructions in their own right, many grammarians try to treat them in terms of other, prototypical constructions, for which categories such as subject were established. It is of course possible to apply the criteria and classify elements in such structures as subjects (see Sections 7.1.3.2 and 8.3), but since such a classification has to remain arbitrary to some extent and since the element in question does not carry all the features of a prototypical subject, the explanatory value of such a classification is limited.439

|| 438 See CGEL (1377–1389) for an overview of structures; also Lambrecht (2001) for a typology of cleft sentences. 439 From a German valency perspective, Järventausta (2003: 781) comes to the same conclusion. Similarly Helbig (2004: 460), who argues against the postulation of perfect mappings (“Isomorphie”) of logical, semantic and syntactic levels in a model of language.

288 | Conclusions and perspectives for syntactic models

9.2.2 Consequences for modelling ‘subjects’ There are several possible ways of dealing with the challenges outlined above in approaches to grammatical theory and description. The first is to ignore them and to keep using the term subject even though there are many even relatively basic structures (such as copular clauses) where some of the criteria fail. There are good reasons for such a treatment since the term is well-established and since many concepts in many linguistic models show no clear-cut boundaries either, rather conforming to prototypes or a point on a gradient scale.440 This is the approach taken by CGEL, the authors of which are obviously aware of all the issues associated with the prototype approach taken and its shortcomings. While CamG does not rely on evidence from the semantic and pragmatic levels and also discards some of the traditional properties (e.g. passivization), the authors nonetheless use a prototype approach in that they include marginal subjects in their definition, too. This may be due to the fact that a model in which (almost) every clause has a subject (and ideally only one, as expressed through CamG’s uniqueness constraint discussed in Sections 2.3.2, 7.1.3.2 and 8.3) appears appealing for its neatness and systematic rigour.441 The second possible way is to restrict the use of the term subject to those cases in which all or most of the prototypical features are united in one constituent (i.e. prototypical active declarative clauses, for instance) and not use it in structures such as existential sentences at all. The advantage of such a solution is that the term is an established shorthand for a range of features which would have to be lengthily listed even where they coincide. The disadvantage is that any decision as to what is still called subject and what is not must remain arbitrary. Thus, if we follow Langacker, who seems to take such a position, we may even want to exclude clausal subjects altogether:

|| 440 One disadvantage of a prototype analysis is that one cannot predict any properties of an element simply due to the fact that it is a subject since it may be a marginal member of the category. 441 If there is no satisfactory definition of subject, the rigour is of course limited, but often grammarians have the (sometimes implicit) aim to keep the number and complexity of rules low, so difficulties can be ‘hidden’ in imperfect assumptions about syntactic relations (see also Herbst 2014 for a critique of the use of the term object in Goldberg’s model of construction grammar).

The concept of subject | 289

Thus complement constructions have to be considered in their own terms. They approximate subject and object constructions in different ways and to different degrees, fully instantiating those constructions only as a special case. (Langacker 2008: 430)442

Should we also exclude non-agentive or non-given nominal subjects? A range of arguments could be brought forward for and against each possible boundary between subjects and non-subjects so that any decision is bound to remain unsatisfactory. The third possible reaction is to do away with the notion of subject altogether443 and instead specify the features separately where necessary. To a limited extent something similar is done in all those approaches that try to keep the levels of syntax, semantics and information packaging separate, as, for instance, pointed out very explicitly by Halliday (1994: 30–34), who only uses the term subject at the syntactic level.444 However, even with such a limitation to the syntactic level, such models run into problems where there is more than one possible subject position, as with extraposition, or where agreement and position do not coincide, as with existentials. Thus the most radical option is to give a set of features for each clause element of a given clause structure (such as active declarative, existential, ...) in a way very similar to the feature structures employed by the standard version of head-driven phrase structure grammar (HPSG; Pollard/Sag 1994) – and quite similarly by Sign-Based Construction Grammar (SBCG; e.g. Boas/Sag 2012) – which will be described briefly below:

|| 442 Complement constructions in Langacker’s terminology are what the present study calls clausal complements. 443 For German, Reis (1982: 195) advocates such an analysis as well; according to her, no relational labels are necessary and nominative NP is all the information that is needed. 444 As mentioned above, CamG’s approach is similar.

290 | Conclusions and perspectives for syntactic models

Fig. 27: Feature structure for she in HPSG (Pollard/Sag 1994: 20) – © 1994 by The University of Chicago Press, reproduced by permission

Figure 27 gives a very detailed picture445 of the features associated with the word she in HPSG. For our purposes, the attribute-value pairs CASE nom, PER 3rd and NUM sing are of relevance here. If we want to combine she with a verb, we need the features to be compatible with the feature structure of the verb. An example of the relevant features of the verb form walks is given in the following figure:446

Fig. 28: Relevant features of walks in HPSG (Pollard/Sag 1994: 28) – © 1994 by The University of Chicago Press, reproduced by permission

In this structure we find nom, 3rd and sing in the subcategorization property, indicating that walks can be combined with she (which is realised in the gram-

|| 445 All following figures give a shorthand notation of the full structure. If they were expanded, the representation would be as feature-rich as the one for she. 446 It has to be stressed that such entries for third person singular verb forms are created by means of a lexical rule in HPSG so they need not be entered into the lexicon separately if they conform to the rule.

The concept of subject | 291

matical model through a process of unification of the two feature structures). Since she walks is no particular challenge to any grammatical model, we shall consider how HPSG deals with existentials next. The feature structures discussed below apply to sentences such as the following: (12)

There is no one absent. (Pollard/Sag 1994: 148)

(13)

There are no students absent. (Pollard/Sag 1994: 148)

The structure for existential there (treated as a pronoun, it appears) is given in the following figure:

Fig. 29: Relevant features of existential there in HPSG (Pollard/Sag 1994: 147) – © 1994 by The University of Chicago Press, reproduced by permission

Note that 3rd person agreement is specified but no number.447 The relevant lexical entry for be448 as used in the existential examples above is reproduced below:

Fig. 30: Relevant features of be as used in existentials in HPSG (Pollard/Sag 1994: 147) – © 1994 by The University of Chicago Press, reproduced by permission

|| 447 In the entry for dummy it in extraposed structures it is of course classified as singular. 448 Again, the lexical entry is “possibly derived by a lexical rule” (Pollard/Sag 1994: 147) and later the finite verb form is derived via another lexical rule.

292 | Conclusions and perspectives for syntactic models

We can see that the subcategorization list specifically selects there, the postverbal noun phrase and the extension. There and the postverbal [NP] are coindexed for number as is signalled by the number 4, so that the [NP], which has number, will influence the number of there, which then again determines the form of the verb.449 Such a treatment is compatible with inheritance views as suggested by CamG or with the additional dependency in word grammar presented in Chapter 8.450 Yet the level of indirectness introduced through this kind of analysis is probably only due to the fact that a pre-verbal [NP] is modelled to determine concord and to avoid a modification of that rule. A model that posits direct agreement between the postverbal [NP] and the verb in existentials may thus be preferred due to a simpler but equally correct mechanism.451 The HPSG analysis should only be taken as an indicator as to what a model can look like that atomizes the subject and lists properties instead – there are countless other possible ways to model the exact technical details. For instance, such an analysis is in principle compatible with a valency approach such as that presented by Jacobs (1994) for German, as observed by Oppenrieder in his discussion of extraposition in German, since it allows a dissociation of various types of valency relations. For extraposition, the clausal complement is seen as the logico-semantic argument whereas the anticipatory element is specified formally (Oppenrieder 2006: 908).452 One major drawback of such a model is that it makes the analysis of language a very complex and technical endeavour. However, since it is the more accurate and more systematic model it is still preferable compared to the other ideas mentioned above. All views presented here on subjects in English can be defended. However, they all presume that there is one single way of analysing a grammatical structure and that grammatical structures necessarily have to be analysed. It will be argued in Section 9.3.3 that there is an alternative way of understanding the phenomena in which subject can be regarded as an emergent notion.

|| 449 Note that the lexical entry of be is not yet specified for tense, number and person. 450 It has to be noted that in later HPSG, a distinction between subjects and other complements is made (see Pollard/Sag 1994: ch. 9 for a detailed discussion of the reasoning behind this decision). 451 The discussion is of course irrelevant for cases of non-concord (see Section 8.4.2 for details), where we would expect there to carry a third person singular feature. 452 For English, we have seen that formal specification usually extends to the clausal element, too.

The concept of subject | 293

9.2.3 A universal category? While the focus of the present study is on English, the idea that the term and concept of subject are universal is such a frequent topos in linguistic literature that it has to be addressed here. As has been shown in Section 2.1, the term was developed in Europe and applied to European languages at different levels of analysis. It is thus not surprising that these different levels of subjecthood coincide in European languages more than in others:453 In English and the familiar languages of Europe, topics are usually also subjects, and comments are predicates: so in John | ran away. But this identification fails sometimes in colloquial English, regularly in certain special situations in formal English, and more generally in some non-European languages. (Hockett 1958: 201)

This becomes even more apparent if one unified concept of subject is postulated and if it is said to have a range of properties at different levels: [I]t seems to me that subjects in some Ls will be more subject-like than those of other Ls in the sense that they will in general, present a fuller complement of the properties which universally characterize b-subjects. Very possibly, for example, European Ls are more subject oriented than those Sino-Tibetan Ls discussed by Li and Thompson (this volume). (Keenan 1976: 307)454

Perhaps a concept that was developed for and most often applied to European (Indo-European is possibly a better term here) languages does not readily lend itself to the study of structurally quite different languages, despite Pullum’s wish that “[i]t would also be undesirable for a relation like SUBJECT itself to be non-universal” (Pullum 1980: 9).455 In fact, why should we expect to find clause elements that share such theoretically distinct properties as agentivity, topicality, agreement and initial position across a variety of languages that are genetically unrelated? When Pullum says “I view it as unfortunate (though I am sure it will not ultimately be harmful) that controversy should still be raging about the status of this basic grammatical notion after two thousand years or more of || 453 As seen in Chapter 2, the topic of the following citation has been called psychological subject in some accounts. 454 L: language; b-subject: subject of a ‘basic sentence’ – a construct Keenan uses to exclude the influence of as many factors as possible in his analysis. See also the brief discussion of Keenan’s criteria at the beginning of Section 2.2.1.3. 455 Of course, even the subjecthood across Indo-European languages is far from uniform. See, for instance, Spanish “null” subjects or the relatively free word order in German combined with comparatively strong case markings.

294 | Conclusions and perspectives for syntactic models

grammatical theory” (Pullum 1980: 16), he ignores the fact that he is referring to over 2,000 years of Western (i.e. mainly European) grammatical theory, which makes it appear as if he regards the terms developed there as primary and likely to be universal.456 Perlmutter, by contrast, appears to differentiate between the universality of an item subject and a universal definition of such an item. While he takes subject to be a theoretical primitive of relational grammar – which is thus applicable to all languages – he states “that there is probably no phenomenon statable in terms of the same notion of subject in every language” (Perlmutter 1982: 324; his italics).457 He concludes that “different notions of subject […] will prove to be necessary” (Perlmutter 1982: 324). Davies/Dubinsky also note that “the full range of ‘subject’ properties is mapped to a single argument in English” (2001: 12), which “creates the illusion of a single set of ‘subject’ properties in English (and other languages like it)” (2001: 13). It thus comes as no big surprise that “[t]he cross-linguistic evidence [...] shows that the notion of subject is epiphenomenal rather than primitive” (Davies/Dubinsky 2001: 13).458 Thus Falk’s analysis provides a good summary of the discussion presented in this section: Problems have arisen because the concept “subject” originates in traditional studies of classical Indo-European languages such as Greek and Latin, languages which are closely related genetically, areally, and typologically. Investing “subject” with theoretical content thus usually depends on either focusing on languages which are typologically similar to classical Indo-European languages or attempting to extend an Indo-European notion to languages which have very different typological properties. (Falk 2006a: 1)

9.2.4 Summary We have seen in the present chapter that the term subject is in fact highly problematic due to the difficulty of defining it satisfactorily. The last hypothesis stated in Section 1.3, namely that the ‘concept of subject in grammatical theory || 456 Halliday makes a similar point when he states that “[o]ne of the concepts that is basic to the Western tradition of grammatical analysis is that of Subject” (Halliday 1994: 30; my italics). 457 Perlmutter even claims that “[c]ertainly there is no reason to assume any such phenomena to exist” (1982: 324), by which he also explicitly rejects Keenan’s (1976) position on the matter. 458 Although universal grammar plays a very important role in the Chomskyan framework, the term subject – not being a primitive of that model – is of limited importance and thus its universality is not an issue for generative grammar. See Section 2.2.1 for a discussion of the status of subjects in generative transformational grammar.

Perspectives for grammatical models | 295

represents an amalgam of properties that can be mapped onto a single constituent in canonical clauses but less so in non-canonical structures’ thus can fully be confirmed. We can now add that for (typologically) different languages, the situation appears to be even more difficult. Evidence for this problem has been available for a long time (see for instance Jespersen’s observations cited in Section 2.1.2) and it is quite remarkable that researchers from theoretical backgrounds as diverse as generative grammar,459 relational grammar, LFG, HPSG and valency grammar460 have all come to similar conclusions. Nonetheless, reference grammars have to some extent ignored these results in favour of an accessible description of language. From the point of view of the accuracy of the description, however, models that discard the notion of subject and list the atomic properties often used for its definition instead have to be preferred.461

9.3 Perspectives for grammatical models 9.3.1 Item-specific selection of subjects The present study set out to show that there are syntactic restrictions on the form of the subject of a clause which are imposed by the valency carrier in the clause, regardless of whether the valency carrier is an active verb, a passive verb, an adjective or a noun (Hypotheses 1 to 3 presented in Section 1.3). In Chapters 4 to 6, we were able to confirm these hypotheses with the help of different clausal realisations of the subject the acceptability of which seems to depend on the valency carrier, as illustrated below for all types of valency carrier: The verb pay, for instance, does not allow a [that_CL] subject but a [to_INF] is fully acceptable. (14) a.

So to say they are a bad team does not pay them the respect they deserve. (newspapers) b. *So that he says they are a bad team does not pay them the respect they deserve.

|| 459 See the quote by McCloskey (1997) at the end of 2.2.1. 460 See the first quote by Jacobs (1994) in Section 2.2.2.2. 461 For consistency’s sake we will have to continue using the term as before in the remainder of this chapter, though.

296 | Conclusions and perspectives for syntactic models

Generally, while we found sufficient evidence for restrictions on both [that_CL] and [to_INF] subjects, the [V-ing] appears to be much more versatile and is acceptable with virtually all verbs that allow a clausal realisation of the subject. In the passive, the verb avoid cannot be combined with a [to_INF] subject but can with a [V-ing], just like in the active clause for the postverbal complement: (15) a. Threatening one with a knife is probably best avoided. (newspapers) b. *To threaten one with a knife is probably best avoided. However, the range of possible clausal subjects in the passive cannot be derived entirely from the postverbal complements in the active clause. Some verbs allow the same complements as passive subjects, some more than occur postverbally in the active, some less, and some simply different ones. Again, [V-ing] can often be used even if it does not occur in the active. While some researchers might want to take this as evidence of the noun-like character of [V-ing] clauses, there are also restrictions on this form. No [that_CL] subject seems to be possible with the adjective smart, whereas the [of_NP_to_INF] is established usage: (16) a.

I haven’t read either, but I think it is smart of HarperCollins to cash in on the classic revival vogue. (GWL 1163) b. *I haven’t read either, but I think it is smart that HarperCollins cashes in on the classic revival vogue.

And for nouns, task shows a restriction on the [that_CL] complement, too. (17) a.

It is the task of pressure groups to pressurise, not to imitate big business by offering glorified bribes. (newsmerge) b. *It is the task of pressure groups that they pressurise, not that they imitate big business by offering glorified bribes.

Interestingly, for subjects in copular sentences with nouns as predicative element, the [to_INF] realisation seems to be almost universally acceptable, so we can observe that they behave differently from verbal valency carriers. More evidence is available in the respective chapters and in the relevant appendices. Other researchers have come to similar conclusions on the nonpredictability of the variation over clausal elements,462 but their studies usually had a focus on postverbal complements. However, CamG’ discussion of the

|| 462 See for instance Faulhaber (2011).

Perspectives for grammatical models | 297

differences between [to_INF] and [V-ing] complements is perfectly compatible with our findings for these elements in subject position: [W]e cannot assign distinct meanings to the form-types and treat the selection as semantically determined. On the other hand, the selection is not random: [...] Prepositional to is characteristically associated with a goal, and a metaphorical association between toinfinitivals and goals is to be found in the fact that they commonly involve temporal projection into the future [...]. Linked with this is the modal feature of potentiality. [...] But it must be emphasised that we are talking here of historically motivated tendencies and associations, not constant elements of meaning. (CamG 1240f)

The modal component associated with [to_INF] subjects was studied in Section 4.2, where we came to similar conclusions, i.e. that there is definitely a tendency for [to_INF] subjects to occur with markers of epistemic modality. Still, only about half of the [to_INF] subjects show such behaviour, so that its predictive power is rather limited and cannot replace item-specific knowledge.463 Langacker’s view is also compatible with our interpretation: The meanings and distribution of to and -ing are complex matters that we can barely touch on here [...]. Both are polysemous and hard to disentangle from the varied constructions they appear in. Convention often dictates a particular choice, and when both are possible, the semantic distinction may be subtle at best. [...] Still, the various meanings of to and -ing center on different prototypes. (Langacker 2008: 439)

Duffley, on the other hand, claims with regard to [V-ing] and [to_INF] that “[i]t will be assumed here that these forms do have an inherent meaning that preexists and is stored outside of any particular use that is made of them” (2003: 329). This is in stark contrast to CamG, Langacker and the perspectives on grammatical models outlined later in Section 9.3.3. For German, many studies also find that the variation in clausal subjects is item-specific (see, for instance, Järventausta 2003: 784).464

|| 463 See Herriman (2000) for a detailed analysis of the modality of matrix clause predicates in extraposed structures. 464 See also the long lists of permissible clausal subjects for verbal, nominal and adjectival predicates in German in Zifonun/Hoffmann/Strecker (1997: 1450ff), which also point in the direction of item-specificity, particularly due to the frequent hedges used in the semantic groupings.

