<<

Verb Valency Patterns Topics in English 71

Editors Bernd Kortmann Elizabeth Closs Traugott

De Gruyter Mouton Valency Patterns

A Challenge for Semantics-Based Accounts

by Susen Faulhaber

De Gruyter Mouton ISBN 978-3-11-024071-9 e-ISBN 978-3-11-024078-8 ISSN 1434-3452

Library of Congress Cataloging-in-Publication Data

Faulhaber, Susen, 1978Ϫ Verb valency patterns : a challenge for semantics-based accounts / by Susen Faulhaber. p. cm. Ϫ (Topics in English linguistics ; 71) Originally presented as the author’s thesis (doctoral Ϫ University of Erlangen and Nuremberg, Germany, 2009) under the title: Semantic aspects of verb valency. Includes bibliographical references and index. ISBN 978-3-11-024071-9 (alk. paper) 1. Grammar, Comparative and general Ϫ Verb phrase. 2. Seman- tics. 3. . I. Title. P281.F38 2011 4251.6Ϫdc22 2011009773

Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.

” 2011 Walter de Gruyter GmbH & Co. KG, 10785 Berlin/New York Cover image: Brian Stablyk/Photographer’s Choice RF/Getty Images Printing: Hubert & Co. GmbH & Co. KG, Göttingen ϱ Printed on acid-free paper Printed in Germany www.degruyter.com Preface

This book, which is a revised version of my doctoral thesis, submitted to and accepted by the Friedrich-Alexander-Universität Erlangen-Nürnberg in 2009 under the title Semantic Aspects of Verb Valency – The Relationship between Meaning and Form, would not have been possible without the help of so many people who supported me in various ways. In the following few lines a number of them are gratefully acknowledged for helping me in vari- ous ways – in the process of collecting and preparing the data and arranging the results into a presentable format or by providing valuable comments throughout different phases of the manuscript as well as providing encou- ragement and moral support. For any errors, shortcomings or inadequacies that may remain in this work the responsibility is, of course, entirely my own. My special thanks go to Thomas Herbst for arousing my interest in the fascinating ways in which behave – or do not behave as expected, for that matter –, for encouraging me to do research in this field and helping me in refining the topic. His support as supervisor of my thesis was truly exceptional, not only in providing numerous valuable suggestions and comments on various stages of the manuscript, but also by always affording the requisite moral support whenever needed. I am very grateful to Michael Klotz for his constructive feedback on various passages of earlier drafts of the manuscript and to Mechthild Habermann for her interest and suggestions concerning possible future research in this field as well as for her willingness to function as deputy supervisor. Bernd Kortmann, the co-editor of this book, provided helpful comments on the manuscript and gave the impetus for several worthwhile changes in the structure of the book for which I am much obliged to him. I would also like to express my gratitude to Birgit Sievert for her interest and support right from the start. I am much obliged to the many native speakers for sacrificing hours of their private time for uncountable interview sessions, patiently going trough ever new batteries of invented sentences: (in alphabetical order) Naomi Bishop, Amy Buer, Gayle Goldstick, Andrea Hunt, Turloch o’Brion, Elena Rippel, and Ned Reif. My special thanks go to David Heath and especially Kevin Pike for not only participating in the interviews but also reading the final manuscript and providing invaluable advice concern- ing both language, style, and also content. vi Preface

Peter Uhrig was a tremendous help concerning all questions of a tech- nical nature – from the subtleties of corpus research to formatting the ma- nuscript – but also in providing invaluable last-minute moral support, for which I am very grateful. I would also like to thank Christian Hauf for proofreading the manuscript, Matthias Limmer for checking example sen- tence references in the appendices, and all other colleagues and friends who contributed to this work and thereby helped to enhance the quality of . Finally my dearest thanks go to my family. My parents Helga and Manfred Schüller always believed in me and supported me in every con- ceivable way, for which I am deeply grateful. Moreover, I cannot thank enough my husband Detlef Faulhaber, who patiently accompanied me through all the ups and downs of writing a dissertation and finalizing a manuscript for publication, always encouraging me, always helping me with unforeseen formatting intricacies and always being my source of ener- gy whenever I seemed to be running out of it. Without their loving and consistent support, I would never have been able to stick to this path and to finish this book. It is dedicated to them.

Erlangen, February 2011 Susen Faulhaber Contents

Preface ...... v Figures ...... x Tables ...... xii Abbreviations ...... xv

1 Introduction ...... 1 1.1 Major questions...... 1 1.2 The model of analysis – overview and important concepts ...... 3 1.2.1 Valency models – a brief overview ...... 3 1.2.2 Valency carriers, complements and adjuncts, and different types of valency ...... 4 1.2.3 Valency patterns ...... 6 1.2.4 Optionality of complements ...... 8 1.2.5 unit (SCU) and complement units (PCUs) ...... 10 1.2.6 Levels of valency and the use of semantic roles...... 11 1.2.7 Valency constructions...... 15 1.2.8 Complement types ...... 16 1.2.9 Participant roles ...... 17 1.3 Design of the study ...... 20

2 The meaning of complements ...... 23 2.1 Complement types as part of a verb’s valency structure ...... 23 2.1.1 Overview ...... 23 2.1.2 Complement types ...... 23 2.1.3 Valency structures – introducing the model used ...... 29 2.2 Alternative realizations of participants ...... 32 2.2.1 Semantic specialization vs. synonymy of complement types .. 32 2.2.1.1 + talk + TOPIC ...... 32 2.2.1.2 AGENT + agree + TOPIC/REFERENCE ...... 36 2.2.1.3 AGENT + forget + ÆFFECTED ...... 47 2.2.2 Polysemous complement types ...... 49 2.2.2.1 AGENT + cheat + REFERENCE ...... 49 2.2.2.2 AGENT + argue + TOPIC/REFERENCE ...... 55 2.2.2.3 AGENT + inform + TOPIC ...... 60 2.2.3 Blocked patterns ...... 64 viii Contents

2.3 Conclusion ...... 67 2.3.1 No stable semantic properties ...... 68 2.3.2 Synonymous and polysemous complement types ...... 78

3 Pattern choice and verb meaning ...... 82 3.1 Assessing the role of verb meaning – theoretical assumptions .... 82 3.2 A of semantically similar verbs ...... 86 3.2.1 The database – introduction and description of methodo- logy ...... 86 3.2.2 Semantically similar verbs and their pattern inventories ... 92 3.2.3 Results of the analysis – quantitative evaluation ...... 114 3.3 Phenomena identified in the analysis ...... 124 3.3.1 Complementation options in a valency framework – an overview ...... 124 3.3.2 Same valency pattern but different participant pattern .... 128 3.3.3 Same pattern restricted to a very specific context ...... 134 3.3.4 Different formal realizations of the same participant ...... 141 3.3.4.1 Differences in optionality ...... 141 3.3.4.2 Different formal realizations of the same parti- cipant – an overview ...... 145 3.3.4.3 Different formal realizations of BENREC ...... 155 3.3.4.4 Different formal realizations of PREDICATIVE ... 159 3.3.4.5 Different formal realizations of TOPIC ...... 165 3.3.4.6 Different formal realizations of ÆFFECTED ...... 170 3.3.4.7 Different formal realizations of PREFERENCE and AIM ...... 189 3.3.4.8 Conclusion...... 191 3.3.5 Different flexibility in the combination of participants ... 193 3.3.6 Frequency-based differences in the choice of complements .. 196 3.3.7 Formal similarity between verbs of opposite meaning .... 207 3.4 Different aspects of verb meaning and pattern choice ...... 208 3.4.1 The participant inventory...... 209 3.4.2 Selection restrictions...... 212 3.4.3 The situation type of the verb ...... 223 3.5 Implications ...... 235

4 The meaning of patterns ...... 249 4.1 The pattern as an additional entity ...... 249 4.1.1 Complement type-independent pattern restrictions ...... 249 4.1.2 Participant mergers ...... 254 Contents ix

4.1.3 Instability of complement-participant correlation ...... 256 4.1.4 Realization of participants dependent on the overall pattern ...... 261 4.1.5 Conclusion ...... 263 4.2 The relationship between pattern and meaning ...... 264 4.2.1 Same valency pattern – same participant pattern? ...... 264 4.2.1.1 Theoretical background: assumptions of con- struction grammar ...... 264 4.2.1.2 The pattern [NP + verb + NP + NP] ...... 267 4.2.1.3 The pattern [NP + verb + for_NP] ...... 272 4.2.2 Same valency pattern – similar verb meaning? ...... 278 4.2.2.1 Verb class studies – a brief review ...... 279 4.2.2.2 Semantic verb groups based on pattern groups .. 284

5 Conclusion ...... 292 5.1 Different approaches towards the -semantics interface – an assessment in the light of empirical findings ...... 292 5.2 Summary of the results ...... 299 5.2.1 , selection restrictions, and participant inventories ...... 299 5.2.2 Competition with other lexical units of the same lexeme .... 302 5.2.3 No stable complement type or pattern meaning ...... 304 5.2.4 Verb meaning not predictable from pattern choice ...... 308 5.2.5 Accounting for alternative valency constructions ...... 309 5.2.6 Idiomaticity in complementation ...... 313 5.3 Implications for a theory of complementation ...... 317

Appendix 1 ...... 327 Appendices 2–5 are available at the following URLs: Appendix 2 ...... htttp://dx.doi.org/10.1515/9783110240788_appendix2 Appendix 3 ...... htttp://dx.doi.org/10.1515/9783110240788_appendix3 Appendix 4 ...... htttp://dx.doi.org/10.1515/9783110240788_appendix4 Appendix 5 ...... htttp://dx.doi.org/10.1515/9783110240788_appendix5

References ...... 329 Index ...... 351 Figures

Figure 1. Levels of disambiguation in the meaning of complements ..... 27 Figure 2. Simplified valency structure of talk (based on VDE: 850) ...... 31 Figure 3. Revised model of talk (‘speech’) ...... 35 Figure 4. Number of example sentences in the BNC for the lemma agree +on and +about in a window-span of 2–5 ...... 41 Figure 5. Simplified valency structure of agree ...... 46 Figure 6. Partial valency structure of forget (based on VDE: 326) ...... 48 Figure 7. Partial model for the valency structure of cheat ...... 51 Figure 8.  Modified model for the valency structure of cheat ...... 54 Figure 9. Partial VDE entry of the verb argue (p. 46) ...... 57 Figure 10. Valency structure of argue ...... 59 Figure 11. Valency structure for argue ‘dispute’ ...... 60 Figure 12. Valency structure for argue ‘discuss’ ...... 60 Figure 13. Valency structure of inform (cf. VDE: 428) ...... 63 Figure 14. Formal overlap, semantic, and unsystematic restrictions ...... 114 Figure 15. Proportion of different types of restrictions ...... 121 Figure 16. Complement block of blame (VDE: 86) ...... 124 Figure 17. Complement block of award (VDE: 61) ...... 127 Figure 18. Complement block of seem (VDE: 729) ...... 127 Figure 19. Frequency of complementation with [to_INF] and [V-ing] for the lemmata imagine and intend ...... 198 Figure 20. Frequency of complementation with [to_INF] and [V-ing] for the lemmata start and begin ...... 199 Figure 21. Make/making as collocates of START/BEGIN +[V-ing]/ [to_INF] ...... 200 Figure 22. Work/working as collocates of START/BEGIN +[V-ing]/ [to_INF] ...... 201 Figure 23. Talk/talking as collocates of START/BEGIN +[V-ing]/ [to_INF] ...... 201 Figure 24. Think/thinking as collocates of START/BEGIN +[V-ing]/ [to_INF] ...... 201 Figure 25. Look/looking as collocates of START/BEGIN +[V-ing]/ [to_INF] ...... 202 Figure 26. Cry/crying as collocates of START/BEGIN +[V-ing]/ [to_INF] ...... 202 Figure 27. Take/taking as collocates of START/BEGIN +[V-ing]/ [to_INF] ...... 202 Figures xi

Figure 28. Valency structure of force (based on VDE: 325)...... 250 Figure 29. Valency structure of hear (based on VDE: 380) ...... 253 Figure 30. Stemma for example sentence (7a) ...... 259 Figure 31. Stemma for example sentence (7b) ...... 259 Figure 32. Stemma for example sentence (8) ...... 260 Figure 33. Formal differences for [NP]+ verb +[NP]+[NP] for lexical units in VDE ...... 269 Figure 34. Participants realized by [for_NP] for lexical units in VDE ... 274 Figure 35. Idealized distribution of participants realized by [for_NP] for lexical units in VDE ...... 276 Figure 36. The want-group (based on the pattern [NP + VHCact + for_NP] in the Erlangen Valency Patternbank) ...... 285 Figure 37. Factors generally considered relevant for verb complemen- tation as opposed to empirical evidence presented in chap- ters 2 and 3 ...... 317 Figure 38. Idealized valency structure with semantic subtype specifi- cations ...... 319 Figure 39. Elements of verb complementation ...... 324 Tables

Table 1. Complement types used in the analysis (VDE: xv–xviii) ...... 18 Table 2. Semantic roles used for the description of participants ...... 19 Table 3. CGEL’s classification of prepositional phrases for comple- mentation (CGEL: 9.60–9.63) ...... 25 Table 4. Herbst’s classification of prepositional phrases for adjective complementation (Herbst 1983: 113–115) ...... 25 Table 5. Pattern choices *[on] (source: VDE) ...... 28 Table 6. Relevant complementation patterns of talk (VDE: 850) ...... 32 Table 7. Subtypes of cheat +[on_X] ...... 50 Table 8. Argue ‘dispute’ and argue ‘discuss’ (based on VDE: 46) ...... 56 Table 9. Selection of blocked alternations between [on_X] and [upon_X] ...... 66 Table 10. Variation between [on_X], [about_X], [over_X], [upon_X], and [of_X] ...... 69 Table 11. Verb groups used in the analysis ...... 87 Table 12. Symbols used for the assessment of pattern acceptability ...... 92 Table 13. Valency patterns of answer, reply, and respond in the sense ‘answer’ ...... 93 Table 14. Valency patterns of assume, suppose, presume, gather, and guess in the sense ‘assume’ ...... 94 Table 15. Valency patterns of choose, select, pick, elect, appoint, nominate, designate, and name in the sense ‘choose’ ...... 96 Table 16. Valency patterns of declare, pronounce, announce, pro- claim, and state in the sense ‘declare’ ...... 97 Table 17. Valency patterns of get, obtain, receive, acquire, buy, and purchase in the sense ‘get’ ...... 98 Table 18.  Valency patterns of hire, rent, and lease in the sense ‘hire’ .... 98 Table 19. Valency patterns of mean, intend, plan, and aim in the sense ‘intend’ ...... 99 Table 20. Valency patterns of love, like, and adore in the sense ‘like’ ... 100 Table 21. Valency patterns of keep, maintain, preserve, and sustain in the sense ‘not lose’ ...... 100 Table 22. Valency patterns of allow, permit, authorize, and entitle in the sense ‘permit’ ...... 101 Table 23. Valency patterns of train, practice, and rehearse in the sense ‘practice’ ...... 102 Tables xiii

Table 24. Valency patterns of propose, suggest, and recommend in the sense ‘propose’ ...... 103 Table 25. Valency patterns of quarrel, argue, dispute, and bicker in the sense ‘quarrel’ ...... 105 Table 26. Valency patterns of remember, recall, recollect, reminisce, and bear in mind in the sense ‘remember’ ...... 106 Table 27. Valency patterns of indicate, show, and demonstrate in the sense ‘show’ ...... 107 Table 28. Valency patterns of start, begin, and commence in the sense ‘start’ ...... 108 Table 29. Valency patterns of phone, ring, and call in the sense ‘telephone’ ...... 109 Table 30. Valency patterns of instruct, teach, train, and educate in the sense ‘teach’ ...... 110 Table 31. Valency patterns of consider, think, reflect, ponder, contemplate, and judge in the sense ‘think’ (part I) ...... 111 Table 32. Valency patterns of consider, think, reflect, ponder, contemplate, and judge in the sense ‘think’ (part II) ...... 112 Table 33. Valency patterns of respect, admire, and esteem in the sense ‘think highly’ ...... 112 Table 34. Valency patterns of bear, tolerate, and endure in the sense ‘tolerate’ ...... 113 Table 35. Valency patterns of urge, press, and push in the sense ‘urge’ .. 113 Table 36. Possible combinations of verb meaning and patterns ...... 126 Table 37. Verbs with obligatory complements which are (contextual- ly) optional with other verbs in the same group ...... 142 Table 38.  Different complements realizing the same participant ...... 146 Table 39. Different formal realizations of TOPIC including possible further conditions for the use of the pattern (+/–) ...... 166 Table 40. Selection restrictions of post verbal [NP]-complements ...... 215 Table 41. Situation types based on CGEL (4.27) and Vendler (1967) .. 226 Table 42. Verb groups with verbs which have the same situation type .. 228 Table 43. Distribution of to death in resultative constructions in the BNC (Boas 2003: 131) ...... 244 Table 44. Distribution of dead in resultative constructions in the BNC (Boas 2003: 130)...... 244 Table 45. Valency patterns of force (based on VDE: 325) ...... 249 Table 46. Verb groups for the valency construction [NP]_AGENT + verb + [NP]_ ÆFFECTED + [NP]_PREDICATIVE ...... 289 xiv Tables

Table 47. Verb groups for the valency construction [NP]_AGENT + verb + [NP]_BENREC + [NP]_ÆFFECTED ...... 290 Table 48. Variability of valency constructions for different lexical units which are semantically similar ...... 320 Table 49. Constructeme AGENT + verb + AFFECTED + PREDICATIVE ... 321 Abbreviations

Abbreviated literature CamG Huddleston, Rodney and Geoffrey Pullum (2002), Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. CGEL Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik (1985), A Comprehensive Grammar of the English Language. LDOCE Longman Dictionary of Contemporary English OED Oxford English Dictionary VDE Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz (2004), A Valency Dictionary of the English. Berlin: Mouton de Gruy- ter.

Sources of examples sentences The book contains numerous example sentences. Their respective sources are giv- en in the form of the following subscripts: BNC British National Corpus QE quoted example (source provided in the respective contexts) VDE Valency Dictionary of English (based on the COBUILD-Corpus) INV invented example sentence NSA example sentence assessed by native speakers

Additional example sentences as well as example sentences judged by native speakers (NSA) together with their respective assessment are given in the online appendices of this book (cf. Contents).

Other abbreviations and conventions [on_X] particle complement type containing the particle on; comple- mentation of on not specified, i.e. the particle can be complemented by different phrases or clauses (e.g. NP, V-ing, wh-to_INF), while no claim is made as to the actual range of different alternative com- plementations [on_NP] particle complement type containing the particle on complemented by a TOPIC participant which can be described by the semantic role topic [on_X]_TOPIC particle complement containing the particle on with the function to realize the participant TOPIC; For a detailed list of complement types and participant roles used see Table 1 and Table 2 in chapters 1.2.8 and 1.2.9.  1 Introduction

1.1 Major questions

The objective of this work is to throw light on the question if – or rather to what extent – different formal uses of a verb are directly connected to its meaning and/or the meaning of its complement(s). An extreme version of this would be syntactic predictability on the basis of semantic rules: tied to each verb and/or complement sense there is one or possibly a limited num- ber of specific formal realizations which are activated alongside their meaning, leading to a one-to-one relationship and link between syntax and semantics. The extreme version in the other direction would be complete arbitrariness between meaning and form. Noël (2003) points out that this relationship has been subject to debate for a considerable time:

Syntax without semantics/No syntax without semantics: in the course of half a century the pendulum of modern linguistics has completed its swing. After a few decades of syntax with as little meaning as possible in the third quarter of the 20th century, it has since become fashionable to adhere to the creed that literally everything in syntax is meaningful and that the linguist’s task is to elucidate the meaning of form within a – so-called “functional”, as opposed to “formal” – theoretical model that coherently links up syntax and semantics. (Noël 2003: 347)

The question addressed here is concerned with the extent of the predicta- bility of form from meaning in the form of a corpus-based as well as cor- pus-driven analysis of complementation patterns of English verbs. The framework chosen for this is valency theory, which allows for a lexical view of language that is unbiased with respect to the question of the role meaning plays for complementation. Of interest is the actual range of com- plementation patterns different verbs can be found with as well as those patterns which do not seem to be available. Which complements a verb combines with is a question of both psycholinguistic and theoretic relev- ance. The occurrence of a pattern in combination with a verb is sometimes described as a “choice” of the speaker in favor of one pattern or against another pattern which should theoretically be possible, i.e. semantically plausible, but does not constitute a grammatically acceptable pattern. Such 2 Introduction a non-occurrence of a pattern has been referred to as a syntactic gap, for instance by Mukherjee (2005: 12–13):

Unfortunately, though, the form *She addressed the children her remarks does not exist. One could hypothesise a ‘syntactic gap’ here. That is to say that the language system as such would allow for this formal realisation – because of the aforementioned structural and semantic analogies – for no apparent reason, it runs counter to the actual norm in language use and is thus not attested.

In this respect also system and norm as introduced by Coseriu ([1970] 1979: 57) are useful concepts, where the former is described as a “system of possibilities” and “a collectivity of freedom rather than requirements” and the latter as the “system of obligatory realizations” which “coerces speakers in their actual choice of expression” in the sense of providing “boundaries of traditional realizations” [translation mine]. Thus while a certain pattern could be considered as part of the system – due to its exis- tence in combination with a semantically similar verb – it is not necessarily part of the speakers’ norm, i.e. they simply do not use the pattern. Such gaps could be either completely idiosyncratic, i.e. a matter of convention, or they could be systematic in the sense that the meaning of the comple- ment type might not match with the meaning of the valency carrier. Thus, instead of assuming irregularities, one could assume semantic reasons for the unacceptability of such a pattern. If this is the case, the concepts of gap and irregularity are misguided, since then the meaning of the verb and the whole construction simply contradict each other, blocking the combination in actual language use. It is this line of thinking that can be found in many approaches as indicated in the quote by Noël (2003) above, which will be discussed below. In that sense irregularities as regards the syntax-semantics interface are just cases for which the semantic explanation has not yet been identified. According to this view such explanations must exist on principle as syntax is determined through meaning. The model of analysis 3

1.2 The model of analysis – overview and important concepts

In order to set the ground for describing the methodology used in the at- tempt to produce empirical evidence for either of the two points of view, the valency framework used as a model for the analysis as well as other relevant notions such as constructions and semantic roles will be briefly introduced.

1.2.1 Valency models – a brief overview

In the framework of valency, the verb is considered to be the most central element of a sentence and the major determinant of its structure. The ori- gins of valency theory are to be found in dependency grammar, especially in the work of Tesnière (1959). Dependency is concerned with the interde- pendencies of the different parts of sentences, which result from the va- lency of the single words that are part of the sentence (cf. Vennemann 1977; Heringer 1993; for an overview see Herbst and Klotz forthcoming 2012). A stronghold of valency research can be found in the description of the German language (e.g. Ágel 2000; Engel 1977; Helbig 1992; Heringer 1996; Welke [2002] 2005; Habermann and van der Elst [1985] 1997) and Romance languages (e.g. Busse and Dubost [1977] 1983; Gärtner 1998). As regards the English language, there are for example Emons (1974, 1978), Allerton (1982), Herbst (1983, 1988), von Randow (1986), Herbst and Schüller (2008), and the Valency Dictionary of English by Herbst, Heath, Roe, and Götz (2004) (henceforth VDE). The main ideas of valency theory also have increasing influence in the context of cognitive grammar (e.g. Croft and Cruse 2004; Fillmore 2007) and they are advanced by a general turn to more lexis-oriented approaches, e.g. Hudson’s Word Gram- mar (e.g. 1984, 1990). Because of its orientation towards item-specific properties, valency theory seems to be predestined to a close cooperation with lexicography which has so far resulted in a considerable number of so- called valency dictionaries, especially for German, e.g. Wörterbuch zur Valenz und Distribution deutscher Verben (Helbig and Schenkel [1969] 1973), Kleines Valenzlexikon deutscher Verben (Engel and Schumacher 1976), Verben in Feldern (Schumacher 1986), VALBU (Schumacher et al. 2004), Wörterbuch der Valenz etymologisch verwandter Wörter (Sommer- feldt and Schreiber 1996), and with VDE (Herbst et al. 2004) also for the English language. In this respect, valency theory – among others – can also be seen as being in the lead of a more general trend: the awareness of the importance of lexical phenomena (cf. Herbst 1999a: 167). 4 Introduction

1.2.2 Valency carriers, complements and adjuncts, and different types of valency

Different models vary with respect to the question as to which word classes they ascribe the property of valency, ranging from valency being a verb- specific property (e.g. Eichinger 1995) to approaches which besides nouns and adjectives also include prepositions and adverbs (Herbst and Schüller 2008; see also Huddleston and Pullum 2002). In order to account for the fact that words with more than one meaning are often distinguished by the choice of patterns they can be found in, a more precise analysis takes the lexical unit as a form-meaning pairing as the actual valency carrier (cf. for instance Cruse 1986: 80). However, what exactly constitutes a lexical unit, i.e. into how many different units a lexeme must be subdivided, is by no means easy to decide (cf. for instance Klotz 2000: 120–126, Sinclair 2004: 24–47, and Àgel 2000: 115–117 for a discussion of this problem). A major distinction is made in valency theory between complements and adjuncts, a distinction which goes back to Tesnière’s (1959) definition of actants and circonstants, with the help of which dependency relation- ships within sentences can be described. Obviously, such a distinction can be found in different shapes in various grammatical models also beyond de- pendency frameworks. In valency theory, it is directly taken up in the dis- tinction between complements – which are demanded by the verb’s valency or at least determined in their form by it – and adjuncts – which are not de- termined in their form by the verb, can often be freely added or omitted, and assume the role of sentence-modifying elements (Tesnière 1959: 238). Tesnière bases his distinction on the following criteria: form (nominal for actants vs. adverbial for circonstants), obligatoriness – which is only seen as a tendency –, and semantic descriptions which are conceptually very similar to semantic roles (1959: 107–111). A distinction between the two is by no means unproblematic (cf. for instance Herbst and Schüller 2008: 113–116). Helbig (1992: 78–87) provides a survey of tests which have been used to differentiate between complements and adjuncts, yet they do not provide a 100% differentiation either. Herbst and Klotz (forthcoming 2012) conclude that “[i]t is fair to say that over the years valency grammarians have spent a considerable amount of time and effort in an attempt to clarify and operationalise the distinction without arriving at a solution which could be generally considered satisfactory”. As a consequence, different ap- proaches vary as to whether the distinction is seen as a gradient, in the sense that the two notions are to be found at opposing ends of a continuum with various intermediate states (Vater 1978: 14–15; Heringer 1984: 35), or The model of analysis 5 whether a prototypical approach is taken (e.g. Engel 1992), where the dis- tinction is relativized by accepting better and worse exemplars (Àgel 2000: 199). Typically, a differentiation between different types of valency – quantit- ative and qualitative valency – is made. Thus lexical units open up a certain number of slots which can or have to be filled by certain (optional or obli- gatory) complements. These complements are determined in their form by the valency carrier. Thus qualitative valency is concerned with a descrip- tion of the complements from a formal-functional point of view, while there is a certain variation with respect to its exact definition in different frameworks: whether complements are described as syntactic-functional categories or with the help of morphological-formal criteria depends on the model chosen. A description ideally tries to integrate both aspects. While Emons, for example, defines different complement classes on the basis of the commutability of the phrases or clauses (1974, 1978), Allerton (1982) basically resorts to traditional functional categories like subject and and excludes most non-NP complement types, while extending the re- pertoire by a number of additional categories such as objoid and perject. His model also differs from other surface-oriented approaches in that he considers active and passive patterns and certain other alternations as “transformationally related structures” (Allerton 1982: 31). In German lin- guistics, the classification typically rests on inflected case, so that Schu- macher (1986) describes complements e.g. as NomE (nominative comple- ment), AkkE (accusative complement) or as different prepositional com- plements.1 However, complements cannot be identified clearly enough on the basis of their commutation properties (Herbst 2003: 66–67; Fischer 1997: 153–154) and a purely functional description has the disadvantage that restrictions as regards the formal realization of the complement do not become clear and have to be indicated in an additional step. To avoid this, the Valency Dictionary of English (VDE), which procedes from the point of view of the lexical item and not from that of an existing sentence, provides a purely formal description of the complements at the outset. This formal description is equipped with additional information as to the functional potential of the complement via subscripts which indicate (1) the potential of a complement to function as subject in an active and/or a corresponding passive sentence (a property traditionally attributed to objects) and (2) the

1. This could be considered problematic when taking into account that a “dass- Satz” must also then be considered a nominative complement. For a justifica- tion of this approach see Schumacher (1986: 22). See also Schumacher et al. (2004) and Helbig and Schenkel (21973) for a similar approach. 6 Introduction order of the complements in the case of two post-verbal noun phrases (thus accounting for the traditional distinction between direct and indirect object) (VDE: xii).

1.2.3 Valency patterns

The formal description as opposed to a functional one raises the overall number of valency patterns (combinations of valency carrier and comple- ments): a pattern including an object may subsume – depending on the verb – a pattern with a noun phrase, a that-clause and a wh-clause in the same position. Although the underlined parts in (1)–(3) are not necessarily iden- tical, they still realize the same semantic function, i.e “what is being thought”.

(1)VDE Barge people are very often alone and have time to think grand though- ts. (2)VDE I don’t think that that is entirely true. (3)VDE I’ve never thought why I like something or why I don’t like something.

A valency pattern can be defined as the simultaneous choice of one or a number of complements in combination with a verb functioning as valency carrier. This makes it a purely formal entity which has to be distinguished from mixed categories such as constructions, which must be seen as form- meaning pairings (see below). A pattern centers on the verb as predicator or valency carrier, i.e. as the element that has valency, and contains further complements whose realization depends on this valency and which are therefore determined in their form by the verb. Adjuncts, which are neither demanded nor determined in their form by the verb, are not part of the pat- tern, even if they can have a certain influence on the choice of pattern (cf. 4.1). The notion of pattern is also evident in the form of so-called Satz- baupläne (‘sentence construction plans’) in Engel (1977) and Engel and Schumacher’s (1976) Kleines Valenzlexikon deutscher Verben or the pat- terns provided, for instance, in learner’s dictionaries such as the Oxford Advanced Learner’s Dictionary (7th edition, henceforth OALD7). Patterns also play an increasingly important role in current approaches towards va- lency (e.g. Herbst 2009, Herbst and Schüller 2008), which is also evidenced by projects like the Erlangen Valency Patternbank (cf. 1.1).With patterns, the goes beyond the description of single complements, as well as beyond the whole range of complements a verb can generally combine with The model of analysis 7

(i.e. its complement inventory). Access to machine-readable corpora, which allows more and more reliable statements about the frequency of co-occur- rence of specific combinations in language use, clearly facilitates the analy- sis of patterns. This is also reflected in an increased interest in the nature of chunks that are larger than a single word. The proposal that a speaker, when constructing a sentence, proceeds with the help of such chunks has been put forward in the form of Sinclair’s idiom principle (1991: 110). Patterns also play an ever more influential role beyond the scope of Sinclair’s research, especially with respect to language acquisition and sentence processing in a cognitive linguistic sense (e.g. Tomasello 2003) but also in more pedagogi- cal endeavors, e.g. the use of patterns in publications such as the Cobuild series’ Grammar Patterns (Francis, Hunston, and Manning 1996) and the so-called Pattern Grammar (Hunston and Francis 2000). With respect to descriptive syntax it is especially valency theory which takes a leading role. The kinds of patterns that are of interest here are rather abstract and purely formal patterns which can be represented for instance as:

(A) [NP]+ verb +[0] (B) [NP]+ verb +[NP] (C) [NP]+ verb +[in_NP]

Pattern (A) is a monovalent pattern which could, for instance, apply to the verb sleep. [NP] stands for a noun phrase complement and [0] is used as a convention to indicate that in this pattern the verb is used without a com- plement in the predicate. Pattern (B) is a divalent pattern which subsumes both monotransitive clauses, e.g. Peter likes Mary, as well as copular claus- es, e.g. Peter is a teacher, as described for instance in the Comprehensive Grammar of the English Language (Quirk, Greenbaum, Leech, and Svartvik 1985; henceforth CGEL). A differentiation of the two could be given by subscripts which indicate the potential of the post-verbal noun phrase to become the subject of a passive equivalent of the clause (cf. VDE) or by indicating the semantic role realized by the noun phrase (see below). Due to the size of the sample, the present analysis had to be confined to active patterns, i.e. passive patterns will not be taken into consideration unless no active examples were available. Consequently the abstract pattern description is always given in the active , where the element func- tioning as subject precedes the verb and other elements follow it, including those which can potentially become the subject of a passive clause.2

2. In order to achieve a conservative estimate of the actual number of patterns a verb can be found with rather than overexaggerating it, a rather generous ap- 8 Introduction

Where necessary, the different syntactic functions are pointed out and a differentiation is provided by means of semantic roles, i.e. [NP]_AFFECTED in the first and [NP]_PREDICATIVE in the second example (for a description of such participant roles see below). Pattern (C) is another divalent pattern, this time including a particle complement: [in_NP] represents a comple- ment headed by the particle in and complemented by a noun phrase, as for instance in the company in He invested in the company (terminology based on Herbst and Schüller 2008: 61–67).3 For a full list of all complement type labels used see below.

1.2.4 Optionality of complements

A valency pattern is characterized by the fact that it contains complements but not adjuncts. As indicated above, it is usually assumed that comple- ments can be optional, which means that they cannot be distinguished from adjuncts by a simple deletion test (cf. Starosta 1988 for a diverging ap- proach). If a sentence is analyzed in terms of a valency-theoretic approach, the non-verbalization of a participant is often referred to as deletion of an optional complement (e.g. Allerton 1982) or as valency reduction (e.g. Welke 1988; Helbig and Schenkel 21973).4 This is unnecessary, however, if valency is regarded as a lexical phenomenon: if an optional complement is not realized, it is neither descriptively necessary nor is it cognitively plausi- ble that this complement is first regarded as “existing” and then as “de- leted”, especially since this implies “verbatim recoverability … [which] is

proach was taken with respect to passivizability: a pattern was also rated as acceptable, when only examples in the were identified in the BNC or accepted by native speakers, while no claims are made here as to a general passivizability of patterns. Strictly speaking, the active and the pas- sive version of a pattern can be described as two different patterns, since there are cases where verbs allow a certain combination of complements only in the active or only in the passive voice. This approach has been adopted in the Er- langen Valency Patternbank (www.patternbank.uni-erlangen.de), which pro- vides online access to the valency patterns of VDE. For a similar treatment of passives, cf. Lexical Functional Grammar, where passives are seen as a lexi- cal phenomenon and not as the result of a syntactic transformation (Bresnan 2001: 30). 3. Alternatively, such a sequence is often analyzed as a prepositional verb (CGEL: 16.5). For a discussion see for instance chapters 2.1.2 and 4.2.2.1. 4. Cf. the notion of null-instantiation of frame elements in FrameNet (Fillmore 2007). The model of analysis 9 by no means a necessary condition for this kind of use” (Herbst 1996: 182– 183). This approach is also supported by the fact that the notion of deletion fails to account for collocational differences between patterns with and without an optional complement (Klotz 2000: 190–209). Taking a strictly lexical point of view, VDE, for instance, indicates the minimum and maxi- mum valency of each lexical unit in active as well as in passive uses, thus accounting for the range of possible uses of each lexical unit (VDE: x). A distinction is usually made between completely optional complements such as a book in He was reading a book, and contextually optional com- plements such as to the proposal in He objected to the proposal, which can only remain unexpressed if the concept has already been mentioned in the context of the sentence (e.g. Allerton 1975, 1982; Herbst 1996, 2003; Herbst and Schüller 2008: 111). For the concept of pattern, this has the consequence that in case of an optional complement, two patterns can be established, one with the optional complement, as in (4), and one without, as in (5).