298 | Conclusions and perspectives for syntactic models

9.3.2 Valency This section will cover the challenges that the findings of the present study pose to valency theory, and what a suitable model to account for these problems may look like.

9.3.2.1 Valency patterns As we have seen in the previous section, the choice of subject is item-specific and thus the approach of valency theory to treat it as a complement of the verb (or of an adjectival or nominal valency carrier) is backed by the data. It is not possible to predict the form of the subject solely on semantic grounds (e.g. by semantic role). However, there are various forms such a valency approach could take and we shall show that a pattern approach is most suitable to account for some problematic structures identified in the present study.465 Traditionally in valency theory, each valency carrier has an inventory of complements that can occur with it. In a basic model, these are given as a simple list of possible realisations for each valency slot in the active clause. Helbig/Schenkel ([1969] 1973) offer such a description in the form of what they call valency frame (‘Valenzrahmen’).466 The order in the valency frame does not specify an actual constituent order in the sentence, which would be difficult anyway due to German word order idiosyncrasies. For English, a simple example would be the verb accuse, which can also be described as having three valency slots. The first would be the subject [NP], the second a postverbal [NP] (traditionally called object) and the third an [of_NP/V-ing] (see VDE 8):467 (18)

The president accused her government of scandalous corruption and incompetence. (VDE 8)

|| 465 See Herbst (2007) for arguments in favour of a valency pattern approach that are not based on subjects but on possible and not possible combinations of postverbal complements. 466 Thus, for the German verb schenken the valency frame is “II. schenken  Sn, Sa, (Sd)” (Helbig/Schenkel [1969] 1973: 180), indicating an obligatory nominantive [NP] (= Sn), an obligatory accusative [NP] (= Sa) and an optional dative [NP] (= bracketed Sd). 467 For English, there are also approaches that do not specify order. Thus, despite the fact that the arguments in Goldberg’s model of argument structure constructions (Goldberg 1995, 2006) are presented in the order of constituents in the active declarative sentence, they do not specify linear order (Newmeyer 2003: 169). While the complement block of the VDE only gives a complement inventory, the canonical order (i.e. without heavy unit shift or similar phenomena) of the postverbal elements (usually for the active declarative clause) is given in the pattern block.

Perspectives for grammatical models | 299

As mentioned in Section 5.1, many valency grammarians opt for a so-called valency reduction in order to be able to account for passive sentences with the same valency inventory as for the active. For the example of accuse we would then reduce the valency by eliminating the first [NP] complement (or, to be precise, by making it optional in the form of a by-phrase):468 (19)

Four leading stockbroking companies are accused of wrongfully compensating privileged clients for losses they sustained in share dealing. (VDE 8)

Since not all postverbal elements can become subjects of a passive clause, the ones that can will have to be indicated in the valency description of the verb.469 Such a simple model is sufficient to describe the valency properties of many verbs but, as discussed in the previous chapters, cannot accurately account for all the linguistic facts. The first such problem is presented by passive clauses in which a subject occurs that does not normally occur as a postverbal complement in active clauses with the verb in question, such as the [V-ing] clause with influence in the following example:470 (20)

Buying a house should not be influenced by the existence of mortgage tax relief, and scrapping it would raise £ 2,800 million a year at the last estimate. (newsmerge)

Cases of this type (more examples can be found in Section 5.2.2.1) show that we cannot derive the passive clause complement inventory from the active clause complement inventory by means of a rule such as valency reduction, since the [V-ing] has to be added specifically in the description of the passive. Thus we would either have to specify complements for their ability to occur in the active or passive clause or posit different inventories for active and passive clauses. The problem of so-called prepositional passives, however, cannot be accounted for in such a model at all. On the evidence presented in (21a–d) any complement

|| 468 The situation is actually slightly more complex given that the second valency slot is only contextually optionally filled according to the VDE, but for the simplicity of illustration such issues will be ignored. 469 This is the approach taken in the VDE, for instance. 470 There are some postverbal complements realised by an -ing form in the corpus, but these seem to be more nominal in character since they occur without their own postverbal complements.

300 | Conclusions and perspectives for syntactic models

inventory-based approach has to fail since it cannot explain why the particle on is impossible in (21a) but perfectly acceptable in (21c): (21) a. b. c. d.

*They agreed on that they would meet. (CGEL 1178) They agreed that they would meet. (CGEL 1178) That they should meet was agreed on. (CGEL 1178) That they should meet was agreed. (CGEL 1178)

In the valency pattern471 approach advocated here, one would give the following patterns for the active and the passive clauses above:472 [NP]subj...agreeact...[that_CL] [that_CL]subj...agreepass...[on] [that_CL]subj...agreepass [that_CL]subj...agreepass...[on]...[by_NP] [that_CL]subj...agreepass... [by_NP]

For the sake of brevity, one could introduce brackets to indicate optionality and thus reduce the four passive patterns to one: [that_CL]subj...agreepass...([on])...([by_NP])

In some cases, such a purely formal specification of the pattern may be misleading. Thus, in analogy to the case of agree, one may be inclined to form a similar passive pattern to cover (22a–b) discussed in Chapter 5: (22) a.

That the moral imperative was not a sufficient condition has already been remarked upon. (CS7 1331) b. That the moral imperative was not a sufficient condition has already been remarked.

The pattern could take the following form: [that_CL]subj...remarkpass...([upon])...([by_NP])

|| 471 As mentioned in Note 165 on page 107, the German concepts of Satzmuster and Satzbauplan are related but not identical to the concept of valency pattern used here. One major difference is that the former are always based on active declarative clauses. 472 The form is a mixture of the styles used in the Erlangen Valency Patternbank (Herbst/Uhrig 2009) and in Herbst/Schüller (2008). The subscript act or pass next to the verb indicates active or passive morphology. The subscript subj indicates subject status and would be redundant if the patterns were read to represent linear order. As discussed below, they are not meant to represent such an order although they are given in the order of a declarative clause pattern for ease of readability.

Perspectives for grammatical models | 301

However, the form of the pattern suggests that the [that_CL] in both variants of the pattern is the same complement, which, one may argue based on the VDE, is not the case since each [that_CL] fills a different valency slot of the verb and corresponds to a different participant role in the two sentences. Thus we need some sort of indication of either the valency slot or the participant role. Since valency slots are often determined based on the participant role of the complements and since the structure of such patterns is much more transparent we shall opt for the inclusion of participant role labels. The representation corresponding to (22a–b) above would thus be the following: [that_CL “topic”]subj...remarkpass...[upon]...([by_NP “remarker”]) [that_CL “content”]subj...remarkpass...([by_NP “remarker”])

As there is formal as well as semantic information, Herbst/Schüller (2008: 139) use the term valency construction (where construction is to be read in the construction grammar sense as a form-meaning pair [Goldberg 1995: 4]) for such a representation. This sort of representation has the further advantage of showing that in our analysis [upon] does not fill an argument slot in the passive pattern since it is not assigned a participant role. The second problem that argues against a complement inventory approach is the case of extraposition. As we have seen in Chapter 7, extraposition is itemspecific, which confirms the fourth hypothesis formulated in Section 1.3. For most cases, it would be possible to describe extraposition in a complement inventory approach if the relevant complements are marked as obligatorily or optionally extraposed. But the complement inventory approach runs into problems where the same complement can occur in extraposed position in one pattern while another pattern allows only the non-extraposed variant, as illustrated for attract in (23a–c): (23) a.

But to do so could attract severe penalties such as deduction of points and, more seriously, a ban from European competitions. (newspapers) b. Labour sources dismissed the claim, pointing out that to take such drastic action would only create martyrs and might even attract more MPs to the rebel cause. (newspapers) c. As I thought it might attract suspicion to walk the same way again, I looked at the shops in Tottenham Lane instead. (fiction)

The non-extraposed variant occurs both in the divalent (23a) and in the trivalent (23b) pattern, but the extraposed variant only occurs in the divalent (23c) version. A verbose valency construction representation of the three examples could take the following form:

302 | Conclusions and perspectives for syntactic models

[to_INF “ATTRACTER”]subj...attractact...[NP “ATTRACTED”] [to_INF “ATTRACTER”]subj...attractact...[NP “ATTRACTED”]...[to_NP “TARGET”] [it]...attractact...[NP “ATTRACTED”]...[to_INF “ATTRACTER”]subj

There are two possible ways of reducing the patterns. One is to introduce a symbol such as (it) to indicate that the complement could also occur in extraposition: [to_INF “ATTRACTER”]subj(it)...attractact...[NP “ATTRACTED”] [to_INF “ATTRACTER”]subj...attractact...[NP “ATTRACTED”]...[to_NP “TARGET”]

The other is to use brackets to indicate the optionality of the [to_NP]: [to_INF “ATTRACTER”]subj...attractact...[NP “ATTRACTED”]...([to_NP “TARGET”]) [it]...attractact...[NP “ATTRACTED”]...[to_INF “ATTRACTER”]subj

We cannot, however, combine all three into the following: *[to_INF “ATTRACTER”]subj(it)...attractact...[NP “ATTRACTED”]...([to_NP “TARGET”])

Such a structure, just like a purely complement based approach where the [to_INF] is marked as optionally extraposed, would wrongly predict that the following construction is possible, too, although it is not attested in the corpus: *[it]...attractact...[NP “ATTRACTED”]... [to_NP “TARGET”] ...[to_INF “ATTRACTER”]subj

Thus if one wants to conflate valency constructions for economy of description, one has to decide in such cases whether one wants to combine the two divalent or the two non-extraposed structures in one valency construction. The decision is to some extent arbitrary but one might prefer the combination of the nonextraposed constructions since the introduction of it and the ‘movement’ of the [to_INF] towards the end of the sentence are much more structure-altering. Here, we shall argue for using the verbose solution where each valency construction is listed individually, an approach also adopted for the valency patterns in the Erlangen Valency Patternbank (Herbst/Uhrig 2009), where only byphrases are treated as an optional part of the pattern (where they are). The reason is that the situation becomes much more complicated and any conflation of patterns thus will become arbitrary when the combinability of postverbal complements is taken into consideration.473

|| 473 See also the discussion in Herbst/Schüller (2008: 140f).

Perspectives for grammatical models | 303

The valency pattern approach outlined here can be used to describe the selection of complements, but it has to be made very clear that valency is just one component of a larger model of grammar. Thus one needs other mechanisms in order to fully describe the structure of sentences. Herbst/Schüller (2008) have argued convincingly for a combination of a valency component with a constituent structure approach and present different sentence types that specify the meaning associated with such sentences and the order in which the subject and other parts of the sentence occur. For instance, the representation of the interrogative 'yes-no-question'-construction is reproduced in the following figure: Interrogative ‘yes-no-question’-construction Meaning: ‘question’ Formal criteria: [op__Subj__rPred]

I.

Typical constituents: (a) Interrogative ‘yes-no-questions’ usually have a subject (b) Interrogative ‘yes-no-questions’ usually have a predicate containing a pre-head functioning as operator

II.

Word order: operator + subject + (rest of predicate)

III. Intonation: rising Fig. 31: Representation of the interrogative ‘yes-no-question’-construction (adapted from Herbst/Schüller 2008: 151)

The formal side of the construction specifies where the subject is positioned in relation to the different parts of the verb (or VHC in Herbst/Schüller’s terms).474 Thus the valency construction does not need to give such information and while the valency constructions listed above have a form that suggests a linear order, this is merely a notational convention for ease of readability.475 We shall also suggest here that valency patterns can be partially lexically filled or have further contextual properties. In this respect we shall again follow work done on the Erlangen Valency Patternbank (Herbst/Uhrig 2009), where contextual and lexical specification are distinguished as two types of additional

|| 474 In Herbst/Schüller’s terminology, the operator is a pre-head in the verbal head-complex (VHC), which in turn is part of the predicate. 475 See also Herbst (2014) for a discussion of what exactly valency constructions specify.

304 | Conclusions and perspectives for syntactic models

information given on a pattern. Contextual specification is illustrated in the following example: (24)

They were at a loss to know what to do with this sullen rebel who kept bursting out against them so unjustly. (CBN 801)

Since at a can hardly be said to fill a valency slot of loss in the first example and since it was decided to make no statement about the exact theoretical status of at a, the neutral term contextually specified was chosen for this sort of phenomenon. The problem is relevant to the description of the valency of loss as the combination at a loss takes a [to_INF] complement whereas loss on its own does not usually allow this form. Thus, “it might be more appropriate to treat such cases in terms of multi-word valency carriers” (Herbst/Uhrig 2010: 131) in a valency framework. Such an analysis is one argument in favour of a theory that allows for constructions or patterns of arbitrary size, such as constructions (e.g. Goldberg 1995, 2006), MacWhinney’s (2014) item-based patterns or Sinclair’s (2004: 24ff) extended units of meaning. In such a framework much of the combinatory properties of such larger units can be described by means of valency, too. The case of of importance (or, more frequently, of [ADJ] importance, as in example (25) below) discussed in Section 6.3.4.1 can be described in such terms: (25)

To encourage succinct writing about science is of great importance – and these winners will make a real contribution to science communication. (newspapers)

A pattern representation with a clausal subject may then take the following form, in which of is simply given in pre-nominal position. Such a pattern can represent the syntactic facts independently of the theoretical treatment, i.e. irrespective of whether one prefers to read it as an instance of contextual specification by of or as a complex valency carrier of importance: [to_INF]subj...of importance [to_INF]subj...of [AdjP] importance476

|| 476 Since the Erlangen Valency Patternbank does not list non-extraposed subjects of adjectives, this pattern is not included, but there are contextually specified patterns for extraposed variants. The pattern without the intervening [AdjP] does not exist in the VDE since it is less common, but there is corpus evidence which shows that it also occurs.

Perspectives for grammatical models | 305

Another type of additional information to include in valency patterns is lexical specification, which can be observed in the following example: (26)

Puzzled, he tore open the envelope. (FPM 2396)

In (26), we could of course describe tear with a valency pattern of the following type: [NP]subj...tearact...[AdjP]...[NP]

However, the only adjective that regularly occurs in the AdjP slot with tear is open, so we can analyse the pattern to be lexically specified and write it as follows: [NP]subj...tearact...[AdjP]:open...[NP]477

In the present study, the case of plain to see discussed in Section 6.3.1.1 and illustrated in (27) below can also be modelled in this way. (27)

Whatever the disciplinary practices across the generations, it is plain to see that without a reasonable amount of co-operation and compliance from the child, parents wouldn't get far in fulfilling their objectives. (B10 23)

The corresponding pattern for the relevant clause would look as follows: [it]...plain...[to_INF]:see

Here, the [that_CL] would be treated as a complement of see in the [to_INF] clause. Again, it may be preferable to treat tear open or plain to see as units of meaning and not as compositional, i.e. as single choices in Sinclair’s idiom principle view of language,478 even though their internal structure is of course transparent. In this case, one may prefer the analysis of the [that_CL] in (27) as a complement of the complex valency carrier plain to see: [it]...plain to see…[that_CL]

|| 477 This pattern is very similar to the pattern for which the Patternbank identifies an idiomatic phrasal verb tear apart, so one could also justify an analysis of tear open in such terms. 478 “The principle of idiom is that a language user has available to him or her a large number of semi-preconstructed phrases that constitute single choices, even though they might appear to be analysable into segments.” (Sinclair 1991: 110)

306 | Conclusions and perspectives for syntactic models

To sum up, we were able to show that a valency pattern model can, with minor additions, account for phenomena that are more idiosyncratic than the categorial information usually indicated by valency. If we add such specifications to our valency description it also becomes clear that item-specificity at the level of grammar and at the level of lexis cannot be fully separated, so that such a model moves closer to construction grammar or similar approaches, where a lexicogrammatical continuum is assumed (see Herbst 2014 for a discussion of the relationship between construction grammar and valency grammar; see also the discussion in Section 9.3.3 below).

9.3.2.2 Long-distance dependencies and multiple valency carriers The term long-distance dependency is a cover-term for a range of phenomena (see Falk 2006b: 316) of which only tough movement is of interest here. It will, however, be argued below that it might also make sense to include a certain passive construction that is not usually treated under this heading. According to Crystal’s definition the term479 describes “a CONSTRUCTION in which a SYNTACTIC RELATION holds between two CONSTITUENTS such that there is no restriction on the structural distance between them (e.g. a restriction which would require that both be constituents of the same CLAUSE)” (Crystal 2008: 501; his emphasis). The following two sub-sections aim to discuss what a valency model may look like in order to account for such structures.480

9.3.2.2.1 Tough movement As shown in Section 7.4, there are good reasons to analyse the subject this route in the following sentence as governed by the verb follow (for which it would have to be analysed as a postverbal complement and not a subject): (28)

In bad weather even this route is tough to follow. (CMD 254)

This becomes clearer if we compare it to the standard extraposed version: (29)

It is tough to follow this route.

|| 479 Crystal’s definition describes the synonymous term unbounded dependency. 480 The representation is also largely compatible with the one given by Matthews for some such structures (Matthews 1981: 185f).

Perspectives for grammatical models | 307

Here, we can analyse tough as a valency carrier that licenses a [to_INF] extraposed subject in which this route is a postverbal complement (traditionally called object) of follow. We can illustrate this in a syntactic representation where the strong arrow shows the valency relation between the two:481 clause

dummy subject

constituent structure

predicate

predicator predicative element

it

is

tough

extraposed subject

to follow this route

valency structure

Fig. 32: Syntactic representation of (29)482

If we come back to the version that shows tough movement, the classic analysis is to keep the relationships as they are even though the ‘postverbal’ complement of the subordinate clause occurs in canonical subject position of the main clause:

|| 481 See Section 6.4.2 for a discussion of the valency relations in copular clauses. 482 The representation is based on Figure 19 on page 174 and is in principle compatible with the model proposed by Herbst/Schüller (2008). The constituent structure is also largely compatible with CamG’s treatment. Only the relevant valency relations are shown since for our discussion it does not matter whether the verb follow is a dependent of the infinitive marker and whether this is a dependent of route or the opposite. The dummy subject it is not governed by a valency carrier in the clause and is only supplied for syntactic well-formedness.