(4) [NP]+ verb +[NP] (e.g. He was reading a book) (5) [NP]+ verb +[0] (e.g. He was reading)

VDE follows this line of , taking as a starting point the potential of the predicator to establish grammatically correct sentences, so that the pattern in which an optional complement is realized (4) stands on equal footing with the pattern where it is absent (5) (and with still further patterns which differ in other ways). Only obligatory complements, which always have to be verbalized in a pattern of a verb meaning in order for the verb to be used in a grammatically correct sentence, are marked as obligatory in VDE. The definition of optionality also rests on what is taken as the basis for this distinction. If the sentence is taken as the basis, complements func- tioning as subjects are for instance generally obligatory in active declara- tive sentences. From a lexical point of view, on the other hand, only those complements are considered as obligatory which have to be realized under all circums- tances. This means that the potential of a verb as regards complementation is taken as the basis for a distinction between obligatory and optional. This can be captured by different types of necessity: valency necessity, resulting from the valency of the predicator, structural necessity, resulting from the choice of sentence type, and communicative necessity, which takes prag- matic aspects into consideration such as whether a certain complement (or ) is necessary because it has a special or decisive communicative value. 10 Introduction

(6a)VDE I admire your courage in saying no. (7)VDE I admire what you’re doing and I’ll help. (6b)NSA *I admire. (6c)NSA *Admires your courage in saying no. (6d)NSA Your courage in saying no was admired (by me). (6e)NSA Admiring someone’s courage is helpful. (8)BNC Admire the elegance of the villa at Langton. B3K 2361

In (6a) the obligatory complement is a noun phrase expressing the “THING ADMIRED” and in (7) it is a wh-clause. While this variation is possible, the complement cannot remain unexpressed as in (6b). (6c) is also ungram- matical, yet the complement expressing the “ADMIRER” can in general be left out, e.g. in a passive construction, as in (6d), or a nonfinite construction as in (6e). The ungrammaticality of sentences such as (6c) can conse- quently not be due to the valency of admire but is to be found in the fact that this active declarative sentence lacks a subject which contradicts the structural requirements for this sentence type (Herbst and Roe 1996: 188– 189). Hence the ungrammaticality of (6c) results from a breach of structural necessities, which do not hold in case of constructions such as (6d) and (6e), or imperative sentence types such as (8).

1.2.5 Subject complement unit (SCU) and predicate complement units (PCUs)

Such considerations have been the basis for the model developed in Herbst and Schüller (2008), in which we combine valency with a linear sentence analysis, thus providing for the level of valency necessity as well as the level of structural necessity. Thus a complement such as your courage in saying no in (6a) above can be assigned to a predicate complement unit (PCU), since it is a complement that occurs in the predicate of the sentence. The complement units can receive a formal description such as [NP] which can be supplemented by a semantic role description such as AFFECTED, i.e. [NP]_AFFECTED. By including a semantic role description, the model is also easily compatible with construction grammar approaches without hav- ing to claim that form and meaning are necessarily fixed (see below). The complement filling the subject slot of a sentence (SCU) differs from other complements in that it strongly depends on the sentence type and not only on the valency carrier. Depending on the verb, and possibly also other contextual factors, the slot can be filled by noun phrases ([NP]), that- clauses ([that_CL]), participle clauses ([V-ing]), or to-infinitive clauses The model of analysis 11

([to_INF]). The formal realizations of this particular slot appear to be less verb-dependent than those of complements occurring in the predicate (PCUs). Moreover, the search for subject complement units is much more difficult in a non-parsed corpus, which is why the classification of subjects in VDE predominantly rests on native speaker assessments. For this reason, the focus here will be exclusively on verbal complementation in the predi- cate. Consequently, valency patterns are usually represented in a reduced form, where only the predicate is formally specified (e.g. [SCU]+ verb +[NP]+[NP]). This is based on methodological reasons and does not entail any theoretical assumptions, neither about the status of the subject com- plement unit nor about the “rest”-pattern as an independent unit. In passive clauses, the complement functioning as the subject functions as 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 pas- sive clauses depends on the verb’s possible predicate complement units and will therefore be included in the analysis.

1.2.6 Levels of valency and the use of semantic roles

It is common in valency-oriented frameworks to distinguish between dif- ferent levels of valency. Syntactic valency, as described above, is under- stood as the concrete realization or filling of the valency slots of a verb (Helbig 1992: 9). From a semantic point of view, these valency slots re- present participants, i.e. semantic concepts involved in the action described by the verb (e.g. the person enacting a certain activity, the thing against which an action is directed, etc.) and realized by complements. The de- scription of participants is part of the semantic valency of a word. Semantic valency is defined by Helbig (1992: 18) as “semantic-functional characteri- zation of the arguments [i.e. valency slots] in relation to the lexicalized predicate in terms of semantic case” [translation mine].5 As a possible third level, the notion of logical valency was discussed as reaching out to the

5. While in German linguistics the term Argument is usually used to describe the semantic equivalent of a complement or the valency slot itself, in English and especially American linguistics, the terms argument and argument structure are used as formal-functional categories. I will use the terms complement for the formal syntactic aspect (in part including functional properties) and partic- ipant for the purely semantic aspect (unless other frameworks are directly quoted). The term participant has been used in this way, e.g. by Halliday (1970), while Croft and Cruse (2004: 269) use the term concept. 12 Introduction fields of pragmatics, which depends on the circumstances of the utterance (Helbig 1992: 18). Helbig (1992: 154–155) extended this finally to six lay- ers including (I) the semantic structure in the sense of the number of participants, (II) inherent semantics influencing the selection of comple- ments, (III) semantic case, i.e. the type of participants, (IV) selection re- strictions of the complements, (V) qualitative valency (function within the clause and morphological representation), and (VI) quantitative valency. The question of how these levels are related to each other, especially the semantic and the syntactic structure, and to what degree they correspond to each other, is a central issue of the present analysis and will be explored in more detail in the chapters to come. Since the focus is on the syntax-se- mantic interface, it is crucial to keep the levels of form and meaning strictly apart in order to be able to analyze how exactly they are related. One of the most difficult problems concerning semantic valency is a methodological one which results from the chosen form of representation, i.e. semantic roles. These can be defined as “a predetermined set of labels that identify arguments according to the semantic relation they bear to their verb; each verb is associated with the relevant list of semantic roles” (Levin and Rappaport Hovav 2005: 35). One of the first to really establish a list of semantic roles and to introduce it to a syntactic analysis of English was Fillmore in his seminal paper “The Case for Case” (1968) (see also Gruber 1965). This approach was originally embedded in the generative frame- work, and Fillmore’s so-called deep cases are to be understood as expan- sions and modifications of morphological case used “to identify the under- lying syntactic-semantic relationship” (Fillmore 1968: 5–7 and 21) and they “comprise a set of universal, presumably innate, concepts which identify certain types of judgments human beings are capable of making … about such matters as who did it, who it happened to and what got changed” (Fillmore 1968: 24). Deep cases were thus not only used to describe sen- tence structures but also to relate semantically parallel sentence structures, i.e. sentences with different surface structures but similar deep structures. The notion of semantic roles, which also appears in Halliday (1967a) about the same time, has since been adopted in various linguistic theories. Helbig (1979), for instance, employed them for the description of a semantic level of valency and also Chomsky ([1981] 1984) introduced what he called ș- roles into his model of syntax. According to Levin and Rappaport Hovav (2005: 4), semantic role lists are “one of the simplest – and, possibly, most commonly adopted – forms of lexical semantic representation, although … a form that also suffers from severe drawbacks”. Irrespective of the frame- work, using semantic roles compels one to make a choice concerning the The model of analysis 13 number and type of roles. The general difficulty is that it is impossible to exclude a certain degree of subjectivity when assigning semantic roles. There is no formal criterion to verify any decision as to what is the most appropriate choice in a specific situation, which is why different models work with different numbers of semantic roles (cf. e.g. Smith 1999: 73). Semantic roles are useful as a means of generalizing the propositions a verb can express by characterizing the different participants according to their function. In order to provide for the possibility of generalization, se- mantic roles should be defined independently of the meaning of the verb (cf. also chapter 3.1). Verb-specific roles such as “ADMIRER” and “THING ADMIRED”, have been criticized for obscuring similarities between different verbs, e.g. Ackerman and Moore (2001: 20) argue that “this kind of con- crete and predicate-specific characterization of thematic roles provides a maximal differentiation of arguments, but raises the question as to what, if any, properties are shared among the arguments of different predicates … [and] whether the properties employed … are grammatically relevant or merely diacritic”. Thus, the number of semantic roles used in a model should be kept to a minimum to enhance options of comparison which are otherwise obscured, rendering the number of roles as unlimited as the num- ber of verbs. The fact that parallel sentences can be related to each other is one of the great benefits of the notion of semantic roles, which is also of crucial importance for the question of the relationship between meaning and form across verbs which will be examined here. However, it has to be taken into consideration that the use of more general roles also entails cer- tain problems. Croft (1991: 157) points out that “[u]nfortunately, the reduc- tionistic approaches all share the problem of vagueness and overgenerality in attempting to account for the richness of typological data”. Consequent- ly, there are also theories which operate with more verb-specific roles de- spite the drawbacks mentioned above, e.g. Fillmore’s FrameNet Project or Goldberg’s (1995, 2006a) participant roles (the latter constituting a separate level besides more general argument roles). In FrameNet, which aims to “document the range of semantic and syntactic combinatory possibilities (valences) of each word in each of its senses” (http://framenet.icsi. Berke- ley.edu), Fillmore tries to strike a happy medium between lemma-specific and generalized semantic roles by establishing different universal frames with certain frame-specific roles (Fillmore 1975: 124). FrameNet provides a semantic description of a considerable number of verbs with the help of such frame elements. These differ from verb-specific participants in that the notion of frame element rests on the idea that they are applicable to seman- tic entities larger than just a single verb’s specific semantic context yet 14 Introduction without aiming at complete generalizability. The verb question, for exam- ple, is assigned to the questioning frame – together with other lexemes such as ask, query, interrogation, and questioning – which has the core frame elements ADDRESSEE, SPEAKER, MESSAGE, and TOPIC. The advantage of such an approach is that it helps to reduce the danger of overgeneralization, i.e. the failure to miss finer nuances of meaning which remain unexpressed if a rather general role label is used for the description. This advantage comes to bear if the purpose is to provide a detailed description but less so, if the goal is a comparison of different verbs and structures as is the case in this work. Obviously, it has to be taken into account that if the roles are a means of describing psychologically relevant information, this is a different question altogether. Still other approaches try to reduce the number of roles by establishing metaroles which subsume other role types. Dowty (1991), for instance, sug- gested PROTO-AGENT and PROTO-PATIENT as thematic proto-roles instead of what he calls role types:

… we may have had a hard time pinning down the traditional role types be- cause role types are simply not discrete categories but rather are cluster concepts, like the prototypes of Rosch and her followers (Rosch & Mervis 1975). And when we accept that arguments may have different ‘degrees of membership’ in a role type, we can see that we really need only two role types to describe argument selection efficiently. (Dowty 1991: 571–572)

Following Tarvainen (1985: 15–16), who argues that the number of roles should depend on the purpose, this present approach will make use of a number of roles that exceeds Dowty’s proto-roles but is still limited, since the objective is mainly to provide a descriptive analysis which allows for a comparison of different verbs. However, if formal differences are to be re- traceable to semantic differences, this might only show when a finer se- mantic classification is used. Therefore, such a classification will be pro- vided where necessary in a second step (for the notion of subtype intro- duced for this purpose see below), after groups for comparison are estab- lished based on more general semantic criteria. It has to be noted here that the description of pattern meaning by means of roles, whether verb-specific or general, has been criticized as insufficient to explain the formal realizations of verbs. In terms of event semantics, for instance, aspectual factors are considered as the basis for verbal pattern choices. This way, grammatical form is still traced back to verb meaning, while now the verb’s immediate context is taken into consideration as well (e.g. Tenny 1994). Such factors will also be taken into consideration. None- The model of analysis 15 theless, for a semantic description of a valency pattern, semantic roles are indispensable.

1.2.7 Valency constructions

Semantic participant patterns are realized by formal valency patterns. Fol- lowing our approach introduced in Herbst and Schüller (2008: 139–141), the combination of valency pattern and participant pattern is referred to as valency construction. The relation between these two entities needs some further clarification. We take it that one valency pattern might realize more than one participant pattern. valency pattern: [SCU]+ call +[NP]+[NP] participant patterns: examples: AGENT + verb + BENREC + AFFECTED He called her a taxi. AGENT + verb + AFFECTED + PREDICATIVE He called her a fool.

At the same time one participant pattern can be correlated with more than one valency pattern. participant pattern: AGENT + verb + AFFECTED + PREDICATIVE valency patterns: examples: [SCU]+ consider +[NP]+[NP] He considered him a suitable candidate. [SCU]+ consider +[NP]+[AdjP] He considered him suitable.

In this sense, the present approach differs from construction grammar and similar approaches (e.g. Fillmore 1988; Fillmore, Kay, and O’Connor 1988; Kay and Fillmore 1999; Goldberg 1995, 2006a; Croft and Cruse 2004) which take the combination of meaning and form to be fixed in the sense that each formal pattern is the representation of a certain meaning and a difference in form is generally accompanied by a difference in meaning. Constructions are defined as “form-meaning pair[s]”, which – similarly to morphemes – are taken to be basic units of language (Goldberg 1995: 4). The construction itself is taken to carry meaning which must be added to the lexical verb meaning and the meaning of the other specific lexical items which are placed into the construction frame. The basic meaning of the lexical units thus remains stable, while different nuances of meaning within different uses of the same verb result from the construction meaning (Goldberg 1995: 11). This definition of construction is mainly concerned with so-called argument structure constructions. These can be said to have abstract positions, which can or must be filled by the formal expression of 16 Introduction the participants of a verb. The lexical items which can fill the slots of a construction need to fulfill certain requirements, i.e. they might be specified both semantically, formally, and sometimes even lexically in the sense that “only verbs can be used in this construction which allow for a ‘fusion’ be- tween the ‘participant roles’ of the verb (e.g. ‘giver’, ‘receiver’ and ‘thing given’ for the verb give) and the argument roles of the construction” (Mukherjee 2005: 51). Acknowledging the usefulness of constructions as an additional level of analysis, the combination of valency pattern and par- ticipant pattern is referred to as valency construction, e.g. He called her a taxi can be described as [NP]_AGENT + call +[NP]_BENREC +[NP]_AF- FECTED.6 Allowing for a separation of meaning and form as well as for a combination of the two has the clear advantage of neither prejudging rela- tions between different semantic patterns and one formal pattern, nor rela- tions between different formal patterns and the same semantic pattern from the start. It is this flexibility that allows an examination of the relationship between the levels of form and meaning in the first place.7

1.2.8 Complement types

The formal repertoire of complement types is limited when no lexically- specified complements are taken into consideration (such as sick in report sick, which cannot be replaced by any other adjective). The fact that such lexically-specified complement types exist shows that a list of abstract complements is – to a certain extent – an idealization. The fact that feel is less limited than report when it comes to complementation with adjective phrases does not mean that it can really be complemented with every possi- ble adjective phrase. The limits are set by what could be called world knowledge, i.e. people are likely to feel good and bad allowing for numerous shades in between,

6. For partially lexically filled constructions see, for instance, Fillmore, Kay, and O’Connor (1988). 7. One sentence can consist of several constructions, i.e. the final construction can have “multiple parents” (Croft and Cruse 2004: 264–265), which also ap- plies to valency patterns as defined above, i.e. they can combine with con- structions such as the negative construction or a question construction. These are covered under the label “sentence types” in Herbst and Schüller (2008), which allows us to emphasize the fact that these reflect general properties of sentence structure and thus to distinguish them from item-specific, i.e. valen- cy properties. The model of analysis 17 while to feel yellow does not appear to be a sensible proposition. The most frequent adjective collocate of feel in the BNC (window span of +1 follow- ing the node) is better with 643 occurrences, followed by guilty (516), sor- ry (501) and good (372). The item with the highest log-likelihood value is guilty (approximately 13% of all uses of guilty in the BNC are in this pat- tern with feel) and mutual information gives soothed as the most important collocate (all 5 occurrences of this word in the BNC are in combination with feel). This clearly shows that taking frequency information into con- sideration provides an additional dimension for the question what kinds of speech chunk are associated with what kinds of verb and to what extent this might be related to meaning. Moreover, the fact that the use of a word trig- gers the choice of another word – as suggested by Sinclair in the Idiom Principle (see above) – is not only effective within complements but can also cross complement boundaries. Despite these drawbacks, the comple- ment will be taken as the basic unit of analysis – both in combination, i.e. as patterns, as well as in isolation – since it is still the unit that can best be described in a fairly specific way without completely losing the possibility of generalization. The various types of complements used are based on the units identified and the labels used in VDE and Herbst and Schüller (2008) and are listed in Table 1 below.

1.2.9 Participant roles

The semantic description of verbs and valency patterns is clearly more dif- ficult (for a discussion see, for instance, Götz-Votteler 2007). As outlined above, semantic roles are generally used in various forms to describe the kinds of propositions expressed by the verb. The kinds of participants a verb has – described by means of semantic roles – of course already pro- vide a partial description of the meaning of the verb itself. Hence, if an activity involves a person who is actively carrying out the action, a person receiving something as a consequence of this and an item that is affected by the action by changing , this activity describes some kind of transfer. Thus separating verb meaning from the verbal participants is as such not possible (for a discussion see also 3.1): “Knowledge of a range of possible participants is part and parcel of the knowledge of the process – not something separate …. You can't have eating without and – it is simply impossible” (Lamb 2001: 179). However, the ver- bal participants do not sufficiently define a verb’s meaning either. Further features like lexical aspect (the situation type of a verb or its Aktionsart) play a role as well. 18 Introduction

Table 1. Complement types used in the analysis (VDE: xv–xviii)

Label Explanation [NP] noun phrase; e.g. I like him. [AdjP] adjective phrase; e.g. He feels better. [V-ing] present participle clause; e.g. He likes dancing with her. [INF] bare infinitive clause, e.g. You can help save the life of dol- phins.VDE [to_INF] to-infinitive clause; e.g. He likes to dance. [that_CL] that-clause, e.g. He thinks that he is great. [SENTENCE] “a sentence or part of a sentence, which is introduced by the verb, which may precede, follow or be inserted in the sen- tence; usually separated by commas” (VDE: xvii), e.g. It changes things, I can promise you.VDE [QUOTE] “a sentence or part of a sentence in inverted commas, which is introduced by the verb, which may precede, follow or be in- serted in the sentence” (VDE: xvii); e.g. “We’ll do it,” she promised him. [Q/S] both [SENTENCE] and [QUOTE] can fill this slot [wh-CL] a clause introduced by a wh-word (when, how, what, …), e.g. I don’t know where he went. [wh-to_INF] to-infinitive clause of which one complement is realized by a wh-word, e.g. I don’t know where to go. [about_X]/ particle phrase (cf. 2.1.2) headed by about, complementation [about_NP]/ of about not specified or a noun phrase, e.g. He is talking [about_V-ing] about emigration; or a present participle clause, e.g. He is talking about going abroad.; can be headed by various differ- ent particles: [on_X], [of_X], [with_X], [as_X] etc. [ADV] “an adverbial which can be realized in various ways... Note that with prepositional phrases covered by ADV the choice of preposition depends on the noun following the preposition not on the word preceding it.” (VDE: xv); e.g. He put the book on the table, He put it there, Let me put it this way.VDE, … [NP (]+[) “a to–infinitive clause preceded by a noun phrase subject” to_INF] (VDE: xvi); e.g. What do you want me to do?VDE or a trivalent pattern with [NP] preceding [to_INF] (cf. 3.3.4.6) [NP V-ing] “an ing-clause preceded by a noun phrase subject” (VDE: xvi), e.g. The government doesn’t want people drifting to London.VDE [so, not, other- This complement type constitutes a certain type of proform wise…] which can replace, for instance, [SENTENCE] and [that_CL]. (cf. e.g. VDE entry of assume [VDE: 54] and CGEL: 12.27) The model of analysis 19

Table 2. Semantic roles used for the description of participants

Role Explanation AFFECTED (AFF.) When we verb, this action has an impact on or concerns AFFECTED. 8 AGENT When an action, i.e. verbing is performed, it is insti- gated volitionally by AGENT. AIM When we verb, we do it to achieve AIM. BASIS When we verb, BASIS is what our activity rests on. BENEFICIARY (BEN.) When we verb, the action is performed for BENEFI- CIARY’s benefit. CAUSE When we verb, we do it because of CAUSE. EFFECTED (EFF.) When we verb, EFFECTED is the outcome or result of this action. GOAL When we verb, we do it in direction of GOAL or some- thing ends up at GOAL. INSTRUMENT (INSTR.) INSTRUMENT is what we use in order to verb. LOCATION (LOC.) When we verb, the action takes place at LOCATION. PURPOSE When we verb, we do it in order to PURPOSE. PREDICATIVE (PRED.) When we verb, PREDICATIVE is either identical with us, describes us or another entity the action is concerned with, i.e. AFFECTED (Herbst and Schüller 2008: 132). RECIPIENT (REC.) When we verb something, RECIPIENT gets it or is in- tended to get it. REFERENCE (REF.) When it comes to REFERENCE, we verb. RESULT When we verb, the outcome of this activity is RESULT. SOURCE When we verb, we have it or take it from SOURCE. TOPIC Our verbing is concerned with or focuses on TOPIC. PRICE When we verb it costs PRICE.

Despite this, a list of semantic roles will be used for the description of complement types in combination with specific verbs, since this seems to be the best method to analyze the connection between verb and pattern and

8. AGENT is a role prototypically filling the subject slot in active declarative clauses. The focus in the present study will be on complements occurring in the predicate. Other roles are for instance AUTHOR (unvolitional), THEME or INSTRUMENT. For a description of these semantic roles as logical actors, see Frawley (1992: 203–210). There are also various approaches which try to de- termine which participant is realized as the subject of a sentence by means of role hierarchies (e.g. Fillmore 1968; Givón 1984). 20 Introduction form and meaning separately. Where necessary and appropriate this list is supplemented by some more verb-specific roles which are taken to be ef- fective on a level subordinate to the participant described by means of the semantic roles indicated in Table 2. Following our convention in Herbst and Schüller (2008), EFFECTED and AFFECTED will be subsumed – where possible – under a combined role of ÆFFECTED. The same applies to BENEFICIARY and RECIPIENT, which are represented as BENREC, and TOPIC and REFERENCE, which are represented as TOPREF, where possible.

1.3 Design of the study

The aim of this research project is to examine to what extent regularity can be found among verbs with similar complementation patterns and verbs with similar meanings, thus providing empirical evidence for the often as- sumed correlation between meaning and form. This relationship is to be approached from two different directions. Both are based on information about the valency properties of verbs as provided in VDE (2004), which itself is based on the COBUILD corpus complemented by native speaker assessments and aims at providing the pattern inventory of the verbs cov- ered as comprehensively as possible, also including rather rare patterns. The information is provided both in the form of a complement inventory, semantically sorted in such a way that alternative realizations of the same semantic role are grouped together, and a list of patterns. This information served as an initial database for this study, which was supplemented by further corpus research with the BNC. Since semantic analyses are natural- ly more problematic than formal ones as regards their reliability, it is all the more important that the findings are based on interpretations of a large amount of data so that the danger of providing purely subjective statements is avoided. This is done by resting all semantic statements on as many au- thentic example sentences as possible, taking into consideration the context, the co-text, style and also, where necessary, information on the respective source of the examples. Patterns which could not be verified with the help of the BNC as well as semantic judgments made in the course of the analyses were checked with a number of British and American native speakers. The validity of native speaker judgments has often been questioned (e.g. Labov 1972: 192–201), since “informant reactions do not always reflect actual usage” (Greenbaum 1977: 5), i.e. sentences are accepted by some speakers, although it is indeed unlikely that they would utter such a sentence themselves, while other sen- Design of the study 21 tences are judged as incorrect, although a different context might change the informant’s opinion. Fillmore (1972: 10) pointed out that native speak- ers might differ from other native speakers in their usage (i.e. deviate from the standard) or simply make mistakes in their judgments, which has to be taken into consideration as well. However, native speaker interviews and tests are the only methodology available – besides introspection – if no authentic example sentences can be identified. Consequently it simply has to be accepted – also with respect to the question with which complemen- tation patterns a verb occurs – that there are fuzzy boundaries between ac- ceptable and not acceptable. Whether they are due to differences in dialect, point in the direction of a possible grammatical change in progress (cf. Hopper and Closs Traugott [1993] 2003: 124–126), or are simply “mis- takes” will not be discussed here, as this would clearly go beyond the goal of examining the extent to which verb complementation patterns can be predicted from verb meaning. If a clear difference is observable between British and American native speakers this is pointed out. However, there are only very few cases where this seems to have played a role in this sam- ple. There is considerably more deviation which does not correlate with this dialect boundary. The following chapter will focus on separate complement types and the question to what extent the acceptability of a certain complement type is related to its meaning. Thus, a number of complement types which often appear to be alternative realizations of one and the same valency slot (based on VDE-data) were examined more closely by looking at verbs which ex- hibit this kind of variability, e.g. We questioned him about it vs. We ques- tioned him on it. If more than one complement type can be used in the ex- pression of a certain participant (in this case [about_NP] and [on_NP] as realizations of TOPIC), they are either synonymous or they “specialize” in the sense that the choice of one possible filling of a slot shapes the meaning of the resulting pattern. To determine a possible synonymy or specialization between such complement types, the actual occurrences of a verb in both patterns in the BNC were compared, complemented by native speaker in- terviews. If differences were identifiable, they were documented and used in order to check whether this specialization can also be found with other verbs exhibiting the same variability. A stable complement type meaning could be taken as the basis for a direct relationship between form and meaning and a predictability of the former from the latter. Closely connected to this comparison of verbs with identical variability within one slot is the search for syntactic gaps, i.e. verbs which deviate 22 Introduction slightly as regards their variability. This is especially interesting with re- spect to verbs with a similar meaning, e.g.:

(1) We questioned him about it. vs. We questioned him on it. (2) We asked him about it. vs. *We asked him on it.

Thus chapter 3 compares the complementation possibilities of semantically similar verbs in order to determine the extent to which semantic similarity entails formal similarity. There are several other attempts in this direction (e.g. Levin 1993; Francis, Hunston, and Manning 1996), yet they all take the complementation possibilities of verbs as a starting point for finding se- mantically similar verbs, which presupposes what is to be examined here: to what extent semantic similarity can be taken to predict complementation possibilities. Whether semantically “plausible” patterns are indeed blocked, will be checked with the help of the BNC (if the pattern is not contained in VDE) and, where necessary, also by means of native speaker interviews. Moreover, the groups are systematically checked with respect to various semantic aspects which are not directly included in the formation of the verb groups, in order to see whether they have an effect on the actual avail- ability of patterns. Thus, the complement can be seen as the decisive unit for both chapter 2 and 3. However, it seems that a pattern is not necessarily always just a sum of the single participants and complements it consists of. A pattern can be a unit of its own. Whether valency information is represented in the form of a complement inventory or whether it is represented in the form of a pattern inventory has certain consequences. A complement inventory reduces re- dundancy as each complement type is only listed once. A pattern inventory, on the other hand lists each complement as often as it occurs in different patterns. The complement-inventory-point-of-view assumes that it is enough to know which participants can be combined and which are the possible realizations of a participant (for a discussion see Herbst 2007). Chapter 2 and 3 will show that this leads to potential overgenerations of patterns which do not seem to exist. Therefore chapter 4 will focus on the role of the pattern, taking data from aforementioned Patternbank into con- sideration. Finally, chapter 5 will sum up the results and provide a possible conclusion to the question to what extent form is determined by meaning. 2 The meaning of complements

2.1 Complement types as part of a verb’s valency structure

2.1.1 Overview

This chapter will address pattern choice from the perspective of the single complement types occurring in the patterns. The meaning of participants is often taken to be an important factor for their possible realization as part of a valency pattern. If a certain complement type is typically an alternative to another complement type (irrespective of the meaning of the verb), i.e. if it regularly commutes with it but not in combination with a specific verb, this restriction might throw light on possible semantic properties of that com- plement type. Such cases could help to identify semantic features of com- plement types which allow for a prediction of their compatibility or incom- patibility with a verb.While blocked complement types might help to single out restrictions which are due to certain facets of meaning of the comple- ment type, alternative patterns can help to single out certain semantic spe- cializations of different complement types. For this purpose, complements that are frequently listed as realizations of the same valency slot in VDE are analyzed with respect to possible semantic differences. If, and only if, both semantic restrictions and semantic specializations of a complement type hold beyond the verb meaning for which they are identified, could one speak of semantic features which are truly inherent in the complement type. On this basis, the question can be asked, as to whether the form of a pat- tern, i.e. the choice of a complement, can be predicted from the com- patibility of the meaning of the participant and the complement.

2.1.2 Complement types

The first part of this analysis focuses on complements which can be said to fulfill the general semantic function of introducing the topic of an action or state (TOPIC) (He talks on butterflies) or functioning as a point of reference (REFERENCE) (He cheats on his income tax return), in this case both ex- pressed in the form of on-prepositional phrases. The complements ex- amined for this purpose are mainly introduced by on, about, of, and over, which are traditionally described as prepositions. Many prepositions, e.g. 24 The meaning of complements since and before, also have subordinating or adverbial uses and are de- scribed as prepositions when they are complemented with noun phrases, as subordinating conjunctions when they are followed by a non-finite clause, and as adverbs when they do not receive any form of complementation. The kind of complementation a word receives is typically regarded as a matter of that word’s valency. However, if a verb can take a range of different complements (e.g. a noun phrase, a clause or no complement at all), this does not lead to a distinction of this verb into different word classes. For this reason, I will assume one single word class particle, which subsumes traditional prepositions, subordinating conjunctions, and a number of ad- verbs. By doing so I follow our approach introduced in Herbst and Schüller (2008), which can be found in a very similar form in the Cambridge Gram- mar of the English Language by Huddleston and Pullum (2002) (henceforth CamG). Particles are taken to be valency carriers and heads of the respec- tive phrases (e.g. [on_X]). Complementation with noun phrases is referred to as prepositional use ([on_NP]), complementation with clauses as subor- dinator use ([on_V-ing]), and the use of the particle on its own as adverb use ([on]). The question addressed with respect to the particle complements men- tioned above is (1) if and to what extent they differ semantically when complementing verbs and (2) if this knowledge can help to predict possible patterns. The meaning of particles referred to in this context is usually not their spatial or temporal meaning, as particle phrases providing this type of information are, most of the time, independent of the verb and consequently outside the scope of valency and complementation: except for a few verbs such as put, they are neither necessary for a grammatically correct use of the verb nor are they determined in their form by the verb and are therefore classified as adjuncts and not as complements (VDE: xxiv). From a dia- chronic point of view the spatial meaning is likely to have been the source of such grammaticalized uses (cf. e.g. Cuyckens 1997: 63; Lakoff 1987: 435), which might still have some of that original spatial meaning. Roh- denburg (2003: 206) argues that “such governed prepositions [i.e. pre- position marking the complement of a lexical ] have become gram- maticalized to varying degrees, [but] they still preserve a variety of subtle semantic constrasts.” However, the analysis in the following chapter will show that it is problematic if not impossible to make any kinds of predic- tions as regards complementation on this basis. Here, the focus is exclusively on valency specific aspects, i.e. what could be referred to as verbal pattern choices. Prepositional uses of particles are often dealt with under the label prepositional verbs, which are analyzed Complement types 25 as combinations of a verb and a preposition complemented by a noun phrase (CGEL: 16.3–16.8). Following VDE, these combinations will not be regarded as new instantiations of a verb besides its use without the particle (cf. also Herbst 1999b: 237; Heringer 1967). Instead, the particle phrase is analyzed as one possible form of complementation of the verb. CGEL provides a semantic classification of the relevant particles in the form of the following five groups (CGEL: 9.60–9.63), while the group ‘subject matter’ is obviously closest to the participants discussed here.

Table 3. CGEL’s classification of prepositional phrases for complementation (CGEL: 9.60–9.63) group associated prepositions subject matter about, on, over, of material, ingredient with, of, out of substance from standard for, at reaction at, to

Herbst (1983) offers a detailed study of such complements and their se- mantic implications for adjectives. The roles identified for the particles under consideration are listed in Table 4, which already shows that on is most versatile when it comes to realizing different participants (EXPE- RIENCER, GOAL, REFERENCE, and TOPIC) and must therefore be described as polysemous, at least with respect to adjective complementation.

Table 4. Herbst’s classification of prepositional phrases for adjective comple- mentation (Herbst 1983: 113–115) semantic function complement introduced by EXPERIENCER on GOAL on, upon REFERENCE about, as to, on, with TOPIC on

In order to obtain a reliable classification of participants, possible para- phrases – as used in Herbst (1983) – will be employed as test criteria, i.e. When it comes to REFERENCE, [Complement 1] VERBs and [Complement 1] VERBs [Complement 2] and the VERBING is concerned with or focuses on TOPIC (cf. Table 2 in 1.2). The distinction between the two, however, clearly appears to be a matter of gradience. While some verbs undoubtedly have a TOPIC in their participant inventory, e.g. in (1), participants of other 26 The meaning of complements verbs are best described as REFERENCE, e.g. in (2). A large group of verbs seems to be rather indeterminate with respect to the classification of this participant, cf. (3).

(1a)NSA Peter talks about butterflies. (1b)NSA Peter talks and his talking is concerned with or focuses on butterflies (= TOPIC) (1c)NSA *When it comes to butterflies, Peter talks. (2a)NSA Peter cheats about his age. (2b)NSA When it comes to his age (= REFERENCE), Peter cheats. (2c)NSA ?Peter cheats and his cheating is concerned with or focuses on his age. (3a)NSA Peter and Mary argue about money. (3b)NSA Peter and Mary argue and their arguing is concerned with or focuses on money (= TOPIC). (3c)NSA When it comes to money (= REFERENCE), Peter and Mary argue.

Certain flexibility as regards the semantic range of complement types has to be taken into consideration in general: in combination with different verbs they can realize different complement types. For a closer examination of the choice of particle, examples for both TOPIC and REFERENCE have been taken into consideration as typically being verbalized by means of the par- ticles on and about, but also upon, of, and over. In cases as example (3) the label TOPIC/REFERENCE (short TOPREF) will be used to indicate the variabil- ity. While dictionaries point out differences in the use of specific verbs which are complemented by on and other particles under the respective verb entries, they do not necessarily point out semantic nuances under the respective entries on the particles themselves. Learner’s dictionaries, for example, do not provide any semantic specifications on the basis of which a pattern choice could be predicted – the Longman Dictionary of Contempo- rary English (42005) (henceforth LDOCE4) defines on simply as “about a particular subject” (emphasis mine), not implying any difference in mean- ing between on and about, while about is defined as “concerning or relating to a particular subject” (LDOCE4: 1146). General dictionaries, on the other hand, do point out two major senses of on, as for instance the Collins Eng- lish Dictionary (82006), which lists (1) “concerned with or relating to” and (2) “used to indicate the basis, grounds, or cause, as of a statement or ac- tion”. The Oxford English Dictionary (OED) also provides two figurative meanings of on and relates them historically to two different spatial mean- ings, on the one hand, MOTION, which was originally marked by the use of the accusative (“in regard to, in reference to, with respect to”), and, on the Complement types 27 other hand, POSITION, which was marked by the use of the dative (“indicat- ing the ground, basis or reason of action, opinion, etc.”) (OED, vol. X: 794, entry of on). However, the OED also points out difficulties in a strict sepa- ration of the two meanings and consequently the differentiation into differ- ent uses:

Even the primary division into senses implying position and those implying motion or direction is difficult to carry out in the figurative uses, in some of which the point of view has gradually changed since they first arose, so that what was originally felt to express a direction of the mind towards some- thing is now felt as a static attitude or mental state. (OED vol. X: 793)

About is defined as “the regular preposition to define the subject matter of verbal activity, as in speak, think, ask, dream, hear, know about” (OED volume I: 39, entry for about). CGEL (9.60), however, points out that “[o]n is chiefly reserved for deliberate, formal linguistic communication (public speaking, lecturing, writing, etc.), and is therefore inappropriate for verbs like chat and quarrel” and explains such meaning distinctions with the help of the following examples: “Thus [1] would suggest she was making a for- mal speech [‘gave a lecture on’], whereas [2] could refer equally to an in- formal conversation or casual allusion.”

(4)QE She spoke [1] on butterflies / [2] about butterflies.

Note that the semantic classification provided by CGEL – at least at this stage – differs from the others. This is due to the fact that CGEL is con- cerned with only one of the two meanings provided so far, i.e. the analysis takes place on a different level (Figure 1 below).

Figure 1. Levels of disambiguation in the meaning of complements

The distinction with respect to formality does indeed account for several restrictions identified in VDE data for the complement type [on_X], as 28 The meaning of complements pointed out in CGEL (9.60) for chat, quarrel, and cry but also for laugh, smile, worry,and fight. Table 5 lists particle complement types available for the participant TOP- REF for a selection of verbs, all of which cannot be complemented with [on_X]. Further semantic distinctions created through the use of the differ- ent particles are indicated in brackets and will, in part, be discussed in more detail below. The example sentences given below are VDE examples which illustrate the use of some of the patterns.

Table 5. Pattern choices *[on] (source: VDE)

verb participant complement in divalent pattern [on_X] [at_X] [about_X] [over_X]

chat TOPIC –– –– 9 –– cry TOPIC (CAUSE) –– 9 9 9 quarrel TOPIC –– –– 9 9 laugh TOPREF (CAUSE) –– 9 99 smile TOPREF (CAUSE) –– 9 99 worry TOPIC –– 9 99 fight TOPIC (CAUSE) –– –– 99

(5)VDE I know you’re crying about Mummy. (6)VDE He laughed about it at the time, but he certainly wasn’t laughing three months later. (7)VDE What are you smiling about? (8)VDE My mother worries about my safety. (9)NSA *He cried / smiled / laughed / worried on this.

Cry, laugh, smile, and worry do not allow complementation with [on_X] as could have been expected from the rather subjective, personal, and private kinds of action these verbs describe: their TOPREF might not be formal enough to justify the use of on which seems to express a TOPIC or REFER- ENCE with a more formal character. The same applies to quarrel and fight, which, however, do not seem to block on but also at. In trivalent patterns, however, complementation with [on_X] is also possible with quarrel. Leaving aside any other restrictions for the moment, the differentiation provided by CGEL appears fitting since quarrel cannot be complemented with [on_X], while fight occurs in this pattern in the BNC exclusively in the rather formal context of election campaigns (cf. 5.2.6). Consequently, it seems reasonable to analyze [on_X] as FORMAL TOPIC and [about_X] as NEUTRAL TOPIC because it is used in reference to both formal and informal Complement types 29 events. Both [on_X] and [about_X] are taken to be realizations of the same valency slot representing one single participant, if it is not possible to com- bine both complement types (as realizations of different participants) in one single sentence (cf. chapter 2.1.3).

(10)NSA *He talked on insects (and) about butterflies.

These assumptions will be taken as the starting point of an analysis of al- ternative realizations. The goal is to find possible semantic differences among formal complement types which could provide a semantic explana- tion for the choice of one pattern over another. What follows from this is the question whether and to what extent complement types are stable in their meaning. Alternative complements cannot only be found with respect to particle complements but also to clausal complements. Contrary to particle com- plements where the meaning of the particle can be seen as contributing to the meaning of the complement – which is also the reason for the analysis of prepositions as theta-role assigning elements in some theories of com- plementation (e.g. Marantz 1984: 17–23) – the lexical elements in clausal complements are often more or less the same (except, for example, dif- ferences in grammatical morphology) and consequently semantically more neutral. Nonetheless a semantic specialization for this type of complement is also assumed in various different frameworks. Since the comparison of the complementation patterns of semantically similar verbs in chapter three provides numerous examples of verbs allowing alternative clausal comple- ment types, a detailed discussion will be provided there. Nonetheless, one case will be discussed in this chapter as well, taking the perspective of a possible specialization of the complement types [V-ing] and [to_INF].

2.1.3 Valency structures – introducing the model used

The idea that one valency slot can have more than one realization was al- ready put forward by Helbig in what he calls the “(partial) non-correspon- dence between semantic and syntactic valency”, i.e. “cases in which se- mantic case can be realized on the surface in alternative ways (although the meaning stays the same)” (Helbig 1992: 15; [translation mine]), a view we also adopted in Herbst and Schüller (2008). The notion of alternative pat- terns as discussed so far has the disadvantage of disregarding finer semantic distinctions as the valency patterns are analyzed as realizing the same par- ticipants and thus appear to be interchangeable. It must be very clear that 30 The meaning of complements treating such instances as semantically alternative patterns does not entail that they are semantically identical patterns. A close semantic analysis might reveal that there are indeed factors influencing the choice of patterns as indicated above in the context of complementation with on and about. When using broadly defined participant roles and thus concentrating exclu- sively on alternative patterns, such aspects would be obscured. Assuming different participants for each semantic difference and describing these by means of finely grained participant roles can possibly help to explain varia- tion in the choice of patterns on purely semantic grounds. However, the obvious disadvantage of such an approach is that it cannot explain which elements cannot co-occur in one pattern because they are too similar in that they fulfill the same broad semantic function. Thus if each semantic differ- ence automatically leads to the establishment of a new construction, restric- tions of semantically similar participants cannot be accounted for. A solu- tion which combines the advantages of both approaches is one which as- sumes more than one level (see also Van Valin [2004: 64] for a similar approach). Alternative complements are considered to belong to on super- ordinate element and it is this super-ordinate element which in this current context will henceforth be referred to as valency slot (formal point of view) or participant (semantic point of view). Finer distinctions will then be ac- counted for on a second level. This framework allows for both generality and specificity as described above. The following model will serve to faci- litate the illustration of the different levels of the semantic analysis. A va- lency structure for the example discussed above could be envisaged as fol- lows (see Figure 2 below): The verb as the central element in a clause opens up around it a number of valency slots (level A) (cf. Bühler [1934] 1982: 295). The verb’s valency defines the number and the type of slots (quantitative and qualitative valency), which can be described by means of fairly general participant roles, e.g. AGENT or TOPIC. Participants are to be understood as correspondents of the syntactic complements of a verb, which are described by means of their morpho-syntactic structure (level B). These can – if necessary – also be assigned a semantic description which is more precise than the participant role (level A) and will be referred to as semantic subtypes. Subtypes (level C) of one participant are elements which are semantically so similar that they cannot co-occur in one and the same sentence (except for when they are coordinated). They are described by means of more specific semantic roles. Dotted lines connecting the verb with the complements indicate that all of the complements are optional and not obligatory (which would be indicated by a solid line) (cf. 1.2.4). Ad- Complement types 31 juncts are not connected, which illustrates their independence of the verb, i.e. they are outside the scope of the verb’s valency.

A Valency Slot I+II [AGENT +CO-AGENT] B Complement [NPpl/group / NP and NP]

A Valency Slot I [AGENT] A Valency Slot II [ADDRESSEE] B Complement [NP] B Complement [to_NP]

A Valency Slot III [TOPIC] talk B Complement [on_NP] (SPEECH) C Subtype [formal TOPIC] B Complement [about_NP] C Subtype [neutral TOPIC] Additional information [PLACE], [TIME] etc. Adjunct [NP], [ADV] etc.

Figure 2. Simplified valency structure of talk (based on VDE: 850)

Adjuncts are not viewed as realizations of participants, either, although they can of course also be described semantically by means of semantic roles. They are not necessary for the depiction of a verb’s valency and have been included here only to demonstrate the different type of relationship as opposed to complements. As pointed out in 1.2.2, the distinction between complement and adjunct is often difficult to draw. In order not to miss any elements that appear to be specific to a verb or which are typically used with it, a rather generous approach will be taken here, in the sense that if in doubt the element will be treated as a complement rather than as an adjunct. With some verbs, the AGENT-valency slot can also be realized by a “par- ticipant group” of AGENT and CO-AGENT, as in They were talking or He and I were talking (VDE: xvi). In the latter case the person addressed is pre- sented in a slightly different, i.e. more agentive way. This process can be described as a merging of participants (for a discussion cf. 4.1.2, see also Heringer 1996: 68). 32 The meaning of complements

2.2 Alternative realizations of participants

2.2.1 Semantic specialization vs. synonymy of complement types

In the following sections, two types of examples will be discussed. On the one hand, there are examples where alternative realizations of the same valency slot indeed express different semantic nuances (2.2.1.1 and 2.2.1.3). On the other hand, there are examples where this is not the case, i.e. where different formal realizations of the same valency slot appear to be interchangeable and thus synonymous (2.2.1.2).

2.2.1.1 AGENT + talk + TOPIC

The participant pattern AGENT + talk + TOPIC can be realized by several different valency patterns, while not all uses of talk occur in all of these patterns. According to VDE, talk ‘speech’ occurs in only two patterns, talk ‘conversation’ in three, and talk ‘topic’ in two but not the same two as talk ‘speech’ (see Table 6). A distinction into different lexical units as in VDE is one possible way to analyze such differences in the sense that a different meaning leads to different patterns or different patterns to different mean- ings.

Table 6. Relevant complementation patterns of talk (VDE: 850)

talk ‘speech’ ‘conversation’ ‘topic’

A [SCU]+ talk +[about_X]P 999 B [SCU]+ talk +[on_X]P 99–– C [SCU]+ talk +[of_X]P –– 9 9

In order to be able to focus on semantic differences between the comple- ment types, such a prior subdivision into – in this case – different types of talking will not be made here. Moreover, at this point no claim is to be made as to which “direction” is considered to be cognitively more relevant, i.e. from verb meaning to pattern choice or vice versa. Either way the com- plement types could have semantic features which make them more or less likely as partners for the expression of certain ideas. Leaving aside pattern (C), which is not really synonymous with the other patterns, the question is what such semantic features could be. As pointed out above, the two pat- terns (A) and (B) differ semantically in that (A) usually refers to a rather Alternative Realizations 33 formal occasion, whereas (B) seems to be more neutral in this respect. Typ- ical examples of (A) are sentences (1) and (2), where the formality (of the situation, i.e. not the text!) can be pinned down, for instance, by such words as experts (1) and lectures (2).

(1)BNC 1992 was a strong contender for best year yet, with eminent experts talking knowledgeably on the myriad of inter-modal transport topics of interest to members.K9W324 (2)BNC …, he gave four lectures in November on “The Aims of Education” and in the same month talked on “Poetry and Drama” at Harvard. EFX 945

However, the complement [on_X] is also found in contexts where formality does not seem to play a role at all as in (3) and (4).

(3)BNC If you and your children have always talked together on all kinds of subjects, you may find it easier to sit down and talk reasonably and quietly about some of the hard decisions that have to be made. BLW 1265 (4)BNC …yeah cos me and Jonathan started talking on it before. KCT 7485

The kind of talking envisaged in (3) appears to be exactly the opposite of parents “lecturing” their children (cf. together), while (4), an example of spoken dialogue, illustrates the use of the pattern in both informal speech and in an informal context. If [on_X] were marked as formal, [about_X] should be a better match in both sentences, and yet it is still not used. Another difference between the two complement types could be that the underlined passage in (5) can also be said to refer to a topic in a more gene- ral way, while (6) appears to be more specific.

(5)NSA Brad said that Jen talked on butterflies. (6)NSA Brad said that Jen talked about butterflies.

In (5), it is not clear whether Brad knows anything about the content of Jen’s talk. Possibly it is just the title of the talk that is known to him. In (6), Brad seems to be more likely to have noticed something about the talk or seems to have been present during the talk. Thus if speakers simply want to mention the topic, they are more likely to use a sentence such as (5). Note that in this case the formality of the situation may be exactly the same “Jen’s talk on butterflies”. It is the perspective of the interlocutor which changes. Hence, [on_X] does not only seem to be more formal but also less specific. [About_X], while being more neutral, implies the potential availa- bility of more information about TOPIC. 34 The meaning of complements

Closely connected to this, there is also a tendency in the BNC that [on_ NP] refers to the title of a talk, often indicated by inverted commas.

(7)BNC …, Alex Seago will talk on “Peter Blake: An English Pop Artist”; Fri- day 18 in the Lower Galleries at 1 pm Charles Hustwick will talk on “Beardsley to Bomberg: British Drawings and Watercolours”, on Sat- urday December 19 in the Auditorium at 1 pm Laurence Bradbury will talk about “Peter Blake: original Pop”; E9W 1039

However, this is not an exclusive of complements introduced by on, either. In the sentence above, complements introduced by on and about are used absolutely interchangeably, in a very similar context and by the same speaker. Neither does the context nor the meaning of the different comple- ments differ in this example sentence – all three complements are titles of talks. It seems rather unlikely that the speaker wanted to indicate that more information is available on one than on the others. A possible explanation might be an avoidance of repetition on the part of the speaker for stylistic reasons, the so-called horror aequi-effect (Rohdenburg 2003), or simply the very high degree of exchangeability of the two patterns for the expression of TOPIC. On the other hand, there is also evidence for priming effects, in the sense that one form can trigger the choice of a following form (e.g. Gries 2005). In both cases, the prediction of the pattern choice might be impossible and the assumption that a difference in form always implies a difference in meaning (e.g. Goldberg 1995: 67) must clearly be questioned (cf. chapter 3.3.4.8). These simple examples already indicate that the meaning of a comple- ment type is by no means easily described. Formality, as a possible feature of on, can, in the case of talk, be used to predict the choice of pattern in only a very limited number of examples. Yet the examples also show that valency structures must be envisaged as multilayered and influenced by various factors. In order to incorporate this in the model outlined above, valency slot II has to be modified in order to include the facets of speci- ficity and the fact that TOPIC can be simply a title of something. Another feature of the valency structure of talk as presented here is the co-occurrence restriction of [about_X] and [on_X], i.e. they belong to the same slot:

(8)NSA *He talked on insects (and) about butterflies. (9)BNC It's very important to be precise and know exactly what you're talking about on these, cos you'll get… FYA 570 Alternative Realizations 35

Sentence (9) is the only example identified in the BNC where the two par- ticle phrases co-occur with talk and it is better analyzed as a chunk rather than as two separate complements. This is in accordance with the high fre- quency of what are you talking about and similar sequences in the BNC, especially in spoken language. Such idiomatic expressions are very likely stored and retrieved from memory as a set piece. This also points towards a necessary extension of the notion of valency carrier, since such expressions can have a valency of their own, which is not necessarily captured by the valency of their single components (cf. Herbst and Schüller 2008: 146– 147). Thus, a co-occurrence of the two complements as described above does not seem to be possible. For this reason both complements are taken to be part of one single valency slot as indicated in Figure 3.

Valency Slot I [AGENT] Valency Slot III [RECIPIENT] Complement [NP] Complement [to_NP]

Valency Slot II [TOPIC] 1 Complement [on_NP] talk Subtype [formal TOPIC] (SPEECH) Subtype [general TOPIC] Subtype [TITLE] 2 Complement [about_NP] Subtype [neutral TOPIC and specific CONTENT] Adjuncts Subtype [TITLE]

Figure 3. Revised model of talk (‘speech’)

Differences in frequency are another phenomenon that needs to be ac- counted for. While a BNC search produced 10,572 sentences with [about_X] directly following the lemma talk (including 8 instances of ['bout_X]), only 23 sentences were found with talk complemented by [on_X] (directly following the predicator). If analyzed according to the criteria mentioned above, these include instances of the subtypes [on_X]_ TITLE, [on_X]_GENERAL, [on_X]_SPECIFIC, and further types which could not be clearly specified. When compared to the number of sentences where talk is immediately followed by an adjunct of place introduced by on, the scarceness of this complement type in combination with talk is even more apparent. With 48 sentences, there are more than twice as many adjuncts 36 The meaning of complements than complements introduced by on, the most frequent combination on the (tele)phone with 29 sentences already outnumbering the cases where on introduces a complement. The sentences do not show signs of a clear spe- cialization. However, even infrequent complement types can express dif- ferent nuances of meaning.

2.2.1.2 AGENT + agree + TOPIC/REFERENCE

An example where specialization of particle complements appears to be even more problematic is provided by agree. Here, three different particle complements can be analyzed as realizations of the valency slot TOPREF, both when the verb is used in the sense ‘being of the same opinion’ and ‘come to conclusion’ (semantic distinctions based on VDE: 23–24). Agree ‘opinion’ specifies a TOPREF which simply identifies the subject matter of, for example, a discussion on which AGENT and CO-AGENT agree. The fol- lowing patterns are available:

(A) [SCU]+ agree +[about_X] (+[with_X]) (B) [SCU]+ agree +[on_X] (+[with_X]) (C) [SCU]+ agree +[upon_X]

In the sense agree ‘conclusion’, on the other hand, the same formal com- plement types [on_X], [about_X], and [upon_X] encompass a resultative facet in that they do not just specify the topic of an but at the same time the result, e.g. a fixed date or a compromise. VDE provides the following examples:

(1a)VDE Pete and my husband have never agreed about these things. (‘opinion’) (2a)VDE They would spend an hour trying to agree about which film they would go to. (‘conclusion’) (3a)VDE All appear to agree on this machine being good value for the money. (‘opinion’). (4a)VDE Currently, government experts are trying to agree on how this should be done. (‘conclusion’)

When exactly the same patterns are available for two different meanings of the same verb, the cause for the semantic difference can neither be the verb nor the complement types since they remain stable in both cases. Thus, if there is a difference it must be identifiable in the context and lexical co-text the patterns are embedded in, which renders the difference a matter of in- terpretation and gradience. Yet the distinction into different lexical units is Alternative Realizations 37 necessary with respect to other patterns associated with the two different uses of the verb, e.g.

[SCU]+ agree +[with_NP]+[that_CL] / [Q/S] (‘opinion’) [SCU]+ agree +[to_INF] / [that-CL] / [wh-CL] / [wh-to_INF] (‘conclusion’)

(5)VDE Many Conservatives agree with the opposition that the poll tax imposes unfair burdens on the poorest sections of society (non-resultative) TO- PREF. Æ “SUBJECT MATTER” (6)VDE The company’s outside directors agreed with Mike to support me (re- sultative) TOPREF.Æ “AGREEMENT”

The classification “SUBJECT MATTER” vs. “AGREEMENT” is to indicate that, while in (5) the complement simply refers to an issue on which there is consent, in (6) possible consequences are inherent in this consent, i.e. the agreement on an action or plan. Thus [on_X] and [about_X] are both able to express two senses (“SUBJECT MATTER” and “AGREEMENT”), i.e. they are polysemous. However, they are also synonymous in that they can ex- press the same two senses – at least in combination with agree. This contra- dicts Goldberg’s (1995: 67) Principle of No Synonymy or Croft’s (2001: 111) principle of , according to which “two grammatical structures [which] occur in the same language to describe the ‘same’ experience, … will differ in their conceptualization of that experience in accordance with the difference in the two structures”. Thus one might expect a specializa- tion of one of the two complement types on one of the two participants associated with agree, and there are indeed examples where this is the case:

(7a)NSA We agreed on Monday. (7b)NSA We agreed about Monday.

While (7a) could be interpreted in such a way that the parties of the agree- ment settled on a specific date, i.e. Monday, for example for a meeting (leaving aside the interpretation of the underlined passage as an adjunct of time), (7b) does not entail such a result, i.e.the parties of the agreement only share the same opinion when it comes to Monday. The sentence could be imagined, for example, in a context where the parties disagree about something which happened on Tuesday, while they consent with respect to the event which took place on Monday. Thus, a possible disambiguation of the two patterns in this case is that [on_X]_TOPREF refers to a compromise or fixed date, whereas [about_X]_TOPREF is not available for this interpre- tation but refers to the subject matter in general. This way, [on_X] and [about_X] would be prototypical realizations of the subtypes AGREEMENT 38 The meaning of complements and SUBJECT MATTER of the participant TOPREF (as the superordinate par- ticipant type). This valency structure would then apply to both lexical units, agree ‘conclusion’ as well as to agree ‘opinion’, a distinction made in VDE and also, for instance, in a very similar way, in LDOCE4. Thus, instead of making the distinction on the level of the verb, it is again possible to defer the difference to the complements. This will be discussed below with re- spect to several authentic as well as modified example sentences. If [on_X] – opposed to [about_X] – implied conclusiveness, example (1b) and (2b) would both imply that a decision is taken, while this is should not hold for (1a) and (2a) (based on the distinction made in VDE).

(1a)VDE Pete and my husband have never agreed about these things. (‘same opinion’) (1b)NSA Pete and my husband have never agreed on these things. (1c)NSA Pete and my husband have not agreed on this. (2a)VDE They would spend an hour trying to agree about which film they would go to. (‘conclusion’) (2b)NSA It took them an hour to agree on which film they would go to.

Sentence (1b) should be less likely for one reason in particular, the lexeme never in combination with the form of the noun phrase these things, which gives the sentence a habitual reading, i.e. the event as such is not conclusive (e.g. Kabakþiev 2000: 134). Sentence (1c) should consequently be more likely, since the habitual reading is removed by turning the action into a conclusive event which is emphasized even more through the singu- lar form of the noun phrase complementing on. However, there is even a tendency among native speakers to prefer on in (1b) over an alternative about and over sentence (1c). Asked for a possible difference in meaning, none of the interviewees was able to provide one. Example sentence (2a) is a VDE example to illustrate the meaning ‘con- clusion’. While TOPREF here clearly refers to a decision, the use of the con- ditional and the verb try in the progressive aspect emphasize an unreal ac- tion in progress, i.e. the sentence does not indicate that a decision has ac- tually been reached. (2b) is more conclusive and thus a more prototypical use of the meaning ‘come to conclusion’, for the simple reason that here a decision is definitely reached in the end which is also expressed through the phrase “take an hour”. Again this could not be confirmed in native speaker interviews. The sentences were judged to be synonymous, similarly to ex- amples (3a) and (3c): (3a) could be interpreted as agree ‘conclusion’ as shows the parallel with (3b). Yet (3c) also appears to be possible. Alternative Realizations 39

(3a)VDE All appear to agree on this machine being good value for the money. (‘same opinion’) (3b)NSA It appears that they all came to the conclusion that this machine is good value for the money. (3c)NSA All appear to agree about this machine being good value for the mon- ey.

Example sentence (4a) below is very similar to example sentence (2a). Both events could be depicted as ongoing with a clearly defined goal, i.e. an endpoint.

(4a)VDE Currently, government experts are trying to agree on how this should be done. (‘conclusion’) (2a)VDE They would spend an hour trying to agree about which film they would go to. (‘conclusion’) (4b)NSA Currently, government experts are trying to agree about how this should be done.

One might argue that the endpoint in example (2a) is less definite than the endpoint in (4a), on the one hand because of the unreal character of the event and, on the other hand, because of its less formal character. Yet the choice of particle in these two examples still appears to be too arbitrary to build a hypothesis on, especially if (4b) is taken into consideration which seems by all means acceptable. Thus, with respect to such examples, an ob- ligatory specification of +[on_X] as AGREEMENT as opposed to +[about_X] as SUBJECT MATTER can by no means be justified. The preceding paragraph presented generality and formality as features of +[on_X] and specificity and neutrality (as opposed to formality) for +[about_X] in combination with the verb talk. If these are complement- type-specific semantic features, i.e. independent of the verb and the pattern, such a difference should also be identifiable for agree. Trying to apply the observed semantic specification for +[on_X] and +[about_X] to the verb agree leads to contradictory results. First of all, these notions would have to be adapted to agree: in combination with talk, formality refers to the type of speech and not to the formality of the situation in which the sentence is uttered. In combination with agree a possible interpretation might be the formality of the agreement, for example whether it is regarded as binding or not. Formality would then coincide with the notion of conclusiveness intro- duced above to differentiate between the meaning ‘opinion’ and ‘con- clusion’. A formal agreement is thus prototypically also conclusive. It has to be taken into account here that this obviously means stretching the no- tion of formality to a considerable degree. Similarly, one could argue that 40 The meaning of complements the difference between general and specific means that a sentence is re- garded as general if it does not yet answer the question “what did they agree”, i.e. what is the specific result or agreement. Instead it only provides information about the field in which there is agreement. In combination with talk, generality was defined as a feature for the complement type +[on_X]. In combination with agree, it is a typical fea- ture of the meaning ‘same opinion’, i.e. what up to here appeared more prototypical of +[about_X]. Specific reference, on the other hand, could be said to entail information on the result of the event, i.e. its conclusion. There is however no 100% correspondence between the features general and non-conclusive and specific and conclusive: one can, for instance, be of the same opinion (non-conclusive sense) with respect to a very specific subject. Whether one agrees with this transfer and reinterpretation of the features from talk to agree or not, it is quite obvious that the features are not iden- tical and they are not tied to the complement type meaning, either. This is not to say that the meaning of about is completely different when used in combination with agree from its use in combination with talk: it still intro- duces something the action is concerned with, which applies to both TOPIC and REFERENCE. Yet it is extremely problematic to decide which of the par- ticles is more appropriate in which contexts on the basis of their meaning. When frequency is taken into account, it becomes obvious that the two complement types again clearly differ, while this time the distribution is completely opposite to that in combination with talk. On is significantly more frequent than about in combination with agree (see Figure 4 below). With an increase in the window span more and more mismatches have to be taken for granted (e.g. where the particle phrase refers to an adjunct of time or place or where about means approximately). The overall number of mismatches in combination with on is higher than that of about, yet the difference in frequency between the two cannot be explained by that. The search for the sequence agree (lemma search) + about yields 48 example sentences with divalent patterns, i.e.where about introduces a complement and not an adjunct. After analyzing them according to the conclusiveness of the complement, the clear majority of the sentences in the sample were non-conclusive (41 sentences) rather than conclusive, i.e. the complement would best be analyzed as TOPREF_SUBJECT MATTER.9

9. See Appendix 2 at htttp://dx.doi.org/10.1515/9783110240788_appendix2 for the example sentences on which the semantic analyses are based. Alternative Realizations 41

number of occurrences of AGREE in the BNC

1,029 on 2 about 69

1,106 3 73 window span 1,315 4 113

1,514 5 148

Figure 4. Number of example sentences in the BNC for the lemma agree +on and +about in a window-span of 2–5

However, there are also sentences which avail themselves just as well or partially even better to an analysis as TOPREF_AGREEMENT. Sentence (8), for example, can be interpreted in such a way that it was not possible that 6 personnel managers had the same opinion when it came to a certain set of applicants (SUBJECT MATTER) but an interpretation as to the common set of applicants being what they finally wanted to choose appears to be at least as likely (AGREEMENT).

(8)BNC … Scott published the first research on interviews showing that 6 per- sonnel managers could not agree about a common set of applicants, a finding that has been replicated innumerable times…. HX4 752 (9)BNC Before we can agree about how to assess or evaluate mathematical attainment or progress in learning, we must clarify what we mean by “doing mathematics”. H88 1375 (10)BNC So we agreed about tea, and invited Father D'Arcy to join us. H9X 152

In (9), the particle complement could also be labeled as AGREEMENT: they want to come to a conclusion, i.e. pin down an evaluating mechanism, and not just share an opinion on it. The wh-clause complementing about is very specific and does not describe a general SUBJECT MATTER. In (10) it is even more obvious that the decision was made to have tea (ergo the invitation to that event), while the sentence clearly does not describe a situation where people shared the same opinion when it comes to tea in general. 42 The meaning of complements

The analysis of the example sentences also supports the view that the distinction between generality and specificity is not relevant in the case of agree +[about_X]. Example sentences were identified which were rather conclusive and general (11a), conclusive and specific (12a), non-conclusive and general (13) and finally non-conclusive and specific (14).

(11a)BNC The French and British could not agree about what to do and the situa- tion drifted; ... CM6 1562 (12a)BNC … departments within the University of London Institute of Education were unable to agree about establishing a language centre because of concern over who should be seen as central. G0W 349 (13)BNC They seemed to have a good relationship, agree about most things, and never, never row -- at least, not in front of the children. CEE 111 (14)BNC I agree about Cantona myself. J1G 692

While in (11a), about what to do specifies an intended AGREEMENT and not just a SUBJECT MATTER on which there is (no) consent among the French and the British, it does not provide any information on the agreement and is thus not as specific as for instance about establishing a language centre in (12a). Compare:

(11b)NSA ?What did the French and the British agree on? - … that they want to do something … (12b)NSA What did the departments within the University of London Institute of Education agree on? – …that they want to establish a language centre…

Most things in (13) even more obviously lacks specificity as it could refer to all kinds of topics and the plural here indicates that it refers to several different things. The habituality of the sentence makes it again a prototypi- cal case of agree ‘opinion’, i.e. it is not conclusive in any way. About Can- tona in (14) is more specific, as the TOPIC, i.e. the former French soccer player Cantona, is identified quite clearly, while the sentence is again not conclusive: there is no decision or conclusion implied on his part. Thus despite the fact that [about_X] is much less frequent than [on_X], it does not appear to be more restricted in its usage in combination with agree than the latter. In the pattern [SCU]+ agree + [what/something/etc.] +[about_X], however, the particle phrase is undoubtedly SUBJECT MATTER, since AGREEMENT is already implied in the other post-verbal complement.

(15)BNC But so little AGREEMENT is agreed about the cause of the disorder SUBJECT MATTER that endogenous opiates may well be involved. B7G 995 Alternative Realizations 43

(16)BNC Well what all the wives do agree about living in their fortified com- pounds guarded round the clock SUBJECT MATTER is that the end of this tour can't come soon enough AGREEMENT. KRM 621

In that sense one might argue for a prototypical meaning of [about_X] in combination with agree but by no means of an exclusive one. The fact that there is no clear dividing line between the meaning of [on_X] and [about_X] can also be attested by their occurrence in very simi- lar contexts. Compare:

(12a)BNC …, departments within the University of London Institute of Education were unable to agree about establishing a language centre because of concern over who should be seen as central. G0W 349 (17)BNC He had mentioned the possibility of Russia introducing its own currency, if the republics did not to agree on establishing an inter-republican bank. HLC 1549

Both particle complements contain a present participle clause [V-ing] with the same verb establish. If it were true that there are no synonymous pat- terns it should be possible to show a difference between the two, which, however, does not manifest itself in the co-text. The lexical items used here are extremely similar and in part identical, i.e. there is no collocational reason for the choice of particle. Both are examples of written text, both of medium perceived level of difficulty. While (17) was published in a period- ical, (12a) was identified as “Miscellaneous – published” (based on BNC file information). The style cannot really be said to differ, either. Both a language center as well as an inter-republican bank are official institutions and the establishment of none of the two was agreed, i.e. both sentences in- clude a marker of negation, thus there do not seem to be any colligational restrictions (cf. Firth 1957) at work either. The use of the two patterns in these examples seems to be absolutely synonymous (conclusive and spe- cific) and the participant would best be described as AGREEMENT. Very similar results can be obtained if the following pairs of sentences are com- pared.

(18)BNC … blessing in church afterwards (if either spouse already divorced, or if neither party can agree on going the whole way in a church wedding); ASD 1053 (19)BNC But I agree about going underground now. EWC 2383 (20)BNC Surviving eyewitnesses are few in number, and give contradictory de- scriptions of the killer, but all agree on his unnatural strength and vi- ciousness. CH0 233 44 The meaning of complements

(21)BNC Table 9.2 shows that for nearly ten years now voters have agreed about her strength of personality and outspokenness. A6F 902 (22)BNC Experts met in Spain in September to discuss the problem but delegates failed to agree on a possible cause. J2N 140 (15)BNC But so little is agreed about the cause of the disorder that endogenous opiates may well be involved. B7G 995

While [on_X] in (18) seems to realize AGREEMENT, [about_X] in (19) does not necessarily imply an AGREEMENT. Here, the AGENT might simply share someone else’s opinion when it comes to going underground. Yet in (20) and (21), the use of the two patterns is again absolutely parallel, however, this time the participants are neither conclusive nor general: they are both specific and non-conclusive, i.e. SUBJECT MATTER. The same applies to (22) and (15). A search for the combination of the participant pattern AGENT + agree + TOPREF_(AGREEMENT/SUBJECT MATTER) containing the noun importance again underlines the arbitrary choice as regards particles in combination with agree. Example sentences contained [on_X], [about_X], [over_X], [upon_X], and [with_X], all clearly realizing TOPREF as the example sentences show. Compare for instance:

(23)BNC Both sides agreed on the importance of continued relief activities and other humanitarian confidence-building measures. HL4 845 (24)BNC Today we must welcome the fact that the three major parties in this country all agree about the importance of the Community. HHW 12744 (25)BNC We also agree over the importance of Staphylococcus saprophyticus, to which we referred, and the difficulty posed by chronic prostatitis. FSY 1729 (26)BNC Although elite theorists have agreed upon the importance of empiri- cally tested research, and their ability to carry it out in a value-free way, in fact this stream of analysis has always been associated with au- thors who take strong value positions. CS3 214 (27)BNC While I agree with the importance that this test places on friendship, I've never agreed with the idea that the more friends the merrier. ED4 1214

Also [on_X] can be found to express SUBJECT MATTER. It can be conclu- sive and general (28), conclusive and specific (29), non-conclusive and general (30), and non-conclusive and specific (31).