308 | Conclusions and perspectives for syntactic models clause

tough-moved subject

constituent structure

predicate

predicator predicative remainder of extraposed subject element

this route is

tough

to follow

valency structure

Fig. 33: Syntactic representation of the relevant part of (28)483

What makes this phenomenon particularly difficult to model is that the socalled ‘movement’ is not restricted locally, so the ‘moved’ element need not occur in the clause one level above the subordinate clause from which it is said to be moved.484 Let us illustrate this with an example: (30)

In parts of Egypt, Bangladesh, Indonesia, China, and other low-lying coastal areas already suffering from poor drainage, agriculture is like-

|| 483 The treatment of (28) is in fact similar to the approach taken by lexical functional grammar (LFG). In LFG, a sentence is modelled with two structures at the same time, a functional structure (‘f-structure’) and a constituent structure (‘c-structure’). At f-structure, the multifunctionality of elements in LDD [=long distance dependency] constructions is represented directly. [...] Such a representation differs from other theories, particularly transformational theory, in not conceptualizing LDD constructions as involving the displacement of an element from its natural position. Instead, the element has two functions, and the c-structural position that it occupies is the canonical position of one of those two functions. (Falk 2006b: 321) 484 In fact, in generative grammar, the item is moved up in the grammatical structure via a series of smaller movements now and thus does not take the direct route: The issue can be formulated in a more theory-neutral way by asking whether the relationship between a gap and its filler is a direct one, or is instead mediated by intervening material. This was a hotly debated topic within generative grammar in the 1970s (sometimes labeled the ‘swooping vs. looping’ controversy). A real measure of progress in the field is that this debate has been definitely settled in favor of ‘looping.’ All generative grammarians now recognize that long-distance filler-gap dependencies are mediated by the intervening material. (Wasow 2001: 309) Since the evidence for the progress can be found in languages that are typologically different from English, the discussion is more or less irrelevant for the model in the present study, so we shall prefer the simpler direct route.

Perspectives for grammatical models | 309

ly to become increasingly difficult to sustain. (www.twnside.org.sg/title2/resurgence/216/cover2.doc) Again, we could analyse agriculture as a postverbal complement of sustain, which leads to a long-distance dependency over multiple levels of embedding:485

Fig. 34: Dependency representation of the relevant part of (30)486

The phenomena discussed so far were described in great detail in early generative grammar in terms of a transformation or movement analysis (e.g. Chomsky 1973, Lasnik/Fiengo 1974). If we allow valency to be at work across the boundaries of the clause in which the valency carrier occurs, we can account for such structures in a valency approach, too. As shown in Section 7.4, however, the subject of a sentence that appears to exhibit tough movement may or may not be a complement licensed by the verb in the extraposed [to_INF] subject. Thus our model should be able to account for such behaviour. Let us examine two examples discussed in Section 7.4 again: (31)

That this may not be the case in certain instances does not take much imagination to comprehend. (HP3 929)

(32)

For this to happen within the past 12 months is hard to believe. (newspapers)

|| 485 This sentence fragment is analysed to contain two nested extraposed subjects, i.e. the clause with become as its main verb is an extraposed subject in the main clause and the clause with sustain as its main verb is an extraposed subject within the become clause. Bars under the constituents illustrate the level of embedding. Since the sentence is structurally ambiguous, different analyses are possible, so one could instead analyse the sustain clause as the extraposed subject of the main clause and the become clause as a standard post-adjectival complement of likely. 486 This and all following diagrams only contain the dependency structure since the constituent structure is not of interest here.

310 | Conclusions and perspectives for syntactic models

In (31) we find a [that_CL] subject that is most convincingly analysed as being licensed by comprehend, given that take does not usually allow [that_CL] subjects.487 In (32) we find a [for_NP_to_INF] subject that is best analysed as being licensed by hard, given that believe does not usually allow a [for_NP_to_INF] postverbal complement.488 Nonetheless, even in the second example, the semantic relationship between believe and the [for_NP_to_INF] corresponds to the relationship that believe has to [that_CL] postverbal complements it licenses in canonical clauses. It will be argued here that both structures can be modelled uniformly with an approach that allows an element to be dependent on more than one valency carrier at the same time. For a formal realisation to occur it is sufficient if one of the two valency carriers licenses this specific formal realisation. We can illustrate this graphically as follows (only the relevant relations are shown; the broken line indicates that the valency carrier does not license this formal realisation):

Fig. 35: Two possible valency carriers for the subject in (31)

Fig. 36: Two possible valency carriers for the subject in (32)

The advantage of such an approach is that with the broken line we can still model that there is a relationship between believe and the clausal subject even though the latter is not formally licensed by believe.

|| 487 Except, of course, in the pattern [that_CL]...takeact...[NP]...by surprise, but there again one would probably prefer to speak of a complex valency carrier, a contextual specification, or, at the very least, of a different lexical unit take. 488 We can see in the following example that hard does allow [for_NP_to_INF] subjects: For us to be reduced to the level of a Balkan state is very hard. (AE8 1066)

Perspectives for grammatical models | 311

If both valency carriers allow a certain complement, we can treat it as valency dependent of both at the same time as in the following sentence: (33)

To go out on penalties is hard to accept. (newspapers)

Fig. 37: Two possible valency carriers for the subject in (33)

To sum up, in the analysis suggested here, we find that the element that occurs as subject (i.e. pre-verbally, determines agreement, etc.) is not necessarily a valency complement of the verb of the main clause and, in the case of copular clauses such as the one given in Figure 33, possibly not even of the predicative adjective, which is one further indicator of the fact that the various properties of what we treat as subjects (e.g. governed by X, agrees with Y) need to be accounted for independently as outlined in Section 9.2.

9.3.2.2.2 Passives and small clauses Now that we have established that a valency model can allow for long-distance dependencies and that one element can occur as valency dependent of two valency carriers, we can apply a similar analysis to another phenomenon we have discussed in Chapter 5, i.e. the passives of sentences with a so-called small clause in postverbal position of the corresponding active clause. The [that_CL] in the following example could be analysed as subject of consider and of fact at the same time: (34) a.

That housing will decline is considered a fact like Isaac Newton’s laws. (housingpanic.blogspot.com)

So we might extend the analysis proposed for tough movement structures to these passives as well, which would result in the following diagram:

312 | Conclusions and perspectives for syntactic models

Fig. 38: Possible dependency relations for (34a)

The reasoning behind such an analysis is that the postverbal structure in the active clause given in (34b) can be analysed as a so-called small clause in which fact acts as a valency carrier and determines the form of the [that_CL] (see Section 5.2.2.2 for details). (34) b. He considered it a fact like Isaac Newton’s laws that housing will decline. Thus consider and fact share a complement in such an analysis. However, unlike for tough movement structures, it is not sufficient if one of the two valency carriers licenses the subject, as indicated by the unacceptability of the following sentence: (34) c.

*For housing to decline is considered a fact like Isaac Newton’s laws.

Given that consider can occur (even in parallel structures) with a [for_NP_to_INF] complement in subject position as long as the second valency carrier allows it, it appears as if the second valency carrier, the noun fact, restricts the possible form of the subject. Due to a lack of data (see the discussion in Section 5.2.2.2) it is impossible to state with any degree of certainty whether the verb, consider in this case, also restricts the subject in terms of valency. Therefore one could also opt for a model in which the subject is not governed by the verb of the main clause and is only a dependent of the postverbal valency carrier, which would result in a diagram of the following form:

Fig. 39: Possible dependency relation for (34a)

Such an analysis appears appealing mainly because the postverbal predication or small clause of the active clause is the only predication remaining in the passive clause, so that the passive verb could be understood as a sort of copula

Perspectives for grammatical models | 313

verb. Additionally, if the quantitative valency of consider in the active clause is determined as 2 (i.e. the postverbal structure is treated as a small clause and thus one complement only), this would mean that there is only one complement left in the passive clause, which is then split into subject position and postverbal position. Accordingly, the subject would not be a complement of the verb but still agree with it, which again is an argument in favour of a multi-level analysis of the type found in HPSG or proposed by Jacobs (1994) or Ágel (2000) (see also section 9.2.2). In a dependency approach, such structures could also be analysed with the help of the catena concept as discussed in detail by Osborne/ Putnam/Groß (2012) or Groß/Osborne (2013). A similar analysis can be proposed for sentences of the type shown in (35) and discussed in more detail in Section 5.2.2.2: (35)

Knocking two small reception rooms into one is virtually guaranteed to create value, as is enlarging a family kitchen so they can eat in it. (newspapers)

Here, the passive verb appears to function in a way similar to semi-auxiliaries, so we would expect the verb create in the postverbal [to_INF] clause to determine the form of the subject as well. In conclusion we can state that there is definitely a long-distance dependency relation between a postverbal valency carrier and the subject in both types of structures. As shown in Section 5.2.2.2, the postverbal valency carrier in the consider type of structures can also occur within a [to_INF] clause with be or within a prepositional phrase with as, which increases the distance between this valency carrier and the subject even more. While the treatment shown in Figure 38 is a possible analysis, we shall treat the simpler version given in Figure 39 as more appropriate as long as there is no counter-evidence. We have thus shown that valency models can account for some problematic cases if we allow valency relations to work over longer distances in the syntactic structure. More research is needed to determine how to model restrictions on long-distance dependencies (such as so-called islands [Ross 1967]) in a valency model.

9.3.2.3 Limitations There are, however, constructions which can less easily be reconciled with a valency analysis. The most important case in the context of the present study is that of existentials. It is of course not impossible to construct a valency pattern representation with some sort of lexical or contextual specification and a corre-

314 | Conclusions and perspectives for syntactic models

sponding sentence type that covers the special agreement patterns, but it has been argued here (in Section 8.4.4) that valency is not the model best suited for describing such structures. Instead, valency is most profitably regarded as a useful and powerful mechanism for the description and theoretical modelling of complementation patterns in many sorts of sentences, but it may not make sense to apply it to all sentences, even if these contain a verb and some other elements around it. Any model that allows for the storage of larger units of meaning (e.g. construction grammar; see Goldberg 1995, 2006) that are not necessarily fixed will be able to account for the idiosyncrasies of structures such as existentials or the lexical and contextual specifications we have seen in the present chapter relatively easily. In the long run, it may make sense to incorporate valency and construction grammar into one approach, and Herbst (2014) presents a model of how these two may be merged and interact to form a more powerful and flexible grammatical model. After all, it will be maintained here that there is no such thing as valency in the language – what we do is describe language in terms of valency, which is an entirely different thing. In addition, valency is of course not a model of grammar as a whole but only of the aspect of complementation and has to be embedded in a larger framework anyway, so constructional approaches (many of which have not yet been extended to a full grammatical model either, however) seem the perfect candidates due to the many similar fundamental assumptions (usage-based, no transformations or deep structures, important role played by storage).

9.3.3 The bigger picture As mentioned in the previous section, valency could be regarded as one component489 of a larger grammatical model. The reason for stressing this is that valency is not necessarily the only determinant of the syntactic structure of a sentence and that there are various other factors that influence what exactly a given sentence looks like, sometimes working against valency.

|| 489 The term component is used in a theory-neutral way here. No claims as to the modular nature of a model of language will be made.

Perspectives for grammatical models | 315

9.3.3.1 Conflicting determinants Let us for instance recall the example of illegal discussed in Section 9.1. We found that illegal followed by an extraposed [to_INF] (or [for_NP_to_INF]/ [of_NP_to_INF]) subject occurred more than 2,000 times in the 1.5 billion word parsed corpora, whereas there was no instance of it followed by a [that_CL] in the 780 million word corpora used for Section 6.3.1.2. We concluded that the structure is not acceptable. However, there is one example in the parsed corpora, where one may want to analyse illegal as occurring in this construction: (36) a.

It is illegal, degrading and inhuman that prisoners in Ireland must carry out slops from their toilet cells each morning. (newspapers)

In fact, this example does not sound particularly problematic to native speakers’ ears, unlike the following version, in which degrading and inhuman was removed: (36) b. *It is illegal that prisoners in Ireland must carry out slops from their toilet cells each morning. We can thus observe that it must be the coordination of illegal with degrading and inhuman in (36a) that renders its use with an extraposed [that_CL] acceptable, even though a valency description of illegal would not contain an extraposed [that_CL] due to the unacceptability of (36b) and the complete absence of other corpus evidence in our large corpora. Thus, in this case, the coordination ‘wins’ over the valency of illegal. However, we cannot say that coordination always takes precedence over lexical valency, but it is important to note that valency is not necessarily stable, although it is probably relatively fixed for illegal. If all constructional variants are relatively rare, we would expect valency to be less stable. This is in line with the position taken by Stefanowitsch (2007: 68), who finds that say is very stable in not allowing ‘ditransitive’ complementation whereas donate and explain are less so: There are conventions that are strong enough to be near-insurmountable, such as [Det N] or the fact that, for example, say does not occur ditransitively under any circumstances that I have been able to determine (cf. Stefanowitsch 2006). However, there are also conventions that are weak enough for speakers to discard them at the first opportunity that presents itself. A linguistic theory should be able to deal with both extremes and with everything in between [...]. (Stefanowitsch 2007: 68)

In order for a linguistic theory to be able to deal with such cases, it must take other factors influencing the syntactic structure into consideration, too.

316 | Conclusions and perspectives for syntactic models

For instance, we also find conflicting factors in the case of extraposition. In Chapter 7, we discussed the different relevant factors, two of which were information structure and weight. If a subject is given, it is more likely to be in canonical position; if it is heavy, it is more likely to be extraposed. In the following examples, the subjects are both heavy and given (as determined from the respective contexts in which they occur): (37) a.

That it was division of labour which gave modern humanity its edge over the Neanderthals is not a completely new idea. (newspapers) b. It is no new idea that RNA copies of genes can be copied back again into DNA and reinserted into new positions. (B7N 888)

Nonetheless, in (37a) givenness seems to take precedence whereas in (37b) weight seems to be the decisive factor. We cannot identify any reason for the preference of one over the other in many such cases apart from the speaker’s choice, but it would be difficult to have these sentences predicted exactly and only in this form by a grammatical model. Textual organisation also plays a role; as it makes sense to have the following relatively close to what it refers to, extraposition is preferable in the following example: (38)

It is essential that any topic should meet all of the following: (HPD 1544)

Countless other factors are relevant to the grammatical structure of English sentences,490 and a model that tries to mechanically predict the structure of sentences, for instance through the order of modules in the grammatical architecture (as some generative models do), is likely to fail in one of the two cases given in (37a–b) above. But even network models such as the one advocated below will have to accept that speakers have a choice: For a speaker in a particular situation, givenness may seem more important than weight (subconsciously, that is), for another speaker, it may be the opposite. And even the same speaker may use the alternative structure on another occasion. We thus have to agree with Gries: Also, it is highly unlikely that we will ever be able to predict native speakers’ behaviour completely flawlessly irrespective of the number of variables we might still want to include in the analysis. (Gries 2003: 16)

|| 490 A whole range of factors is discussed in detail in the volume Determinants of Grammatical Variation in English edited by Rohdenburg/Mondorf (2003).

Perspectives for grammatical models | 317

Besides the obvious point that there is variation across native speakers we cannot even satisfactorily account for the variation of one and the same native speaker if we do not introduce a non-deterministic element that allows for the speaker’s choice into our model.

9.3.3.2 A cognitive approach The model advocated here corresponds to a representation used by Bybee (1995, 2007: 323–326) for the storage of linguistic forms. One of the premisses of the model is that a large amount of storage takes place in the acquisition of a language, which is consistent with present-day usage-based theory: Usage based accounts of acquisition assume that learning takes place by generalising over concrete usage events (see Tomasello 2003 for a summary). They do not draw a distinction between universal and innate core grammar, which is acquired by deduction, and the periphery, which has to be learnt by induction. Instead, it is supposed that all properties of languages can be acquired from the input by powerful generalization abilities in connection with social cognition. (Behrens 2007: 201)

Thus input leads to storage, and stored items are not stored in isolation but are connected to existing stored items and structures and thus form a network with them. Bybee illustrates the relationships with two diagrams that represent two views onto the network, i.e. both diagrams should be understood as snippets taken out of the huge network that is our mental representation of language:

Fig. 40: Simplified possible organization of NPs (adapted from Bybee 2007: 324)

318 | Conclusions and perspectives for syntactic models

Fig. 41: Simplified possible organization of NPs (adapted from Bybee 2007: 325)

Speakers can then access the network and can retrieve individual, stored instances of language or abstractions over these instances that emerge from the network. In Figure 40, common nouns emerge from stored usage; in Figure 41, determiners emerge. According to Bybee, who follows Langacker (1987) in this respect, “schemas are formed at various levels of abstraction” (Bybee 2007: 325). She gives an example of what such levels of abstraction may look like for English NPs based on the figures above: 1. Very specific: my mother, my computer, the car, a problem, an idea 2. Partially general: [my + NOUN], [POSS PRO + mother] 3. More general: [POSSESSIVE + NOUN] 4. Fully general: [DETERMINER + NOUN] (Bybee 2007: 325)

The main appeal of such a model is that it allows us to treat item-specificity and regularity using the same mechanisms and that there is a very strong focus on idiosyncratic language use: Language is a mixture of regularity and idiosyncrasy. By training and inclination, linguists are better equipped to deal with the former than the latter, with the consequence that far more effort goes into the formulation of general rules than into the patient elucidation of their limitations (cf. Gross 1979). The notion of a usage-based model represents an attempt to redress this imbalance, and to overcome the problems it engenders. The central claim is that a reductive account of grammatical constructions is unworkable: a speaker’s conventional knowledge of a construction is not given by any single structure (such as a prototype or high-level schema). Its cognitive representation is more adequately treated as a full schematic network, where specific structures co-occur with categorizing schemas extracted to describe their commonality at various levels of abstraction. (Langacker 1987: 411)

Perspectives for grammatical models | 319

We can now apply the model to structures we have looked at in the present study. It follows from what we have just discussed that similar structures are connected by more links, dissimilar structures by fewer or none. In Figure 42 below, there are sentence fragments with an extraposed [that_CL] subject with the adjective important, with an extraposed [to_INF] subject also with the adjective important and with an extraposed [to_INF] subject with the adjective illegal. 491 To keep the diagram at least partially readable, only two types of connections are shown:492 the connections between identical valency carriers (dotted line) and the connections between identical first words (possibly heads) of the clausal subjects, i.e. to or that, (dashed line). We can see that the sentences in which important occurs with a [to_INF] extraposed subject are densely connected to both other groups since they share the form of the subject with one group and the valency carrier with the other. The speaker is said to abstract over individual instances along dense paths, which in this figure would not lead the speaker to produce a sentence with illegal and a [that_CL] extraposed subject. This does of course not mean that it is impossible to produce such sentences; the opposite is the case, as we have shown with the invented example in Section 9.1, but such a use is not conventionalised and thus much less likely to be produced subconsciously. In this way we can understand valency as emergent from such networks of stored and interconnected, conventionalised language use.493 Let us reconsider the problem of obligatory extraposition and the subject status of as if-clauses discussed in Section 7.1.2. We found that there is no agreement among theorists as to the treatment of sentences such as the following:

|| 491 All sentence fragments are taken from actual corpus evidence from the BNC. 492 In the fully detailed version of the diagram, all sentences are connected because they share it and a form of be. 493 As of now, many questions as to the nature of the mental representation of such structures are still open, and it may well turn out that the network model has no neural correlates (see Ullman et al. 2005 for neurological evidence in favour of a model that separates lexicon and grammar; see also MacWhinney’s 2005 commentary for an opposing view), but experiments on morphology (Bybee 1995) and phonology (Bybee/Scheibman 1999) have shown that the model shows a behaviour similar to that of humans.