(28)BNC … ; and the loss to creditors of failing to agree on new terms is thus potentially bigger. ABJ 3745 (29)BNC Yeah, I thought, I thought we both agreed on Lotus. KBM 174 You want a Fiat Uno? KBM 175 (30)BNC If you agree on the details and choose a theme you can then weave them into your planning at every stage. ADK 171 Alternative Realizations 45

(31)BNC My concern is with the nature of his reputation, and the striking extent to which both his admirers and his detractors are unable to agree on what kind of writer Derrida is, or even what he is actually saying. A1A 492

In (28), on new terms does not provide specific information yet the cha- racter of the event is conclusive: the situation describes a decision-making process. In (29) the intended result of the decision-making process is speci- fied clearly. Example (30) neither specifies an intended result nor does it provide specific information on the topic of the consent. With (31), the list can be seen as completed with an example that is specific and not conclu- sive. The example sentences do not indicate any kind of restriction, despite the fact that the two patterns with [on_X] and [about_X] do not seem to be absolutely synonymous as demonstrated above. A distinction into different lexical units as in VDE (‘opinion’ and ‘conclusion’) is thus a convenient and legitimate way of describing the two possible meanings the patterns can express. Yet the difference in meaning cannot be deduced from the choice of particle or, vice-versa, the choice of particle is not necessarily dependent on the meaning the speaker wants to express. Ambiguity arises only in those cases where for example the AGENT has the power or is in the position to decide something in the first place and where the complement type refers to something a decision can be made on.

(32)BNC One of the impressive things about all this is that everyone agrees about his modesty, his lack of show. CL2 688 (33)BNC Right you know if you've got any questions now er this is the stage at which we agree on a date for you to start. JA2 1255 (34)BNC … he agrees about the date; ... G1A 327

This is, for example, rather unlikely in (32) since modesty is more likely the SUBJECT MATTER and not a final agreement, as, for instance, something such as a date (33). However, as has already been shown, this cannot be taken as a rule but rather as an indicator in the right direction (cf. 34). Figure 5 below tries to capture the findings on agree in a valency struc- ture stemma. Valency slot I (AGENT) and II (CO-AGENT) can also be rea- lized in the form of a complement realizing both participants at the same time (valency slot I+II) as in the case of talk. Valency slot III incorporates both possible readings, i.e. subtypes AGREEMENT and SUBJECT MATTER as they block each other in the same sentence. The blocking however is only in operation with respect to particle complement types. An agreement can also be formalized by other complement types, i.e. [that_CL] and [QUOTE/ 46 The meaning of complements

SENTENCE]. In more general terms this agreement is best classified as EFFECTED (valency slot IV): “When we agree, EFFECTED is the outcome or result of this action”. Thus, the subtype meaning of TOPIC AGREEMENT coincides with the meaning EFFECTED. Consequently, valency slot IV and the subtypes AGREEMENT of valency slot III are incompatible. However, slot IV should theoretically be compatible with the other subtype of slot III, i.e. SUBJECT MATTER.

Valency Slot I + II [AGENT] + [CO-AGENT] Complement [NPpl/group] / [by NPpl/group] [NP and NP] / [by NP and NP]

Valency Slot I [AGENT] Valency Slot II [CO-AGENT] Complement [NP]/[by_NP] Complement [with_NP]

Valency Slot III [TOPREF] agree 1 Complement [on_NP] Subtype [AGREEMENT] Subtype [SUBJECT MATTER] 2 Complement [about_NP] Subtype [SUBJECT MATTER] Subtype [AGREEMENT]

Valency Slot IV [EFFECTED] Adjuncts 1 Complement [that_CL] 2 Complement [Q/S]

Figure 5. Simplified valency structure of agree

The only examples found in the BNC are sentences where the particle phrase might better be classified as an adjunct (35) or takes an intermediate position between complement and adjunct status as in (36) and (37).

(35)BNC I agree that, on the basis of the new evidence, the judge's order com- mitting these two defendants to prison should be set aside. FCB 217 (36)BNC On regional affairs, the summit agreed that the prospect of Israel set- tling emigré Soviet Jews in the occupied territories was “an extremely serious development ....” HKR 1665 Alternative Realizations 47

(37)BNC On Cambodia, the two sides agreed that the UN “framework document” [see p. 37654] was a “good foundation” for the achievement of an all- embracing political settlement. HL7 843

This underlines the flexibility in the expression of semantic notions be- tween adjunct and complement status (cf. 4.1.3). However, there are exam- ple sentences where [about_X] co-occurs with [that_CL] in trivalent pat- terns, which justifies the classification as two different participants. To complicate matters, an example was found where TOPREF_SUBJECT MATTER (valency slot III) was coordinated with EFFECTED (valency slot IV) as if they were realizations of the same role.

(38)BNC Not everyone was agreed, however, that the Hooligans represented a novel development, nor about the scale of the outrages. EDE 644

Sentence (38) illustrates a passive use of agree. Differences between active and passive uses of a pattern have so far not been discussed, while it cannot be excluded that this also plays a role. To summarize: there is no clear specialization between complements in- troduced by about or on in combination with agree and one could at best speak of prototypicality effects but clearly not of a clear division between the two on the basis of semantic rules.

2.2.1.3 AGENT + forget + ÆFFECTED

Forget can be complemented with [V-ing] and [to_INF] to express ÆF- FECTED and here the choice indeed expresses a difference in meaning. For this reason the two complements appear as representations of different par- ticipants (indicated by Roman numerals below) in the VDE-entry of forget. In the VDE note block, the difference is described as “A person can forget somethingII or doing somethingII, i.e. not remember it [emphasis mine]”, as illustrated in (1), as opposed to “A person can forget doing somethingIII, i.e. not do it [emphasis mine]”, as illustrated in (2) (VDE: 327).

(1)VDE I’ll never forget being in Mexico City at the church of the Virgin of Guadalupe. (2)VDE Ellen had thoughtlessly forgotten to bring a torch.

In this present context, a prior division into lexical units is again avoided in order to be able to compare the different complement types. Since the two complements cannot co-occur, they are depicted as representing the same 48 The meaning of complements valency slot in Figure 6. Forget provides a clear case of a semantic specia- lization of [to_INF] as opposed to other possible clausal complement types which can combine with forget. The divalent pattern [NP]+ forget +[to_INF] is clearly more frequent than the pattern with [V-ing] (1,025 occurences in the BNC as opposed to 22). Most example sentences with [V-ing] are of the type I will never forget verbing or imperatives as in (3).

(3)BNC Forget talking to your plants…if you want them to look their best try a bit of good oldfashioned pampering. K1M 2141

Note that the difference in meaning determined for the active declarative patterns seems to be non-existent between the imperative patterns. This underlines the influence the choice of grammatical constructions can exert on the meaning of complements.

Valency Slot I [AGENT] Complement [NP] forget

Valency Slot II [ÆFFECTED] 1 Complement [NP], [V-ing], [that_CL],[wh-CL], [wh to_INF], [NP V-ing] Subtype [INFORMATION] Adjuncts 2 Complement [to_INF] Subtype [TASK]

Figure 6. Partial valency structure of forget (based on VDE: 326)

Even within the subtype INFORMATION (“not remember it”), i.e. compared to the complement type [NP] (with 4,864 occurences in the BNC), [V-ing] is very infrequent. This means that when speakers want to express that they forgot INFORMATION, it is more likely that a noun phrase is chosen than a pattern including [V-ing], which is certainly due to the fact that INFORMA- TION is only very rarely an activity expressed by a complement type includ- ing a predicate such as [V-ing]. In general it can be said that forget – other than agree and more so than talk – is indeed a case where a semantic specialization can be identified, and thus a different meaning for the two patterns established. Alternative Realizations 49

2.2.2 Polysemous complement types

The following sections will discuss cases where one and the same comple- ment type can either realize different semantic nuances within one valency slot or serve as a realization of different valency slots in combination with the same verb.

2.2.2.1 AGENT + cheat + REFERENCE

The verb cheat can be distinguished into two lexical units, i.e. cheat ‘be dishonest’ and cheat ‘sex’ (VDE: 128), which differ in their complementa- tion patterns, e.g. only the ‘be dishonest’ sense can occur with an [NP]- complement as in (1) and the ‘sex’-sense always demands complementation with [on_NP] as in (2), while this complementation pattern can also occur with the other sense, cf. (3):

(1)VDE Everybody tries to cheat her. (2)VDE I asked him if he’d ever cheat on me and added that he didn’t seem excited by me anymore. (3)VDE So you can’t even cheat on your income tax.

From a semantic point of view, a distinction into two lexical units, while useful for lexicographic purposes, might again not be necessary. The victim of a fraud is not always the cheater’s sexual partner, as can be seen in ex- amples (4) and (5).

(4)BNC The truth is probably that Ceauúescu cheated on everyone, and that Pacepa, involved on the pro-Russian side, was unaware of the extent to which Ceauúescu was also double-crossing Moscow. CCK 428 (5)BNC If a public servant cheats on his wife does this make him more likely to cheat on the electorate? G1X 2071

When the context is taken into consideration in (4), everyone is much more likely to refer to political than to sexual partners. In (5), the parallel be- tween his wife and his electorate is obvious. Although one could argue that the two are different, and placing them in such a parallel way is done for comical reasons, i.e. as a stylistic device, the causal connection would not be considered adequate if the two acts of cheating were regarded as com- pletely different types of event. When a semantic description is added to the pattern form, two problems emerge with respect to pinning down the meaning of the complement types 50 The meaning of complements or predicting patterns on the basis of such semantic factors. For once, the participant VICTIM, can be expressed by two different complement types: [NP] and [on_NP]. This matter, including the lack of absolute synonymy between sentences such as (6) and (7) (and also between He cheated his electorate and He cheated on his electorate where the cheating seems to be of different kinds), will be addressed below.

(6)VDE Everybody tries to cheat her. (7)VDE I asked him if he’d ever cheat on me and added that he didn’t seem excited by me any more.

Secondly, the complement type [on_NP] can be a realization of at least two different subtypes, irrespective of whether a distinction is made between cheat ‘be dishonest’ and cheat ‘sex’: [on_NP] can be an OCCASION at which AGENT cheats, a MANIPULATED ENTITY, and also the VICTIM (if cheat ‘sex’ is taken into consideration). The patterns of interest are illu- strated by the examples below as well as in Table 7, while some of them are ambiguous.

Table 7. Subtypes of cheat +[on_X]

OCCASION MANIPULATED VICTIM ENTITY

(10)BNC cheating on nuclear (5)BNC cheats on his wife weapons (8)BNC cheating on his 1931 income tax return (5)BNC cheat on the electorate (9)BNC cheating on the materials (11)BNC cheating on Celeste

(8)BNC In 1931 Andrew Mellon was accused by Congressman Louis McFadden of cheating on his 1931 income tax return... EBV 1822 (9)BNC ‘They're right bastards who cheated on the materials…’ CJT 2270 (10)BNC If Iran is cheating on nuclear weapons, the International Atomic Energy Agency has yet to catch it at it. CRC 1468 (5)BNC If a public servant cheats on his wife does this make him more likely to cheat on the electorate? G1X 2071 (11)BNC ‘How did you feel about cheating on Celeste?’ JY9 3218

While the dividing line between the different participants or subtypes of participants is not a clear-cut one, differences do exist and are covered by one and the same complement type. Example (8) is ambiguous in the sense that Andrew Mellon could have manipulated his income tax return or he could have cheated on the occasion of completing his tax return, i.e. when Alternative Realizations 51 it came to completing it, Andrew cheated. Whether making such a distinc- tion is necessary from the point of view of sentence processing is of course arguable. Similarly, materials in (9) could be manipulated but whether the complement here is intended to express the entity itself or rather stress the occasion depends on the speaker’s intention. Sentence (10), however, shows that it could indeed make a difference: when Iran is cheating on nuclear weapons, the only possible interpretation is that the weapons are the OCCASION at which the deception takes place, i.e. when it comes to nuclear weapons, Iran is cheating, and not Iran is manipulating their nuc- lear weapons. The status of these semantic entities is by no means unproblematic. From a purely semantic point of view, the difference could be seen as equally significant as the difference between separate participants of a verb. Thus the line of reasoning that these are three different valency slots could, from that point of view, be considered justified. The analysis as subtypes, however, is again preferable since the participants cannot co-occur. If they are analyzed as different valency slots, an additional restricting mechanism would be necessary to rule out this kind of co-occurrence. Further examples suggest that this participant might also license a realization by means of the complement type [about_X], [over_X] or [at_X], illustrated in Figure 7 and by examples (12)–(14) below.

Valency Slot I [AGENT] Complement [SCU]

Valency Slot II [REFERENCE] 1 Complement [on_NP] cheat Subtype [OCCASION] Subtype [MANIPULATED ENTITY] Subtype [VICTIM] 2 Complement [about_NP] Subtype [OCCASION] Adjuncts 3 Complement [over_NP] Subtype [OCCASION] 4 Complement [at_X] Subtype [OCCASION]

Figure 7. Partial model for the valency structure of cheat 52 The meaning of complements

(12)BNC If we cheated about not having it back you know, when par- ents came in June but we don't need it immediately you know, it doesn't have to be photocopied so that we all… F7G 744 (13)BNC And of course, as I'm the Editor and set up the rules for these features, I can cheat a little bit over costs ... CGH 52 (14)VDE I watched Yul cheat at Monopoly.

A classification into subtypes yields the following picture: the examples found for [over_X], [about_X], and [at_X] clearly seem to be instances of the subtype OCCASION, while [at_X] seems to be restricted to certain con- texts such as games. Sentences such as (15) and (16) instantly trigger a reading with Mary being the REFERENCE of the fraud and not the VICTIM.

(15)NSA Peter cheated over Mary. (16)NSA Peter cheated about Mary.

It is difficult to assign a general semantic role for valency-slot II due to the possible variation found within it (REFERENCE has been chosen as most neutral role description). While there are four different parallel patterns with cheat +[on/about/over/at_X], they only partially cover the same se- mantic range (but still cannot co-occur). The use of realizations other than [on_X] seems to be severely restricted (one match for cheat +[about_X] and six matches for +[over_X] compared with 72 for +[on_X] in the BNC) and only [on_X] seems to be eligible for all nuances of meaning. Another problem can be demonstrated with the help of the following pair of example sentences:

(17)NSA Peter cheated on costs. (18)NSA Peter cheated on Mary.

Sentences like (17) and (18) are not ambiguous: (17) clearly triggers a REF- ERENCE reading, whereas (18) only allows a VICTIM reading, unless no additional comment pointing in a different direction is given. Thus on could also be seen as neutral in that the meaning of the complement is derived from the following lexeme: if the lexeme has the feature –human, the only possible interpretation is REFERENCE, and if the lexeme has the feature +human it is VICTIM. However, the classification of VICTIM is also proble- matic because there are two types of VICTIM in the process of cheating, a problem which can be solved by deferring it to the lexical unit in saying that there are two types of cheating as was done in VDE (p. 128): Alternative Realizations 53 cheat (‘be dishonest’) He cheated Mary. / He cheated the electorate. cheat (SEX) He cheated on Mary.

Alternatively one could distinguish between two types of VICTIM, e.g. VIC- TIM_SEX and VICTIM_FRAUD. As these are again incompatible in one and the same pattern, they must be seen as subtypes of one and the same parti- cipant VICTIM. Yet this does not yet solve the problem posed by sentences such as

(5a)BNC If a public servant cheats on his wife does this make him more likely to cheat on the electorate? G1X 2071 (5b)NSA If a public servant cheats on his wife does this make him more likely to cheat the electorate?

Either it must be assumed that the sentence includes two different lexical units of cheating – ‘sex’ in the first instance and ‘be dishonest’ in the second – although the second pattern is typically associated with cheat ‘sex’ and not with cheat ‘be dishonest’. Alternatively one must assume that [on_X] allows for another subtype besides VICTIM_SEX. In either case it is world knowledge that tells us that the two activities described in (5) are not the same. This would imply that [on_X] is a realization of two different valency slots, i.e. VICTIM and REFERENCE, i.e. [on_X]_VICTIM_SEX as op- posed to [on_X]_REFERENCE_OCCASION and [on_X]_REFERENCE_MA- NIPULATED ENTITTY, although they block each other. Yet to analyze VIC- TIM and REFERENCE as being in the same slot is also problematic, since combinations of [NP]_VICTIM_FRAUD and [over_X]_REFERENCE_OCCA- SION are possible as illustrated in (19) and (20):

(19)BNC Ralph claims that James cheated him VICTIM/FRAUD over the sale of Carinish Court REFERENCE/OCCASION. CKB 3235 (20)BNC Former club chairman Brian Hillier, who last year served a prison sen- tence for cheating the taxman VICTIM/FRAUD over cash payments to players REFERENCE/OCCASION, told the jury at Bristol Crown Court, To- bin had tried to blackmail him. K1J 769

The model used here is not able to capture this kind of incompatibility ex- cept with the help of an external blocking mechanism (see Figure 8 below). What such “external mechanisms” could look like outside this analytical model, i.e. what has to be imagined from a cognitive point of view is diffi- cult to answer. However, the most likely solution seems to be storage of this kind of knowledge, since any rules which could capture such pheno- mena appear to be more complex than the information itself. 54 The meaning of complements ] ] ] FRAUD FRAUD SEX ] ] / / / Adjuncts GOODS VICTIM VICTIM VICTIM VICTIM [of_NP] [of_NP] of_NP] [out [on_NP] [NP] IV [ III III [ Complement Complement Complement Complement 2 1 Valency Slot Slot Valency Subtype [ Valency Slot Valency 1 Subtype [ Subtype [ 2 II 1 and II 2 incompatible with III 1 III 1 with 2 incompatible II II 1 and cheat ] ] ] ] ] ] ] AGENT

[NP] REFERENCE OCCASION ENTITY MANIPULATED OCCASION OCCASION OCCASION [over_NP] [at_NP] [ [on_NP] [about_NP]

II [  Valency Slot I Slot Valency [ Complement Complement Modified model for the valency structure of cheat  Complement Complement Subtype Subtype [ [ Valency Slot Slot Valency 1 2 Complement 4 Complement Subtype Subtype [ Subtype 3 [ Subtype Figure 8. 8. Figure Alternative Realizations 55

Again it is interesting to take differences in frequency into considera- tion. In 45 of the 71 example sentences identified in the BNC with cheat +[on_X], the particle complement is obviously a realization of the partici- pant VICTIM, while 26 are best analyzed as REFERENCE.10 The availability of other complement types for the realization of the role REFERENCE could lead to the assumption that [about_X] is the preferred realization of REFERENCE, since [on_X] is already predominantly marked as the prototypical realization of VICTIM. If the filler of the complement type is animate, it does not even allow for a different interpretation. How- ever, in the BNC only one single example sentence with cheat +[about_X]_ REFERENCE_OCCASION was found as opposed to 26 for cheat +[on_X]_ REFERENCE_OCCASION. This shows that even if there were room for specialization, i.e. since it would avoid ambiguity because otherwise different kinds of roles have to be realized by the same complement type, no use is made of this option unless speakers are forced to in order to avoid ambiguity as in the case of examples of the following type, where the only possible interpretation is “in reference to Mary”:

(21)NSA Peter cheated about Mary.

This sentence was marked as incorrect by native speakers and only ac- cepted when an explanation of the context was provided. This apparent inacceptability of [about_X]_REFERENCE in combination with +animate is another factor which speaks for a combinatorial meaning of the various aspects and against a fixed pattern or complement type meaning. The re- strictions can neither be accounted for on the basis of the pattern [SCU]+ verb +[on_X] nor the complement type [on_X]. Thus, it is part of the lexi- cal meaning of the filling of the complement type (i.e. the complement of the particle) which accounts for restrictions and pattern preferences respec- tively. Similar cases can be found with argue and inform.

2.2.2.2 AGENT + argue + TOPIC/REFERENCE

The verb argue can mean ‘dispute’, ‘discuss’, ‘claim’ (i.e. argue that…), and ‘propose’ (i.e. argue in favor of…). These differences in meaning are again accompanied with differences in the verb’s complementation patterns

10. See Appendix 2 at htttp://dx.doi.org/10.1515/9783110240788_appendix2 for the example sentences on which the semantic analysis is based. 56 The meaning of complements

(VDE: 46). Of interest here are only the first two lexical units, which both specify TOPREF and which can both be complemented with the same com- plement types. In the sense argue ‘dispute’, all three complement types – [about_X], [over_X], and [on_X] – realize the same valency slot, which can be labeled TOPREF. In the sense argue ‘discuss’, they appear to realize two different participants: TOPREF, realized by [over_X] and [about_X], and BASIS (cf. the label SUPPORT in FrameNet), realized by [on_X]. The pat- terns available for the two lexical units are illustrated in Table 8. This time, an analysis with two valency slots is necessary because TOPREF and BASIS can co-occur, although in some cases it might be arguable whether the latter is really dependent on the verb or better analyzed as an adjunct.

Table 8. Argue ‘dispute’ and argue ‘discuss’ (based on VDE: 46)

argue ‘dispute’ argue ‘discuss’ [SCU]+ argue +[over_NP] 99 [SCU]+ argue +[about_X] 99 [SCU]+ argue +[on_NP] –– 9 [SCU]+ argue +[with_NP]+[about_X] 9 –– [SCU]+ argue +[with_NP]+[on_NP] 9 ––

From a purely semantic point of view, the main difference between argue ‘dispute’ and argue ‘discuss’ appears to be the “aggressiveness” of the event. As can be seen in examples (1)–(4), i.e. when only divalent patterns are taken account of, the distinction between the two lexical units is a mat- ter of gradience. There is indeed a difference, however, when other patterns are included in the analysis as illustrated by sentences (5) and (6).

(1)VDE They talked and argued about the money for about an hour and a half. = ‘dispute’ (2)VDE Although experts argue about the importance of different risk factors for heart disease the consensus is that risk factors have an accumulative effect. = ‘discuss’ (3)VDE We had been arguing a lot over little things. = ‘dispute’ (4)VDE The summit’s close was delayed for five hours while ministers argued over and eventually accepted the tougher wording. = ‘discuss’ (5)VDE Sometimes he argued with her because he was a man, a peasant, and couldn’t swallow taking orders from a woman. = ‘dispute’ (6)VDE The parties do not even argue the case in front of each other. = ‘dis- cuss’ Alternative Realizations 57

Argue ‘dispute’ implicitly presupposes an opponent, whereas the more neutral argue ‘discuss’ does not and cannot occur in trivalent patterns. Nevertheless, both can take plural subjects which can cover the AGENT as well as a potential OPPONENT as in (1) and (3) above. This can again be taken as an argument in favor of treating them as one single lexical unit while accepting slight semantic differences. Yet the pattern choices are even more complex. As illustrated in Figure 9, [on_X] is available for ar- gue ‘dispute’ in trivalent patterns only (while [over_X] can occur in diva- lent patterns only and [about_X] can occur in both divalent and trivalent patterns). Argue ‘discuss’ – even if complemented with [on_X] – does not seem to take any additional complements except for the AGENT and ex- presses a different participant than the other two complement types.

Figure 9. Partial VDE entry of the verb argue (p. 46)

This is indicated in VDE by different argument slots (A-II vs. B-III and B- IV) (see above). Indeed, when the participants expressed by these patterns are defined, it becomes clear that the complement type [on_X] is not al- ways synonymous with the other two complement types.

(7)BNC I would argue with the author on the merits of the wing-down as op- posed to the crab cross-wind landing techniques TOPREF, ... CAU 562 (8)BNC “It is a truth universally acknowledged that a single man in possession of a good fortune must be in want of a wife,”said James. GV8 3309 “I'm not going to argue with Jane Austen on that one TOPREF,” Matthew said. GV8 3310

In (7) and (8), [on_X] seems to be best described as TOPREF. This inter- pretation is, however, rather unlikely for the examples below where BASIS is a much better classification of the underlined sequences: 58 The meaning of complements

(9)VDE The utmost confusion is caused when people argue on different statis- tical data BASIS. (10)BNC But since the biologists operate within the quantitative framework of statistically based natural science, while the socio-cultural anthropolo- gists mostly argue on the basis of intuition BASIS, communication be- tween the two sides is very difficult. H10 379 (11)BNC The reformers argue their case on the assumption that voting beha- viour experienced under the current mode of election would most like- ly continue under a new mode BASIS: this is, as Geoffrey Alderman has pointed out, a most unlikely hypothesis. J57 1749 (12)BNC …; Hume seems to agree that we understand propositions about unob- served objects, although he does argue on independent grounds BASIS that they are mostly false. F9K 222

Other examples found in the BNC support the hypothesis that on can have at least one additional meaning, i.e. BASIS. However, example (10) is the only example sentence found in the BNC with a divalent use. Thus even for argue ‘discuss’ this pattern seems to be rather rare. The trivalent examples are best analyzed as BASIS as well. Example (11) is best classified as an instance of what was defined as the lexical unit argue ‘discuss’ with [on_X] as a realization of BASIS, since the assumption (…) is clearly not the topic of the controversy. Example (12) illustrates the use of the lexical unit argue ‘claim’ (following the distinctions made in VDE), where [on_X] is again clearly BASIS and not TOPREF. Example (10), the divalent example, is debatable on the grounds that [on_X] seems to depend on the choice of the phrase the basis rather than on argue. In that sense it could be analyzed as an adjunct. The same applies in varying degrees to examples (11) and (12), where [on_X] could be ana- lyzed as a complement (semantically it fulfills the exact same function as the other particle phrases which were classified as complement) or as an adjunct (as they are fairly mobile, could be left out, and one might argue that the form, i.e. on, depends more on the following noun than on the pre- dicator). Note that the different meanings of [on_X] described here are the pro- totypical meanings of on as presented in the Oxford English Dictionary (cf. 2.1.2). Thus, in combination with argue, [on_X] appears to be polysemous. A distinction of this kind could neither be verified for talk nor for agree. Although a co-occurrence of the two complements [about_X] and [on_X], or [over_X] and [on_X] was not found in the BNC, it is clearly plausible from a semantic point of view, since the two interpretations of [on_X] are semantically rather different, and invented sentences were also accepted by Alternative Realizations 59 native speakers. For this reason [on_X] must be seen as a realization of a different participant and thus a different valency slot. At the same time, if used in a divalent pattern, [on_X] seems to fulfill the same function as [about_X] and [over_X] yet there is obviously no co-occurrence of two [on_X] complements realizing different participants in one and the same sentence. Figure 10 illustrates the valency structure of the verb argue. Comple- ment types which are interrelated, i.e. whose realization depend on one another, are marked gray. According to VDE (and supported by the lack of findings in the BNC) one influential factor is the presence or absence of valency slot II, i.e. OPPONENT. If there is one, there is also more likely a TOPREF of the discussion than a BASIS of someone’s point. Yet, as has al- ready been pointed out above, a plural subject can take the function of OP- PONENT as well, which is why this condition alone is not sufficient.

Valency Slot I [AGENT] Valency Slot II [OPPONENT] Complement [NP] Complement [with_NP]

Adjuncts argue Valency Slot III [TOPREF]

1 Complement [about_NP] 2 Complement [over_NP] 3 Complement [on_NP] Valency Slot IV [BASIS] Complement [on_NP]

Figure 10. Valency structure of argue

Figure 11 and Figure 12 depict argue separated into two lexical units. This does not solve the problem either: the restriction cannot be explained on the basis of the meaning of the complement types and would therefore demand further external regulating mechanisms. Hence neither solution is complete- ly satisfactory, which shows that an explanation of the valency of a verb on the basis of participant roles as for instance attempted in FrameNet cannot be sufficient. 60 The meaning of complements

Valency Slot I [AGENT] Valency Slot II [OPPONENT] Complement [NP] Complement [with_NP]

argue Adjuncts 'dispute' Valency Slot III [TOPREF]

1 Complement [about_NP] 2 Complement [over_NP] if [on_X] Æ 3 Complement [on_NP] [with_X]

Figure 11. Valency structure for argue ‘dispute’

Valency Slot I [AGENT] Valency Slot II [BASIS] Complement [NP] Complement [on_NP]

argue Valency Slot [TOPREF] 'discuss' Adjuncts

1 Complement [about_NP] only one 2 Complement [over_NP] complement 3 Complement [on_NP] [on_NP]

Figure 12. Valency structure for argue ‘discuss’

This clearly weakens any position which rests exclusively on semantic criteria. Quite obviously syntactic co-occurrence restrictions are not always rooted in the meaning of the respective complements. There is a clear idio- syncratic component, which can only be accounted for by assuming an important role for the storage of such information in any cognitive plausible model of language.

2.2.2.3 AGENT + inform + TOPIC

According to VDE, the verb inform allows variation in the realization of TOPIC in three different trivalent patterns: Alternative Realizations 61

(A) [SCU]+ informs +[NP]+[on_NP] (B) [SCU]+ informs +[NP]+[of_X] (C) [SCU]+ informs +[NP]+[about_X]

There is indeed a semantic difference between, on the one hand, [on_X] and, on the other hand, [of_X], [about_X], and also complementation with a [(that)_CL], which has a similar semantic function (VDE: 429). While [of_X] and [about_X] seem to be interchangeable to a considerable degree, this is not necessarily true of [on_X]. The number of sentences with the latter is also much smaller in comparison to [about_X] and [of_X]. In the BNC, there are about 900 sentences containing the respective patterns with [of_X] and slightly more than 300 with [about_X] but only 42 sentences with [on_X], which might indicate that [on_X] is much more restricted in its use. A close analysis of the sentences, however, shows that no such ex- planation can be found in the context in which those complements occur. There were both sentences where inform refers to fairly general topics (nouns such as matter, issue, subject, and topic were taken as indicators here) as well as sentences where the complement describes something ra- ther specific, so that a specialization in this field is unlikely. A difference can be pointed out, however, when these sentences are compared with a sample of the sentences complemented with [of_X], as these can often be paraphrased by a sentence with [that_CL]. Compare the following sentences:

(1a)BNC Over a cognac he gloomily informs us of the Japanese surrender. FYV 2664 (1b)NSA Over a cognac he gloomily informs us that the Japanese surrendered. (1c)NSA ?Over a cognac he gloomily informs us on the Japanese surrender.

Although most native speakers accepted sentence (1c), it was sponta- neously marked as incorrect by some and corrected by replacing on with of. However, when asked for the possibility of this sentence in a different kind of context, (1c) was partly accepted with a slight difference in meaning in that this modified sentence is more likely to include more detailed informa- tion. If this were true, [on_X] focuses on the content, whereas [of_X] just names the topic or even realizes AFFECTED (i.e. the same participant as [that_CL]). This is contrary to the combination of these complements with the verb talk, where [about_X] was analyzed as focusing on the content and [on_X] just named the topic (described as TOPIC_CONTENT vs. TOPIC_ THEME) (cf. 2.2.1.1). It has to be taken into account, however, that this in- terpretation does not reflect the general tenor of the native speaker inter- views which clearly speaks in favor of a synonymy between [of_X] and 62 The meaning of complements

[on_X] for those native speakers who found the sentence with [on_X] ac- ceptable. Moreover [of_X] also shows signs of polysemy, i.e. not all sen- tences can be paraphrased by [that_CL], which means that they can also focus on the content as in example (2a).

(2a)BNC As well as providing multi-line switchboard status information, the screen informs the proprietor or receptionist of the status of the rooms, for exam- ple, whether that early morning alarm has actually been set, .... HX4 115

Here, the status of the rooms is modified by additional information and thus clearly focuses on the content as it implies further details. This sentence cannot be paraphrased by [that_CL] but easily by [on_X].

(2b)NSA ?..., the screen informs the proprietor or receptionist that the rooms are in a certain status, for example, whether that early morning alarm has actually been set, ... (2c)NSA …, the screen informs the proprietor or receptionist on the status of the rooms, for example, whether that early morning alarm has actually been set, ...

Examples like (3) and (4) finally show that there seem to be no general lexical restrictions either and it will prove rather difficult to define a clear- cut boundary between the two complement types.

(3)BNC … there were also a whole series of committees with representatives from the regions who needed to be consulted and informed on the views of headquarters. CRD 111 (4)BNC The Home Office informs local forces of its views by advisory circulars of which there are about 100 a year. G1H 1707

Thus, [of_X] seems to be available for both TOPIC and AFFECTED and must therefore be considered polysemous, i.e. there are two polysemous valency patterns: [x]+ informs +[x] +[of_X]_TOPIC as in example (2), where the underlined passage in bold type cannot be paraphrased by [that_CL], and [X]+ informs +[X] +[of_X]_AFFECTED, as in example (1), where this is possible. If polysemy is taken as an explanation for this difference in mean- ing, however, it must be pointed out that the difference is certainly only a very fine one. Figure 13 illustrates the findings in a valency structure stemma. Besides AGENT and BENREC, the valency slots of interest are TOPIC (valency slot III) and AFFECTED (valency slot IV). [On_NP], which is a realization of the Alternative Realizations 63 participant TOPIC can also express a different semantic nuance, i.e. WRONGDOER (a label used in FrameNet), as in (5).

(5)BNC They talked about the glory of the family and then taught children to inform on their parents. A7A 531

WRONGDOER is here taken to be a subtype of TOPIC, since it cannot co- occur with any other type of topic. VDE solves the difference in meaning by including the pattern [NP]A/[by_NP]+ inform +[on_NP]P/[against_NP]P as a subsense (a notation which is used inVDE [p. xiv and 428] for mean- ings of a verb “which apply to a very small number of patterns”).