320 | Conclusions and perspectives for syntactic models

Fig. 42: Aspects of a network representation of extraposed sentences

Perspectives for grammatical models | 321

(39) a.

It is perfectly clear that Charles and Diana are less and less comfortable in each other’s company. (CEN 3385) b. It appears that Charles and Diana are less and less comfortable in each other’s company. c. It appears as if Charles and Diana are less and less comfortable in each other’s company.

For (39a), all models discussed (e.g. CamG, CGEL and Herbst/Schüller 2008) agree that it should be treated in terms of extraposition. For (39b), CamG rejects an extraposition analysis because there is no non-extraposed counterpart whereas CGEL and the VDE treat it as an instance of obligatory extraposition. Since as if-clauses as in (39c) can never occur in canonical subject position, they are not treated as subjects by CamG, and CGEL also treats them under the heading of complementation by an adjunct (CGEL 1174f); the VDE, however, treats them as extraposed subjects. Figure 43 tries to illustrate that all three positions are possible abstractions from a network model, depending on how far one would like to take generalisations.494 The figure represents not a network of actual exemplars of language use but of abstractions over these exemplars. The only reason for using this form of representation is readability, so one could imagine a set of densely connected instances instead for each construction. The size of these smaller-scale networks (i.e. the frequency of the construction) is represented by font size. The dashed lines between the instances of clear, obvious and helps together with the dashed/dotted lines between the instances of is and some of the dashed lines between the instances of [that_CL]s form a comparatively dense network that represents the relationship between extraposed and non-extraposed variants.495

|| 494 Larger font size indicates higher frequency, but the font size is not directly proportional to frequency. For readability, only third person present tense verb forms are given, but the constructions were searched with a lemma in the corpus. Actual frequencies with relevant queries for the BNC (100 million words): it {be/V} clear that: 1806; it {be/V} obvious that: 473; it {help/V} that: 18; it {seem/V} that: 1863; it {appear/V} that: 936; it {seem/V} as if: 221; it {appear/V} as if: 13; it {look/V} as if: 515. The non-extraposed variants were searched in the parsed corpora (1.5 billion words): clear –csubj--> [any word] –complm--> that: 102; obvious: –csubj--> [any word] –complm--> that: 93; help% –csubj--> [any word] –complm--> that: 46. As can be seen from the queries, all numbers are very simple approximations in that the BNC queries do not account for negation, modal verbs and the like, and the queries in the parsed corpora may include other copula verbs or irrelevant structures due to parsing errors. 495 It must be made very clear that due to the fact that only a small portion of the network is shown, some items appear to be relatively isolated, but due to the high frequency of, for in-

322 | Conclusions and perspectives for syntactic models

This is the generalisation CamG makes. The solid connections between different instances of it and the dashed ones between [that_CL]s create an additional link to so-called obligatory extraposition in the sense of CGEL, i.e. to it seems that and it appears that. Together with the part mentioned above, we can thus cover what CGEL calls subject.496 The wider use in the VDE can be captured if we add the dashed lines between instances of seems and appears, and the dotted lines between [as_if_CL]s to the solid connections between instances of it. We can conclude that all three positions taken by the different grammatical models are possible generalisations over the instances in the network. The differences between the models stem from the criteria they apply for their decision as to where the limit of the generalisation is to be found, but all models are compatible with a network view. There are some further properties of such network models related to frequency that will be discussed here. Bybee (1995) shows that type frequency and token frequency have very distinct influences on the predictions the model makes. A high token frequency leads to a strong entrenchment of the relevant structure, so even structures with few connections to other structures are very stable in such a model. Also, high token frequency seems to lead to a certain independence and separate storage as illustrated by the case of the phonological reduction of I don’t compared to less frequent structures such as we don’t discussed in Bybee/Scheibman (1999). As we have seen in Chapter 8, existential structures show some very distinct grammatical properties. We can thus imagine existentials in terms of such a comparatively independent and separate tightly-knit network. Of course existentials are bound to have connections to other sentences with a copula verb and with a noun phrase,497 but since they are considerably more frequent than I

|| stance, help, the non-extraposed version will be densely connected to other instances of help without a clausal subject and form a large network with them. 496 Reality is of course more complex since there are sentences with seem and appear that have a non-extraposed version if the verbs are used in certain divalent patterns (see discussion in Section 7.1.2). 497 For existential there we would have to assume separate storage due to its reduced phonological form, which does of course not exclude connections to locative there.

Perspectives for grammatical models | 323

Fig. 43: Non-extraposed and extraposed [that_CL] subjects together with the analogous structure with an [as if_CL] in a network representation.

324 | Conclusions and perspectives for syntactic models

don’t we would expect them to be stored separately, to be deeply entrenched and accordingly to show different features in the model.498 On the other hand, a high type frequency leads to the generation of generalisations which, if they are strong and relatively abstract, could be understood as rules. For morphology, Bybee (1995) shows that if there are many instances of a certain type of past tense formation, this type comes to be interpreted as a rule and is generalised by analogy to novel verbs. Similarly, if we regard valency as emergent from the network around specific lexical items (see Herbst’s valency realisation principle [Herbst 2011, 2014]) we can regard argument structure constructions (Goldberg 1995) as emergent over structures across a range of lexical items, as a generalization at a higher level of abstraction. Accordingly, if we have many valency carriers that can be used in extraposed and in nonextraposed sentences, the extraposition rule can emerge (such a ‘rule’ need not be represented as such but may simply be an ad-hoc abstraction that leads to analogy) that allows us to create a non-extraposed version of a sentence with a valency carrier that we have only heard used in sentences with extraposition. Thus even without explicit knowledge of grammatical rules, speakers will be able to recognise regularities that emerge from their stored networks of the language use they experienced, as long as a sufficient number of types show identical behaviour. In the context of his discussion of grammaticality cited in Section 3.3, Sampson uses a catchy metaphor to describe a network model that seems to be largely compatible with the view presented here: [T]he grammatical possibilities of a language are like a network of paths in open grassland. There are a number of heavily used, wide and well-beaten tracks. Other, less popular routes are narrower, and the variation extends smoothly down to routes used only very occasionally, which are barely distinguishable furrows or, if they are used rarely enough, perhaps not even visible as permanent marks in the grass; but there are no fences anywhere preventing any particular route being used, and there is no sharp discontinuity akin to the contrast between metalled roads and foot-made paths – the widest highway is only the result of people going that way much more often and in far greater numbers than in the case of narrow paths. (Sampson 2007: 10f)

|| 498 In the BNC there are about 200,000 instances of there_EX0 + {be/V} (existential there followed by a form of be) and more than 52,000 of I followed by a form of do followed by a form of not, which is a very generous way of counting I don’t (since it includes I did not and others). If we limit our search to the spoken part of the BNC, there are still about 50 % more existentials than instances of our generously counted I don’t.

Perspectives for grammatical models | 325

However, Sampson’s account does not explain why we would want to exclude illegal with an extraposed [that_CL] subject (as discussed above and in Section 9.1) from our model description of English clausal subjects.499 But we can show how this can be done if we use Stefanowitsch’s (2006) method of accounting for negative evidence with the help of frequency, which is compatible with our network representation.500 In Figure 44, we shall consider only one aspect of the network, frequency. Every dot represents one instance of use.501 No connecting lines between items are necessary, even though one should imagine several connecting lines from every dot to every other dot in the same box, fewer connecting lines from every dot in a box to every dot in the adjacent box and even fewer lines (for it and possibly the copula verb) from every dot in the image to every other dot in the image. In the boxes for extraposed [that_CL] and [to_INF] subjects with important, we can observe that both occur very frequently, so what we would find is a densely connected bifocal network. Both structures can be easily accessed since they are strongly entrenched. In the case of selfish, only four instances with a [that_CL] extraposed subject were found in the parsed corpora, compared to roughly 85 with a [to_INF]502 clause. If we consider the figures on corpus size given in Section 9.1, this means that some native speakers possibly have never heard selfish constructed with a [that_CL] extraposed subject, whereas they probably have come across it with a [to_INF] a few times. So the [to_INF] examples could form a small network and the [that_CL] might or might not be present at all. However, an application of analogy with other adjectives may lead to the production of selfish with a [that_CL] if the need arises, basically because there is no evidence contradicting such a use. Actual evidence against the use of illegal with a [that_CL] does of course not exist either, but the high frequency of illegal with a [to_INF] (and other infinitival constructions) creates a very dense and – as we have seen above – more independent network of uses, which is a strong competitor for a [that_CL] use. If this way of expressing a clausal subject

|| 499 Sampson may not want to exclude it, though – see the discussion in Section 3.3. 500 As we have seen in Section 9.1, we cannot use Stefanowitsch’s calculation due to limitations on the corpus evidence, but we can show how his concept of significant absence (Stefanowitsch 2006: 62ff) can emerge from a network model representation. 501 In the corpora, frequency differences are much bigger than can be sensibly depicted in such a diagram, but since the idea here is to demonstrate how the mechanism works, this inaccurate representation should be sufficient. 502 This number excludes [for_NP_to_INF] and [of_NP_to_INF] and instances where coordination may have influenced the valency pattern.

326 | Conclusions and perspectives for syntactic models

Fig. 44: Schematic representation of the frequency distribution of extraposed subjects with three adjectives (not to scale)

with illegal is so highly frequent, it will be so strongly entrenched (and thus the combination of the infinitival constructions with illegal so strongly associated) that this route is chosen automatically in the network unless the language user deliberately deviates from the norm. Thus the high frequency of the alternative construction in a way pre-empts the non-conventional construction. In the case of selfish, the much lower frequency of occurrence of extraposed [to_INF] subjects is not sufficient to pre-empt a [that_CL] use as strongly as in the case of illegal. Thus with a few simple assumptions we can make the network model predict valency restrictions based on corpus data, provided that the data available is rather large scale. We can also conclude that the notion of subject in this view of the language is emergent just like all other concepts, “rather than given a priori or by design” (Bybee 2010: 2). So basically what exactly counts as a subject is once again a matter of how far one would like to extend the generalisations possible in the model. In the present study, a case was made against using the term with exis-

Summary | 327

tentials since existential there shares fewer properties with other subjects and thus shows less integration with other subjects in a network graph.503

9.4 Summary In this chapter we were able to show that the relationship between corpus data and native speaker data is problematic and that once a certain threshold of corpus size is reached, the corpus may permit a much better description of the language use of native speakers than native speaker intuition, even for rare phenomena. Furthermore we showed that we were able to confirm our hypotheses presented at the beginning of the present study: Subjects are indeed valency complements and are governed by valency carriers in the clause, no matter whether the valency carrier is an active verb, a passive verb, an adjective or a noun because they are formally restricted by the valency carrier. Extraposition of subjects is also item-specific and depends not only on the valency carrier but also on other complements present in the clause. The idiosyncratic nature of existentials led us to conclude that the term subject is not as ideally applicable to existentials as to canonical sentences and that it may make sense not to give any element in an existential the name subject at all. Since criteria for subject status do not necessarily coincide, a model that attempts to describe all aspects of their grammatical behaviour will have to spell out the individual properties independently, as for instance suggested by frameworks such as HPSG (Pollard/Sag 1994). As for valency theory we were able to show that valency can only be satisfactorily described if we give up the notion of complement inventories and go for patterns instead, which is in line with Herbst’s (2007) conclusions on the basis of postverbal complements. These patterns need a semantic representation as in Herbst/Schüller’s (2008) valency constructions, which include a participant pattern in addition to the valency pattern. Valency needs to be described for active and passive clauses independently (or at least a model must allow for such descriptions). Long-distance dependencies need to be accounted for by not restricting valency patterns to the clause in which the valency carrier occurs and by allowing a dependent to be governed by more than one valency

|| 503 As mentioned before, such a model does not per se restrict the kinds of generalisations people can make, so the decision to regard existentials as a separate construction is of course arbitrary to some extent, too.

328 | Conclusions and perspectives for syntactic models

carrier. Given that the criteria for complement status do not necessarily coincide, it is also preferable to describe the properties of such elements individually in a model similar to that presented by Jacobs (1994) or Ágel (2000) for German. Valency is, however, only a descriptive device that is unlikely to correspond to language organisation in the mind. Bybee (1995, 2007) proposed a network model of language that performed well in the prediction of morphological and syntactic rules and the limitations of these rules. This model appears to be wellsuited for the sort of evidence we found in relation to subjects. It also allows for the individual behaviour of language items to occur side by side with regularity that arises from generalisations over items in the network. This behaviour also makes it compatible with other cognitive linguistics approaches such as construction grammar in the sense of Goldberg’s (2006) revised model.504 Of course there are still many open questions as to the exact nature of such a network, the most pressing of which is probably where and why generalisations stop, i.e. why it is not acceptable to construct sentences such as, say, *I want that he leaves. Using Stefanowitsch’s (2006) model of negative entrenchment we have been able to show that frequency seems to play an important role, but it does not seems to be the only factor at work (see the quote from Stefanowitsch 2007 at the beginning of Section 9.3.3). All in all, what the present study showed is that item-specificity plays an extremely important role in language use. What is striking is not so much that these idiosyncrasies exist – exceptions are said to confirm the rule, after all – but the scale at which these phenomena occur. Their high frequency suggests that the ‘exceptions’ may in fact be the rule. The scope of ‘rules’ (such as the prediction of the syntactic form of a complement from a semantic representation) appears to be much more limited than many linguists may have thought – or may still think if they do not work with corpora systematically.

|| 504 The original model (Goldberg 1995) did not allow for storage of constructions that were predictable on the basis of smaller constructions (e.g. regular past tense forms) whereas the revised version accounts for the counter-evidence presented by Bybee (1995, 2007) and others.

References Aarts, B. 1992. Small Clauses in English: The Nonverbal Types. Berlin: Mouton de Gruyter. Aarts, B. 2000. Corpus Linguistics, Chomsky and Fuzzy Tree Fragments. In Corpus Linguistics and Linguistic Theory: Papers from the Twentieth International Conference on English Language Research on Computerized Corpora (ICAME 20) Freiburg im Breisgau 1999, ed. C. Mair and M. Hundt. Amsterdam: Rodopi, 5–13. Aarts, B. 2011. Oxford Modern English Grammar. Oxford: Oxford University Press. Aarts, F. 1987. Dutch Progress in English Syntax: Zandvoort’s Handbook of English Grammar and after. In One hundred years of English Studies in Dutch universities, ed. G. H. V. Bunt, E. S. Kooper, J. L. Mackenzie, and D. R. M. Wilkinson. Amsterdam: Rodopi, 67–79. Aarts, F., and J. Aarts. 1982. English Syntactic Structures: Functions and Categories in Sentence Analysis. Oxford: Pergamon. Abney, S. P. 1987. The English Noun Phrase in its Sentential Aspect. PhD Thesis: Massachusetts Institute of Technology. Ágel, V. 1993. Ist die Dependenzgrammatik wirklich am Ende? Valenzrealisierungsebenen, Kongruenz, Subjekt und die Grenzen des syntaktischen Valenzmodells. Zeitschrift für germanistische Linguistik 21:20–70. Ágel, V. 2000. Valenztheorie. Tübingen: Narr. Albert, R., and N. Marx. 2010. Empirisches Arbeiten in Linguistik und Sprachlehrforschung. Tübingen: Narr. Alexiadou, A., and E. Anagnostopoulou. 2001. The Subject-in-Situ Generalization and the Role of Case in Driving Computations. Linguistic Inquiry 32, no. 2: 193–231. Alexiadou, A., and E. Anagnostopoulou. 2007. The Subject-In-Situ Generalization Revisited. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from SyntaxSemantics, ed. U. Sauerland and H.-M. Gärtner. Berlin/New York: Mouton de Gruyter, 31– 59. Allerton, D. J. 1975. Deletion and Proform Reduction. Journal of Linguistics 11, no. 2:213–37. Allerton, D. J. 1982. Valency and the English Verb. London: Academic Press. Anderson, J. M. 1997. Preliminaries to a History of Sentential Subjects in English. Studia Anglica Posnaniensia: An International Review of English Studies 31:21–28. Aristotle. 1924. Aristotle’s Metaphysics: a revised text with introduction and commentary. Trans. W. D. Ross. Oxford: Clarendon Press. Aristotle. 1989. Aristotle in 23 Volumes. Vols. 17, 18. Trans. H. Tredennick. Cambridge, MA: Harvard University Press. (Orig. pub. 1933 by William Heinemann, London.) Aristotle. 1994. Metaphysics: Books Z And H. Translated with a Commentary by D. Bostock. Oxford: Clarendon Press. Askedal, J. O. 2003. Das Valenz- und Dependenzkonzept bei Lucien Tesnière. In Dependency and Valency / Dependenz und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung, ed. V. Ágel, L. M. Eichinger, H.-W. Eroms, P. Hellwig, H. J. Heringer, and H. Lobin. Handbücher zur Sprach- und Kommunikationswissenschaft 25.1. Berlin: Walter de Gruyter, 80–99. Aston, G., and L. Burnard. 1998. The BNC Handbook: Exploring the British National Corpus with SARA. Edinburgh: Edinburgh University Press. Atkins, S., J. Clear, and N. Ostler. 1992. Corpus Design Criteria. Literary and Linguistic Computing 7, no. 1:1–16. https://doi.org/10.1515/9783110589801-349

330 | References

Bakker, D., and A. Siewierska. 2007. Another Take on the Notion Subject. In Structuralfunctional Studies in English grammar: In Honor of Lachlan Mackenzie, ed. M. Hannay and G. J. Steen. Amsterdam: Benjamins, 141–58. Behaghel, O. 1909. Beziehungen zwischen Umfang und Reihenfolge von Satzgliedern. Indogermanische Forschungen 25:110–42. Behrens, H. 2007. The acquisition of argument structure. In Valency: Theoretical, Descriptive and Cognitive Issues, ed. T. Herbst and K. Götz-Votteler. Berlin: Mouton de Gruyter, 193– 214. Berndt, R. 1991. Fact or not Fact - that is the Question in the Semantic Interpretation of Gerundive Nominals. In Languages in Contact and Contrast: Essays in Contact Linguistics, ed. V. Ivir and D. Kalogjera. Berlin: Mouton de Gruyter, 41–59. Bernstein, J. 2001. The DP Hypothesis: Identifying Clausal Properties in the Nominal Domain. In The Handbook of Contemporary Syntactic Theory. Malden, MA: Blackwell, 536–61. Biber, D. 1993. Representativeness in Corpus Design. Literary and Linguistic Computing 8, no. 4:243–57. Biber, D., S. Johansson, G. Leech, S. Conrad, and E. Finegan. 1999. Longman Grammar of Spoken and Written English. Harlow: Longman. [LGSWE] Bloomfield, L. 1935. Language. London: George Allen & Unwin. (Orig. pub. 1933.) Boas, H. C., and I. A. Sag. 2012. Sign-Based Construction Grammar. Stanford, CA: CSLI. Boeckx, C., ed. 2006. Agreement systems. Amsterdam: Benjamins. Bolinger, D. 1977. Meaning and Form. London: Longman. Breivik, L. E. 1983. Existential ‘there’: A Synchronic and Diachronic Study. Bergen: University of Bergen. Breivik, L. E., and A. E. Martínez Insua. 2008. Grammaticalization, Subjectification and NonConcord in English Existential Sentences. English Studies 89, no. 3:351–62. Bresnan, J. 1982. Control and Complementation. Linguistic Inquiry 13, no. 3:343–434. Bresnan, J. 2001. Lexical-Functional Syntax. Oxford: Blackwell. Bresnan, J., and R. M. Kaplan. 1982. Lexical-Functional Grammar: A Formal System for Grammatical Representation. In The Mental Representation of Grammatical Relations, ed. J. Bresnan. Cambridge, MA: MIT Press, 173–281. Burnage, G., and G. Baguley. 1996. The British National Corpus. Library and Information Briefings 65:1–15. Bybee, J. 1985. Morphology: A Study of the Relation between Meaning and Form. Amsterdam: Benjamins. Bybee, J. 1995. Regular Morphology and the Lexicon. Language and Cognitive Processes 10, no. 5:425–55. Bybee, J. 2007. Frequency of Use and the Organization of Language. New York: Oxford University Press. Bybee, J. 2010. Language, Usage and Cognition. Cambridge: Cambridge University Press. Bybee, J., R. Perkins, and W. Pagliuca. 1994. The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. Chicago: University of Chicago Press. Bybee, J., J. Scheibman. 1999. The effect of usage on degrees of constituency: the reduction of don’t in English. Linguistics 37, no. 4:575–96. Cardinaletti, A. 1997. Subjects and clause structure. In The New Comparative Syntax, ed. L. Haegeman. London/New York: Longman, 33–63. Cardinaletti, A. 2004. Toward a Cartography of Subject Positions. In The Structure of CP and IP, ed. L. Rizzi. Oxford/New York: Oxford University Press, 115–165.

References | 331

Cardinaletti, A. 2014. Cross-linguistic variation in the syntax of subjects. In Linguistic Variation in the Minimalist Framework, ed. M. Carme Picallo. Oxford: Oxford University Press, 82– 107. Cer, D., M.-C. de Marneffe, D. Jurafsky, and C. D. Manning. 2010. Parsing to Stanford Dependencies: Trade-offs between Speed and Accuracy. In Proceedings of the 7th International Conference on Language Resources and Evaluation (LREC’10), ed. N. Calzlari, K. Choukri, B. Maegaard, J. Mariani, J. Odijk, S. Piperidis, M. Rosner, and D. Tapias. Valletta, Malta: ELRA, 1628–32. Chomsky, N. [1955] 1975. The Logical Structure of Linguistic Theory. New York: Plenum Press. Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, N. 1973. Conditions on Transformations. In A Festschrift for Morris Halle, ed. S. R. Anderson and P. Kiparsky. New York: Holt, Rinehart and Winston, 232–85. Chomsky, N. [1981] 1993. Lectures on Government and Binding: The Pisa Lectures. 7th ed. Berlin: Mouton de Gruyter. Chomsky, N. 1982. Some Concepts and the Consequences of the Theory of Government and Binding. Cambridge, MA: MIT Press. Chomsky, N. 1986. Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, N. 2004. The Generative Enterprise Revisited: Discussions with Riny Huybregts, Henk van Riemsdijk, Naoki Fukui and Mihoko Zushi. Berlin: Mouton de Gruyter. Clear, J. 1992. Corpus Sampling. In New Directions in English Language Corpora: Methodology, Results, Software Developments, ed. G. Leitner. Berlin: Mouton de Gruyter, 21–31. Coates, J. 1983. The Semantics of the Modal Auxiliaries. Beckenham: Croom Helm. Collins, P. 1992. Cleft Existentials in English. Language Sciences 14, no. 4:419–33. Contini-Morava, E., and Y. Tobin, eds. 2000. Between Grammar and Lexicon. Amsterdam: Benjamins. Coseriu, E. 1973. Probleme der strukturellen Semantik. Tübingen: Narr. Cowart, W. 1997. Experimental syntax. Thousand Oaks, CA: Sage. Croft, W. Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford/New York: Oxford University Press. Croft, W., and D. A. Cruse. 2004. Cognitive Linguistics. Cambridge: Cambridge University Press. Cruse, D. A. 1986. Lexical Semantics. Cambridge: Cambridge University Press. Crystal, D. 2003. The Cambridge Encyclopedia of the English Language. 2nd ed. Cambridge: Cambridge University Press. Crystal, D. 2008. A Dictionary of Linguistics and Phonetics. 6th ed. Malden, MA: Blackwell. Crystal, D. 2011. Internet Linguistics: A Student Guide. London: Routledge. Culicover, P. W., and R. S. Jackendoff. 2005. Simpler Syntax. Oxford: Oxford University Press. Culy, C. 1998. Statistical Distribution and the Grammatical/Ungrammatical Distinction. Grammars 1:1–13. Dąbrowska, E. 2004. Language, Mind and Brain: Some Psychological and Neurological Constraints on Theories of Grammar. Edinburgh: Edinburgh University Press. Dauses, A. 2000. Ökonomie und Kybernetik natürlicher Sprachen: Universelle Gesetze des Sprachwandels. Stuttgart: Steiner. Davies, A. 2003. The Native Speaker: Myth and Reality. Clevedon: Multilingual Matters.

332 | References

Davies, W. D., and S. Dubinsky. 2001. Remarks on Grammatical Functions in Transformational Syntax. In Objects and Other Subjects: Grammatical Functions, Functional Categories, and Configurationality, ed. W. D. Davies and S. Dubinsky. Dordrecht: Kluwer, 1–19. Di Sciullo, A. M., and E. Williams. 1987. On the Definition of Word. Cambridge, MA: MIT Press. Dik, S. C. 1983. Auxiliary and Copula be in a Functional Grammar of English. In Linguistic Categories: Auxiliaries and Related Puzzles. Vol. 2: The Scope, Order, and Distribution of English Auxiliary Verbs, ed. F. Heny and B. Richards. Dordrecht: Reidel, 121–43. Dowty, D. 1991. Thematic Proto-Roles and Argument Selection. Language 67, no. 3:547–619. Duffley, P. J. 2003. The Gerund and the to-Infinitive as Subject. Journal of English Linguistics 31, no. 4: 324–352. Egan, T. 2008. Non-finite Complementation: A Usage-based Study of Infinitive and -ing Clauses in English. Amsterdam: Rodopi. Eisenberg, P. 2006. Grundriss der deutschen Grammatik. Vol. 2: Der Satz. 3rd. ed. Stuttgart: Metzler. Emons, R. 1974. Valenzen englischer Prädikatsverben. Tübingen: Niemeyer. Emons, R. 1978. Valenzgrammatik für das Englische: Eine Einführung. Tübingen: Niemeyer. Engel, U. 2004. Deutsche Grammatik. Neubearbeitung. München: Iudicium. Erdmann, P. 1976. There: There Sentences in English: A Relational Study Based on a Corpus of Written Texts. München: Tuduv. Erdmann, P. 1987. It-Sätze im Englischen. Heidelberg: Winter. Erdmann, P. 1997. The for…to construction in English. Frankfurt am Main: Lang. Eroms, H.-W. 2000. Syntax der deutschen Sprache. Berlin: De Gruyter. Falk, Y. N. 2006a. Subjects and Universal Grammar: An Explanatory Theory. Cambridge: Cambridge University Press. Falk, Y. N. 2006b. Long-Distance Dependencies. In Encyclopedia of Language & Linguistics, ed. K. Brown. Oxford: Elsevier, 316–23. Faulhaber, S. 2011. Verb Valency Patterns: A Challenge for Semantics-Based Accounts. Berlin: De Gruyter Mouton. Fillmore, C. J. 1968. The Case for Case. In Universals in Linguistic Theory, ed. E. Bach and R. T. Harms. New York: Holt, Rinehart and Winston, 1–88. Fillmore, C. J. 1992. ‘Corpus linguistics’ or ‘Computer-aided armchair linguistics’. In Directions in Corpus Linguistics: Proceedings of Nobel Symposium 82. Stockholm, 4–8 August 1991, ed. J. Svartvik. Berlin: Mouton de Gruyter, 35–60. Fillmore, C. J. 2007. Valency issues in FrameNet. In Valency: Theoretical, Descriptive and Cognitive Issues, ed. T. Herbst and K. Götz-Votteler. Berlin: Mouton de Gruyter, 129–60. Fillmore, C. J., P. Kay, and M. C. O’Connor. 1988. Regularity and Idiomaticity in Grammatical Constructions: the Case of let alone. Language 64, no. 3:501–38. Fischer, K., and A. Stefanowitsch. 2006. Konstruktionsgrammatik: Ein Überblick. In Konstruktionsgrammatik: Von der Anwendung zur Theorie, ed. K. Fischer and A. Stefanowitsch. Tübingen: Stauffenburg, 3–17. Fischer, K. 1997. German English Verb Valency: A Contrastive Analysis. Tübingen: Narr. Francis, W. N. 1965. A Standard Corpus of Edited Present-Day American English. College English 26, no. 4:267–73. Freeze, R. 1992. Existentials and Other Locatives. Language 68, no. 3:553–95. Gabelentz, G. von der. 1869. Ideen zu einer vergleichenden Syntax: Wort- und Satzstellung. Zeitschrift für Völkerpsychologie und Sprachwissenschaft 6:376–84.

References | 333

Gabelentz, G. von der. 1891. Die Sprachwissenschaft: ihre Aufgaben, Methoden und bisherigen Ergebnisse. Leipzig: Weigel. Garside, R. 1987. The CLAWS Word-tagging System. In The Computational Analysis of English: A Corpus-based Approach, ed. R. Garside, G. Leech, and G. Sampson. London: Longman, 30–41. Garside, R. 1996. The Robust Tagging of Unrestricted Text: The BNC Experience. In Using Corpora for Language Research: Studies in the Honour of Geoffrey Leech, ed. J. Thomas and M. Short. London: Longman, 167–80. Garside, R., G. Leech, and A. McEnery, eds. 1997. Corpus Annotation: Linguistic Information from Computer Text Corpora. London: Longman. Garside, R., and N. Smith. 1997. A Hybrid Grammatical Tagger: CLAWS4. In Corpus Annotation: Linguistic Information from Computer Text Corpora, ed. R. Garside, G. Leech, and A. McEnery. London: Longman, 102–21. Gast, V., and F. Haas. 2011. On the distribution of subject properties in formulaic presentationals: A diachronic-typological approach. In Impersonal Constructions: A cross-linguistic perspective, ed. A. Malchukov and A. Siewierska. Amsterdam/Philadelphia: Benjamins, 127–166. Giparaitė, J. 2006. English [V [NP XP]] adjunct constructions as Small Clauses. In Contrastive Studies and Valency: Studies in Honor of Hans Ulrich Boas, ed. P. C. Steiner, H. C. Boas, and S. J. Schierholz. Frankfurt am Main: Lang, 15–23. Goldberg, A. E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Goldberg, A. E. 2006. Constructions at Work. Oxford: Oxford University Press. Greenbaum, S. 1988. Good English and the Grammarian. London: Longman. Gries, S. Th. 2003. Towards a corpus-based identification of prototypical instances of constructions. Annual Review of Cognitive Linguistics 1:1–27. Grimshaw, J. 1979. Complement Selection and the Lexicon. Linguistic Inquiry 10, no. 2:279– 326. Groß, T. 2003. The Valency of Non-Verbal Word Classes: The Adjective. In Dependency and Valency / Dependenz und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung, ed. V. Ágel, L. M. Eichinger, H.-W. Eroms, P. Hellwig, H. J. Heringer, and H. Lobin. Handbücher zur Sprach- und Kommunikationswissenschaft 25.1. Berlin: Walter de Gruyter, 835–842. Groß, T., and T. Osborne. 2013. Katena und Konstruktion: Ein Vorschlag zu einer dependenziellen Konstruktionsgrammatik. Zeitschrift für Sprachwissenschaft 32, no. 1:41–73. Haegeman, L. 1994. Introduction to Government and Binding Theory. 2nd ed. Malden, MA: Blackwell. Halliday, M. A. K. 1991. Corpus Studies and Probabilistic Grammar. In English Corpus Linguistics: Studies in Honour of Jan Svartvik, ed. K. Aijmer and B. Altenberg. London: Longman, 30–43. Halliday, M. A. K. 1994. An Introduction to Functional Grammar. 2nd ed. London: Arnold. Hannay, M. 1985. English Existentials in Functional Grammar. Dordrecht: Foris. Hauser-Suida, U., and G. Hoppe-Beugel. 1972. Die Vergangenheitstempora in der deutschen geschriebenen Sprache der Gegenwart: Untersuchungen an ausgewählten Texten. München: Hueber. Hausser, R. 2001. Foundations of Computational Linguistics: Human-Computer Communication in Natural Language. 2nd ed. Berlin: Springer.

334 | References

Hawkins, J. 1994. A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press. Helbig, G. 1988. Zum Verhältnis von Grammatik und Lexikon (aus der Sicht der Sprachwissenschaft und des Fremdsprachenunterrichts). Deutsch als Fremdsprache 25, no. 3:160–67. Helbig, G. 1992. Probleme der Valenz- und Kasustheorie. Tübingen: Niemeyer. Helbig, G. 2004. Kleinere Schriften zur Grammatik. Ed. Horst Sitta, Bernd Skibitzki, Johannes Wenzel, and Barbara Wotjak. München: Iudicium. Helbig, G., and G. Heinrich. 1978. Das Vorgangspassiv. 2nd ed. Leipzig: VEB Verlag Enzyklopädie. Helbig, G., and W. Schenkel. [1969] 1973. Wörterbuch zur Valenz und Distribution deutscher Verben. 2nd ed. Leipzig: VEB Verlag Enzyklopädie. Herbst, T. 1983. Untersuchungen zur Valenz englischer Adjektive und ihrer Nominalisierungen. Tübingen: Narr. Herbst, T. 1999. English Valency Structures: A First Sketch. Erfurt: Erfurt Electronic Studies in English. http://webdoc.gwdg.de/edoc/ia/eese/artic99/herbst/main1.html (accessed June 20, 2012). Herbst, T. 2007. Valency Complements or Valency Patterns? In Valency: Theoretical, Descriptive and Cognitive Issues, ed. T. Herbst and K. Götz-Votteler. Berlin: Mouton de Gruyter, 15–35. Herbst, T. 2009. Introduction: The Erlangen Valency Patternbank. http://www.patternbank.unierlangen.de/Patternbank_Introduction.pdf (accessed June 20, 2012). Herbst, T. 2011. The status of generalizations: Valency and Argument Structure. In Argument Structure – Valency and/or Constructions? Special Issue of ZAA, ed. T. Herbst and A. Stefanowitsch. Tübingen: Stauffenburg, 347–67. Herbst, T. 2014. The valency approach to argument structure constructions. In Constructions Collocations Patterns, ed. T. Herbst, H.-J., Schmid, and S. Faulhaber. Berlin: De Gruyter Mouton, 167–216. Herbst, T., and S. Faulhaber. 2011. Optionen der Valenzbeschreibung. In Grammatik und Korpora 2009: Dritte Internationale Konferenz. Mannheim, 22.-24.09.2009, ed. M. Konopka, J. Kubczak, C. Mair, F. Šticha, and U. H. Waßner. Tübingen: Narr, 411–28. Herbst, T., D. Heath, I. Roe, and D. Götz. 2004. A Valency Dictionary of English. Berlin: Mouton de Gruyter. [VDE] Herbst, T., and M. Klotz. 2002. Meeting and Kissing as Valency Problems: Some Remarks on the Treatment of Reciprocity and Reflexivity in a Valency Description of English. In Reflexives and Intensifiers: The Use of Self-forms in English. Special Issue of ZAA, ed. E. König. Tübingen: Stauffenburg, 239–49. Herbst, T., and I. Roe. 1996. How Obligatory are Obligatory Complements? An Alternative Approach to the Categorization of Subjects and Other Complements in Valency Grammar. English Studies 77, no. 2:179–99. Herbst, T., and S. Schüller. 2008. Introduction to Syntactic Analysis: A Valency Approach. Tübingen: Narr. Herbst, T., and P. Uhrig. 2009. The Erlangen Valency Patternbank. http://www.patternbank.uni-erlangen.de (accessed June 20, 2012). Herbst, T., and P. Uhrig. 2010. Valency information online – research and pedagogic reference tools. In eLexcicography in the 21st Century: New Challenges, New Applications, ed. S. Granger and M. Paquot. Louvain-la-Neuve: Presses universitaires de Louvain, 125–36. Heringer, H. J. 1970. Theorie der deutschen Syntax. München: Hueber. Heringer, H. J. 1996. Deutsche Syntax dependentiell. Tübingen: Stauffenburg.