Valency Slot I [AGENT] Valency Slot II [BENREC] Complement [NP] Complement [NP] [by_NP]

Valency Slot III [TOPIC] inform 1 Complement [of_NP] 2 Complement [about_NP] Adjuncts 3 Complement [on_NP] Subtype [TOPIC] Valency Slot IV [AFFECTED] Subtype [WRONGDOER] 1 Complement [that_CL] 2 Complement [of_NP]

Figure 13. Valency structure of inform (cf. VDE: 428)

The difference in meaning results from clear contextual constraints for the application of this verb use. The possible commutation with [against_X] emphasizes the negative connotation attributed to the participant. Another difference seems to be that this use is not possible in trivalent patterns which express BENREC. In the present context, inform provides an example of a situation where two complement types which can be used synonymously – [of_X] and [on_X] – can realize two and three different semantic nuances respectively, all within the same valency slot TOPIC, while [of_X] must also be seen as an alternative realization of [that_CL] in the valency slot labeled AF- FECTED. These cases pose a problem to theories which assume a direct relationship between semantic participant and certain formal types of com- 64 The meaning of complements plementation. Not only can certain complement types express different semantic nuances within one single valency slot, they can even express different participants with one and the same verb. This clearly contradicts a one-to-one relationship between form and meaning.

2.2.3 Blocked patterns

As has been demonstrated above it is problematic to pin down the semantic properties of complement types outside the context of a specific verb. Con- sequently, it is problematic to predict the acceptability of patterns on the basis of the meaning of a complement type. Nonetheless there seem to be cases where such semantic features of complement types might be said to be responsible for their incompatibility with certain verbs and consequently the blocking of certain patterns. In 2.2.1 it has been shown that the comple- ment type [about_X] is a prototypical realization of TOPIC and REFERENCE. However, not all verbs specifying a TOPIC or REFERENCE can be comple- mented with [about_X] as a realization of this participant type. The verb unite, for instance, is a case in point. Unite can be complemented with +[on_X] and [over_X] specifying REFERENCE: The number of example sentences found for these two patterns (12 for the pattern with on and 3 for pattern with over, only considering complements and not adjuncts) does not provide a sound basis for analyzing differences between the two patterns. A comparison of the two sets of examples shows that [on_X] is again found with general as in (1) and (2), as well as with specific REFERENCE as in (3) and (4).

(1)BNC ‘Women do unite together on projects and support each other -- the fact that feminism survives, and the existence of projects like Spare Rib, show how well women work together.’ ARW 871 (2)BNC If the partners were united on policy, why have two parties, two or- ganizations and two leaders? FRB 1727 (3)BNC Mr Wallace acknowledged that his own ideas for a Scottish parliamen- tary council were unlikely to proceed, but a campaign for a multi-option referendum was ‘the only thing we can all unite on as it encompasses all our differences’. K5M 5778 (4)BNC After a tough debate within the Miners' Federation in 1911 -- the miners were not united on the baths' efficacy …. EG0 1918

There are no “general” [over_X] examples, yet with only three examples no representative claims can be made here. Alternative Realizations 65

(5)BNC They would unite with the workers over their struggle for economic improvements only so that ‘by smuggling in the Marxian doctrine, [they could] transform it into the ideological struggle of classes.’ FB1 1603 (6)BNC Farmers and environmentalists unite over CAP reforms. J2V 2 (7)BNC The community can unite over that proposition while disagreeing about which alternative would be more and which less just. JXJ 829

What is remarkable, however, is that there is not a single example sentence with [about_X], which is otherwise a typical realization of this participant type. This could be due to a specification of the respective participant of unite in such a way that it is always and necessarily different from other TOPIC participants so far discussed, e.g. in that it includes resultative as- pects, i.e. the basis on which the unification actually takes place. Thus the question is whether certain particles carry a specific kind of meaning which could, for instance, be important for the expression of a certain type of situ- ation. For the argument to hold, [about_X] would have to be intrinsically non-conclusive and indeed, non-conclusive verbs such as argue or question can be complemented with [on_X], [over_X], and [about_X]. However, there is counter-evidence with respect to the verb agree as presented above. [On_X] in combination with the verb agree is indeed more likely to express conclusiveness than [about_X]. Despite this, it has been shown that agree can be used still fairly freely with both of these complement types and also [about_X] was found in conclusive contexts, as in:

(8)BNC So we agreed about tea, and invited Father D'Arcy to join us. H9X 152

Although unite is semantically fairly similar to agree (both have similar participant patterns: AGENT + verb + REFERENCE_“RESULT/AGREEMENT”), it is not found in a pattern with [about_X]. REFERENCE in the unite sense is different in that it is obligatorily conclusive, while in the agree sense it depends on the speaker’s intention, yet this does not explain why it is cho- sen in combination with agree in conclusive contexts in the first place. This current blocking can therefore not satisfactorily be explained on the basis of an incompatibility of the complement and a conclusive quality of the verb. It does not seem to be the case that there is a causal relationship between this situation type distinction and the choice of particle. Other blocked patterns can be found with respect to the alternation be- tween [on_X] and [upon_X]. There is no doubt about the fact that these two complement types are fairly similar and it is indeed the case that [on_X] can be replaced by [upon_X] with respect to almost all verbs specifying TOPIC or REFERENCE with an obvious stylistic restriction: upon appears to 66 The meaning of complements be less likely and even impossible in colloquial English and texts produced on a lower stylistic level. Thus [upon_X]_TOPREF is also less frequent than [on_X]_TOPREF. However, a comparison of the verbs in VDE allowing an alternation between these two complement types and those which can only be complemented with [on_X] shows that there are indeed a number of cases where this complement type appears to be blocked. This holds for the participant type TOPIC/REFERENCE but also for other semantic roles or in specific collocations or idioms. It is not possible to identify a semantic fea- ture which distinguishes those which allow variation between [on_X] and [upon_X] from those which do not allow this variation. Table 9 lists those verbs, which are incuded in VDE as only allowing [on_X] and where no example sentences with [upon_X] could be found in the BNC. Asterisked examples were unanimously considered as unaccept- able by native speakers, while examples marked with ‘?’ indicate that the general tendency was that the sentence is not acceptable. Interestingly, these sentences were typically accepted by American native speakers who thought that these sentences would be acceptable in British English, while British native speakers unanimously rejected all of these sentences. This is probably also connected to the stylistic difference between on and upon which might be reflected in stereotypes of different dialects of English.

Table 9. Selection of blocked alternations between [on_X] and [upon_X]

verbs which refuse complementation with [upon_X] specific role

argue ?They argued upon this subject matter. REFERENCE finish ?His year finished upon a high note when he beat RESULT world No. 1 Edberg. travel ?Currently more than 150,000 passengers travel INSTRUMENT upon city trains every day. talk *He talked upon butterflies. TOPIC question ?They questioned him upon this affair. TOPIC quote ?Do not quote me upon this. REFERENCE tell ?You could tell upon his face. REFERENCE cheat *She cheated upon him/her tax statement. TOPIC/VICTIM

Restrictions cannot only be found with respect to complements such as [upon_X] but also, for instance, when it comes to complementation with clauses and the expression of the participant ÆFFECTED. While allow and permit can be complemented with [V-ing], only allow can also be comple- mented with [to_INF], even if allow +[V-ing] is comparatively rare and Alternative Realizations 67 many of the examples are relatively nominal in character (cf. also Quirk et al. 1972: 11.24).

(9)BNC The partial transparency of the rock allows focussing up and down to assess grain packing. H9S 541 (10)VDE Gently pour the butter into the sieve and allow to drain through undis- turbed. (11a)VDE Congressional negotiators have imposed limits which may permit build- ing only two more of the bombers. (11b)NSA *Congressional negotiators have imposed limits which may permit to build only two more of the bombers.

Similarly, endure and tolerate can be complemented with [V-ing] express- ing ÆFFECTED but only endure can also express this participant by means of [to_INF]. Sentence (13b) was clearly rejected by the majority of native speakers.

(12)BNC I had to endure watching him dismantle all the safeguards Ayling had created. GWG 1010 (13a)BNC But Carteret's wife, who frequented health spas, could not endure to live with him or he with her: there were no children. CD2 1061 (14)BNC They used Hilbert's wine glasses because none of them could tolerate drinking from plastic or cartons but they ate off paper plates of which they had bought a hundred. CDB 1682 ? (13b)NSA *But Carteret’s wife, who frequented health spas, could not tolerate to live with him or he with her: there were no children.

Since chapter 3 will discuss in detail the connection between the semantics of the verb and their pattern choices and the question to what extent these rest on and are predictable from the meaning of the verb, it will suffice here to point out that this irregular blocking does not point in the direction of a unified complement type meaning, despite the fact that certain differences have been identified.

2.3 Conclusion

There are various approaches with respect to how the choice of comple- ments is typically analyzed. While some focus on syntactic aspects, as ma- nifested in the idea of frames (e.g. Chomsky 1984), oth- ers focus on semantic aspects in the sense that the meaning of the verb and/or the complement type causes a certain choice of patterns (e.g. Pinker 1989). In the context of cognitive grammar (e.g. Langacker 1987), also 68 The meaning of complements aspects such as the intention of the speaker are taken into consideration (cf. Hamawand 2002: 2–6). Constructional approaches, finally, assume an addi- tional level, i.e. the construction (e.g. Goldberg 1995, 2006a): A verb can occur in a specific construction when it does not have any participants which contradict the roles provided by the construction. Meaning is then created as the combination of construction meaning and verb meaning. In a way, this approach is also semantic yet it is not confined to the meaning of the verb. Particle phrases, however, are often treated separately in such theories, while most attention is given to noun phrases and the question which functions they can take (see below). In this chapter the focus has been widened by looking at particle phrases functioning as complements and clausal complements. Consequently the analysis also goes beyond the functions of subject and object. The semantic analyses have produced three major findings which will be discussed in reference to the theories men- tioned above: no stable semantic properties, synonymous complement types, and polysemous complement types.

2.3.1 No stable semantic properties

Neither particle nor clausal complements have a fixed meaning, i.e. they are not generally associated with one specific participant. Even if there are certain tendencies, i.e. in the sense that particle phrases headed by on or about are prototypical realizations of TOPIC and/or REFERENCE, there are no stable semantic features which would allow a prediction of the choice of one over the other across verbs. Van Valin (2004: 65) demonstrates that different realizations of noun phrases must be described by very different semantic roles depending on the meaning of the verb.

(1)QE The farmer killed the duckling. Agent Patient (2)QE The rock broke the window. Instrument Patient (3)QE The lawyer received the summons. Recipient Theme (4)QE Many tourists saw the accident. Experiencer Stimulus (5)QE The clown amused the child. Agent Experiencer

While the interesting factor is the high congruence between same form and same construction which nonetheless results in a difference in meaning, one might expect that with particle complements – which are less neutral than [NP]-complements because of the semantic content the particle contributes to the sentence – a semantic difference between realizations of the same form is at least less likely. Conclusion 69

Table 10. Variation between [on_X], [about_X], [over_X], [upon_X], and [of_X]

complement verb participant / subtype (if available) [on_X] talk TOPIC_GENERAL; TOPIC_FORMAL; TOPIC_TITLE agree TOPIC/REFERENCE_AGREEMENT TOPIC/REFERENCE_THEME argue TOPIC/REFERENCE BASIS [different participant] inform TOPIC_SPECIFIC; TOPIC_CONTENT TOPIC_WRONGDOER cheat REFERENCE_MANIPULATED ENTITY REFERENCE_OCCASION VICTIM_SEX [different participant] [about_X] talk TOPIC_NEUTRAL TOPIC_CONTENT agree TOPIC/REFERENCE_THEME TOPIC/REFERENCE_AGREEMENT argue TOPIC/REFERENCE inform TOPIC_SPECIFIC TOPIC_CONTENT cheat ?REFERENCE_OCCASION (based on only 1 example) [over_X] talk BLOCKED (phrasal verb: talk over, see below) agree BLOCKED argue TOPIC/REFERENCE inform BLOCKED/strongly dispreferred (only 1 BNC example) cheat REFERENCE_OCCASION [upon_X] talk BLOCKED argue BLOCKED or at least strongly dispreferred inform TOPIC_SPECIFIC TOPIC_WRONGDOER cheat BLOCKED or at least strongly dispreferred agree REFERENCE_OCCASION [of_X] talk BLOCKED argue BLOCKED inform TOPIC_CONTENT; TOPIC_SPECIFIC AFFECTED [different participant] cheat GOODS as in He cheated him of his fortune agree BLOCKED 70 The meaning of complements

This is also why a differentiation is made in various approaches between “direct” and “indirect arguments” in the sense that direct arguments are noun phrase arguments which receive their semantic role directly from the verb and indirect arguments are noun phrase arguments where it is the par- ticle which instigates the role choice together with the verb (cf. Tenny 1994: 9; Marantz 1984: 17–23). In valency theory, the particle is taken to be part of the complement (e.g. Heringer 1967; Emons 1978; Herbst et al. 2004) but also here the range of participants a particle phrase can express is clearly more limited than that of noun phrases because of the meaning of the particle. Welke (2005: 53–54) points out that in some theories, e.g. Dowty (1982: 56–57), prepositional phrases are not analyzed as arguments at all. Instead, the concept is restricted to relational terms such as direct and indirect object. Welke himself concludes that “noun phrases are more typi- cal arguments than prepositional phrases [because they are] autonomously encoded [translation mine]” (Welke 2005: 54), i.e. the meaning of the pre- position itself reveals the semantic relation to the rest of the sentence. He- ringer (1996: 157) denies such a strict division, however, he acknowledges an “affinity of complements to be rather realized as noun phrases and that adverbials rather have an affinity towards prepositional phrases” [transla- tions mine]. Since particle complement types have a limited semantic range which is decisive for their availability for the realization of a specific par- ticipant, it seems to be sensible to make such a distinction between noun phrase complements and particle complements. On the other hand, there are patterns such as He gave a book to him, where the particle phrase to him corresponds to the first post-verbal [NP]-complement in He gave him a book, which can both be described as BENREC. The findings of the preced- ing paragraphs summarized in Table 10 on the preceding page show that there is great variability inherent in the different particle complement types in combination with a selection of different verbs, i.e. the semantic range is wider than one might have expected. From this it follows that the meaning of the particle alone will simply not suffice to classify a complement semantically. The analysis so far concentrated on what was broadly defined as TOPIC and REFERENCE. The range of particles which had to be included for these roles is also striking. Thus there are particles which can express several different participants (or at least subtypes) and participants which can be realized by a range of different particle complements. This clearly speaks in favor of a more neutral complement type meaning. Although there are in part certain parallels, there is no such thing as a clearly definable meaning for complement types on the basis of which a Conclusion 71 possible occurrence of the type in a pattern could be predicted, i.e. it was not possible to identify stable semantic features which make the comple- ment types candidates for a selection on the basis of their specific meaning across verbs. Thus semantic properties identified throughout the analysis were not complement-type-specific but differed from verb to verb. One complement type can entail different and sometimes even contradictory semantic facets and the specifications made with respect to one verb, e.g. talk, must not be regarded as universal features of the complement types [about_X] and [on_X]. Even for talk it could be shown that a differentiation into ‘formal’ and ‘neutral’ is not sufficient to explain the difference be- tween, and thus predict the choice of [SCU]+ talks +[on_NP] vs. [about_NP]. Beyond talk, they are not very helpful in determining possible semantic differences between, for instance, agree +[on_X] and agree +[about_X]. One and the same complement type can – in combination with the same verb – realize different participants as in the case of inform +[of_X] (TOPIC and AFFECTED) or different subtypes of one and the same participant as in the case of talk +[on_X] (TOPIC_GENERAL, TOP- IC_SPECIFIC and TOPIC_ TITLE) or even both different subtypes and differ- ent participants as in the case of cheat +[on_X] (REFERENCE_OCCASION and REFERENCE_MANIPULATED ENTITY vs. VICTIM_SEX). This is in line with Croft (2001: 113–117) who argues against a “1:1 cor- respondence between linguistic form and linguistic meaning”, assuming that either formal differences exist, e.g. distributional differences, or that forms are polysemous. The kinds of subtypes which emerge are closely connected to other factors such as verb meaning and other participants in the respective patterns. In addition, there are verbs which do not accept the whole range of particle complement types available for other verbs specify- ing comparable participants (e.g. unite). With respect to clausal complements, there also seems to be a strong tra- dition according to which the choice of complement type directly affects the meaning of the sentence and is directly connected to specific facets of meaning of the complement types (e.g. Verspoor 1990; Wierzbicka 1988; see also, for instance, CGEL: 16.40). Indeed, a slightly different situation can be observed in the case of complementation with [V-ing] and [to_INF]. While here the complement type is not absolutely independent of the verb meaning, either, there is at least a certain tendency that [to_INF] has a more purposive reading as opposed to [V_ing], e.g. forget + [to_INF] is confined to a TASK reading. At the same time there are examples, where such a clear differentiation appears to be much more difficult if not impossible. Com- pare: 72 The meaning of complements

(6)VDE I love looking around the shops to see what other retailers are doing and selling. (7)VDE My childhood was a happy one; I just loved to read.

Since chapter 3 will cover the differences between these two complement types and others such as [that_CL], the discussion of the subject will be deferred until there (cf. 3.3.4.6). If the meaning of complement types played a crucial role in pattern availability, the fact that not all of the complement types are available for all verbs with the respective participant pattern should be explicable on the basis of a difference in meaning of these complement types. It has been demonstrated above that while many verbs allow TOPREF to be realized by one or a number of particle phrases out of a fairly confined list of particles, not all verbs with this participant accept all of the particles listed. For in- stance, while [on_NP] and [over_NP] are typical realizations of TOPREF, they are not freely available to all verbs with this participant. Compare:

(8) He argued on it? Æ possible, but semantically different from about (9) *He talked over it. Æ only as phrasal verb He talked it over.

One explanation might be that defining these participants as TOPIC or REF- ERENCE is a semantic under-specification in the sense that there are more features differentiating them. If this were the case, the difference could account for the fact that verbs are not compatible with all complement types generally available for a certain slot. In order for a complement type to be an appropriate realization of a participant and consequently to fill a valency slot in a certain valency pattern, the semantics of the complement type should then be compatible with the participant. There are two possible perspectives: (1) either complement types have a specific meaning or show at least a strong tendency towards a specific meaning and are therefore available as the realization of a certain participant, or (2) complement types must be seen as rather neutral fillers with only a broad semantic contour which arbitrarily combine with a certain range of verbs. As the preceding paragraphs have shown, even a close semantic analysis did not reveal any specific semantic facets for particle complement types which hold beyond its usage in combination with a specific verb. In the case of agree a differ- ence between the complement types could not even be found in combina- tion with the same verb. If the meaning of the complement type cannot be specified (i.e. if it has to be taken as much more neutral than is often sug- gested), and if consequently a correlation between certain participants and complement types cannot be confirmed (i.e. if complement types are avail- Conclusion 73 able for the realization of a certain range of participant types in a rather arbitrary way), the patterns available to a verb cannot be predicted on the basis of the meaning of the complement. This clearly shows that potential rules as to the usage of specific particles do not present themselves as easily available. At best they must be seen as verb-specific rules which is, of course, contrary to the idea of providing a general mechanism. According to Chomsky (1986) the verb hit, for instance, specifies an Agent and a Patient as arguments. Categorial selection or c-selection then determines which form these arguments will have as complements of the respective verb, e.g. a noun phrase for the Patient of hit (Chomsky 1986: 86–87). According to Chomsky (1986: 86) himself, however, c-selection is redundant: “If hit s-selects a patient, then this element will be an NP. If c- selection is redundant, in general, then the lexicon can be restricted to s- selection.” Thus the lexicon should be exclusively specified through s- selection (i.e. semantic selection) which is then represented also on every syntactic level because of the projection principle, while a further formal specification is not arranged for.

Let us assume that if a verb (or other head) s-selects a semantic category C, then it c-selects a syntactic category that is the “canonical structural realiza- tion of C” (CSR(C)). Take CSR(patient) and CSR(goal) to be NP; then hit c-selects NP. (Chomsky 1986: 87)

The question that remains is what counts as a canonical realization of a category and how alternative realizations of the same argument can be ac- counted for, especially when the range of alternative realizations is not the same for different verbs. VDE provides numerous examples of verbs which allow more than one formal complement as the realization of a participant, even if we leave aside those cases where there is a strong tendency in favor of one of the possible alternatives in terms of frequency. Concerning per- suade, Chomsky’s second example from Knowledge of Language, a Patient and Goal are s-selected. C-selection allows for the second argument to be a that-clause, a to-infinitive clause or a particle phrase with the particle of ([of_NP]) as an alternative to the noun phrase solution. Nonetheless Chomsky explains:

In short, the lexical entry for persuade need only indicate that it s-selects two complements, one a goal, the other a proposition. All other features of the VP headed by persuade are determined by general principle of UG …. If we succeed in eliminating recourse to c-selection as well as phrase struc- ture rules, thus reducing syntactic representations at D-structure to projec- 74 The meaning of complements

tions of semantic properties of lexical items, it will follow that the comple- ment of any head in a syntactic representation must be s-selected by it, be- cause there is no other way for the position to exist. (Chomsky 1986: 88– 90)

Klotz (2000) exemplified in reference to corpus evidence that there are clear restrictions as regards alternative realizations of complements (Klotz 2000: 117–119). Chomsky’s principles allow the generation of the follow- ing sentences quoted from Klotz (2000: 116–117):

(10)QE They persuaded John that he should go to college. (11)QE They persuaded John to go to college. (12)QE They persuaded John of the importance of going to college. However, it does not explain restrictions as the following ones:

(13)QE Blair would no longer permit Ginny to come to Seymour House BUT NOT*that she should come to Seymour House. (14)QE I answered her that I'd completed all the case notes and filed them away BUT NOT *to have completed all the case notes and filed them away. (15)QE It was a newsreel film which decided him to go to the front. BUT NOT *that he should go to the front. (16)QE *The general commanded his soldiers of an immediate departure.

Examples (13) and (14) show that there are restrictions as regards the choice of clause type even if the argument expressed is still a PROPOSITION. Hence, the semantic role of the argument cannot be taken to predict its realization. Example (15) demonstrates that a similar verb meaning, in this case “persuade someone” and “decide someone”, cannot serve as a safe indicator for possible patterns. Example (16) finally shows that the choice of particle is also hard to pin down to such general case-marking rules as stated above. Instead we have to accept that there are different particles and their choice clearly depends on the verb, i.e. the verb valency, e.g. to ask for something, to decide on something, to talk someone into something, and never of something in these cases (Klotz 2000: 117). Chomsky (1986) addresses similar problems by pointing out that cases likes these constitute violations of the case filter, a Universal Grammar principle: The fact that ask and wonder c-select a clause but only ask can be complemented by a noun phrase but not wonder, although both s-select a proposition, is thus explained by the fact that ask is a , which assigns objective case, while the other two verbs are intransitive and thus cannot be complemented by a noun phrase in objective case. This shows that a reduction to semantic aspects is not sufficient to account for the for- Conclusion 75 mal usage of a verb. With respect to the example above, the non-semantic lexical information provided does not explain the restrictions in case of sentential complementation.11 A general semantic approach on the basis of the meaning of the partici- pant appears to be problematic with respect to the data analyzed here, espe- cially also the differences in frequency in cases where the same range of complement types is available. Also Rappaport Hovav and Levin (1998) show that lexemes with parallel constructions are problematic from a pro- jectionist point of view:

In an approach which takes argument expression to be determined from a lexical semantic representation via linking rules, verbs with multiple options for the expression of arguments have to have multiple lexical semantic re- presentation. If such variation is the rule rather than the exception – and re- cent studies show that the phenomenon is indeed widespread – then the lex- icon must contain a vast number of verbs with multiple lexical entries. This result is undesirable, however, if only that it seems counterintuitive to have to posit that … we are dealing with six different verbs sweep, nine different verbs whistle, and six different verbs run. (Rappaport Hovav and Levin 1998: 98–99)

Thus what they consider to be the consequence is the necessity of a distinc- tion of verbs into many different lexical units (“verbs with multiple lexical entries”). Even if from a valency point of view the idea of storage is much less problematic, it is counterintuitive, also from that point of view, to posit so many different lexical units for verbs which do apparently not differ in meaning. Rappaport Hovav and Levin (1998: 99) consequently conclude that there must be “linguistic principles governing the range of meanings associated with a verb”, which leads to the possibility of establishing gen- eral verb classes which share the same patterns. However, as will be shown below, such verb classes are clearly not unproblematic either (cf. 4.2.2). This directly leads to the question whether it is indeed the verb that is stored in combination with a “list” of its complement types or whether whole patterns are the unit which is actually stored. According to construction grammar, a difference in form, i.e. the choice of one complement type over another if both are available for the verb un-

11. Note that Chomsky (1995: 191 and 220) himself finally pleaded for an elimi- nation of d-structure and alongside with this the theta-criterion and the projec- tion principle in its previous form: “We therefore derive one major conse- quence of the Projection Principle and ș-Criterion at D-Structure, thus lend- ing support to the belief that these notions are indeed superfluous.” 76 The meaning of complements der consideration, implies a difference in meaning or discourse function (e.g. Goldberg 2006a: 9). If this difference is implied in the form of the construction – which is made up of formal constituents – the substance of the difference must be connected to those forms and thus ought to be identi- fiable unless it results out of their configuration in the pattern. The latter is certainly a relevant aspect in connection with trivalent patterns, which have more than one complement in the predicate. However, when only one sin- gle complement follows the verb and a certain meaning triggers the choice of a certain form, a stability of form and meaning pairs would have to be considered a precondition or at least favorable to an error-free pattern choice. Again, such stable semantic features could not be identified. In cognitive approaches, emphasis is put on the role of the speaker who produces a certain sequence: the speaker or conceptualizer creates the over- all meaning of a sentence by combining such factors as, for example, pers- pective (objective vs. subjective) with the pure semantic content of the participants in the action or situation described. This choice of perspective ultimately motivates the choice of a syntactic pattern (cf. Hamawand 2002: 5). Which kind of complement type is chosen then depends on what seman- tic options these carry with them for a possible conceptualization. This again presupposes that complement types have specific semantic charac- teristics from which their employment in certain sentences as well as their restriction in others follows. Yet this approach allows for combinatorial effects of verb meaning, complement meaning, and the context, which pro- vides a possibility to account for much more aspects discussed here than approaches which isolate such components. The results of the analysis of complement types show that there are ex- amples where the existence of one participant demands or blocks the occur- rence of another one. The participant is not generally blocked (which could be explained via semantic rules) but with respect to a specific complement type and a specific combination of participants, e.g. the availability of the valency construction [SCU]_AGENT + argue ‘dispute’ +[on_X]_TOPREF +[with_X]_OPPONENT, although [on_X] does not occur in divalent patterns. There, TOPREF can only be expressed by [about_X] or [over_X]. For such occasions additional external mechanisms have to be posited, which are hard to formulate as general rules. This can be taken as a clear indication that the meaning of the complement type cannot be seen as the sole deter- minant of pattern form, which already indicates a major problem of seman- tic roles. Semantic role hierarchies are not sufficient for solving the prob- lem of the possibility of co-occurrence and restrictions, either. Moreover, even if there seems to be a certain potential for predicting certain patterns Conclusion 77 in reference to specific contexts, it is not possible to relate this to any gen- eral semantic features of the particles. Klotz (2000) also provides several analyses of parallel valency construc- tions, where the decision as to which of the alternative constructions is chosen seems to be closely tied to the context but not to the semantics of the complement type. Hence we are dealing with a specialization in combi- nation with a specific verb and in relation to the specific environments this verb occurs in. For example, in the case of the construction [SCU]_AGENT + ride +[on_NP]_“VEHICLE” vs. [SCU]+ ride +[in_NP] _“VEHICLE”, bus and train exhibit different preferences: While it is more likely to ride on a train (82.1%) than in a train (17.9%), there is no such difference with re- spect to the distribution of bus (47.8% for in and 52.2% for on) (Klotz 2000: 135). Similarly, it is unlikely to complain at or about pain (only 1.7% of all sentences analyzed as expressing illnesses and symptoms occur in the pattern with about and none in the pattern with at) – instead com- plain of pain is found in the BNC much more often (40% of all sentences analyzed as expressing illnesses and symptoms occur in the pattern with of). Lack and problem, on the other hand, occur both in the pattern [SCU]+ complain +[at_NP] and [SCU]+ complain +[about_NP] (Klotz 2000: 143; for an analysis of these and further examples see Klotz 2000: 133–179). Yet although these distributional differences are based on the co-text, the types of example already indicate that the co-textual indicators have only very limited power when it comes to predicting pattern choice. It does not seem to be the case that the selection of particles can be deduced from any generally applicable rule since the types of restriction or preferences are extremely closely bound to the specific contexts they occur in. Thus, it seems that speakers, when planning an utterance, are influenced towards choosing one pattern over another parallel pattern by the lexical choices they make, i.e. semantic features in the lexemes occurring as com- plements are responsible for the choice of complement type. However, this is clearly not the case with all verbs allowing alternative realizations of one participant, and again, there is no obvious connection to the meaning of the complement type. In part it simply seems that when more than one pattern is available to express the same proposition, it is possible that different preferences develop for the different pattern options. As a consequence it might be possible to distinguish the patterns semantically. However, this does not mean that the formal patterns specialize because they are semanti- cally different. At the same time a specialization does not necessarily entail the loss of an originally more general meaning or a second meaning. In the context of grammaticalization this phenomenon is described as layering, 78 The meaning of complements i.e. “[the] persistence of older forms and meanings alongside newer forms and meanings” (Hopper and Closs Traugott 2003: 124). A similar pheno- menon is very likely to be at work here as well. An interesting methodology for showing which lexical items tend to co- occur with which constructions was introduced by Stefanowitsch and Gries (2003), i.e. collostructional analysis. Collostructions are based on a calcu- lation of the strength of association between a specific verb and a specific abstract construction – as opposed to all other verbs and all other possible constructions. The fact that individual verbs can differ in this respect clear- ly shows that the knowledge we have about the company words keep – in the sense of Firth’s (1957) famous quote – is not confined to specific lexi- cal items in the traditional sense of collocation but can also be extended to the more abstract level of complementation. There appear to be certain tendencies that semantically similar verbs have similar collostructional preferences (e.g. Gries and Stefanowitsch 2004, 2006; Gries, Hampe, and Schönefeld 2005). However, just as the choice of potential collocators of a word can often be limited on the basis of semantic considerations but not pinned down on that basis, this does not mean that collostructional prefe- rences can really be predicted. At any rate, it does not seem to be plausible to speak of rules under such circumstances since the number of rules or the number of exceptions to a more general rule could be expected to be as memory intensive as storage of individual co-occurrences or patterns. While this of course does not yet prove that patterns are stored, to some extent at least it counterbalances the original advantage of rules, i.e. saving memory capacity.

2.3.2 Synonymous and polysemous complement types

Complement types can be synonymous in combination with a specific verb, i.e. when more than one complement type is available for the realization of a valency slot, this does not necessarily lead to a differentiation of the two into different semantically specialized subtypes. Although complete synonymy between alternative patterns expressing the same participants might be considered uneconomic and consequently rather unlikely, patterns where exactly this seems to be the case have been discussed above. Alternative complement types have been taken as repre- sentations of the same participant if they cannot co-occur and express a semantically similar notion. This allows for semantic variation on a sub- type-level, yet several of the examples identified did not exhibit any dif- ferences on this level either. Agree for example can be complemented with Conclusion 79 both +[on_NP] and +[about_NP] to express TOPREF_AGREEMENT or TOP- REF_THEME. If it were the case that the form of a pattern is a projection of an under- lying semantic level, as projectionist approaches suggest (e.g. Chomsky 1965, 1984), one would expect a considerable congruence between form and meaning. Yet if no difference can be identified between two valency patterns, the question arises as to the nature of the projection: how can the choice of one valency pattern expressing a certain participant pattern, over another valency pattern expressing the same participant pattern, be ex- plained on the basis of the meaning of the participant when the semantic participant pattern is identical? As has already been pointed out above, construction grammar assumes a difference in meaning whenever a difference in form is given as a matter of principle. As the examples show, it is not always possible to find such a difference. Native speakers did not always seem to have made such a dis- tinction – at least not consciously. In part they also gave contradictory ex- planations when asked to describe the difference in meaning between such alternatives. This is not to say that if there is a choice between different comple- ments, this choice is always completely arbitrary. As the analyses above have shown, there are obviously cases where the choice of one over the other leads to a clearly different meaning (as in cheat on someone vs. cheat about someone). However, this is not always the case, and, besides, even if a difference seems to exist, it appears that it can be overridden by other fac- tors (e.g. priming, avoidance of repetition etc.). Also the analysis of corpus examples has shown that there are cases where the same content – often even in combination with the same lexemes – is conveyed with different valency patterns. This poses a problem for theories which do not allow for synonymy between different constructions. Moreover, not only can different complements have the same meaning, the analysis above also produced cases where one complement can have different meanings in combination with the same verb. As indicated in 2.3.1, this is in line with the general realization that there is no stable com- plement type meaning across verbs, yet it is also interesting from the point of view of the composition of valency structures. Ramchand (1998: 69) argues that there are languages where “within one lexical predicate, the ‘same’ participant appears to have semantic properties of different semantic relations, depending on other syntactic properties of the construction”. In English there are similar cases as has been shown above for cheat, where [on_X] must be classified as a realization of VICTIM or OCCASION. Thus 80 The meaning of complements one and the same complement can be polysemous in combination with one and the same verb. The two senses can either be similar and exclude each other in the same sentence both for formal reasons (e.g. no repetition of two particle phrases headed by on) and semantic reasons (because they express a similar concept), from which follows that they are subtypes of one va- lency slot. This is, for example the case with talk +[on_X] (TOPIC_FORMAL, TOPIC_GENERAL, TOPIC_TITLE) and, agree +[on_X] and agree +[about_X] (TOPREF_AGREEMENT and TOPREF_THEME). Even if they fill the same va- lency slot, they clearly differ from a semantic point of view. Alternatively, they can have semantic properties which suggest an interpretation as reali- zations of different valency slots. In the latter case the co-occurrence is only restricted on formal grounds, while alternative realizations of the two participants represented by the valency slots are possible. This is the case with argue +[on_X] (TOPREF and BASIS) and inform +[of_X] (TOPIC and AFFECTED). This is in line with observations made by Dowty (1991), who also claims that one argument (corresponding to what here has been called va- lency slot) can have features of different roles (Dowty 1991: 549). Con- structionist approaches can effectively cope with semantic differences be- tween parallel realizations of the same general participant, as they are then simply treated as different constructions. However, they cannot account for formally identical patterns which exhibit differences in meaning in the form of different subtypes, as in the case of cheat [on_X]_ MANIPULATED ENTI- TY or VICTIM or inform [on_X]_TOPIC or WRONGDOER. An alternative solution to this classification problem is to speak of two different lexical units ‘cheat’ and ‘inform’, which can be differentiated between on the basis of contextual semantic factors. Consequently, the two lexical units can then be said to occur exclusively in the respective con- texts. However, such an analysis has the clear disadvantage that the verb must be stored twice in the mental lexicon with its idiosyncratic pattern inventories, although the meaning of the verb hardly changes – as it is the context which results in the different readings. Moreover, this does not solve the problem of verbs which have truly parallel realizations of the same participant, without a clear specialization into subtypes, as for in- stance agree, where the participant TOPREF can imply that an agreement has been reached or simply provide the theme of a possible agreement, while both nuances can be expressed with [on_NP] and [about_NP]. Both patterns would have to be stored separately if they cannot be explained as the projection of different meanings. This means that speakers possibly access whole combinations of verbs and complements in the sense of pat- Conclusion 81 terns (cf. chapter 4). In any case, it is very difficult to imagine how such information could be accounted for by general rules. Another argument against a direct projection of semantic participant onto one specific complement type can be illustrated by examples like (1) which show that two different participants, TOPREF_THEME and TOPREF_ AGREEMENT, are coordinated as if they were realizations of the same role.

(1)BNC Not everyone was agreed, however, that the Hooligans represented a novel development, nor about the scale of the outrages. EDE 644

This also clearly speaks against the original concept of case roles as intro- duced by Fillmore (1968), where “[only] noun phrases representing the same case may be conjoined” (Fillmore 1968: 22). This again indicates that the dividing line between participants does not seem to be absolute and that complement types are extremely flexible in their application to participant structure. 3 Pattern choice and verb meaning

3.1 Assessing the role of verb meaning – theoretical assumptions

When examining to what extent meaning and form are related in a sen- tence, the meaning of the valency carrier is unquestionably at the core of the matter. An examination concentrating on adjectives as valency carriers conducted by Herbst (1983) demonstrates that synonymous adjectives do not necessarily have the same valency structures, i.e. the patterning poten- tial of an adjective cannot be inferred from the patterning potential of another semantically similar adjective.

(1a)QE *He is keen for going to Cornwall. (1b)QE He is keen on going to Cornwall. (2a)QE She is eager for success. (2b)QE *She is eager on success. (3a)QE *She is ambitious for fame. (3b) QE *She is ambitious on fame. (Herbst 1983: 323)

Although examples (1), (2), and (3) could be labeled as AGENT + verb + GOAL, possibly paraphrased by “What he wants is to go to Cornwall/suc- cess/fame”, they obviously differ in their patterning behavior (Herbst 1983: 138, 155, and 323). Looking at various adjectives and their complementa- tion patterns, Herbst concludes:

A mutual correspondence between semantically defined groups of adjec- tives and syntactic patterns exhibited clear correlations between the mean- ing of adjectives and their occurrence in specific patterns. Evidence for a 1:1 correspondence of, on the one hand, complements (in a specific seman- tic function) and, on the other hand, semantically defined groups of adjec- tives could be found in only three [out of 34] cases … (Herbst 1983: 245; [translation mine])

With verbs, however, it is often taken for granted that it is their meaning that determines their potential to be complemented or not, what kinds of complements are allowed and how many. Considering the central role of verbs in the structure of sentences, this has much wider implications for a model of grammar. Assessing the role of verb meaning 83

Different senses of one and the same verb often have different valency properties, i.e. which valency patterns a verb allows is connected to the sense in which the lexical item is used. This has been amply shown with the examples in chapter 2, where a distinction into different units was based on different pattern inventories (cf. VDE). Thus there are verbs, where a dif- ference in meaning is reflected in a difference in the patterns they allow. However, this does not necessarily imply that the reverse is true as well, i.e. that sameness of pattern means sameness of meaning or that verbs sharing the same meaning accept the same patterns. Yet this point of view has been proposed in various discussions on the question of how meaning is linked to form, ranging from rather cautious comments to hypothesizing a one-to- one dependency between the two. Levin (1993: 5) for instance suggests that “[k]nowing the meaning of a verb can be a key to knowing its behavior ... because particular syntactic properties are associated with verbs of a certain semantic type”, and Goldberg (2006a: 58) argues that “[s]emantically simi- lar verbs show a strong tendency to appear in the same argument structure constructions”. According to Haegeman (1991: 35) “[w]hether a verb is transitive or not is not a matter of mere chance; it follows from the type of action or state expressed by the verb, from its meaning” and Pinker (1989: 62) even claims that “[s]yntactic argument structures of verbs are predicta- ble from their semantic structures, via the application of linking rules”. Attempts to establish semantic verb groups which also share the same cha- racteristics as regards their formal usage have been made for instance by Levin (1993) or with a slightly different objective by Francis, Hunston, and Manning (1996) and Hunston and Francis (2000) and also onomasiological dictionaries such as Schumacher (1986). This chapter focusses on the relationship between verb meaning and formal verb complementation in a similar way. Verbs which can be used as synonyms or which are semantically at least similar are compared with respect to the valency patterns in which they can occur in order to show to what extent – or if at all – the usage of a verb can be predicted from its meaning. There are two possible scenarios: (1) If verbs which can be labeled as semantically similar also show congruence in their choice of valency pat- terns, one can conclude that meaning is a good indicator of verbal usage and that it is at least likely that meaning plays a decisive role in how the participants of a verb are realized, i.e. linked to the respective valency complements (and possibly adjuncts). (2) If there is no such congruence, or if it does not exceed a satisfactory threshold, the unpredictability of valency patterns which follows from this has to be accounted for. Then, meaning 84 Pattern choice and verb meaning cannot be accepted as the (sole) instigator of pattern choice. This outcome will obviously strengthen the notion of storage of lexical items in combina- tion with their valency information or even the storage of whole patterns. If the occurrence of a verb in a certain amount of patterns can be predicted from the verb’s meaning but not all possible formal realizations of that verb, one could assume that verb meaning has the function of facilitating storage, i.e. of reducing the number of combinations which have to be ac- tively learned. This mechanism would not work for the unpredictable cases, which would then solely rest on storage. In a sense, one could, of course, argue that as soon as only some aspects are predictable, nothing really is, because the speaker cannot know which patterns can be formed on the basis of a semantic rule and which not (cf. Herbst and Klotz 2003: 138–139 for a discussion of this in connection with their delineation of collocations). The question of whether verb meaning can be used to predict a verb’s formal usage is obviously tied to the question of what exactly is meant by verb meaning and, more specifically, whether one should or could make a distinction between the verb meaning and the participant inventory of a verb. This latter question is again inseparably tied to the question of how verb-specific one defines the participants of a verb. It is possible to identify verbs with the same broad participant inventory which are not synonymous. The underlined parts in sentences (4) and (5) could both be regarded as “THING JUDGED” (Herbst and Schüller 2008: 128) yet there are still obvious differences, e.g. in the intensity or in the formality of context for possible uses of the verbs regard and think. With respect to examples (6) and (7), where he in both cases could be described as “PERPETRATOR” and her in both cases as “VICTIM”, a difference in verb meaning is even more obvious.

(4)QE ... and I suppose that most of us in this room must be regarded as com- pulsive readers ... (Lodge, Small World: 27) (5)QE She did not think it necessary to tell him that the University adminis- tration had been obliged to circulate all members of staff with a memo- randum asking them, … (Lodge, Nice Work: 113) (6)INV He killed her. (7)INV He hurt her.

At the same time there are verbs which have great semantic similarity and are exchangeable in many different contexts but still differ with respect to their participant inventory. For instance blow, explode, and go off are syn- onymous to a very high degree in contexts such as the following: Assessing the role of verb meaning 85

(8)VDE News was coming through that the power station at Chernobyl had blown. (9)VDE Two bombs have exploded in a shopping centre in Belfast. (10)VDE One civilian died, another escaped just in time and in the third case the bomb did not go off.

However, go off does not map an instigator of the action, i.e. the subject is usually realized by the entity which explodes, which explains why the pat- tern [NP]+ verb +[NP] is blocked:

(11a)VDE Somebody had blown a tire on the 14th Street Bridge. (12)VDE Moneymore is only four miles from Magherafelt, where the IRA ex- ploded a van bomb last week. (11b)NSA *Somebody had went off a tire on the 14th Street Bridge.

Still one could argue that they are similar in meaning. This “problem” of similarity could be overcome by using verb-specific classifications such as “KILLER” and “KILLEE”, “HURTER” and “HURTEE”, and “OBJECT EX- PLODED” and “BLOWN” and “PERSON BLOWING/EXPLODING OBJECTS”. Yet as indicated in chapter 1, such roles do not allow for any kind of compari- son and are thus no help in trying to account for either regularity or irregu- larity in the relationship of form and meaning. Croft (1991: 158) summa- rizes that “taken to its logical conclusion, a fine-grained analysis of the- matic roles will result in a unique case frame for almost every verb in a natural language”. Such participant descriptions are not role descriptions in a traditional sense either, i.e. roles which actually identify relations be- tween complements and a predicator. The semantic labels used above are rather descriptions of elements involved in a scene. To account for this difference Tarvainen (1985: 13–15), for instance, clearly differentiates between what he calls categorical roles and relational roles (see also Dür- scheid 1999: 183–184). If verb meaning is assumed to project its role in- ventory, the roles need to be more abstract, since otherwise a distinction between verb meaning and a verb’s participants is rather pointless: to say that a “KILLER” kills a “KILLEE” does not express how these participants are semantically connected to the verb. Describing them as the AGENT of an activity and an entity being AFFECTED by it, leaves room for a certain gene- ralization between verbs and makes this rather artificial division between verb meaning and a verb’s participant inventory less circular. Thus in the remainder of this chapter the focus will be on relational roles as used for instance in chapter 2, while I will also resort to more verb-specific, de- scriptive roles when they can help to illustrate differences between verbs. A 86 Pattern choice and verb meaning sharp division between verb meaning and participant inventory will cer- tainly not be possible, because the participants involved in an event clearly influence or even dominate the character of this event and consequently also the verb meaning. To facilitate the discussion of the findings of the analysis which follows, this rather artificial division has nonetheless been made, which, however, has to be regarded as a purely methodological and not as a systematic, theory-based decision.

3.2 A comparison of semantically similar verbs

3.2.1 The database – introduction and description of methodology

The following analyses will put to the test different semantic aspects which might play a role in the choice of patterns, i.e. to investigate how accurately such aspects can be used to predict the occurrence of a verb in a specific pattern. The basis for this analysis is a database of semantically similar verbs, i.e. verbs which show at least a semantic overlap as regards some of their lexical units and can thus be used to express similar activities or states or generally have a similar pragmatic function. There are different re- sources for identifying such semantic relatedness. On the one hand, there are online-resources such as WordNet, which provides information specifi- cally on how words are related from a semantic point of view, i.e. whether they are synonyms, troponyms, hypernyms etc. On the other hand, there are numerous synonym dictionaries and thesauri which try to list the semanti- cally most similar words for each lemma included. The source which was taken as the starting point in this analysis, however, is VDE, which is not primarily concerned with the different meanings verbs can have but rather their different valency patterns, or more specifically the patterns in which the different lexical units of a verb can occur. To identify these different lexical units, the authors of VDE employed semantic labels, e.g. para- phrases and synonyms. These were taken as the basis for a preliminary list of semantically similar verb uses, which was then manually thinned so that it included only those verbs which were rated most comparable. The re- sulting groups were cross-checked with the help of The Oxford Dictionary of Synonyms and Antonyms (2007), the Oxford Learner’s Thesaurus – A dictionary of synonyms (2008), the Longman Language Activator (1993), and WordNet, and – where appropriate – they were supplemented by words listed in these sources. The resources used differ in part to a considerable degree with respect to what they include as a synonym and what not. This is due to diverging judgments as regards the semantic similarity of certain A comparison of semantically similar verbs 87 words on the side of the lexicographers but also, quite simply, to a different choice of lemmata included in these dictionaries. At any rate one has to be aware that the lists of verbs given in these dictionaries are not to be taken as synonymous in every possible context: there are stylistic or collocational differences or even fine semantic differences, which are not necessarily indicated. This has also led to the assumption of different types of syno- nymy (Lyons 1981: 50–51) or different degrees of synonymy, e.g. total and partial synonymy (Bußmann 1983: 525–526). To some degree the same in- evitably applies to the verb groups chosen here, despite efforts to cancel out such differences as far as possible. The final database consists of 22 verb groups, which are illustrated in Table 11.

Table 11. Verb groups used in the analysis

‘shared meaning’ verbs expressing this ‘shared meaning’ 1 ‘answer’ answer, reply, respond 2 ‘assume’ assume, suppose, presume, guess, gather 3 ‘choose’ choose, select, pick, elect, appoint, name, nominate, designate 4 ‘declare’ declare, pronounce, announce, proclaim, state 5 ‘get’ obtain, receive, get, acquire, buy, purchase 6 ‘hire’ hire, rent, lease 7 ‘intend’ mean, intend, plan, aim 8 ‘like’ love, like, adore 9 ‘not lose’ keep, maintain, preserve, sustain 10 ‘permit’ allow, permit, authorize, entitle 11 ‘practice’ train, rehearse, practice 12 ‘propose propose, suggest, recommend 13 ‘quarrel’ quarrel, argue, dispute, bicker 14 ‘remember’ recall, recollect, remember, reminisce 15 ‘show’ indicate, show, demonstrate 16 ‘start’ start, begin, commence 17 ‘teach’ instruct, teach, train, educate 18 ‘telephone’ phone, ring, call 19 ‘think’ consider, think, reflect, ponder, contemplate, judge 20 ‘think highly’ respect, admire, esteem 21 ‘tolerate’ bear, endure, tolerate 22 ‘urge’ press, urge, push

For the formal comparison of these verbs, or more precisely these lexical units, the patterns in which they occur were extracted from VDE (when 88 Pattern choice and verb meaning included) or from the BNC. VDE already provides a very consistent data- base since it is based on the COBUILD corpus and cross-checked by native speakers. Remaining gaps were cross-checked with the help of the BNC, and where no example sentences could be identified at all, native speaker interviews were carried out. This is in line with Rudanko (1989: 13) who argues in reference to infinitivial complementation that “the two criteria [i.e. native speaker intuition and recorded usage] should be kept in balance and checked against each other: neither isolated instances or recorded usage that are felt to be deviant by native speakers … nor idiosyncratic intuitive judgments can form a reliable basis for the study of … complementation”, which certainly also applies to other types of complementation. For the native speaker assessments, semistructured interviews were con- ducted with a number of British and American native speakers. In order to get reliable results and to cope with the amount of data, the sentences were divided into several test batteries with the goal to get an assessment of each sentence by at least four native speakers. They were given a questionnaire with invented examples or authentic sentences from the corpus-based VDE or the BNC which were modified by exchanging the verb or the pattern and – where necessary – by making further, e.g. stylistic adjustments. A context was provided where necessary. Alternatively, if available, they were given authentic examples found on the internet. Since the authorship in such ex- ample sentences it is often not clear, they have to be treated with caution and were only rated as acceptable, if this was confirmed by a native speak- er. For the assessment, a four point scale was used: native speakers were asked whether the sentence is acceptable and whether they could imagine using such a sentence themselves, or whether they think the sentence is not acceptable under any circumstances.12 If no clear judgment was possible, the sentence could be labeled marginally acceptable or unlikely. In the rare cases where a native speaker was completely undecided, i.e. was not able to even indicate a tendency in favor or against acceptability, this assessment was not taken into consideration in order to avoid a forced answer. The ac-

12. To ensure that the judgments were made on the basis of the pattern and not on stylistic or contextual reasons, the interviewees were always encouraged to point out the mistake or to correct the sentence. If the reason why the sen- tence was considered unacceptable was not connected to the syntactic pattern of the sentence, the pattern was rated as acceptable and – where appropriate – the sentence was used in the modified form in further interviews. Since pat- terns without specific elements such as particles are difficult to find in an un- parsed corpus, invented example sentences and acceptability judgments occa- sionally had to be used for such patterns as well. A comparison of semantically similar verbs 89 ceptability of a combination of a verb and a certain pattern was conse- quently judged according to the weighted average of the native speaker assessments (for example sentences see Appendix 3).13 Since the objective of this comparison is to find out whether a pattern is conventionalized or not, a clear tendency is sufficient to make that point. It is important to note that the groups of verbs are not compared in their entirety but only with respect to that lexical unit which is indeed compa- rable (i.e. the lexical unit with the “verb group meaning”). This means, although the verbs are clearly not synonymous in their entirety, they all share certain elements of meaning in which they at least overlap. By fo- cusing on these lexical units, semantic dissimilarities between the verbs have been neutralized to a high degree from the outset. The fact that the verbs within one group are not necessarily of equal status (some verbs might be more similar than others or there might be different subgroups within one group) is taken into account in the course of the analysis, as well as other semantic differences which might lead to a different pattern in- ventory. It is part of this examination to single out those aspects in which these verbs possibly differ semantically in order to examine whether these differ- ences can be related to their formal patterning in a systematic way as well

13. For the four answering options described above the weights were defined as follows: fully acceptable = 1, marginally acceptable = 0.75, highly unlikely = 0.25 and completely unacceptable = 0. If no clear answer in reference to this scale was given (e.g. if native speakers did not even indicate a tendency), such answers were not taken into consideration in order to avoid a falsifica- tion of the results. For the interpretation of the data, the following rules were defined for the weighted average of the answers provided by native speakers: 0.66 ” x ” 1 = acceptable. 0.33” x < 0.66 = undecided: no clear judgment available. This applies to sentences where native speakers disagreed about the acceptability of the pat- tern in such a way that no tendency could be observed. Where appropriate, these cases are commented on in the respective chapters. These sentences are labeled with a question mark (?). x <0.33 = unacceptable. Sentences are labeled with an asterisk (*). Appendix 3 at htttp://dx.doi.org/10.1515/9783110240788_appendix3 pro- vides a list of all example sentences testifying the occurrence of a verb in a pattern, including both authentic BNC and VDE examples and sentences as- sessed by native speakers (supplemented with their assessment). Appendix 3 also includes further native speaker assessed example sentences which are used in the argumentation in chapter 3 as a whole but which go beyond the 22 verb groups defined above. 90 Pattern choice and verb meaning as the other way around: are formal differences explicable on the basis of semantic differences? Thus the verb groups chosen for this analysis all exhibit semantic similarity and possible differences in meaning, which are to be identified. No claim is to be made about synonymy in a strict sense with respect to any of the semantic groups established (for a similar ap- proach see Hudson, Rosta, Holmes, and Gisborne 1996: 440). This database differs from other efforts to establish verb groups, as men- tioned above, in a very important way: first of the all, the basis is semantic not formal. Although this leads to the problems pointed out above, it must still be considered a necessary decision. Both the Pattern Grammar by Hunston and Francis (2000) and Levin’s (1993) verb groups are based on pattern overlap and common alternations respectively, and the verbs were only subsequently described semantically in order to show that the resulting groups can be given semantic labels. An a priori sorting on the basis of formal similarity in order to find out whether there are verbs comparable from a semantic point of view but different from a formal point of view is problematic: verbs which do not allow the pattern or alternation although they are semantically similar will never occur in any of the groups, i.e. possible gaps, which are of interest here, would never be identified, when starting the analysis from the point of view of the pattern. Moreover, the semantic groups in such studies often appear rather “forced” in the sense that a limited number of verbs are put into groups which often remain ra- ther heterogeneous with rather fuzzy boundaries (cf. 4.2.2).14 Another rea- son for this present approach is that the semantic labels used for the groups (i.e. the common denominator of all the verbs within one group) are the result of a semi-automatic procedure based on conditions disconnected with the question at hand. This means that the verbs in this database were not chosen in order to prove or disprove any theory connected to the interface between meaning and form. Instead the semantic labels were either origi- nally given in the process of the compilation of VDE as a lexicographic feature to facilitate the identification of the different lexical units or – as regards the dictionaries used – to provide lists of verbs with a similar com- municative function. Consequently the list serves as a more objective data- base in order to see to what extent it is possible to predict the pattern inven-

14. Gries and Stefanowitsch (2006: 9) show that an analysis on the basis of a simple cluster analysis – similar but more complex than the approach taken in Levin (1993) – yields more promising results, although with respect to verbs occurring in the ditransitive construction “a considerable number of verbs are clustered in ways which are not semantically motivated or coherent at all”. A comparison of semantically similar verbs 91 tory of a verb by comparing it to other verbs with which it shares certain semantic aspects. Also the notion of “gap” needs to be commented on as it is also poten- tially problematic. The term as used here only refers to such patterns which are attested with semantically similar verbs and not with patterns that do not occur with the verb in general. However, a corpus cannot provide nega- tive evidence, i.e. just because a pattern-verb combination was not found does not mean that it does not exist (cf. Chomsky 1957). However, if a pat- tern was not found in the corpus and was found unacceptable by native speakers, the combination at least has to be considered highly unconven- tional. Stefanowitsch (2008; see also 2006) convincingly argues that nega- tive entrenchment can also occur on the basis of non-occurrence in the data: if the non-occurrence of the verb in the construction is significant – on the basis of a collostructional analysis –, it cannot be a matter of chance and thus can be taken as negative evidence. Hunston and Francis (2000: 3) as- sume that having a pattern means that the verb regularly occurs in it. How- ever, since the purpose of this study is to evaluate the formal similarity between semantically similar verbs, it appeared more problematic to “miss” any patterns than to include accidental patterns. In order to avoid that the argument is based on gaps in finding the patterns rather than actual gaps in the language, cases where only one example was identified were included but have to be treated with caution. This is also reflected in a differentiated labeling system which also allows for undecided cases (see below). In order to provide an overview of the extent of structural similarity as opposed to difference between the verbs in the respective verb groups, all 22 verb groups will be briefly described, including information on the pat- terns they can occur in. The table given at the beginning of each paragraph indicates all patterns that were identified for the different lexical units in the respective groups. A number of symbols and subscripts were used to classify the acceptability of patterns, which are listed in Table 12 below. A general overview of the results of this comparison is given in chapter 3.2.3. Chapter 3.3 provides a detailed discussion of different example sentences which illustrate various phenomena identified in this comparison of seman- tically similar verbs. Taking such a large number of example sentences into consideration is regarded as a necessary step in this discussion since look- ing at actual language data provides important insights – especially with respect to idiosyncrasies and irregularities – which often go unnoticed in other approaches. 92 Pattern choice and verb meaning

Table 12. Symbols used for the assessment of pattern acceptability

symbol explanation 9 examples identified in VDE and/or the BNC 91 only one authentic example identified 9NSA construed sentences or internet examples rated as acceptable by native speakers ? no examples identified in VDE and/or the BNC; contradictory native speaker judgments e.g. pattern only compatible with a different lexical unit of the verb 9=responsibility under consideration (as described in the subscript, which is based on VDE labels). Note that there are many patterns which are available to more than one lexical unit of the same verb. This is only indicated, if the pattern is not available in the ‘group’- meaning. (cf. 3.3.3) –– / ? –– no examples identified in VDE and/or the BNC; internet exam- ples or construed sentences rated as not acceptable by native speakers; irrespective of the weighted average, the pattern is ad- ditionally marked by a question mark, if it was assessed as com- pletely acceptable by one native speaker but rejected by three or more others. This applies to a small number of cases where the positive assessment simply appears to be a mistake. Yet since there might also be cases which are doubtful, this convention is used. For the assessment of the results, such cases are counted as not acceptable following the rules defined above. ––P no examples in VDE/BNC, not accepted by native speakers; re- striction due to a difference in the participant inventory, i.e. no other patterns with respective participant(s) were identified (cf. 3.4.1) ––S no examples in VDE/BNC, not accepted by native speakers, re- striction due to selection restrictions (cf. 3.4.2)

3.2.2 Semantically similar verbs and their pattern inventories

Verb group 1 (‘answer’ with answer, reply, and respond): There is consi- derable overlap between answer, reply, and respond. The verbs have in common the participants AGENT (“the person who answers”), EFFECTED (the “answer” itself), INSTRUMENT (i.e. “the choice made to provide an an- swer”), and CAUSE. However there are also several restrictions, e.g. in that reply and respond cannot express AFFECTED.

(1)NSA *Let her reply/respond how she yearns for them. A comparison of semantically similar verbs 93

Table 13. Valency patterns of answer, reply, and respond in the sense ‘answer’

# SCU + ‘answer’ answer reply respond 1 +[0] 9 9 9 2 +[(that)_CL]_EFFECTED 9 9 9 3 +[QUOTE/SENTENCE]_EFFECTED 9 9 9 4 +[by_V-ing]_INSTRUMENT 9 9 9 5 +[with_NP]_INSTRUMENT 9 9 9 6 +[to_NP]_CAUSE/BENREC > 9 9 9 <+[by_V-ing]_INSTR./EFFECTED. 1 NSA 7 +[to_NP]_CAUSE +[with_NP]_INSTR. 91 9 9 8 +[to_NP]_CAUSE +[QUOTE]_EFFECTED 9 9 9NSA 9 +[to_NP]_BENREC/CAUSE 9 9 9 10 +[NP]_AFFECTED/BENREC 9 –– –– 11 +[wh-CL]_EFFECTED 9 –– ––

12 +[NP]_BENREC +[NP]_AFFECTED 9 ––P ––P 13 +[NP]_BENREC +[QUOTE]_EFFECTED 9 –– –– 14 +[NP]_BENREC +[that_CL]_EFFECTED 9 –– ––

15 +[NP]_AFFECTED +[to_NP]_BENREC 9 ––P ––P 16 +[NP]_AFF./BENREC +[with_NP]_INSTR. 9 –– ––

As explained in Table 12 above, restrictions such as the latter are marked by a subscript P. However, there are also restrictions, which cannot be ex- plained in reference to the meaning of these verbs. Reply and respond, for instance, other than answer, cannot express BENREC as a noun phrase di- rectly following the verb but only as a particle phrase, i.e. in terms of Levin (1993), reply and respond cannot participate in the so-called dative alterna- tion.

(2)NSA When no one *responded/*replied her, she came stamping into the room with a scowl on her face.

These restrictions and other restrictions will be discussed in detail in chap- ter 3.3. For this reason, the following sections will mainly introduce the verb groups, providing a more general overview of the semantic similarity of the respective verbs as regards the participants they express as well as the respective patterns they were found to be associated with, without dis- cussing separate examples.

Verb group 2 (‘assume’ with assume, suppose, presume, gather, and guess): When comparing the verbs in this group, it is notable that they are not semantically similar to the same degree: assume and suppose are more 94 Pattern choice and verb meaning similar than for example assume and gather. The first three verbs can be found in the same set of patterns. Nonetheless all five verbs have been compared to achieve a broad coverage of verbs expressing the meaning ‘assume’. Participants all five verbs have in common are AGENT, ÆFFEC- TED (“the assumption”), and SOURCE (“on what the assumption is based”).

Table 14. Valency patterns of assume, suppose, presume, gather, and guess in the sense ‘assume’

# [SCU]+ ‘assume’ assume suppose presume gather guess 1 +[Q/S]_ÆFFECTED 9 9999 2 +[so/not/…]_ÆFFECTED 99 999 3 +[(that_)CL]_ÆFFECTED 99 999 4 +[from_NP]_SOURCE > 99 999 <+[(that_)CL]_ÆFF. 5 +[from_X]_SOURCE > 9 9 99 9 <+[NP]_ÆFFECTED NSA 1 NSA 6 +[NP]_ÆFFECTED 9 9NSA 9rare 9collect 9 7 +[NP to_INF]_ÆFF.+PRED. 9 99––P 9 8 +[NP]_ÆFFECTED > 9 9 9 –– –– <+[AdjP]_PREDICATIVE rare rare P ? ? 9 +[wh-CL]_ÆFFECTED 9NSA –– –– –– 9

10 +[about_NP]_REFERENCE ––P ––P ––P ––P 9 11 +[as to_NP]_REFERENCE ––P ––P ––P ––P 9 ? 12 +[at_NP/wh-CL]_TOPIC ––P ––P ––P ––P 9

Again, several restrictions are due to differences in the participant invento- ries of these verbs. Guess, for instance, can express both a REFERENCE and a TOPIC (cf. patterns 10–12), which is not possible with any of the other verbs in this group, and gather cannot express PREDICATIVE, which is pos- sible with the other verbs in this group (pattern 8). Yet again there are re- strictions which cannot be explained this way, e.g. not all verbs allow the same range of realizations of the participant ÆFFECTED (cf. chapter 3.3.4.6).

Verb group 3 (‘choose’ with choose, select, nominate, designate, elect, appoint, and name): The verbs in this group all have the participants AGENT, ÆFFECTED (“the entity chosen”), PURPOSE, and SOURCE. From a semantic point of view, one could argue that choose and select represent more general types of choosing and that the others are more specific, e.g. confined to the context of politics and the like. Indeed choose and select share considerably more valency constructions compared to the other verbs A comparison of semantically similar verbs 95 in this group (cf. also Dixon 1991 for a semantic classification of these verbs). However, when comparing all patterns, it becomes quite clear that there are not two completely separate groups (choose and select vs. nomi- nate, designate, elect, appoint, and name) but that there are also several patterns that are acceptable to the verbs of the first group and to some verbs of the second group, such as [SCU]+ ‘choose’ +[from_NP] (which is ac- ceptable with all verbs except for designate and name) and that patterns cut across both groups without completely covering any of them, such as [SCU]+ ‘choose’ +[to_INF] (which is only possible with choose and elect). Thus the picture presenting itself is at any rate more complex (see Table 15 on the following page).

Verb group 4 (‘declare’ with declare, pronounce, announce, proclaim, and state): The verbs in group 4 have in common that someone (AGENT) can ‘declare’ something (EFFECTED), or ‘declare’ someone/something (AF- FECTED) as something (PREDICATIVE). Only very few of the restrictions identified in this group can be explained on the basis of semantic factors. The majority of restrictions are due to differences in the formal realization of the verbs’ participants, e.g. PREDICATIVE (cf. 3.3.4.4) (see Table 16 be- low).

Verb group 5 (‘get’ with obtain, receive, get, acquire, buy, and purchase): Similar to the ‘choose’-group this verb group can be seen as consisting of two subgroups, where the first group appears to have a more general mean- ing of ‘getting’, (get, obtain, receive, acquire) and the second one a more specific type, i.e. prototypically in exchange of money (buy and purchase). The latter can be found in the exact same constructions. Note that buy can occur in several other patterns which are specific for the notion of “ex- change goods for money”. These were not taken into consideration here. Participants all six verbs have in common are AGENT (“the buyer”), AF- FECTED (“the item bought”), BENREC (“the ultimate recipient”), and SOURCE (“the seller”) (see Table 17 below).

Verb group 6 (‘hire’ with hire, rent, and lease): These three verbs are a case in point for the claim that semantically similar verbs exhibit structural similarity with respect to the patterns they can be found in. All three verbs can express the same participants: AGENT, AFFECTED, PURPOSE, BENREC, SOURCE, and PRICE. Nonetheless, the verbs are not equally acceptable in all of these patterns (see Table 18 below). For example, despite the fact that pattern (9) is included in VDE for lease, native speakers disagreed with respect to the acceptability of this combination (cf. 3.3.4.3). 96 Pattern choice and verb meaning P P S P S P S call NSA 9 9 9 9 9 9 9 –– –– –– –– –– –– –– –– –– 9 9 name name P S ? ? NSA NSA NSA NSA –– –– –– 9 –– –– ? –– –– –– –– ? ? –– –– 9 9 9 9 designate designate ? S S P S P S ? ? ? ? NSA NSA NSA NSA NSA NSA NSA –– –– –– –– –– –– –– –– in the sense the in ‘choose’ 9 9 9 9 9 9 9 nominate nominate name name S S S S 1 rare 9 9 9 –– –– –– –– 9 –– –– –– –– 9 ,and appoint NSA –– –– –– 99 9 9 9 9 9 –– –– –– –– ? ? elect 9 , designate ? P NSA NSA NSA NSA 999 9 9 9 999 9 harass –– –– –– flowers –– pick 9 9 9 9 9 9 nominate , 1 ? NSA NSA NSA NSA –– –– 9 999 9 9 9 9 9 ? 9 9 9 9 select , appoint P , elect NSA NSA 99999 99 9 99999 9 9 9 9 9 99 99 99 9 9 9 –– 9 9 choose choose pick , . select , PRED ./ GOAL . SOURCE SOURCE SOURCE ./ ÆFF SOURCE / SOURCE choose PURPOSE ÆFF PRED ÆFFECTED . +[NP]_ +[to_NP]_ ÆFFECTED ÆFFECTED ÆFF / SOURCE ÆFFECTED ÆFFECTED . +[for_NP]_ . +[from_NP]_ . of_NP]_ +[out .+[among_NP]_ . +[as_NP]_ ÆFFECTED ÆFFECTED ÆFFECTED ÆFF ÆFF ÆFF ÆFF ÆFF ÆFFECTED BENREC Valency patternsValency of  +[NP]_ +[NP]_ +[NP]_ +[from_NP]_ +[NP]_ +[0] +[for_NP]_ +[on_NP]_ 7 +[NP]_ Table 15. 15. Table + ‘choose’ [SCU] # 1 +[NP]_ to_INF]_ 2 +[NP]_ (]+[) 4 +[NP 5 6 8 8 10 11 12 +[wh-to_INF]_ 14 16 +[among_NP]_ 17 +[to_INF]_ 18 13 +[wh-CL]_ 13 3 +[NP]_ 15 +[between_NP and NP]_ and 15 +[between_NP A comparison of semantically similar verbs 97 1 1 9 9 9 / / / NSA 9 9 9 9 9 –– –– –– –– –– –– 9 9 9 P P P 99 99 99 / / / ? 9 9 9 9 9 9 –– –– –– –– –– –– ? y P P 1 99 99 99 / / / 99 999 999 9 9 9 9 9 –– –– –– –– 9 –– –– candidac 9 1 99 99 99 eech / / / ? p NSA NSA 9 9 s –– –– –– –– –– –– 9 9 9 9 in the sense ‘declare’ state the sense in 99 99 99 / / / 99 99 9 99 9 9 9 9 9 9 9 9 9 99 cricket cricket –– –– , and 9 9 9 9 9 declare pronounce announce proclaim state . PRED PREFERENCE announce, proclaim announce, TOPIC , . +[as_NP]_ ÆFF ÆFFECTED ÆFFECTED PREDICATIVE pronounce ÆFFECTED ÆFFECTED AIM , BENREC EFFECTED +[about_NP]_ ./ +[NP]_ ./ PREFERENCE PREDICATIVE declare PRED / +[Q/S]_ ÆFFECTED . +[inof/against_NP]_ favor +[wh-CL]_ +[that_CL]_ +[Q/S]_ +[as_AdjP]_ / [that_CL]_ / +[NP]_ +[for_NP]_ +[to_NP]_ ÆFFECTED ÆFF ÆFFECTED PREFERENCE ÆFFECTED .+[AdjP]_ ÆFFECTED TOPIC BENREC BENREC BENREC ÆFFECTED ÆFF ÆFFECTED ÆFFECTED ÆFFECTED ÆFFECTED Valency patternsValency of  9 +[NP]_ 12 +[to_NP]_ 12 +[on_NP]_ 13 +[ what/sth/16 etc.]_ +[for_NP]_ 17 +[wh-to_INF]_ 22 14 +[NP]_ 14 19 +[0] 20 of/against_NP]_ favor +[in +[to_INF]_ 21 Table 16. 16. Table ‘declare’ [SCU]+ # +[NP]_ 1/2 +[wh-CL]_ 3/4 to_INF]_ 8 +[NP(]+[) +[to_NP]_ 15 +[REFLPRON]_ 18 10 +[to_NP]_ 10 +[NP]_ 11 5/6 +[NP]_ 5/6 7 +[NP]_ 98 Pattern choice and verb meaning NSA NSA NSA 9 9 9 99 99 –– –– quire buy buy quire purchase ? P 99 9999 9 99 9 9999 9 99 9 99 9 99 9 hire rent rent hire lease –– in the sense ‘get’ sense the in 1 purchase purchase 9 , and buy , 99 999 9 999 9 9 9 9 9 9 9 9 9 9 get obtain get obtain receive ac in the sense ‘hire’ the sense in , acquire lease lease AFFECTED receive , , and SOURCE BENREC SOURCE PURPOSE BENREC PRICE rent +[NP]_ , BENREC obtain , get hire AFFECTED TIME BENREC / +[for_NP]_ +[to_INF]_ +[from_NP]_ +[to_NP]_ +[at_NP]_ +[for_NP]_ +[from_NP]_ +[NP]_ PURPOSE BENREC AFFECTED AFFECTED AFFECTED AFFECTED AFFECTED AFFECTED AFFECTED AFFECTED AFFECTED BENREC  Valency patternsValency of patternsValency of 3 +[NP]_ 4 +[NP]_ #[SCU]+ ‘get’ ‘get’ 17. Table #[SCU]+ 1 +[NP]_ 2 +[NP]_ 18. Table 1 +[0] 2 +[NP]_ 5 +[NP]_ 9 +[NP/REFLPRON]_ 3 +[NP]_ 4 +[for_NP]_ 6 +[NP]_ 7 +[NP]_ 8 +[NP]_ +[to_NP]_ 10 # [SCU]+ ‘hire’ ‘hire’ # [SCU]+ A comparison of semantically similar verbs 99

This shows that although there is as such no semantic contradiction be- tween the construction meaning and the verb meaning, such a use is not necessarily automatically fully acceptable.15

Verb group 7 (‘intend’ with mean, intend, plan, and aim): The only par- ticipants all verbs in this group attestedly occur with are AGENT (“the per- son having or expressing an intention”) and ÆFFECTED (“the intention”).