References | 335

Herriman, J. 2000. Extraposition in English: A Study of the Interaction between the Matrix Predicate and the Type of Extraposed Clause. English Studies 81, no. 6: 582–599. Hill, A., ed. 1962. Proceedings of the Third Texas Conference on Problems of Linguistic Analysis in English. Austin: The University of Texas. Hockett, C. 1958. A Course in Modern Linguistics. New York: Macmillan. Hoffmann, S., S. Evert, N. Smith, D. Lee, and Y. Berglund Prytz. 2008. Corpus Linguistics with BNCweb: a Practical Guide. Frankfurt am Main: Lang. Hölzner, M. 2007. Substantivvalenz: Korpusbasierte Untersuchungen zu Argumentrealisierungen deutscher Substantive. Tübingen: Niemeyer. Horrocks, G. 1987. Generative Grammar. London: Longman. Hoye, L. 1997. Adverbs and Modality in English. London: Longman. Huddleston, R. D. 1984. Introduction to the Grammar of English. Cambridge: Cambridge University Press. Huddleston, R. D., and G. K. Pullum, eds. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. [CamG] Hudson, R. 1984. Word Grammar. Oxford: Blackwell. Hudson, R. 1990. English Word Grammar. Oxford: Blackwell. Hudson, R. 1999. Subject-Verb Agreement in English. English Language and Linguistics 3, no. 2:173–207. Hudson, R. 2010. An Introduction to Word Grammar. Cambridge: Cambridge University Press. Jacobs, J. 1994. Kontra Valenz. Trier: WVT. Jacobs, J. 2003. Die Problematik der Valenzebenen. In Dependency and Valency / Dependenz und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung, ed. V. Ágel, L. M. Eichinger, H.-W. Eroms, P. Hellwig, H. J. Heringer, and H. Lobin. Handbücher zur Sprach- und Kommunikationswissenschaft 25.1. Berlin, New York: Walter de Gruyter, 378– 99. Järventausta, M. 2003. Das Subjektproblem in der Valenzforschung. In Dependency and Valency / Dependenz und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung, ed. V. Ágel, L. M. Eichinger, H.-W. Eroms, P. Hellwig, H. J. Heringer, and H. Lobin. Handbücher zur Sprach- und Kommunikationswissenschaft 25.1. Berlin, New York: Walter de Gruyter, 781–94. Jespersen, O. 1909–1949. A Modern English Grammar on Historical Principles. 7 vols. London: George Allen & Unwin; Copenhagen: Munksgaard. Jespersen, O. 1924. The Philosophy of Grammar. London: George Allen & Unwin. Jespersen, O. 1933. Essentials of English Grammar. London: George Allen & Unwin. Johansson, S. 1991. Times Change, and so do Corpora. In English Corpus Linguistics: Studies in Honour of Jan Svartvik, ed. K. Aijmer and B. Altenberg. London: Longman, 305–14. Johnson, D. E. 1977. On Keenan’s Definition of ‘Subject Of’. Linguistic Inquiry 8, no. 4:673–92. Kaltenböck, G. 2004. It-extraposition and non-extraposition in English: A study of syntax in spoken and written texts. Vienna: Braumüller. Karlsson, F. 2007a. Constraints on Multiple Initial Embedding of Clauses. International Journal of Corpus Linguistics 12, no. 1:107–18. Karlsson, F. 2007b. Constraints on Multiple Center-Embedding of Clauses. Journal of Linguistics 43, no. 2:365–92. Kay, P., and C. J. Fillmore. 1999. Grammatical constructions and linguistic generalizations: The What’s X doing Y? construction. Language 75:1–33.

336 | References

Keenan, E. L. 1976. Towards a Universal Definition of Subject. In Subject and Topic, ed. C. N. Li. New York: Academic Press, 303–33. Kibbee, D. A., ed. 2010. Chomskyan (R)evolutions. Amsterdam: Benjamins. Kiparsky, P., and C. Kiparsky. 1970. Fact. In Progress in Linguistics: A Collection of Papers, ed. M. Bierwisch. The Hague: Mouton, 143–73. Kjellmer, G. 2001. ‘It Comes Time’: A Look at Existential it. English Studies 82, no. 4:328–35. Klein, D., and C. D. Manning. 2003. Accurate Unlexicalized Parsing. In Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics, 7–12 July 2003, ed. Association for Computational Linguistics. Sapporo, 423–30. Kleinke, S. 2002. Englische Komplementstrukturen: Schematische und prototypische Bedeutungen. Frankfurt am Main: Lang. Klotz, M. 2000. Grammatik und Lexik: Studien zur Syntagmatik englischer Verben. Tübingen: Stauffenburg. Klotz, M. 2007. Valency Rules? The case of verbs with propositional complements. In Valency: Theoretical, Descriptive and Cognitive Issues, ed. T. Herbst and K. Götz-Votteler. Berlin: Mouton de Gruyter, 117–28. Koerner, E. F. K. 1989. Practicing Linguistic Historiography: Selected Essays. Amsterdam: Benjamins. Kortmann, B. 2006. Syntactic Variation in English: A Global Perspective. In The Handbook of English Linguistics, ed. B. Aarts and A. McMahon. Malden, MA: Blackwell, 603–24. Koya, I. 1992. Subjecthood and Related Notions: A Contrastive Study of English, German and Japanese. Basel: Birkhäuser. Kratzer, A. 1996. Severing the external argument from its verb. In Phrase Structure and the Lexicon, ed. J. Rooryck and L. Zaring. Dordrecht: Kluwer, 109–137. Krug, M. G. 2000. Emerging English Modals: A Corpus-based Study of Grammaticalization. Berlin: Mouton de Gruyter. Kuhn, T. S. 1962. The Structure of Scientific Revolutions. Chicago: University of Chicago Press. Labov, W. 1972. Sociolinguistic Patterns. Philadelphia: University of Pennsylvania Press. Lakoff, G. 1987. Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago: University of Chicago Press. Lambrecht, K. 2001. A framework for the analysis of cleft constructions. Linguistics 39, no.3:463–516. Langacker, R. W. 1987. Foundations of cognitive grammar. Vol. 1: Theoretical prerequisites. Stanford, CA: Stanford University Press. Langacker, R. W. 2005. Construction Grammars: Cognitive, Radical, and less so. In Cognitive Linguistics: Internal Dynamics and Interdisciplinary Interaction, ed. F. J. Ruiz Mendoza de Ibáñez and M. S. Peña Cervel. Berlin: Mouton de Gruyter, 101–59. Langacker, R. W. 2008. Cognitive Grammar: A Basic Introduction. Oxford: Oxford University Press. Langacker, R. W. 2009. Investigations in Cognitive Grammar. Berlin: Mouton de Gruyter. Lasnik, H. 1995. Case and Expletives Revisited: On Greed and Other Human Failings. Linguistic Inquiry 26, no. 4:615–33. Lasnik, H., and R. Fiengo. 1974. Complement Object Deletion. Linguistic Inquiry 5, no. 4:535– 71. Law, P. 1999. On the Passive Existential Construction. Studia Linguistica 53, no. 3:183–208. Leech, G. 1991. The State of the Art in Corpus Linguistics. In English Corpus Linguistics: Studies in Honour of Jan Svartvik, ed. K. Aijmer and B. Altenberg. London: Longman, 8–29.

References | 337

Leech, G., M. Hundt, C. Mair, and N. Smith. 2009. Change in Contemporary English: A Grammatical Study. Cambridge: Cambridge University Press. Lees, R. 1960. A Multiply Ambiguous Adjectival Construction in English. Language 36, no. 2:207–21. Lehmann, H. M., P. Schneider, and S. Hoffmann. 2000. BNCweb. In Corpora Galore: Analyses and Techniques in Describing English, ed. J. Kirk. Amsterdam: Rodopi, 259–66. Lodge, D. J. 1984. Small World: An Academic Romance. Harmondsworth: Penguin Books. Lyons, J. 1991. Chomsky. 3rd ed. London: Fontana Press. MacWhinney, B. 2005. Commentary on Ullman et al. Brain and Language 93, no. 2:239–42. MacWhinney, B. 2014. Item-based patterns in early syntactic development. In Thomas Herbst, Hans-Jörg Schmid & Susen Faulhaber (eds.), Constructions Collocations Patterns. Berlin/ Boston: De Gruyter Mouton, 33–70. Mair, C. 1988. Extraposed Gerundial Subject Clauses in Present-Day British English: An Investigation of the Corpus of the Survey of English Usage. Arbeiten aus Anglistik und Amerikanistik 13, no. 1:51–63. Mair, C. 1990. Infinitival Complement Clauses in English: A Study of Syntax in Discourse. Cambridge: Cambridge University Press. Manning, C. D., and H. Schütze. 1999. Foundations of Statistical Natural Language Processing. Cambridge, MA: MIT Press. Marantz, A. 1984. On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. Marx-Moyse, J. 1983. Untersuchungen zur deutschen Satzsyntax: es als vorausweisendes Element eines Subjektsatzes. Wiesbaden: Steiner. Marcus, M. P., B. Santorini, and M. A. Marcinkiewicz. 1993. Building a Large Annotated Corpus of English: The Penn Treebank. Computational Linguistics 19, no. 2:313–30. Marneffe, M.-C. de, B. MacCartney, and C. D. Manning. Generating Typed Dependency Parses from Phrase Structure Parses. In Proceedings of the 5th International Conference on Language Resources and Evaluation (LREC 2006). Genoa: ELRA, 449–454. Marneffe, M.-C. de, and C. D. Manning. 2008. Stanford Typed Dependencies Manual. Stanford, CA: Stanford University Technical Report. http://nlp.stanford.edu/software/dependencies_manual.pdf (accessed June 20, 2012). Martínez Insua, A. E. 2002. On the Nature of the Verb in Present Day English (Existential) ThereConstructions: Formal and Communicative Implications. Revista Alicantina de Estudios Ingleses 15:133–52. Martínez Insua, A. E., and I. M. Palacios Martínez. 2003. A Corpus-Based Approach to NonConcord in Present Day English Existential There-Constructions. English Studies 84, no. 3:262–83. Mathesius, V. 1929. Zur Satzperspektive im Modernen Englisch. Archiv für das Studium der neueren Sprachen und Literaturen 84, no. 155:202–10. Matthews, P. 1981. Syntax. Cambridge: Cambridge University Press. Matthews, P. 1993. Grammatical Theory in the United States from Bloomfield to Chomsky. Cambridge: Cambridge University Press. Matthews, P. 2007a. Syntactic Relations: A critical survey. Cambridge: Cambridge University Press. Matthews, P. 2007b. The scope of valency in Grammar. In Valency: Theoretical, Descriptive and Cognitive Issues, ed. T. Herbst and K. Götz-Votteler. Berlin: Mouton de Gruyter, 3–14. McArthur, T. 1992. The Oxford Companion to the English Language. Oxford: Oxford University Press.

338 | References

McCloskey, J. 1996. Subjects and Subject Positions in Irish. In The Syntax of the Celtic Languages: A Comparative Perspective, ed. R. D. Borsley and I. Roberts. Cambridge: Cambridge University Press, 241–83. McCloskey, J. 1997. Subjecthood and Subject Positions. In Elements of Grammar: Handbook in Generative Syntax, ed. L. Haegeman. Dordrecht: Kluwer, 197–235. McEnery, A., and A. Wilson. 1996. Corpus Linguistics. Edinburgh: Edinburgh University Press. Michel, J.-B., Y. K. Shen, A. P. Aiden, A. Veres, M. K. Gray, The Google Books Team, J. P. Picket et al. 2011. Quantitative Analysis of Culture Using Millions of Digitized Books. Science 331, no. 6014:176–82 Miller, P. H. 2001. Discourse constraints on (non)extraposition from subject in English. Linguistics 39, no. 4:683–701. Mindt, I. 2007. The Valency of Experiential and Evaluative Adjectives. In Valency: Theoretical, Descriptive and Cognitive Issues, ed. T. Herbst and K. Götz-Votteler. Berlin: Mouton de Gruyter, 101–15. Mindt, I. 2011. Adjective complementation: An empirical analysis of adjectives followed by thatclauses. Amsterdam: Benjamins. Modrak, D. K. W. 2001. Aristotle’s Theory of Language and Meaning. Cambridge: Cambridge University Press. Mohr, S. 2005. Clausal Architecture and Subject Positions: Impersonal Constructions in the Germanic Languages. Amsterdam: Benjamins. Nagle, S. J. 2003. Double Modals in the Southern United States: Syntactic Structure or Syntactic Structures? In Modality in Contemporary English, ed. R. Facchinetti, M. Krug, and F. Palmer. Berlin: Mouton de Gruyter, 349–71. Napoli, D. J. 1989. Predication Theory: A Case Study for Indexing Theory. Cambridge: Cambridge University Press. Neef, M., and H. Vater. 2006. Concepts of the Lexicon in Theoretical Linguistics. In Advances in the Theory of the Lexicon, ed. D. Wunderlich. Berlin: Mouton de Gruyter, 27–55. Newmeyer, F. J. 1986. Has There Been a ‘Chomskyan Revolution’ in Linguistics? Language 62, no. 1:1–18. Newmeyer, F. J. 2003. Theoretical implications of grammatical category-grammatical relation mismatches. In Mismatch: Form-function incongruity and the architecture of grammar, ed. E. J. Francis and L. A. Michaelis. Stanford, CA: CSLI Publications, 149–78. Nuyts, J. 2001. Epistemic Modality, Language, and Conceptualization: A Cognitive-Pragmatic Perspective. Amsterdam: Benjamins. Oehrle, R. 1979. A Theoretical Consequence of Constituent Structure in Tough Movement. Linguistic Inquiry 10, no. 4:583–93. Oosten, J. H. v. 1984. The Nature of Subjects, Topics and Agents: A Cognitive Explanation. Bloomington: Indiana University Linguistic Club. Oppenrieder, W. 1991. Von Subjekten, Sätzen und Subjektsätzen: Untersuchungen zur Syntax des Deutschen. Tübingen: Niemeyer. Oppenrieder, W. 2006. Subjekt- und Objektsätze. In Dependenz und Valenz / Dependency and Valency: Ein internationales Handbuch der zeitgenössischen Forschung, ed. V. Ágel, G. Ungeheuer, H. Steger, H. E. Wiegand, and A. Burkhardt. Handbücher zur Sprach- und Kommunikationswissenschaft 25.2. Berlin, New York: Mouton de Gruyter, 900–913. Osborne, T. 2007. The Weight of Predicates: A Dependency Grammar Analysis of Predicate Weight in German. Journal of Germanic Linguistics 19, no. 1:23–72.

References | 339

Osborne, T., M. Putnam, and T. Groß. 2012. Catenae: Introducing a Novel Unit of Syntactic Analysis. Syntax 15, no. 4:354–396. Paikeday, T. M. 1985. May I Kill the Native Speaker? TESOL Quarterly 19, no. 2:390–95. Paikeday, T. M. 2003. The native speaker is dead! An informal discussion of a linguistic myth with Noam Chomsky and other linguists, philosophers, psychologists, and lexicographers. Brampton, Ontario: Lexicography Inc. Palmer, F. 1971. Grammar. Harmondsworth: Penguin Books. Palmer, F. 1986. Mood and Modality. Cambridge: Cambridge University Press. Palmer, F. 1990. Modality and the English Modals. 2nd ed. London: Longman. Patten, A.L. 2016. Well-formed lists: specificational copular sentences as predicative inversion constructions. English Language and Linguistics 22, no. 1: 77–99. Perlmutter, D. 1982. Syntactic Representation, Syntactic Levels, and the Notion of Subject. In The nature of syntactic representation, ed. P. Jacobson and G. K. Pullum. Dordrecht: Reidel, 283–340. Perlmutter, D., and P. Postal. 1974. Lectures on Relational Grammar. Amherst: LSA Linguistic Institute, University of Massachusetts. Pollard, C., and I. A. Sag. 1994. Head-Driven Phrase Structure Grammar. Chicago: University of Chicago Press. Pollock, J.-Y. 1989. Verb Movement, Universal Grammar and the Structure of IP. Linguistic Inquiry 20, no. 3:365–424. Postal, P. 1971. Cross-over Phenomena. New York: Holt, Rinehart and Winston. Postal, P. 1974. On Raising. Cambridge, MA: MIT Press. Postal, P. 1976. Avoiding Reference to Subject. Linguistic Inquiry 7, no. 1:151–82. Proisl, T., and P. Uhrig. 2012. Efficient Dependency Graph Matching with the IMS Open Corpus Workbench. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC’12). Istanbul: ELRA, 2750–56. Pullum, G. K. 1980. Syntactic Relations and Linguistic Universals. Transactions of the Philological Society, 1–39. Pullum, G. K. 2009. Lexical Categorization in English Dictionaries and Traditional Grammars. Zeitschrift für Anglistik und Amerikanistik 57, no. 3:255–73. Quirk, R., and S. Greenbaum. 1970. Elicitation Experiments in English: Linguistic Studies in Use and Attitude. London: Longman. Quirk, R., S. Greenbaum, G. Leech, and J. Svartvik. 1972. A Grammar of Contemporary English. London: Longman. Quirk, R., S. Greenbaum, G. Leech, and J. Svartvik. 1985. A Comprehensive Grammar of the English Language. London: Longman. [CGEL] Reis, M. 1982. Zum Subjektbegriff im Deutschen. In Satzglieder im Deutschen: Vorschläge zur syntaktischen, semantischen und pragmatischen Fundierung, ed. W. Abraham. Tübingen: Narr, 171–211. Renouf, A. 1987. Corpus Development. In Looking Up: An Account of the COBUILD Project in Lexical Computing, ed. J. McH. Sinclair. London: Collins COBUILD, 1–22. Riordan, B. 2007. There’s two ways to say it: Modeling nonprestige there’s. Corpus Linguistics and Linguistic Theory 3, no. 2:233–79. Rizzi, L. 2006. On the Form of Chains: Criterial Positions and ECP Effects. In Wh-Movement: Moving On, ed. Lisa Lai-Shen Cheng and Norbert Corver. Cambridge, MA/London: MIT Press, 97–133.