Table 19. Valency patterns of mean, intend, plan, and aim in the sense ‘intend’

# [SCU]+ ‘intend’ mean intend plan aim 1 +[to_INF]_ÆFFECTED 9999 2 +[for_NP to_INF]_ÆFFECTED 9999NSA 3 +[NP (]+[) to_INF]_ÆFFECTED 9 9 9 9 4 +[that_CL]_ÆFFECTED 9e.g. imply 9 9 9rare 5 +[NP]_ÆFFECTED 9 9 9 –– 6 +[V-ing]_ÆFFECTED 9imply 9 9 –– 7 +[NP]_ÆFFECTED +[as_NP]_PRED. 999 ––P 8 +[NP]_ÆFF. +[for_NP]_BENREC 999 ––P 9 +[NP]_BENREC +[NP]_ÆFFECTED 9 9NSA 9NSA ––P 10 +[NP]_ÆFF. +[to_NP]_BENREC 9 9 ? ––P ? 11 +[for_NP]_AIM ––P –– 9 9 12 +[on_X]_AIM ––P 9 9 –– 13 +[NP]_ÆFFECTED +[at_NP]_AIM ––P ? –– 9

14 +[at_X]_AIM ––P –– –– 9

Mean, intend, and plan without aim show a much higher overlap of valency constructions: they also express PREDICATIVE (“a certain role intended for ÆFFECTED”), PURPOSE, and BENREC (“for whom something is intended”). However, also among these verbs further restrictions can be identified which are not due to a difference in their participant inventory, for instance with respect to the formal realization of AIM (patterns 11, 12, and 14), which is compatible with intend, plan, and aim (cf. chapter 3.3.4.7), as well as a different acceptability of certain combinations of participants, in this case ÆFFECTED and AIM (pattern 13) (cf. 3.3.5).

15. Note that VDE also lists a quadrivalent pattern for the verb hire ([NP]+ verb +[NP] +[NP] +[to_INF]) which could be said to realize the participant pattern AGENT+ hire + BENREC + AFFECTED + PURPOSE. Due to the sparse evidence for this pattern even for hire, it was not included in the present analysis. 100 Pattern choice and verb meaning

Verb group 8 (‘like’ with love, like, and adore): Participants that can be assigned to all three verbs in group 8 are AGENT, AFFECTED (“the person, thing, activity etc. loved”), and CAUSE (“the reason for which someone or something is being loved”). Also structurally love, like, and adore pattern in a very similar way, although there are still three exceptions, none of which can be accounted for in reference to possible semantic differences between these verbs but rather to a difference in optionality (cf. 3.3.4.1) and further idiosyncratic restrictions.

Table 20. Valency patterns of love, like, and adore in the sense ‘like’

# [SCU]+ ‘like’ love like adore 1 +[NP]_AFFECTED 99 9 2 +[V-ing]_AFFECTED 99 9 3 +[to_INF]_AFFECTED 999rare 4 +[NP V-ing]_AFFECTED 99 9 5 +[NP]_AFFECTED +[for_NP]_CAUSE 99 91 6 +([it)] +[when/if_CL]_AFFECTED 999NSA 7 +[it] +[for_NP to_INF]_AFFECTED 99 –– 8 +[NP to_INF]_AFFECTED 99 –– 9 +[0] 99 ?––

Verb group 9 (‘not lose’ with keep, preserve, maintain, and sustain): Par- ticipants the verbs in group 9 have in common are AGENT (“the person act- ing in a certain way as not to lose something”), AFFECTED (“the object not to be lost”), and LOCATION (“the place where this object is kept”).

Table 21. Valency patterns of keep, maintain, preserve, and sustain in the sense ‘not lose’

#[SCU]+ ‘not lose’ keep preserve sustain maintain 1+[NP]_AFFECTED 99 9 9 2+[NP]_AFF.+[ADV]_LOCATION 99 9 9 3+[NP]_AFF.+[as_NP]_PREDICATIVE 99 9 9 4+[NP]_AFF.+[AdjP]_PREDICATIVE 9 9 –– 9 5+[NP]_AFF.+[for_NP]_PURPOSE 99 9 ––P 6+[NP]_AFF.+[to_NP]_GOAL 9 ––P 9 ––P

Preserve appears to be somewhat more limited in the actual expression of [ADV] than keep, maintain, and sustain, which is why VDE lists pattern (2) in Table 21 as +[NP]+[in_NP], which is here subsumed under [ADV]. It has to be taken into account, however, that this might lead to further restric- A comparison of semantically similar verbs 101 tions. Three of the four restrictions can be explained on the basis of seman- tic differences: preserve and maintain do not express the participant GOAL and maintain does not express PURPOSE either. The other restriction is due to the fact that sustain, unlike the other three verbs in the group, is limited as regards the formal realization of PREDICATIVE. (cf. 3.3.4.4)

Verb group 10 (‘permit’ with allow, permit, entitle, and authorize)

Table 22. Valency patterns of allow, permit, authorize, and entitle in the sense ‘permit’

# [SCU]+ ‘permit’ allow permit authorize entitle 1 +[NP]_ÆFFECTED + >….. 99 9 9 <+[for_N/V-ing]_PURPOSE 2 +[NP]_ÆFF. +[to_NP]_BENREC 99 9 9 3 +[NP(]+[)to_INF]_ÆFFECTED 99 9 9 4 +[NP]_ÆFFECTED 99 9 –– 5 +[V-ing]_ÆFFECTED 99 9 ––P 6 +[that_CL]_ÆFFECTED 9admit 91 9rare ––P 7 +[for_NP to_INF]_ÆFFECTED 9 91 9NSA –– 8 +[NP]_BENREC +[NP]_ÆFFECTED 99 –– 9 title 9 +[0] 99 ––P ––P 10 +[of_NP]_ÆFFECTED 99 –– ––P 16 11 +[NP]_ÆFF. +[into_NP]_GOAL 99 ––P ––P 12 +[for_NP/V-ing]_ÆFFECTED 9 –– –– ––P 13 +[for_NP V_ing]_ÆFFECTED 9rare –– –– ––P ? 14 +[to_INF]_ÆFFECTED 9 –– –– ––P

Looking at the pattern inventories of these verbs, this group might serve as an example of the hypothesis that the more general a meaning, the more patterns can be used to express it and the more restricted the meaning, the more restrictions can also be found as regards the choice of patterns com- patible with it. Entitle allows only patterns the other three verbs can be found in as well, authorize again allows only patterns the more general allow and permit occur in, and finally permit does not have any patterns in addition to those allow can be used in as well. Participants all four verbs have in common are AGENT, ÆFFECTED (the “thing permitted”), BENREC

16. With allow this pattern is described in VDE (p. 30) as +[N]+[ADV] (often: into N). Since it is impossible to trace back all possible realizations of [ADV], this present analysis is confined to this frequent realization. 102 Pattern choice and verb meaning

(the “person addressed”), and PURPOSE. Nonetheless not all constructions expressing theses roles are compatible with all verbs (cf. 3.3.4).

Verb group 11 (‘practice’ with practice, train, and rehearse): The mem- bers of this group are closely linked to those in group 19 (instruct, teach, train, and educate), as both deal with a similar concept but from different perspectives. While ‘practice’ is concerned with the attempt to improve some kind of skill by working on it by oneself, ‘teach’ is concerned with helping to create or foster a skill in somebody else. Both of these meanings are entailed in the verb train, which is why it is included in both groups. This difference between train and the other verbs in this present group and the ‘teach’ group clearly has to be taken into consideration in the discussion of possible restrictions. Patterns of train which only occur in the sense train ‘teach’ are therefore not taken into consideration in a first step.

Table 23. Valency patterns of train, practice, and rehearse in the sense ‘practice’

# [SCU]+ ‘practice’ train practice rehearse 1 +[0] 99 9 2 +[to_INF]_ÆFFECTED 99 9 3 +[for_NP]_AIM 99 9 4 +[NP]_ÆFFECTED 99 9 5 +[with_NP]_CO-AGENT/INSTRUMENT 99 9 6 +[as_NP]_PREDICATIVE 99 91 7 +[wh-to_INF]_ÆFFECTED ? 9 9NSA 8 +[V-ing]_ÆFFECTED –– 9 91 ? ? 9 +[on_NP]_ÆFFECTED –– 9 –– ? ? 10 +[to_NP]_“STANDARD” 9 ––P ––P 11 +[NP]_ÆFFECTED +[for_NP]_AIM 9teach 9 9teach

The participants these verbs have in common are AGENT, ÆFFECTED (“the activity exercised”), AIM (“what is to be achieved”), PREDICATIVE (as in He trained as a doctor), CO-AGENT, and INSTRUMENT (“someone or something – such as a training device – AGENT trains with”). With one exception, the restrictions cannot be accounted for on a semantic basis. Moreover, this group serves as an example for a different preference of some verbs for some patterns: some verb pattern combinations were only attested by native speakers or found very rarely in the BNC. At the same time native speakers did not provide a unanimous judgement about some of the patterns that are described as restrictions here. A comparison of semantically similar verbs 103

Verb group 12 (‘propose’ with propose, suggest, and recommend): Pro- pose, suggest, and recommend exhibit a structural overlap for the majority of their patterns and participants. All three have AGENT, EFFECTED (“the proposal”), BENREC, PREDICATIVE (describing the role or properties of the proposal), and finally PURPOSE. This group shows great variety as regards the realization of EFFECTED, and it is in this area that the verbs pattern slightly differently. Again a semantic explanation cannot be given (cf. 3.3.4.6).

Table 24. Valency patterns of propose, suggest, and recommend in the sense ‘pro- pose’

# [SCU]+ ‘propose’ propose suggest recommend 1 +[NP]_EFFECTED 99 9 2 +[wh-CL]_EFFECTED 99 9 17 3 +[Q/S]_EFFECTED 99 9 4 +[(that_)CL]_EFFECTED 99 9 5 +[V-ing]_EFFECTED 99 9 6 +[NP]_EFF. +[as_NP]_PREDICATIVE 99 9 7 +[NP]_EFF. +[as_AdjP]_PRED. 9 9NSA 9 8 +[NP]_EFF. +[for_NP]_PURPOSE 99 9 9 +[to_NP]_BENREC +[that_CL]_EFF. 99 9 10 +[NP]_EFFECTED +[to_NP]_BENREC 99 9 11 +[so/otherwise]_EFFECTED 9NSA 9 9NSA 12 +[to_INF]_EFFECTED 9 –– 9 13 +[NP to_INF]_EFFECTED 9 9rare 9 ? 14 +[NP V-ing]_EFFECTED –– 9 9 15 +[QUOTE]_EFF. +[to_NP]_BENREC ? 9 ––

Verb group 13 (‘quarrel’ with quarrel, argue, dispute, and bicker): Al- though quarrel, argue, dispute, and bicker exhibit considerable similarity as regards their patterns, the overlap with respect to valency constructions appears comparatively small. The participants they all express are AGENT, TOPIC (“what the argument is concerned with”), and CO-AGENT (“the coun- terpart of the AGENT in the conflict”). Most restrictions can be accounted for in reference to a difference in verb meaning: dispute is the only verb that expresses ÆFFECTED. However, also in this group, there are restrictions

17. Herbst (2009: 65) points out that in patterns like this, words like suppose “take the function of a kind of commenting adverbial” and consequently “one may question the wisdom of analyzing it as the governing verb”. 104 Pattern choice and verb meaning which cannot be explained in such a way, e.g. as regards the variation in the expression of TOPIC (see Table 25 below) (cf. 3.3.4.5).

Verb group 14 (‘remember’ with remember, recall, recollect, reminisce, and bear in mind). The overlap of patterns and constructions among re- member, recall, recollect, reminisce, and the multi-word combination bear in mind is rather small. There are only two patterns, both containing the participant ÆFFECTED (“what is remembered”), in which all five verbs can occur. However, the overlap between remember, recollect, and recall is considerably larger (see Table 26 below). Only remember can be used in a different sense, i.e. remember ‘TASK’ as in (3), which is not included since it belongs to a different lexical unit of remember:

(3)VDE You should remember to replace your child’s toothbrush every three months or so. = TASK

There is considerably more structural similarity when reminisce and bear in mind are not taken into consideration as indicated above. Some of the re- strictions with respect to these two verbs can be accounted for by differ- ences in the participant inventories of remember, recall, and recollect, on the one hand, and reminisce and/or bear in mind on the other, e.g. remi- nisce does not occur with SOURCE, CAUSE, and PREDICATIVE but has a par- ticipant none of the other verbs have (CO-AGENT). However, not all of the restriction can be explained this way, either, e.g. the variability in the reali- zation of ÆFFECTED.

Verb group 15 (‘show’ with indicate, show, and demonstrate): All three verbs have the participants AGENT, ÆFFECTED (“the object presented to someone”), BENREC (“the person who is shown something”), PREDICATIVE (“a certain property of ÆFFECTED which is pointed out”), and possibly LO- CATION. All of the restrictions identified result from the fact that indicate, show, and demonstrate differ with respect to the kinds of formal com- plements they allow in the realization of these complements (cf. 3.3.4) (see Table 27 below). A comparison of semantically similar verbs 105 ? ? –– –– 9 9 9 9 9 9 9 9 ? P P P P NSA 9 9 9 9 9 –– –– –– –– –– –– 9 P P claim 9 9 9 9 9 9 9 –– –– 9 discuss/claim 9 P P P P NSA 9 9 9 9 9 9 –– –– –– –– 9 quarrel argue bicker dispute dispute bicker argue quarrel AGENT - in the sense the sense ‘quarrel’ in bicker CO + , and AGENT TOPIC , dispute TOPIC TOPIC ÆFFECTED , argue +[about_NP]_ +[over_NP]_ +[on_NP]_ +[NP]_ quarrel +[between_NP and NP]_ and +[between_NP AGENT AGENT AGENT AGENT AGENT - - - - - TOPIC TOPIC CO CO CO CO CO ÆFFECTED ÆFFECTED ÆFFECTED Valency patternsValency of  # [SCU]+ ‘quarrel’ [SCU]+ # Table 25. 25. Table 1 +[0] 2 +[about_NP]_ 6 +[with_NP]_ 7 +[with_NP]_ 8 +[NP]_ 3 +[over_NP]_ 4 +[with_NP]_ 5 +[with_NP]_ 9 +[that_CL]_ +[with_NP]_ 10 +[NP]_ 11 106 Pattern choice and verb meaning P P P P P P P P 9 / NSA NSA 9 9 –– –– –– –– –– –– –– –– –– –– –– –– –– ? 9 9 P P P P 99 / 9 9 9 9 –– –– –– –– –– –– –– –– –– –– –– in the sense ‘remember’ the in P P P 99 / ? NSA –– –– 99 99 9 9 9 9 9 –– –– –– –– ? –– –– ? 9 ? k P P 99 bear in mind mind in bear / –– –– –– –– –– call bac ? 9 , and P 99 / 99 99 99 99 99 99 9 99 99 99 9 99 99 –– –– –– 9 , reminisce remember remember recall recollect reminisce in mind bear recollect , TOPIC

ÆFFECTED recall , SOURCE CAUSE PREDICATIVE BENREC TOPIC ÆFFECTED +[about_NP]_ ? remember +[that_CL]_ +[for_NP]_ +[from_NP]_ +[to_NP]_ AGENT AGENT - - ÆFFECTED ÆFFECTED ÆFFECTED SOURCE ] +[about_NP]_ CO CO TOPIC ÆFFECTED . +[as_NP/V-ing]_ ÆFFECTED ÆFFECTED ÆFF ÆFFECTED ÆFFECTED ÆFFECTED Valency patternsValency of  what/sth/… # [SCU]+ ‘remember’ [SCU]+ # Table 26. 26. Table +[that_CL]/+[if/wh-CL]_ 1+2 3 +[0] 4 +[QUOTE]_ 6 +[NP]_ 9 +[about_NP]_ 10 +[V-ing]_ +[NP]_ 13 +[NP]_ 14 +[with_NP]_ 16 +[with_NP]_ 17 +[over_NP]_ 18 7 +[ 8 +[from_NP]_ 5 +[wh-to_INF]_ V-ing]_ +[NP11 +[NP]_ 12 +[NP]_ 15 A comparison of semantically similar verbs 107

Table 27. Valency patterns of indicate, show, and demonstrate in the sense ‘show’

# [SCU]+ ‘show’ indicate show demonstrate 1 +[NP]_ÆFFECTED 99 9 2 +[(that)_CL]_ÆFFECTED 99 9 3 +[wh-CL]_ÆFFECTED 99 9 4 +[NP]_ÆFF. +[as_NP]_PREDICATIVE 99 9 5 +[NP]_ÆFFECTED +[to_NP]_BENREC 99 9 6 +[NP]_ÆFF.+[with_NP]_INSTRUMENT 99 9 7 +[to_NP]_BENREC +[that_CL]_ÆFF. 99 9 8 +[NP to_INF]_ÆFFECTED 99 9 9 +[wh-to_INF]_ÆFFECTED 99 9 10 +[NP]_ÆFFECTED +[ADV]_LOCATION 9 9 ? 11 +[NP V-ing]_ÆFFECTED 99 ? 12 +[NP]_BENREC +[wh-CL]_ÆFFECTED 99 –– 13 +[to_NP]_BENREC +[wh-CL]_ÆFF. 9 –– 9 14 +[to_NP]_BENREC > 9 –– 9 < +[wh-to_INF]_ÆFFECTED NSA 15 +[QUOTE]_ÆFFECTED 9 –– –– 16 +[NP]_ÆFFECTED +[as_AdjP]_PRED. ? 9NSA –– 17 +[NP]_ÆFFECTED +[AdjP]_PRED. –– 9 –– ? ? 18 +[NP]_BENREC +[NP]_ÆFFECTED –– 9 –– 19 +[NP]_BENREC +[that_CL]_ÆFF. –– 9 –– 20 +[NP]_BENREC +[wh-to_INF]_ÆFF. –– 9 ––

Verb group 16 (‘start’ with start, begin, and commence): Leaving aside stylistic differences, start, begin, and commence are very similar both with respect to the kinds of participants they express – AGENT, ÆFFECTED (“what is started”), SOURCE (“the place the activity originates”), INSTRU- MENT, PREFERENCE (“when it comes to this, AGENT starts doing some- thing”), and PREDICATIVE (“a role AGENT assumes”) – as well as the kinds of valency patterns they can occur in. Although there is a clear similarity between the semantic function of INSTRUMENT and ÆFFECTED, the fact that [by V-ing] can co-occur with [NP]_ÆFFECTED (pattern 12) shows that they are not identical. The group exhibits restrictions as well as differences as regards their preference for certain valency patterns in order to express certain participant patterns (cf. 3.3.6). 108 Pattern choice and verb meaning

Table 28. Valency patterns of start, begin, and commence in the sense ‘start’

# [SCU]+ ‘start’ start begin commence

1 +[0] 99 9NSA 2 +[NP]_ÆFFECTED 99 9 3 +[to_INF]_ÆFFECTED 99 9 4 +[V-ing]_ÆFFECTED 99 9 5 +[by V-ing]_INSTRUMENT 99 9 6 +[from_NP]_SOURCE 99 9 7 +[QUOTE]_EFFECTED 99 9NSA 8 +[with_NP]_REFERENCE 99 9 9 +[NP]_ÆFF.+[with_NP]_REFERENCE 99 9 10 +[as_NP]_PREDICATIVE 99 9 11 +[NP]_ÆFF.+[as_NP]_PREDICATIVE 99 9 12 +[NP]_ÆFF.+[by V-ing]_INSTRUMENT 99 9NSA 13 +[on_NP]_TOPIC 99 ––P 14 +[NP]_ÆFFECTED+[from_NP]_SOURCE 9 –– –– 15 +[for_NP]_GOAL 9 ––P ––P

Verb group 17 (‘teach’ with instruct, teach, train, and educate): There are only three participants that are expressed by instruct, teach, train, as well as educate. These are AGENT, BENREC (“the person who gets the lesson”), and REFERENCE (“the subject of this lesson”). However, only about a third of the restrictions can be accounted for this way. Among the restrictions iden- tified, some are indeed due to semantic difference – e.g. train does not ex- press TOPIC (“what the lesson is about”) and instruct and teach have no PURPOSE – or differences in the optionality of certain participants (train always needs BENREC) (cf. 3.3.4.1). Other restrictions are idiosyncratic and cannot be explained in reference to the participant inventories of these verbs (see Table 30 below).

Verb group 18 (‘telephone’ with phone, ring and call): Phone, ring,and call can all be used to express an AGENT, AFFECTED (“the person sum- moned”), TOPIC (“what the summoning is about”), PURPOSE, and LOCA- TION, while the latter in this case typically refers to a telephone number. This verb group is a case in point for semantically similar verbs also being structurally similar: phone, ring, and call have the same structural potential. Not only do they have the same formal patterns, they also have the same valency constructions, i.e. the meaning of the patterns in the sense of par- ticipant roles realized by the formal complements is identical. Yet even with such indubitable similarity in meaning and obvious similarity in the A comparison of semantically similar verbs 109 structural occurrences, there are minor differences, which, however, do not figure on the abstract formal surface of valency patterns (cf. 3.3.6).

Table 29. Valency patterns of phone, ring, and call in the sense ‘telephone’

# [SCU]+ ‘telephone’ phone ring call 1 +[0] 999 2 +[NP]_AFFECTED 999 3 +[about_NP]_TOPIC 999 4 +[for_NP]_PURPOSE 999 5 +[NP]_AFFECTED +[at_NP]_LOCATION 999 6 +[NP]_AFFECTED +[on_NP]_LOCATION 999 7 +[NP]_AFFECTED +[about_NP]_TOPIC 999 8 +[NP]_AFFECTED +[for_NP]_TOPIC 999

Verb group 19 (‘think’ with consider, think, reflect, ponder, contemplate, and judge): The verbs in this group are not of equal status semantically. Obviously, there are several types of thinking, e.g. ‘having an opinion’ or ‘thinking a thought’ or ‘considering something’ (compare also VDE). On this basis one could divide the group into subgroups such as (1) reflect, ponder, contemplate, and think in the sense ‘thinking thoughts’, (2) think and consider in the sense ‘consider something’ or (3) think, consider, and judge as ‘having an opinion on something’. That think is part of all three groups and, more importantly, that the dividing line between these different lexical units is by no means clear-cut, provide arguments for grouping them together. What becomes clear throughout the analysis is that there is con- stant overlap between these categories not just for think but also for many of the other verbs. Only one valency construction can be found with all six verbs: AGENT + ‘think’ +[that_CL]_EFFECTED, i.e. think that something is the case. However, many of the verbs have the same participants but differ with respect to their formal realization or with respect to how they can be combined (cf. 3.3.4) (see Table 31 and Table 32 on the following pages).

Verb group 20 (‘think highly’ with respect, admire, and esteem): Respect, admire, and esteem all express AGENT, ÆFFECTED (“the person, object or quality admired”), PREDICATIVE (“a quality for which the person or object is admired”), and REFERENCE (“when it comes to this, AGENT is full of admiration”). None of the formal restrictions identified can be related to differences in the participant inventories of these verbs (see Table 33 be- low). 110 Pattern choice and verb meaning P P 1 9 9 / / –– –– 9 9 9 –– –– –– –– –– –– –– –– –– –– 9 ? –– –– t P P P P P P P direc 99 99 / / ? ractice –– –– 9 99 9 –– –– –– –– –– 9 ? p –– –– –– –– –– –– ? 9 ?/ ?/ P 99 99 / / NSA 99 9 9 9 9 9 9 9 9 9 9 –– –– –– –– 9 9 y lo P order p 99 / ? NSA order –– –– –– 99 9 99 9 99 –– –– –– –– –– –– 9 em ? ? –– / 9 9 9 9 instruct teach teach instruct train educate 9 TOPIC REFERENCE in the sense ‘teach’ sense the in educate SOURCE ,and +[in_NP]_ ” +[about_X]_ TOPIC PREDICATIVE , train BENREC +[from_NP]_ PREDICATIVE AMOUNT SOURCE ÆFFECTED , teach PURPOSE BENREC ]_“ TOPIC ÆFFECTED / +[NP]_ +[as_NP]_ ÆFFECTED ÆFFECTED instruct ” ? ” ? +[to_INF]_ more/sth… +[from_NP]_ +[for_NP]_ +[to_NP]_ ÆFFECTED ÆFFECTED +[that_CL]_ +[wh-to_INF]_ / / +[on_NP/wh-to_INF]_ +[about_X]_ +[NP]_ +[NP]_ ÆFFECTED +[ BENREC BENREC TOPIC ÆFFECTED STANDARD ÆFFECTED ÆFFECTED PREDICATIVE ÆFFECTED ÆFFECTED ÆFFECTED BENREC BENREC BENREC BENREC BENREC BENREC BENREC BENREC BENREC  Valency patternsValency of +[NP]_ Table 30. 30. Table 13 +[NP]_ 13 7 +[NP]_ 8 +[NP]_ +[NP]_ 10 11 +[as_NP]_ 11 +[NP]_ 12 +[NP]_ 15 +[that_CL]_ 16 +[to_NP]_“ 17 +[wh-CL]_ 19 +[wh-to_INF]_ 20 +[NP]_ 21 1/2 +[NP]_ 1/2 3 +[NP]_ 4/5 +[NP]_ +[0] / 9 +[NP]_ 14 +[about_X]_ 18 # [SCU]+ ‘teach’ [SCU]+ # 6 +[NP]_ 22 +[to_INF]_ 22 A comparison of semantically similar verbs 111 1 . assess . assess NSA –– –– 9 9 9 9 9 9 9 9 assess assess assess –– –– –– –– –– g g 9 ? e. e. 9 9 9 9 9 9 1 1 ? ? 9 9 9 9 9 9 9 –– –– –– –– –– –– –– –– –– –– 9 9 P P P P P P 9 9 9 9 9 9 9 9 9 9 9 –– –– –– –– –– –– –– –– –– –– in the sense ‘think’ (part I) ‘think’ sense the in judge judge P P P P P 9 9 9 9 9 9 9 mirror mirror mirror –– –– –– –– –– –– –– –– –– –– –– 9 9 9 , and ine g hr.v. p 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 –– –– –– ima 9 9 contemplate , 1 1 1 1 9 9 9 9 9 9 9 9 9 9 , ponder –– –– –– –– 9 9 9 9 consideration consideration consider think reflect ponder contemplate judge judge contemplate ponder reflect think consider 9 9 . think , reflect , PRED ? . PREDICATIVE consider PREDICATIVE PRED PREDICATIVE ./ PREDICATIVE ÆFFECTED ÆFF ]_ +[NP]_ ÆFFECTED . +[as_NP/AdjP]_ TOPIC TOPIC ÆFFECTED TOPIC ÆFFECTED .+[AdjP]_ . +[as_NP]_ . +[as_AdjP]_ ÆFFECTED ÆFF ÆFFECTED TOPIC ÆFFECTED ÆFFECTED ÆFFECTED ÆFFECTED ÆFF ÆFF ÆFF Valency patternsValency of # [SCU]+ ‘think’ ‘think’ # [SCU]+ Table 31. 31. Table 1 +[0] 2 +[that_CL]_ 4 +[wh-CL]_ 5 +[wh-to_INF]_ 9 +[ so/not/otherwise +[of_X]_ 10 +[on_X]_ 11 +[about_NP]_ 12 7 +[Q/S]_ 3 +[NP]_ 6 +[V-ing]_ 8 +[to_INF]_ +[NP]_ 16 +[NP]_ 17 +[NP]_ 18 +[of_NP]_ 19 13 +[upon_X]_ 13 +[over_NP]_ 14 +[NP]_ 15 +[NP(]+[)to_INF]_ 20 112 Pattern choice and verb meaning P ? NSA –– –– assess ? –– 9 9 9 9 –– –– ? –– –– –– –– –– –– ? 9 9 –– ? –– –– –– –– –– in the sense ‘think’ (part II) (part II) ‘think’ sense the in NSA NSA 999 999 999 –– 9 9 respect admire esteem respect admire ? 9 –– –– –– –– judge judge , and ? 9 9 –– –– –– , contemplate in the sense ‘think highly’ highly’ ‘think sense the in P 9 9 9 –– –– –– , ponder consider consider think judge contemplate ponder reflect esteem esteem , and . CAUSE ÆFF think , reflect , TOPIC admire , TOPIC TOPIC PREDICATIVE PREDICATIVE PREDICATIVE consider respect TOPIC

. +[of_NP]_ > +[for_NP/V-ing]_ ÆFF +[as_NP]_ +[in_NP]_ +[NP]_ . +[about_NP]_ ÆFF . +[on_NP]_ . +[on/upon_NP]_ /etc.]_ ÆFFECTED BENREC BENREC ÆFF ÆFF etc]_ ÆFFECTED ÆFFECTED ÆFFECTED ÆFFECTED ÆFFECTED what Valency patternsValency of patternsValency of  < +[Q/S/NP/that_CL/wh-CL]_ # highly’ ‘think [SCU]+ 5 +[wh-CL]_ 23 +[NP]_ 23 Table 32. 32. Table ‘think’ # [SCU]+ +[ what 21 +[NP/ 22 +[Q/S]_ 24 33. Table 1 +[NP]_ 2 +[NP]_ 6 +[NP]_ 3 +[NP]_ 4 +[NP]_ 25 +[to_NP]_ 25 A comparison of semantically similar verbs 113

Verb group 21 (‘tolerate’ with bear, tolerate, and endure): Bear, tolerate, and endure express AGENT and AFFECTED (“the state of affairs that is tole- rated”). This can be expressed in three different ways with all three verbs. The complement type +[with_NP] could be described as REFERENCE (“when it comes to this, AGENT must be tolerant or enduring”) but it could also be seen as a subtype of AFFECTED as it cannot co-occur with it and semantically basically fulfills a very similar role. Nonetheless tolerate and endure, which do have AFFECTED, are not found in this pattern (see Table 34 below).

Table 34. Valency patterns of bear, tolerate, and endure in the sense ‘tolerate’

# [SCU]+ ‘tolerate’ bear tolerate endure 1 +[NP]_AFFECTED 99 9 2 +[NP V-ing]_AFFECTED 99 9 3 +[V-ing]_AFFECTED 99 91 ? 4 +[to_INF]_AFFECTED 9 –– 9 5 +[with_NP]_AFFECTED 9 –– ––

Verb group 22 (‘urge’ with urge, press, and push): Urge, press, and push all express AGENT, AFFECTED (“the person on who pressure is exerted”), AIM (“what is to be achieved by this”), and INSTRUMENT (“what is used to exert pressure”). None of the restrictions can be accounted for on a seman- tic basis.

Table 35. Valency patterns of urge, press, and push in the sense ‘urge’

# [SCU]+ ‘urge’ urge press push 1 +[NP]_AFFECTED 999 2 +[to_INF]_AIM 999NSA 3 +[for_NP]_AIM 999NSA 4 +[NP]_AFFECTED +[to_INF]_AIM 999 5 +[NP]_AFFECTED +[for_NP]_AIM 999 6 +[NP]_AFFECTED +[into_NP/V-ing]_AIM 999 7 +[NP]_AFFECTED +[upon_NP]_AIM 999NSA 8 +[NP]_AFFECTED +[with_NP]_INSTRUMENT 999 9 +[NP]_AFFECTED +[about_NP]_AIM 91 9 –– ? 10 +[NP]_AIM/AFF. +[on_NP]_AFF./REFERENCE 99–– ? 11 +[that_CL]_AIM 99–– 12 +[QUOTE]_AIM 99–– ? 13 +[upon_NP]_AFFECTED +[that_CL]_AIM 9 –– –– 14 +[NP]_AFFECTED +[QUOTE]_AIM 9 –– –– 114 Pattern choice and verb meaning

3.2.3 Results of the analysis – quantitative evaluation

The analysis of the 22 verb groups reveals that a significant number of the patterns actually do occur with all of the verbs or at least with several members of a group. Semantically similar verbs can be found in almost 70% of all plausible pattern-verb combinations, i.e. combinations which are attested with semantically similar verbs (cf. Appendix 1). There are even groups and subgroups in which all verbs allow the exact same range of patterns, while this is clearly the minority in this sample and very likely also in general.