340 | References

Rizzi, L., and U. Shlonsky. 2006. Satisfying the Subject Criterion by a non subject: English Locative Inversion and Heavy NP Shift. In Phases of Interpretation, ed. M. Frascarelli. Berlin/New York: Mouton de Gruyter, 341–362. Rizzi, L., and U. Shlonsky. 2007. Strategies of Subject Extraction. In Interfaces + Recursion = Language? Chomsky's Minimalism and the View from Syntax-Semantics, ed. U. Sauerland and H.-M. Gärtner. Berlin/New York: Mouton de Gruyter, 115–160. Rohdenburg, G. 1992. Bemerkungen zu infiniten Konstruktionen im Englischen und Deutschen. In: New Departures in Contrastive Lingustics – Neue Ansätze in der Kontrastiven Linguistik: Proceedings of the Conference Held at the Leopold-Franzens-University of Innsbruck, Austria, 10–12 May 1991, ed. Ch. Mair and M. Markus. Innsbruck: Innsbrucker Beiträge zur Kulturwissenschaft, 187–207. Rohdenburg, G. 2003. Cognitive Complexity and horror aequi as Factors Determining the Use of Interrogative Clause Linkers in English. In Rohdenburg & Mohndorf (2003), 205–49. Rohdenburg, G., and B. Mondorf, eds. 2003. Determinants of Grammatical Variation in English. Berlin: Mouton de Gruyter. Rosenbaum, P. 1967. The Grammar of English Predicate Complement Constructions. Cambridge, MA: MIT Press. Ross, J. R. 1967. Constraints on Variables in Syntax. PhD Thesis: Massachusetts Institute of Technology. Rothstein, S. D. 2001. Predicates and their Subjects. Dordrecht: Kluwer. Rowling, J. K. 1997. Harry Potter and the Philosopher’s Stone. [HP1]. London: Bloomsbury. Rowling, J. K. 1998. Harry Potter and the Chamber of Secrets. [HP2]. London: Bloomsbury. Rowling, J. K. 1999. Harry Potter and the Prisoner of Azkaban. [HP3]. London: Bloomsbury. Rowling, J. K. 2000. Harry Potter and the Goblet of Fire. [HP4]. London: Bloomsbury. Rowling, J. K. 2003. Harry Potter and the Order of the Phoenix. [HP5]. London: Bloomsbury. Rowling, J. K. 2005. Harry Potter and the Half-Blood Prince. [HP6]. London: Bloomsbury. Rowling, J. K. 2007. Harry Potter and the Deathly Hallows. [HP7]. London: Bloomsbury. Rupp, L. 2005. Constraints on nonstandard -s in expletive there sentences: a generative– variationist perspective. English Language and Linguistics 9, no. 2: 255–288. Rychlý, P. 2007. Manatee/Bonito: A Modular Corpus Manager. In 1st Workshop on Recent Advances in Slavonic Natural Language Processing. Brno: Masaryk University, 65–70. Sadziński, R. 2006. Diathesen und Konversen. In Dependenz und Valenz / Dependency and Valency: Ein internationales Handbuch der zeitgenössischen Forschung, ed. V. Ágel, G. Ungeheuer, H. Steger, H. E. Wiegand, and A. Burkhardt. Handbücher zur Sprach- und Kommunikationswissenschaft 25.2. Berlin: Mouton de Gruyter, 963–73. Sampson, G. 1987. Evidence against the ‘Grammatical’/’Ungrammatical’ Distinction. In Corpus Linguistics and beyond: Proceedings of the 7. International Conference on English Language Research on Computerized Corpora, ed. W. Meijs. Amsterdam: Rodopi, 219–26. Sampson, G. 2001. Empirical Linguistics. New York: Continuum. Sampson, G. 2007. Grammar without Grammaticality. Corpus Linguistics and Linguistic Theory 3, no. 1:1–32. Schierholz, S. J. 2001. Präpositionalattribute: Syntaktische und semantische Analysen. Tübingen: Niemeyer. Schierholz, S. J. 2005. Einige grundlegende Überlegungen zur Corpuslinguistik. In Corpuslinguistik in Lexik und Grammatik, ed. F. Lenz and S. J. Schierholz. Tübingen: Stauffenburg, 1–14.

References | 341

Schmid, H. 2008. Tokenizing and part-of-speech tagging. In Corpus Linguistics: An International Handbook, Vol. 1, ed. A. Lüdeling and M. Kytö. Berlin: Mouton de Gruyter, 527–51. Schmidtke-Bode, K. 2014. Complement Clauses and Complementation Systems: A CrossLinguistic Study of Grammatical Organization. PhD Thesis: Friedrich-Schiller-Universität Jena. Schneider, G. 2009. Hybrid Long-Distance Functional Dependency Parsing: A hybrid deepsyntactic Dependency Grammar parser for English combining statistical performance and formal grammar-based competence approaches. Saarbrücken: Südwestdeutscher Verlag für Hochschulschriften. Schumacher, H., J. Kubczak, R. Schmidt, and V. de Ruiter. 2004. VALBU - Valenzwörterbuch deutscher Verben. Tübingen: Narr. Seuren, P. A. M. 1999. The Subject-Predicate Debate X-rayed. In History of Linguistics 1996: Traditions in linguistics worldwide, ed. D. Cram, A. Linn, and E. Nowak. Amsterdam: Benjamins, 41–55. Siepmann, D. 2005. Discourse Markers across Languages: A Contrastive Study of Second-level Discourse Markers in Native and Non-Native text with Implications for General and Pedagogic Lexicography. London: Routledge. Siewierska, A. 1984. The Passive: A Comparative Linguistic Analysis. London: Croom Helm. Sinclair, J. McH. 1991. Corpus Concordance Collocation. Oxford: Oxford University Press. Sinclair, J. McH. 2004. Trust the Text. London: Routledge. Sinclair, J. McH., and A. Mauranen. 2006. Linear Unit Grammar: Integrating Speech and Writing. Amsterdam: Benjamins. Somers, H. L. 1984. On the Validity of the Complement-Adjunct Distinction in Valency Grammar. Linguistics 22:507–30. Somers, H. L. 1987. Valency and Case in Computational Linguistics. Edinburgh: Edinburgh University Press. Stefanowitsch, A. 2006. Negative Evidence and the Raw Frequency Fallacy. Corpus Linguistics and Linguistic Theory 2, no. 1:61–77. Stefanowitsch, A. 2007. Linguistics beyond Grammaticality. Corpus Linguistics and Linguistic Theory 3, no. 1:57–71. Stefanowitsch, A. 2010. Keine Grammatik ohne Konstruktionen. Abstract. Paper presented at the 46. Jahrestagung des Instituts für Deutsche Sprache, Mannheim. Stefanowitsch, A., and S. Th. Gries. 2003. Collostructions: Investigating the Interaction of Words and Constructions. International Journal of Corpus Linguistics 8, no. 2:209–43. Stein, G. 1979. Studies in the Function of the Passive. Tübingen: Narr. Steinthal, H. 1855. Grammatik, Logik und Psychologie: Ihre Principien und ihr Verhältniss zu einander. Berlin: Dümmler. Sternefeld, W. 1985. Deutsch ohne grammatische Funktionen: Ein Beitrag zur Rektions- und Bindungstheorie. Linguistische Berichte 99:394–439. Stubbs, M. 1995. Collocations and Semantic Profiles: On the Cause of the Trouble with Quantitative Studies. Functions of Language 2, no. 1:23–55. Svartvik, J. 1966. On Voice in the English Verb. The Hague: Mouton. Szmrecsányi, B. M. 2004. On Operationalizing Syntactic Complexity. In Le Poids des Mots: Actes des 7es Journées internationales d’Analyse statistique des Données Textuelles, ed. G. Purnelle, C. Fairon, and A. Dister. Louvain-la-Neuve: Presses universitaires de Louvain, 1031–38.

342 | References

Teubert, W. 2003. Die Valenz nichtverbaler Wortarten: das Substantiv. In Dependency and Valency / Dependenz und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung, ed. V. Ágel, L. M. Eichinger, H.-W. Eroms, P. Hellwig, H. J. Heringer, and H. Lobin. Handbücher zur Sprach- und Kommunikationswissenschaft 25.1. Berlin: Walter de Gruyter, 820–35. Tesnière, L. 1953. Esquisse d’une Syntaxe Structurale. Paris: Klincksieck. Tesnière, L. [1959] 1965. Éléments de Syntaxe Structurale. 2nd ed. Paris: Klincksieck. (Inside front matter gives 1966 as year of publication.) Townsend, S. 2003. The Secret Diary of Adrian Mole Aged 13 3/4. New York: HarperCollins. Trudgill, P., and J. Hannah. 2008. International English: A Guide to Varieties of Standard English. 5th ed. London: Hodder Education. Uhrig, P., and T. Proisl. 2011. The Treebank.info Project. Paper presented at ICAME 32, Oslo. Ullman, M. T., R. Pancheva, T. Love, E. Yee, D. Swinney, and G. Hickok. 2005. Neural Correlates of Lexicon and Grammar: Evidence from the Production, Reading, and Judgment of Inflection in Aphasia. Brain and Language 93, no. 2:185–238. Verplaetse, H. 2003. What you and I want: A Functional Approach to Verb Complementation of Modal want to. In Modality in Contemporary English, ed. R. Facchinetti, M. Krug, and F. Palmer. Berlin: Mouton de Gruyter, 151–89. Wasow, T. 1997. End-Weight from the Speaker’s Perspective. Journal of Psycholinguistic Research 26, no. 3:347–61. Wasow, T. 2001. Generative Grammar. In The Handbook of Linguistics, ed. M. Aronoff and J. Rees-Miller. Oxford: Blackwell, 295–318. Wasow, T. 2002. Postverbal Behavior. Stanford, CA: CSLI Publications. Wedin, M. V. 2000. Aristotle’s Theory of Substance: The Categories and Metaphysics Zeta. Oxford: Oxford University Press. Wegener, H. 1990. Komplemente in der Dependenzgrammatik und in der Rektions- und Bindungstheorie: Die Verwendung der Kasus im Deutschen. Zeitschrift für germanistische Linguistik 18:150–84. Welke, K. M. 1988. Einführung in die Valenz- und Kasustheorie. Leipzig: VEB Bibliographisches Institut. Welke, K. M. 2002. Deutsche Syntax funktional: Perspektiviertheit syntaktischer Strukturen. Tübingen: Stauffenburg. Welke, K. 2011. Valenzgrammatik des Deutschen: Eine Einführung. Berlin: De Gruyter. Williams, E. 1981. Argument Structure and Morphology. The Linguistic Review 1:81–114. Williams, E. 1984. There-insertion. Linguistic Inquiry 15, no. 1:131–53. Winston, M. E. 1976. Did a (Kuhnian) Scientific Revolution Occur in Linguistics? Proceedings of the Biennial Meeting of the Philosophy of Science Association, no. 1:25–33. Wittgenstein, L. 1953. Philosophical Investigations. Oxford: Blackwell. Wunderlich, D., ed. 2006. Advances in the Theory of the Lexicon. Berlin: Mouton de Gruyter. Zandvoort, R. W. 1965. Handbook of English Grammar. 3rd ed. London: Longman. (Orig. pub. 1957.) Zandvoort, R. W. 1972. Handbook of English Grammar. 6th ed. London: Longman. (Orig. pub. 1957.) Zhang, G. P. 2015. It is suggested that…or it is better to…? Forms and meanings of subject itextraposition in academic and popular writing. Journal of English for Academic Purposes 20: 1–13.

References | 343

Zifonun, G., L. Hoffmann, and B. Strecker. 1997. Grammatik der deutschen Sprache. 3 vols. Berlin: De Gruyter.

Dictionaries Cambridge English Pronouncing Dictionary (2011), 18th ed. (editors: P. Roach, J. Setter, and J. Esling), Cambridge: Cambridge University Press. [EPD18] Longman Dictionary of Contemporary English (2009) 5th ed. (editorial director: M. Mayor), Harlow: Pearson. [LDOCE5] Oxford Advanced Learner’s Dictionary (2007) 7th ed. (chief editor: S. Wehmeier), Oxford: Oxford University Press. [OALD7]

Corpora The British National Corpus, version 3 (2007), distributed by Oxford University Computing Services on behalf of the BNC Consortium. [BNC] The Corpus of Contemporary American English: 450 million words, 1990-present (2008-), published by M. Davies, available online at http://corpus.byu.edu/coca/. [COCA] [For a description of further, unpublished corpora used see Section 3.2.2.1.]

Index A Valency Dictionary of English (VDE) 4, 31, 56, 59, 71ff., 95ff., 99ff., 108ff., 112ff., 118ff., 129ff., 134, 136ff., 142, 144ff., 149, 152, 161, 165f., 169, 179ff., 185ff., 221f., 224ff., 248, 282, 298f., 301, 304, 321f. Aarts, B. 9, 54, 105, 107ff., 171, 242, 254 Aarts, F. 13, 93f., 128 Aarts, J. 13, 93f., 128 acceptability 55, 59, 67ff., 79f., 91, 102, 114, 116f., 136ff., 143, 147ff., 152, 156f., 174f., 184, 191, 195f., 211f., 230, 245, 249, 258, 284, 295 accusative 38, 190, 298 active 3, 12, 30f., 44, 46, 71–91, 93ff., 107ff., 113ff., 120f., 169, 197, 202, 221f., 224f., 231, 245, 260, 281, 288f., 295f., 298ff., 311f., 327 adjective 4, 29, 47, 56, 82, 123ff., 149ff., 157, 160, 162, 164, 166, 171, 179ff., 184, 204, 215, 220f., 229ff., 233ff., 265f., 282f., 295f., 304f., 311, 319, 325ff. adjunct 29, 31ff., 38, 43, 45, 101, 106, 128, 164, 204f., 207f., 239, 241, 252, 261, 280, 321 adverb 64, 82, 157, 198, 240, 258 adverbial 38, 85, 157, 175, 178, 185, 198, 205, 240f., 249 Ágel, V. 8, 32, 50, 313, 328 agent 12, 44, 46, 50f., 82, 114, 197, 199, 202, 261f., 286 agentive 45f., 285f., 289 agentivity 46, 197, 199f., 202, 262, 285f., 293 Allerton, D. J. 2, 29, 33, 94 Anderson, J. M. 253, 259 anticipatory it 171, 179, 187ff., 193, 195, 198ff., 202ff., 212, 219 argument (see also complement) 3, 8, 15, 20f., 23, 26, 30, 32f., 50, 128, 180f., 199, 201, 229, 260, 273ff., 292, 294, 298, 301, 324 Aristotle 6f., 26

https://doi.org/10.1515/9783110589801-365

Behaghel, O. 209 Behrens, H. 317 BNC 55f., 58f., 62, 64, 76, 84ff., 102, 104, 110, 116, 129, 135, 146, 157, 204f., 241, 252, 256ff., 269, 279f., 319, 321, 324 Breivik, L. E. 241, 243, 253, 269, 276 Bresnan, J. 20, 23, 38, 95 Bybee, J. 19, 82, 187, 236, 317ff., 322, 324, 326, 328 CamG 8f., 13ff., 23, 26f., 37ff., 47, 49, 51, 64, 71, 82, 93f., 119, 124, 127ff., 157, 159, 173ff., 182ff., 196f., 199, 201f., 204ff., 209ff., 225, 229ff., 234, 237f., 240ff., 246ff., 257ff., 263, 265, 267ff., 275, 288f., 292, 296f., 307, 321f. canonical 4, 13, 37, 39ff., 50, 64, 75, 86, 96, 103, 136, 149f., 153f., 159, 161f., 169f., 172, 176, 179, 185ff., 189, 193, 195, 203, 209ff., 213ff., 219ff., 228, 230, 232f., 236f., 243, 254f., 285, 287, 295, 298, 307f., 310, 316, 321, 327 Cardinaletti, A. 24 CGEL 9, 12ff., 26, 36ff., 44ff., 51, 53, 71, 82, 93f., 96, 110, 112ff., 120, 123f., 130, 147, 159, 171ff., 177f., 182, 185ff., 191, 197ff., 206ff., 212f., 229, 232, 237f., 240ff., 251ff., 258, 260f., 263, 268f., 287f., 300, 321f. Chomsky, N. 2, 13, 16ff., 25, 28, 31, 42, 54f., 67f., 94, 105, 124, 210, 231, 233, 253, 270ff., 274, 284, 309 clausal role 46, 202, 286 clausal subject 2ff., 35, 39, 43, 45f., 49f., 55f., 60, 64, 71–236, 279, 286, 288, 296f., 304, 310, 319, 322, 325f. cleft 204, 230, 287 cleft existentials 271 Coates, J. 86ff. cognitive 7, 209, 235, 317f., 328 Collins, P. 217f., 239, 243ff., 250f., 271 complement 1f., 13ff., 23, 27ff., 38f., 43f., 47f., 50, 56, 60, 66, 68, 71, 76ff., 86, 91, 93ff., 111ff., 123, 127, 130, 132f., 136ff.,

346 | Index

146, 151f., 157, 164ff., 172, 177, 179ff., 184, 186ff., 190f., 193, 197, 201ff., 211, 213, 215, 217ff., 226ff., 233ff., 247f., 250, 267, 273, 275, 283, 289, 292f., 296ff., 301ff., 309ff., 327f. complement inventory 95f., 115, 120, 235, 298ff. complementation 3, 58, 76, 94, 156, 230, 241f., 314f., 321 complex transitive 94, 124, 166f. concord 11f., 39, 44, 61, 126, 158, 172, 253, 256, 260, 267ff., 276, 292 conjunction (see also particle) 6, 62ff., 125 constituency 33f., 51 constituent 4, 7, 12ff., 17, 26f., 33f., 36, 38, 40f., 52, 63, 93, 106, 108f., 113, 125, 163f., 170, 173f., 178, 189ff., 209f., 212, 229, 236, 241, 245, 263, 265, 274, 276, 285, 288, 295, 298, 303, 306ff. construction 12, 14, 19, 27, 37ff., 43, 55, 59f., 68, 86, 89, 101, 106, 109f., 112, 119, 124, 128, 130, 132ff., 137f., 144, 146, 149, 154, 157f., 171ff., 177, 182ff., 189, 196, 201, 206, 214, 216,220, 224, 229f., 232, 236ff., 241, 244ff., 261, 268, 274ff., 281, 283, 287ff., 297f., 302ff., 306, 308, 313ff., 318, 321, 324ff. contextual specification 275, 304, 310, 313 contextually optional 29f., 165, 299 coordination 37, 40ff., 85, 100, 136, 183f., 190f., 193f., 196, 200f., 204, 208, 213, 222f., 258, 267, 315, 325 copula 3, 34, 71, 78, 83, 87, 90f., 123–67, 182, 222, 232, 259, 269, 271, 312, 321f., 325 copular 3f., 11, 14, 34, 105f., 123–67, 182, 192, 281, 285f., 288, 296, 307, 311 Crystal, D. 53, 58, 61, 306 Dąbrowska, E. 281 Davies, W. D. 19, 22, 28, 294 declarative 3, 14, 30f., 38, 40, 46, 71, 94f., 120, 163, 191, 202f., 237, 288f., 298, 300 definite 10, 36, 200, 243f. deletion 29, 107, 114, 206, 234, 272

dependency 2, 13, 16, 22, 29, 33ff., 51f., 64, 127, 206, 229, 267, 280, 292, 306, 308f., 311ff., 327 determiner 2, 63, 67, 267, 270, 318 direct object 41, 44, 94, 97, 113, 244 dislocation 147f., 171, 173ff., 219, 240 divalent 29, 97, 108, 123, 166, 224, 226f., 273, 301f., 322 Dowty, D. 45ff., 199, 261, 286 Dubinsky, S. 19, 22, 28, 294 Duffley, P. J. 88, 297 Eisenberg, P. 33, 124, 189, 191, 201, 203 Emons, R. 2, 29, 33, 188 Engel, U. 33f., 96, 188, 235 Erdmann, P. 172, 181, 206, 218ff., 225f., 239, 272f. Erlangen Valency Patternbank 56, 59, 96, 129ff., 134, 139f., 142, 146f., 149, 153, 161, 169, 222, 236, 275, 300, 302ff. Eroms, H.-W. 33ff., 124, 126f. existential 4f., 42, 51, 61, 198, 230, 237–77, 285f., 288f., 291f., 313f., 322, 324, 327 extraposition 4, 28, 42, 51f., 56, 64, 75, 79, 81, 83, 87ff., 100, 103, 106, 112f., 118f., 129ff., 142, 144, 146ff., 151, 153, 155, 157, 160f., 164, 167, 169–236, 252ff., 259, 264, 273ff., 280ff., 285f., 289, 291f., 297, 301f., 304, 306f., 309, 315f., 319ff. Falk, Y. N. 24, 50, 294, 306, 308 Faulhaber, S. (see also Schüller, S.) 1, 46, 78, 91, 95, 97, 100, 107f., 114ff., 152, 202, 296 Fillmore, C. J. 8, 45, 54, 158, 246, 276, 286 Fischer, K. 29, 33f., 188 frequency 2, 54, 59, 61, 86f., 89, 116, 138, 153, 177, 217, 219, 225, 241f., 252, 269, 279, 283, 321f., 324ff., 328 frequent 38, 55, 60, 63, 74, 76, 82, 87, 93f., 104, 129f., 142, 145, 171f., 214, 218f., 225, 232f., 293, 297, 322, 326 Garside, R. 62, 64 Gast, V. 238, 241, 262, 276 generative grammar 2f., 5, 9, 16ff., 24ff., 30, 52, 66, 91, 94f., 105, 107ff., 123, 169,

Index | 347

181, 190, 210, 228f., 234, 236, 239, 241, 247, 259, 262ff., 266, 271f., 274, 294f., 308f., 316 gerund 11, 14, 88, 119, 175, 214, 216 Goldberg, A. E. 19, 74, 89, 109, 128, 158, 187, 202, 274, 276, 288, 298, 301, 304, 314, 324, 328 gradient 25, 82, 198, 234, 288 Greenbaum, S. 55, 58ff., 65f., 206, 284 Gries, S. Th. 54, 316 Haas, F. 238, 241, 262, 276 Halliday, M. A. K. 8, 44, 51, 54, 178, 188, 287, 289, 294 Hannay, M. 241, 248, 251, 256, 265f. Hawkins, J. 210 head 10, 13, 22, 26, 34, 67, 125, 167, 189, 201, 203, 240, 248, 250, 284, 289, 303, 319 Helbig, G. 2, 8, 29ff., 95, 124ff., 164, 202, 273, 287, 298 Herbst, T. 2ff., 12, 27, 29ff., 34f., 41, 46, 56, 59, 64, 71, 94ff., 107, 109, 112ff., 121, 125, 129, 132, 136, 149, 153, 159, 163ff., 179, 187f., 197, 201f., 215, 236, 238ff., 254, 263, 268, 273, 275, 286, 288, 298, 300ff., 306f., 314, 321, 324, 327 Heringer, H. J. 29f., 33f. Huddleston, R. D. 2, 9, 13f., 37, 124, 128, 178, 188f., 193ff., 199, 201, 204 Hudson, R. 2, 165, 262, 267 imperative 41, 50, 114f., 300 impersonal 130, 184ff., 220, 224 indefinite 10, 29, 174, 243, 254, 261 indirect object 44, 197 infinitive 21, 62, 66, 88, 110, 113, 151, 172, 229, 231, 243f., 253, 283, 307 interrogative 14, 39f., 71, 136, 191ff., 195f., 200ff., 224, 254, 257f., 303 intransitive 14f., 46, 124, 240, 251, 259 inversion 38ff., 136, 154, 159, 191, 196, 200f., 203, 238f., 242, 256f. Jacobs, J. 32f., 35, 292, 295, 313, 328 Järventausta, M. 35, 287, 297

Jespersen, O. 5, 8ff., 36, 46, 82, 94, 105, 171f., 177, 181, 206, 219, 237, 295 Kaltenböck, G. 157f., 171, 174, 198, 206, 214, 217ff., 221, 231 Keenan, E. L. 25f., 50, 293f. Kiparsky, C. and P. 213ff., 284 Klotz, M. 27, 46, 54, 152, 202, 273 Langacker, R. W. 19, 68, 119, 288f., 297, 318 lemma 62, 64, 321 lexical unit 14, 71f., 74, 89, 108, 130, 137, 139f., 142, 144ff., 150, 161, 169, 222, 226, 281, 310 lexicon 18f., 30, 35, 128, 170, 290, 306, 319 LGSWE 58, 83, 86, 94, 247, 263 locative 159, 238ff., 243, 254f., 257f., 261, 271, 322 Mair, C. 3, 54, 86f., 125, 173, 175, 178, 214, 216f., 219, 221, 231, 234 Marantz, A. 23, 27, 50f. Martínez Insua, A. E. 241, 252f., 269, 276 matrix 37, 109, 119, 171, 184, 194, 204ff., 214, 221, 225, 230f., 297 Matthews, P. 5, 31, 33f., 51f., 124, 175, 179, 203, 260, 306 McCloskey, J. 20ff., 28, 295 meaning 6ff., 12, 32, 42ff., 67, 77, 82, 86ff., 104, 114ff., 128, 134, 138, 144ff., 159, 181, 198, 216, 239ff., 243, 245f., 252, 254, 260, 275f., 280, 284, 297, 301, 303ff., 314 modal verb 58, 81ff., 182, 215, 242, 258, 268, 321 modality 71, 77, 81f., 87, 91, 133, 176, 181f., 215f., 297 modifier 157, 250f. monovalent 15, 126, 226 negation 82, 130, 136, 145, 152, 321 Newmeyer, F. J. 16, 38, 81, 191, 283, 298 nominative 2, 30, 35, 38f., 190f., 289 non-canonical 4f., 37, 170, 236, 276, 285, 287, 295 noun 2, 4, 17, 29, 38, 46f., 56, 58f., 62, 67, 76, 82, 107, 123, 125, 129f., 138ff., 143f.,

348 | Index

146ff., 153ff., 161ff., 172, 174, 189, 191, 204, 210, 214, 216, 220f., 230f., 235f., 242, 244, 248, 267, 269f., 273, 275, 286, 292, 295f., 312, 318, 322, 327 object (see also direct object and indirect object) 3, 10, 12ff., 23, 32, 38, 41, 47, 49, 51, 93f., 97, 113, 127ff., 177, 185, 197, 229, 234, 246f., 259f., 262, 288f., 298, 307 obligatoriness 2, 31ff., 35, 41f., 193ff., 200, 222 obligatory 2, 4, 13, 22, 27, 29ff., 41f., 97, 100, 103, 118, 129, 133, 136, 178ff., 182, 184ff., 188, 191, 195f., 208, 215, 217f., 220f., 223ff., 235f., 240, 254, 273, 298, 319, 321f. operator 39f., 191f., 242, 303 Oppenrieder, W. 14, 23, 39, 46, 49, 185, 292 optional 4, 29, 31, 114, 128, 130, 147, 165, 169, 179, 206f., 213, 215, 221, 225, 235, 298f., 302 optionality 30, 96, 128, 165, 193, 206, 300, 302 Palmer, F. 38f., 82, 87f. participant pattern 327 participant role 46, 121, 202, 234, 301 particle 107, 111f., 114f., 126, 157, 239f., 253, 300 passive 3, 12, 20, 25, 30, 44f., 49, 51, 58, 93– 121, 197, 202, 225, 247, 259ff., 281, 295f., 299ff., 306, 311ff., 327 passivization 18, 44, 94f., 112, 119, 197, 200, 202, 260, 262, 288 patient 46f., 50, 199 periphery 8, 18f., 317 Perlmutter, D. 20, 294 personal pronoun 39, 174, 195, 254f. plural 44, 47, 158, 172, 190f., 193, 256, 268f. Pollard, C. 289ff., 327 Postal, P. 19f., 173, 228 postmodification 242, 247, 249ff., 269 postmodifier 106, 156, 210, 248f., 255, 280 postverbal 1f., 14, 39, 56, 71, 78, 80, 91, 93f., 96ff., 100ff., 104ff., 116ff., 120, 123, 128, 152, 157f., 164, 166, 190, 202f., 205, 211,

224, 231ff., 238, 241, 243, 246f., 253, 255ff., 260ff., 266ff., 273f., 286, 292, 296, 298f., 302, 306f., 309ff., 327 pragmatic 13, 15, 31, 36, 51, 145, 209, 213, 236, 244, 270f., 275f., 287f. predicate 3, 6ff., 10ff., 15, 17f., 24, 26ff., 37, 40, 51, 95, 98, 110, 123ff., 157f., 173, 179, 184, 206, 212ff., 221, 231, 237, 265, 270f., 274, 284, 293, 297, 303 Predicate Complement Unit (PCU) 98f., 101ff., 117f., 163, 238 predication 11, 17, 22, 24, 27, 105, 110, 124, 128f., 237, 239, 251, 262, 266, 271, 274f., 312 predicative 3, 11, 14f., 34, 47, 90, 106f., 121, 123ff., 150, 154, 158, 160ff., 167, 173, 179f., 186, 229, 231, 243, 265, 296, 311 predicator 13, 15, 38f., 127, 179 premodifier 142, 156 preposition (see also particle) 64, 93, 112f., 115, 230, 240, 272 prepositional 38, 43f., 93f., 96, 105, 109, 112f., 116, 120, 157, 210, 224, 299, 313 presentational 237f., 242, 251f., 256, 274, 276 pronoun (see also personal pronoun) 2, 24, 40, 44, 171f., 174, 177, 183, 192f., 195f., 198, 200, 203, 212, 244, 254, 259, 267, 283, 291 prop it 198, 203 proposition 27, 45, 237, 260 pseudo-cleft 287 psychological 7f., 30, 37, 51, 267, 273, 277, 284, 293 Pullum, G. K. 9, 13f., 25, 37, 43, 66, 124, 128, 178, 188f., 193ff., 199, 201, 204, 293f. putative 135f. question 10ff., 68, 158, 192, 194f., 203, 250, 253f., 257f., 303 Quirk, R. 9, 12f., 82, 86, 124, 171, 174, 188, 196ff., 204, 284 referent 20, 46, 48f., 80, 165, 180, 198, 244, 261, 267 referential 174, 196, 275, 283 reflexive 44, 196, 200, 259

Index | 349

relative clause 39, 71, 176, 210, 244ff., 268, 271 relative pronoun 244, 268 Rizzi, L. 24 Roe, I. 2, 27, 30f., 35, 41, 202 Rothstein, S. D. 5, 26f. Sag, I. A. 289ff., 327 Sampson, G. 16, 59, 61, 66f., 324f. Satzbauplan 96, 300 Satzmuster 96, 300 Schenkel, W. 2, 126, 298 Schierholz, S. 29, 57f., 279, 284 Schüller, S. (see also Faulhaber, S.) 3, 12, 29f., 34f., 46, 64, 71, 94, 96f., 109, 112, 121, 159, 163f., 188, 197, 201f., 238ff., 254, 263, 268, 273, 286, 300ff., 307, 321 semantic 7f., 12ff., 23, 26ff., 31f., 39, 45ff., 50ff., 71, 74, 78, 80f., 86, 91, 95ff., 102, 105, 107, 109, 112, 115, 118, 121, 123ff., 133f., 136f., 145f., 152, 157, 164, 181, 186, 198f., 201ff., 215, 229f., 235f., 241, 245, 251ff., 263, 271, 274, 276, 284, 287f., 292, 297f., 301, 310, 327f. semantic role 13, 23, 26f., 31, 46, 50, 91, 95, 97, 121, 164, 199, 202, 298 semantics 8, 36, 81, 86, 91, 133, 145, 153, 202, 216, 289 Seuren, P. A. M. 7f. Shlonsky, U. 24 Sinclair, J. McH. 19, 54, 109, 158, 276, 304f. small clause 21, 105, 108ff., 128, 255, 263ff., 271, 311f. spoken 44, 53, 56, 94, 173ff., 177, 205, 217ff., 269f., 324 Stefanowitsch, A. 19, 54f., 66ff., 283, 315, 325, 328 storage 30, 89, 146, 159, 187, 236, 252, 275, 297, 314, 317ff., 322, 324, 328 Subject Complement Unit (SCU) 34, 98f., 101ff., 117f., 163, 238 subjunctive 136 subordinate clause 13f., 37, 48, 152, 169, 181, 193, 204f., 207, 210ff., 229f., 307f.

tag question 40, 43f., 174, 193, 196, 200, 214, 253f., 258 tense 18, 23f., 39, 77, 79, 215f., 292, 321, 324, 328 Tesnière, L. 2, 8, 29f. theme 8, 44f., 51, 198, 200, 258, 261 theta role (see also semantic role) 2, 21ff., 27f., 274 traditional grammar 8f., 12, 18, 20, 22, 28, 50, 82, 93, 100, 105, 156, 158, 169, 179, 189, 235, 260f., 271, 274, 276, 288, 294 trivalent 97, 108, 166, 224, 227, 301 Uhrig, P. 56, 59, 64, 96, 129f., 132, 149, 153, 236, 275, 300, 302ff. valency carrier 3f., 29, 32f., 35, 56, 105, 111, 124ff., 164ff., 179, 186f., 203f., 220ff., 225f., 231, 235f., 268, 273, 275, 295f., 298, 304ff., 309ff., 319, 324, 327 valency construction 109, 121, 301ff., 327 valency pattern 96, 109, 115, 120, 144, 226, 235, 275, 298, 300, 302f., 305f., 313, 325, 327 valency slot 31, 35, 93, 95ff., 102, 105, 108, 114, 116, 156, 166, 179f., 228, 232f., 261, 273, 275, 298f., 301, 304 valency theory 1f., 5, 13, 22, 29ff., 33, 35, 52, 94, 109, 121, 128, 156, 163, 178f., 187, 235, 263, 273ff., 292, 295, 298, 306, 309, 327, 334 verbless clause 48, 109 Wasow, T. 209f., 308 Welke, K. M. 2, 25, 29f., 32ff., 95, 126 Williams, E. 19, 23, 263ff., 271f., 274 word class 62ff., 254f. written 32, 36, 41, 44, 53, 56, 110, 177, 217ff., 240, 249, 268f., 281 Zandvoort, R. W. 9, 12, 36