(x) number of analyzed pattern-verb combinations x proportion in percent different meaning (34) undecided (39) 3% unsystematic restrictions (193) 3% 16%

semantic restrictions (123) 10%

68% overlap (844)

Total: 100% (1,233)

Figure 14. Formal overlap, semantic, and unsystematic restrictions

At the same time a considerable number of restrictions have to be ac- counted for, i.e. those valency patterns which are acceptable to one or some but not all members of the groups. Such restrictions fall into different cate- gories: (1) restrictions which can be accounted for with respect to the par- ticipant inventories of the verbs, (2) restrictions which can be accounted for by other semantic aspects, such as a difference in the lexical aspect of the verbs or their selection restrictions, (3) restrictions which do not apply to the verb with the restriction in general but only to the lexical unit under consideration, i.e. the pattern is acceptable when the verb is used in a dif- ferent sense (‘different meaning’ in Figure 14), and finally, (4) restrictions which cannot be accounted for on a semantic basis and are therefore unsys- tematic restrictions from the point of view taken here. The remaining 3% A comparison of semantically similar verbs 115 are cases where the respective pattern-verb-combination is not attested in VDE or the BNC and where native speakers gave contradictory assess- ments (‘undecided’). 1) Differences in the participant inventory: the comparison of semanti- cally similar verbs in the same group revealed that patterns realizing certain participant types in combination with some verbs in the group consistenly do not occur with other verb(s) of the same group. Such restrictions can be found in almost every verb group. In the ‘answer’-group, answer has a participant ÆFFECTED, subsuming the subtypes AFFECTED, i.e. a question which is answered (typically rea- lized by [NP] or [wh-CL]) and EFFECTED, i.e. the answer itself (realized by [that_CL] and [QUOTE/ SENTENCE]). AFFECTED and EFFECTED are tak- en to be subtypes of ÆFFECTED because they cannot co-occur in one pattern (cf. 2.1.3). Dixon (1991: 151), for instance, refers to this participant as MESSAGE, i.e. “the question which was asked, and the answer that might be given”. Respond and reply can only be complemented with EFFECTED as illustrated in (2) and (3) but not with AFFECTED. Moreover, respond does not necessarily describe the verbalization of an answer but can also express nonverbal behavior. Answer, like respond, can express both nuances, while the verbal response might be regarded as more prominent.

(1)VDE I answered that I hadn’t actually tried to do this EFFECTED. (2)VDE Mitterrand replied that the question of sanctions had to be decided by the European Community summit in June EFFECTED. (3)VDE I will respond that we get our money from central government EF- FECTED. (4a)VDE A small book dealing with the history of insurance plaques will, I hope, answer all your questions AFFECTED. (4b)NSA *A small book dealing with the history of insurance plaques will, I hope, respond/reply all your questions. (5a)VDE Let her answer how she yearns for them AFFECTED. (5b)NSA *Let her reply/respond how she yearns for them.

In the ‘think’-group, ponder and reflect do not have a participant PREDICA- TIVE, i.e. they differ from other verbs in this group in that they cannot ex- press the mental evaluation of someone or something. This accounts for the restrictions of all patterns realizing this participant in combination with the other verbs in this group, e.g. *to ponder or reflect someone (as) clever. Similarly, the participant type PREDICATIVE is also restricted in the ‘not lose’-group: keep, maintain, and preserve can occur in the pattern [SCU]+ 116 Pattern choice and verb meaning verb +[NP]+[AdjP], while sustain is the only verb in this group which can- not occur with this participant.18

(6)VDE It is generally not advisable to take B complex in the evening as it can keep some people awake. (7)VDE The Greek government says this collection of marble heads and torsos is of the utmost importance to the Greek cultural heritage and should be preserved/*sustained intact rather than sold off as individual items. (8)BNC The relatively large bonus was preferred by the majority of longer-term savers and the idea of a substantial reward for maintaining savings un- touched is clearly popular. GXA 85

Moreover, one can keep/preserve/sustain something for a purpose (AF- FECTED + PURPOSE) but not maintain it, and neither maintain nor preserve seem to have a participant GOAL.

? (9)VDE Please read these notes carefully and keep/* maintain them AFFECTED for future reference PURPOSE. (10)VDE I made up my mind to do everything in my power to have the area AF- FECTED preserved for posterity PURPOSE. (11)BNC The purpose of community intervention is not simply to observe but to ensure that treatment AFFECTED is sustained for those whose illness, in terms of either severity or nature, warrants this PURPOSE. FSY 1576 (12)BNC No matter how interesting the music and the dance movements, if the passages are too long the dancer rarely has sufficient stamina to sustain the dance AFFECTED to its proper climax GOAL. A12 28 (13)VDE When you get news, do you keep/*maintain/*preserve it AFFECTED to yourself GOAL. (14)BNC …-- they believed he was an ancestor in a white skin who was obsessed with privacy simply to maintain his secret. HH3 2587

Sustain can have two different meanings in the second pattern: the meaning illustrated by example (12) but also ‘suffer an injury to a certain body part’. In example (13) the meaning of keep is idiomatic in the sense of ‘keeping a secret’ where maintain is very rare, yet does not seem to be completely impossible as can be seen in (14). The problem with the examples provided by the ‘not lose’-group is that there is only one pattern each that expresses

18. VDE labels this sense of keep as ‘remain’. This is similar to ‘not lose’ in that it focuses on an AFFECTED, while there is a clear notion of preventing an ac- tion from happening, which is absent from the ‘not lose’ notion. Yet as three of the four verbs can occur in this pattern in the same sense, it has been in- cluded here. A comparison of semantically similar verbs 117 the respective participant. Thus the whole argument whether the verbs have the participants GOAL and PURPOSE or not rests on the acceptability of the verb in one single pattern. Since it cannot be excluded that there are other possible causes for this restriction, the argument is much weaker than when a whole range of different patterns with one and the same participant is not acceptable as in the case of ponder. Hence it is not without problems to say that a pattern cannot occur because there is no semantic equivalent in the form of a participant pattern that could be expressed by the valency pattern. For once, there is the problem of negative evidence, i.e. the fact that the verb cannot be found in a pattern which is used for the realization of this participant in combination with semantically similar verbs does not neces- sarily prove that it does not exist. This is counterbalanced by native speaker interviews. If native speakers do not accept such a sentence either, this can be taken as a very clear indication that the participant under consideration is not conventionally expressed in combination with this verb. Moreover, one might also argue along the lines of Stefanowitsch (2008), that if no examples can be found despite statistical probabilities to the contrary, the pattern could indeed be negatively entrenched and thus blocked. Nonethe- less, the argumentation is to a certain degree circular: because no such com- bination appears to be conventionalized, the non-existence of the partici- pant, which is taken as proof for this hypothesis in the first place, is also considered as the semantic explanation for the non-occurrence. Obviously this necessitates a rather artificial separation of verb meaning and partici- pant structure. Thus it must be very clear that this is done here in order to avoid missing any possible semantic explanations for syntactic gaps. In- deed, if a participant is consistently not expressed, this is a certain semantic regularity and must therefore be taken into consideration in this compari- son. At the same time, one has to be aware that which participants are asso- ciated with a lexical unit and which are not is, to a certain degree, already arbitrary in itself: from a purely semantic point of view, semantic partici- pants might fit into the general scene described by a verb but the verb never occurs in a corresponding pattern. It is by no means clear how a speaker knows that uses such as those exemplified in sentences (15)–(17) are not possible with mean.

(15)VDE That’s precisely what Labour should aim/*mean at: the replacement of a market mechanism with planning mechanisms. (16)VDE Aim/*mean for clarity and precision; remove words that obscure your directions. (17)BNC And I think she intends/*means on having quite a few drinks. KD3 528 118 Pattern choice and verb meaning

This can be even more problematic, when alternative explanations apply. One could say that the verb mean differs from the other members of its group (aim, intend, and plan) in that it does not have a participant AIM and therefore cannot occur in sentences such as (15) to (17). However, an alter- native explanation would be that the participant ÆFFECTED has a different status in combination with mean as opposed to the other verbs in the group, since it cannot occur in a pattern without it (at least not in the sense ‘in- tend’). This also applies to the monovalent pattern, which is possible with plan and which does not express ÆFFECTED (and obviously not AIM, either) and which is not compatible with mean.

(18)VDE But nothing would happen before they reached the lonely plains of West- ern Tibet, so at least he had time to plan/*mean.

Another interesting perspective is provided by cases which are undecided, as in case of the ‘declare’-group. While declare, announce, proclaim, and state all have a participant BENREC (even if they differ as regards the range of different formal realizations they allow), the situation with pronounce is more difficult. There is only one pattern expressing BENREC in which pro- nounce appears to be acceptable, illustrated by example (19a). Other sen- tences with complements expressing this participant were either rejected or native speakers gave diverging judgments as to their acceptability.

(19a)NSA His followers then pronounced to the world BENREC that he had tri- umphed over death ÆFFECTED. (20)VDE In her Christmas ’91 broadcast, she declared to her people: “With your prayers and your help, and with the love and support of my family, I shall try to serve you in the years to come.” (21)NSA *Grenfell told the interviewing officers that Vernage once pronounced to him: “Whatever police officer gets in my way, they are getting it.” (22)VDE The worst part about being a bargain addict is that odd need to an- nounce to the world how cheap you are. (19b)NSA ?His followers then pronounced to the world why he had triumphed over death.

(19a) and (19b) were judged acceptable by American native speakers, but rejected by British native speakers. Strictly speaking, one would then have to argue that pronounce has a participant BENREC in American English but not in British English. This clearly shows that what must be considered as part of a word’s participant inventory might not be completely fixed. Ra- ther than the participant being the prerequisite for a verb to occur in a cer- tain pattern, it is not unlikely that processes such as analogy to other similar A comparison of semantically similar verbs 119 verbs play a role here. If it is not entirely implausible that a certain partici- pant is associated with an event or situation described by the verb, it cannot be excluded that such a verb is eventually used in a pattern that realizes this participant. This could be seen as an example where Goldberg’s idea of an independent construction meaning that allows for verbs to occur in a con- struction as long as verb and construction do not contradict each other se- mantically (1995, 2006a) appears very fitting. At the same time, one must not forget that the acceptability judgments above only rest on a limited number of native speakers and the fact that no examples were found for this pattern in the BNC clearly shows that this use – at least in British English does not seem to be acceptable, which clearly raises the question, why – if semantically plausible – such a use was not accepted by British native speakers in the first place. Also in the ‘practice’-group a case of analogy can be identified. Re- hearse, which cannot express BENREC in a divalent pattern, was found with this participant in trivalent patterns. In these patterns, the meaning of re- hearse is ‘teach’ rather than ‘practice’.

(23)VDE He told me to train her BENREC for the Oaks PURPOSE. (24)BNC I was a player in a drama written and cast by Jean-Claude, who had not rehearsed me BENREC for it PURPOSE. FAT 2085 (25)VDE It aims to run a special course next month to train people BENREC in the skills they need to launch a business and make it succeed REFERENCE. (26)BNC John was so sure of an encore that he rehearsed the girls BENREC in a burlesque version of a quartet from the Gaiety Theatre in London REF- ERENCE, and in a later scene they performed … B34 106

When rehearse can occur in the patterns [SCU]+ verb +[NP]+[for_NP] and [SCU]+ verb +[NP]+[in_NP] on the basis of its similarity to train, it is not clear as to why it does not have more structural parallels with train. Com- pare, for instance, sentence (27), where the only interpretation of she could be AFFECTED:

(27)NSA *She was rehearsed by her father. (28)BNC This facility is useful for rehearsing a piece of music, or for setting the EQ or effects on any of the tracks for mixdown without having to labo- riously switch by hand. C9L 2383

Examples identified for [SCU]+ rehearse +[NP]+[to_NP] differ from those found for train in this pattern: 120 Pattern choice and verb meaning

(29)BNC I read the other day a well praised first novel in which the narrator -- who is both sexually inexperienced and an amateur of French literature -- comically rehearses to himself BENREC the best way to kiss a girl without being rebuffed ÆFFECTED. G1A 554 (30)BNC After all, we BENREC 're trained to a superbly technical level of scientific competence STANDARD. A17 398

This clearly shows that the semantic similarity does not lead to a full ac- ceptance of BENREC in the sense that it can be realized in the same way. This raises the question where BENREC in combination with rehearse comes from and – if it is a regular member of its participant inventory (which was denied by native speakers, i.e. the pattern was considered rather unusual) –, how come it is not available in other patterns as well? These cases clearly show that even if these restrictions seem to be con- nected to the meaning of these verbs, we can by no means speak of a regu- lar link between meaning and form in the sense that it is rule-based in any way. Quite the contrary, what is acceptable and what seems to be restricted is very idiosyncratic in such cases and the acceptability of the patterns might be heavily influenced by factors connected to the example sentences such as word choice and style, as pointed out above, or simply be speaker- dependent. Predicting such cases appears to be very problematic. Nonethe- less it appears to be more systematic to point out such difference which might be directly related to the participants the verbs express than to sub- sume them under other restrictions. In this context it is also important to note that different formal valency patterns do not automatically entail a semantic difference as is assumed in theories such as construction grammar, where a difference in form is gener- ally taken to imply a difference in meaning (e.g. Goldberg 1995: 67). In- stead, the objective was to exclude that the same general concepts can be expressed by means of other complement types. In reference to the ‘think’- example presented above, this means that in this sense only ponder and reflect, which never occur with PREDICATIVE, are considered to differ se- mantically from, for instance, consider, while think, which can express PREDICATIVE with [AdjP] or [NP] but not with phrases headed by the par- ticle as, does not. Only if no other pattern can express a semantically com- parable concept, were the verbs considered to differ in their participant inventory. These cases were indicated with a subscript P (for participant) in the tables above, which applies to 9% of all analyzed plausible verb-pattern combinations and can account for 32% of all restrictions in this sample (see Figure 15 on the following page). A comparison of semantically similar verbs 121

(x) number of analyzed pattern-verb combinations (restrictions) x proportion in percent

participant (111) 32%

55% unsystematic restrictions (193)

3% selection restrictions (12) 10% different meaning (34)

Total: 100% (350)

Figure 15. Proportion of different types of restrictions

(2) Other semantic restrictions: the impact of semantic factors which are typically considered to feature prominently in the linking of meaning to form such as selection restrictions and aspectual factors do not seem to play an important role with respect to this sample. Only about 3% of all restric- tions can be accounted for by selection restrictions. In order to do justice to approaches which attribute such an important role to this type of features, a closer analysis of both lexical aspect as well as selection restrictions, which goes beyond the syntactic gaps identified here, will be provided in chapters 3.4.2 and 3.4.3. (3) Same pattern – different verb meaning: 3% of all verb-pattern- combinations (i.e. 10% of all restrictions) do not trigger the respective group-meaning but are associated with a different lexical unit of the same verb, e.g. entitle when complemented with two noun phrases can only mean ‘give a title to someone’ but not ‘permit’ as in allow +[NP]+[NP]. Ob- viously, this does not lead to any pattern restrictions. Nonetheless, this phe- nomenon is of interest when considering the idea of a one-to-one relation- ship between meaning and form (cf. 3.3.3). (4) Unsystematic restrictions: more than half of all restrictions (55%) are neither related to differences in the participant inventory nor could they be related to other differences in meaning between the verbs. Many of these restrictions are cases where the verbs have the same participants but do not allow the same range of complement types for their formal realization (e.g. reply can be complemented with [to_NP]_BENREC but not with [NP]_BEN- 122 Pattern choice and verb meaning

REC). Another phenomenon are combinatorial restrictions, where different but semantically similar verbs have the same participants but cannot com- bine these participants in the same way (e.g. it is not possible to combine recommend with BENREC and EFFECTED at the same time although it can combine with both in divalent patterns, i.e. separately, and although such a combination in a trivalent pattern is possible with suggest in the same sense) (cf. 3.3.5). As an interim result, one can safely say that the amount of formal dis- crepancies does not support the hypothesis that the patterns a verb allows can be systematically linked to the meaning of the verb via semantic rules. Many of the verbs in the groups can, in part, occur in the same patterns and it is clearly also the meaning of these verbs that to a certain degree delimits the kinds of complements that can occur with it. Yet even beyond semantic similarity, it is quite obvious that many verbs have a certain amount of patterns in common, i.e. there is of course a considerable overlap with re- spect to a subset of all possible patterns simply because the overall number of patterns is limited. At the same time, this limitation is not as severe as one might assume. The Patternbank lists more than 2,000 active verbal valency patterns, while these, of course, also include fairly infrequent pat- terns, patterns with which only a very limited number of verbs can occur, as well as patterns, where in part also the lexical filling is fixed.19 A great number of verbs can be complemented by an [NP]-complement, for in- stance. However, taking all complementation types existing in the English language into account, total pattern overlap does not seem to be the default situation, not even amongst extremely similar verb meanings. Of all 22 groups only one group could be identified where all members of the group were found in the exact same set of valency constructions, i.e. phone, ring, and call. The apparent similarities among these verbs can have several rea- sons. First, they are used to refer to a very confined type of activity, and especially phone does not have any other senses that could interfere. In that sense, the synonymy is indeed very strong. Another factor that might play a role is that these verbs are, in comparison to other verbs in this study, com- paratively “new” or – like call – have not been used in this sense for a very long time compared to other verbs. Thus what can be observed with respect to the patterns these verbs allow might also be connected to the fact that all

19. The large number of patterns as opposed to the much smaller number of argu- ment structure constructions is also due to their formal basis. While Goldberg (1995, 2006a) uses labels such as oblique or resultative phrase as formal la- bels, the Patternbank uses much more fine-grained formal descriptions which would be subsumed under one construction in the Goldbergian sense. A comparison of semantically similar verbs 123 we can see is the early history of these verb uses and they might well de- velop different uses in the course of their future usage (Habermann, per- sonal communication). In addition to the ‘phone’-group, subgroups within six further groups were found to be identical as regards their occurrence in different valency constructions: reply and respond, suppose and presume, lease and rent, get, obtain, buy, and purchase, love and like, and respect and admire. Com- pared to the overall number of lexical units evaluated, this figure is rather small. Thus, irrespective of whether some modifications could be made to some of the verb groups, it will certainly be problematic to account for the fact that no more than these six components of groups can be found to pat- tern in the same way. This is clearly not very convincing for a theory where the semantics of a verb determines its formal patterns. It unmistakably shows that identity of pattern choice between two or more verbs is rather unusual. Moreover, the number of unsystematic restrictions – which is higher than the number of restrictions for which a semantic explanation could be identified – is clearly too high to be regarded as peripheral. These findings clearly point in the direction of storage of pattern information, as indicated by Herbst (2009: 62):

What is idiosyncratic about valency is that not all verbs (Klotz 2000, 2007) or adjectives (Herbst 1983) with a particular semantic feature necessarily occur in all valency constructions compatible with their meaning. This is why an important role has to be attributed to storage.

What exactly storage means from a cognitive perspective is by no means clear. Behrens (2007: 209), for instance, concludes in reference to Bybee’s (2006)20 usage-based grammar model that “[i]t is as yet not known whether we simply store more and more tokens upon repeated usage, or whether we store more repeated information on a more general and abstract level when available, or whether we do both.”

20. In the original text, Behrens (2007) quotes Bybee (2005), a manuscript of the article which was later published as Bybee (2006). 124 Pattern choice and verb meaning

3.3 Phenomena identified in the analysis

3.3.1 Complementation options in a valency framework – an overview

The following chapter provides a discussion of the different phenomena pointed out above, which can clearly be taken to support this claim. Gener- ally speaking, the association between verb meaning and pattern is not ne- cessarily a straightforward one-to-one relationship. Instead there are differ- ent possibilities:

− All uses of a verb can be analyzed as representing one single sense (ir- respective of whether there is one single pattern or several ones). − The different uses of a verb lead to a distinction into different lexical units, i.e. the difference in meaning is reflected in a different choice of valency patterns realizing them. − The different uses of a verb lead to a distinction into different lexical units but these cannot completely be distinguished on the basis of dis- tinct sets of valency patterns: certain valency patterns are available to more than one lexical unit of a verb and can even be realizations of dif- ferent participant patterns.

In the first case, there is a close relationship between verb meaning and pattern. The verb meaning entails a certain number of participants and for each participant there is exactly one formal realization. All uses of the verb express the same meaning. Blame, for example, occurs in three different valency patterns (one divalent and two trivalent ones, indicated by D, T1, and T2 respectively in the VDE-entry in Figure 16 below). They all express the sense “believe or say that someone or something is responsible for something bad” (VDE: 86), while each participant (indicated by Roman numerals) is realized by one distinct complement type.

Figure 16. Complement block of blame (VDE: 86) Phenomena identified in the analysis 125

Also the second case speaks in favor of a one-to-one relationship between form and meaning: different senses of one and the same verb can be distin- guished by their occurrence in different patterns. The verb conclude can be used to express that something gets finished or to express the activity of thinking in the sense that people reckon that something is the case from “evidence provided” (VDE: 170). These two senses are strictly separated by the patterns with which they can be expressed: there is not one single pattern both senses are associated with, e.g. conclude +[NP] is never inter- preted as ‘think’ and conclude +[that_CL] never as ‘finish’ (VDE: 169). If, in general, there were a one-to-one relationship between the participants of a verb and its complements, this is what one would expect. Yet looking at the data here and in VDE in general, it is quite clear, that this is not neces- sarily the case. One and the same pattern can be used to express different meanings (or nuances of meaning) of one and the same verb, i.e. its use in one sense of the verb does not need to block it for other senses of the same verb (third case). Breed, for instance, can mean ‘produce young’, ‘rear them’ (1), ‘produce effects’ (2), and ‘shape the character of someone’ (3) (VDE: 96). All senses except for ‘produce young’ can be expressed by the same pattern +[NP].

(1)VDE We breed barn owls in captivity and we release them back into the wild. (2)VDE Ask them why they carry a knife, and they tell you it’s self-defence. That might be true in some cases, but it just breeds more violence. (3)VDE All you do is breed more and more sophisticated criminals.

On the basis of this pattern alone, a distinction into lexical units might not even be considered necessary. However, taking the whole range of possible uses of this verb into account, the distinction is indispensable on the basis of differences in the pattern inventories that can be assigned to each sense respectively, i.e. in sum the patterns to be found associated with each sense differ. Moreover, ‘produce effects’ and ‘shape the character of someone’ are also distinguished from the other two senses in that the subject slot is usually filled with a non-animate participant, i.e. THEME rather than AGENT, and the fact that also ÆFFECTED can be non-animate. Thus a difference can also be identified in the participant inventories of these lexical units. Hence one and the same valency pattern can be the formal expression of different participant patterns and be associated with different senses of one and the same verb. Thus from a difference in meaning it must not necessarily fol- low that there is also a formal difference. The fact that different meanings are not distinguished formally is not au- tomatically in conflict with semantic determinism. Yet if the formal uses of 126 Pattern choice and verb meaning a verb are determined by meaning, one would expect that the same meaning is associated with the same formal properties as regards the occurrence in patterns. The comparison of semantically similar verbs shows that this is simply not the case. This does not mean that there is no such thing as blocking at all, however. In several of the groups, cases were identified where a verb can be distinguished into different lexical units (on the grounds discussed above) and a pattern is associated with only one of the them but blocked for the others (the second case described above) – while semantically similar verbs in the same group were found to accept the pat- tern in the “blocked” sense. Thus a pattern which is freely available to one verb in a certain sense is not available to another semantically similar verb in the same sense, since it is used in a very specific context with that verb (or even a completely different meaning). Again this speaks against a direct relationship between meaning and form. Besides the question as to whether a verb has one or several meanings and whether this is reflected in the patterns they can be complemented with or not, there is still another dimension as indicated inTable 36. In the ex- amples discussed so far each participant is expressed by means of a specific complement type and each participant pattern consequently by one single valency pattern, leading to a clearly distinguishable set of valency construc- tions (form-meaning pairs) (Type A). However, there are numerous cases where in order to express a participant more than one complement type is available, whether the verb has only one single sense (B1) or several senses (B2) and (B3).

Table 36. Possible combinations of verb meaning and patterns

1 2 3 one verb more than one more than one meaning verb meaning, verb meaning, no distinct patterns distinct patterns A participants realized by one (A1) (A2) (A3) distinct complement type B all or some participants are realized by several different (B1) (B2) (B3) complement types

In the case of award (B1), valency slot III (BENREC) can be expressed with either [NP] or with [to_NP]. Phenomena identified in the analysis 127

Figure 17. Complement block of award (VDE: 61)

In fact, there is not a single example in VDE of (A2), i.e. where a verb is distinguishable into different lexical units, while the senses are distin- guished by distinct patterns and each participant is realized by only one single complement type. This clearly shows that the matrix in Table 36 above depicts theoretic possibilities which do not necessarily reflect reality. What is more, the (B)-types are clearly no exceptions. Quite the contrary, the clear majority of entries in VDE exhibit this form of variability. This can even go as far as in the case of seem, where THEME can be expressed by fifteen different complement types and PREDICATIVE by eight complement types, while not all of them can be combined freely (e.g. only [AdjP], [to_INF], and [like_NP/V-ing] can occur in a trivalent pattern) (see below). Hence there is not always a one-to-one relationship between participant and complement and participant pattern and valency pattern respectively.

Figure 18. Complement block of seem (VDE: 729)

The fact that they cannot be freely combined speaks against the notion of a complement inventory as regards the representation of a verb’s valency 128 Pattern choice and verb meaning information in the mental lexicon and stresses the role of patterns as an entity in their own right (cf. chapter 4). One and the same participant pattern can be expressed by different va- lency patterns with one and the same verb. This means that there are paral- lel valency constructions. Herbst (2009) introduces the term constructeme to refer to “the set of all valency constructions that share the same partici- pant structures” (www.patternbank.uni-erlangen.de). This is not to say that constructions in a constructeme are completely synonymous but they clear- ly fulfill the same semantic function (cf. also chapter 5). Differences be- tween them can for instance be described on the level of subtypes or in part also by clausal roles. Again, if one assumes that it is the meaning of a verb which determines how it can be used, i.e. in which pattern it can occur, one would expect that such variability in the expression of a participant should also be mirrored by semantically similar verbs. There are numerous examples where this is not the case. The question to be asked here is to what extent semantically similar verbs exhibit the same properties when it comes to complementa- tion. What the analysis clearly reveals is that these properties cannot be in- ferred from the properties of semantically similar verbs. While the meaning of verbs obviously limits the kinds of construction that are plausible, not all plausible constructions are indeed used. Such instances will be the focus of the following chapters.

3.3.2 Same valency pattern but different participant pattern

The sample shows that there are several groups of semantically similar verbs which allow the exact same patterns, yet the participants expressed by these patterns differ despite the general semantic similarity of the verbs. The fact that one and the same complement type can represent different participants as such is not unusual (compare the discussion of possible roles of +[NP] by Van Valin [2004] in 2.3.1). Also Hunston and Francis (2000: 125) note that “two utterances with the same pattern may demand different role mappings if the lexical items in them are different”. Thus there is no congruence of form of complement and participant type as has already been discussed in chapter 2, e.g. the pattern [SCU]+ verb +[NP]+[NP] can be both a realization of AGENT + verb + BENREC + ÆFFECTED and AGENT + verb + ÆFFECTED + PREDICATIVE as in He gave her a present and He con- sidered her a role model. In this case the two verbs clearly differ. Another example which is often quoted in this context is He called her a taxi vs. He called her a fool. In this case the same verb is associated with these two Phenomena identified in the analysis 129 different valency constructions yet here call can be seen as representing two different lexical units (call ‘order’ and call ‘identify/label’), i.e. the difference in meaning could explain the different choice of participants expressed by the pattern. Alternatively, one could of course also argue that the difference in meaning is a result of a different choice of pattern in com- bination with a certain context.What is remarkable, however, – especially from the point of view of predictability – is that there are cases where the same form receives different interpretations despite its combination with lexical items that are semantically similar or even synonymous. If one be- lieves in a direct linking-relationship between meaning and form, one would expect that the same participant should be expressed by the same complement type when verb meaning and circumstances are the same. In part, this phenomenon can be deduced to a difference in the participant inventory of the verb, i.e. a pattern is ambiguous in combination with one verb because it has two participants that could be expressed with it, where- as another verb has only one of those two participants, which consequently renders the pattern unambiguous. Yet even in these cases, it is difficult to speak of a predictable pattern-participant association. While all three verbs answer, reply, and respond can occur in sentences with a participant BENREC, there still seems to be a slight difference be- tween respond and the other two verbs. Answer, reply, and respond can all be complemented with [to_NP].

(1)VDE Pasternak replied to him in a personal letter. (2)VDE Those responsible must now answer to the courts. (a) When we verb (something), BENREC gets it or the action is performed for BENREC’s benefit. (3)VDE He added that there could be times when force had to be adopted to respond to force. (4)VDE He would reply to such comments with honest, inappropriate looks of deep appreciation. (5)BNC To this the Tory Party answers with obscure economic jargon which most of its members seem to have no understanding of themselves.AHN 2004 (b) When we verb, we do it because of CAUSE.

While to him and to the courts are best paraphrased by (a), to force and to such comments and to this would be better paraphrased by (b) since they seem to fulfill the function of CAUSE rather than BENREC (since the two cannot cooccur in one and the same sentence, they could be analyzed as subtypes of one and the same participant). Thus one could say that to him in (1) expresses that entity which the reply (intrinsically expressed through the 130 Pattern choice and verb meaning verb) is intended for. This also applies to to the courts in (2), even if the meaning of answer slightly differs from that of reply and respond in this pattern. However, it does not apply to to force in (3), which describes what instigated the response. Since some native speakers accepted respond in very similar contexts as reply in this pattern, respond was analyzed as hav- ing a participant BENREC in chapter 3.2.2. Indeed, respond can also occur with more typical BENREC complements as in examples (6) and (7).

(6)BNC But I will ask dir-- , the director to re-- , respond to you on this Mr [gap:name] , er, er, did you want to also put this to vote?JJG 427 (7)BNC If they don't respond to me in the next forty eight hours I'm going to send them a second letter er tel-- . J9X 972

At the same time, the clear majority of complements of this type in combi- nation with respond are non-animate objects to which someone responds, for which a description as CAUSE appears to be more fitting and some na- tive speakers also clearly rejected a BENREC interpretation in combination with respond. When the perspective from the meaning of the com- plement to the meaning of the verb, one could also say that respond in this pattern more typically means ‘react to something’ than ‘answer someone’, whereas both of these senses can be found with answer and reply. Howev- er, the participant CAUSE seems to have a different status among the three verbs. There appears to be a tendency for respond to allow a greater variety of CAUSE-complements than is the case for reply and answer, where nouns such as question(s) or letter(s) are very typical collocates. Thus although answer is, as such, more frequent than respond, it seems to be much rarer in this pattern. Despite the fact that the verbs appear to be very similar and occur in the same pattern, when taking the lexical co-text into account it becomes quite obvious that analogies between the two verbs on a less ab- stract level than the pattern are clearly not possible. Press has the participants AIM, AFFECTED, and REFERENCE, and the pat- tern [SCU]+ verb +[NP]+[on_NP] receives two different interpretations with this verb, which are illustrated below.

(8)BNC ‘Don't press it AIM on him AFFECTED’ said Lady Grubb. FSP 1133 (9a)VDE I mean I never pressed him AFFECTED on it REFERENCE.

Example (8) would not be described adequately by the participant pattern AFFECTED + REFERENCE. The notion of affectedness of it is clearly con- nected to its position directly following the predicate, which generally puts a certain focus on the participant. This can be described by the clausal role Phenomena identified in the analysis 131

FOCUSSED AFFECTEDNESS (Herbst and Schüller 2008: 160), which is added to the actual participant role (cf. chapter 4.1.1 for a discussion of clausal roles). The semantically similar verbs push and urge do not allow both interpretations.

? (10)VDE Every night she conferred with John Cullen, trying to urge/* push speed AIM on him AFFECTED, instead of the perfection he desired. (11)NSA Michael and David pushed more gin AIM upon Angel AFFECTED, intently watching her drink. (9b)NSA I mean I never *pushed/?urged him on it.

Push was not found acceptable in sentence (10) but can occur in the con- struction illustrated by sentence (11) with the variant upon. It cannot occur in the second valency construction (9b), which could be “explained” by the fact that it does not have a participant REFERENCE (see discussion in 3.2). Urge is compatible with the first construction illustrated in (10), while na- tive speakers did not agree with respect to the second construction (9b). In any case, a rule-based connection between the meaning of these verbs and the acceptability or restriction of the interpretation of this pattern is difficult to imagine. There are also cases where verbs with the same meaning occur in the same valency pattern and despite the fact that they also have all the relevant participants, the valency patterns realize different participant patterns. Thus although both indicate and show can express ÆFFECTED and LOCATION with the pattern [SCU]+ verb +[NP]+[ADV], and both verbs have the par- ticipant BENREC, only show can have an additional interpretation of this pattern: [VC I] is only acceptable with show, whereas [VC II] is compatible with show and indicate and possibly also with demonstrate as indicated in (14):

[VC I] [SCU]_AGENT + show + [NP]_BENREC + [ADV]_LOC/DIRECTION [VC II] [SCU]_AGENT + show/indicate + [NP]_ÆFF. + [ADV]_LOC/STATE

(12)VDE Our son Ted had shown/*indicated/*demonstrated them BENREC around the farm LOC/DIRECTION before lunch. (13)BNC So it'll pause and show it ÆFFECTED on the screen LOC/STATE. G4K 549 (14)VDE Brian, the leader for the day, chose a favorite nursery rhyme, Little Boy Blue, and led the class in reciting it by using a pointer to indi- cate/?demonstrate the words ÆFFECTED on the chart LOC/STATE. 132 Pattern choice and verb meaning

Similarly, only teach can realize both participants ÆFFECTED and BENREC ([VC I] and [VC II]) with the pattern [SCU]+ verb +[NP], despite the fact that also instruct, train, and educate have these participants:

[VC I] [SCU]_AGENT + teach/instruct/train/educate +[NP]_BENREC [VC II] [SCU]_AGENT + teach +[NP]_ÆFFECTED

(15)VDE From 8.30 am to 10 am on Tuesdays I teach students BENREC and take them round the ward. (16)VDE Then the judge will instruct the jury BENREC. (17)VDE He is director of Youth Alive Ministries, an organisation which helps to educate and train young people BENREC. (18)VDE The aristocrats might educate the democrats BENREC. ? (19)VDE Science teachers included those teaching/?instructing/* educating/ *training physics, chemistry, biology and mathematics ÆFFECTED.

As regards the functional description of this post-verbal noun phrase, ap- proaches differ. According to CGEL (10.7), [VC I] includes a direct and [VC II] an indirect object, while Aarts and Aarts (1982: 138) classify ob- jects which directly follow the verb phrase generally as direct object, which again underlines the problematic status of this category (see, for instance, Herbst and Faulhaber forthc.). Eitherway [VC II] is not possible with train and educate. With train, BENREC is obligatory and with educate ÆFFECTED must be realized as a [that-CL], which explains these restrictions. Native speakers disagreed with instruct and they were fairly consistent in their judgment, i.e. either they always demanded a BENREC or they consistenly accepted ÆFFECTED in this slot. Consequently, even if the formal pattern itself is not restricted, the type of lexical filling allowed in the pattern is. With train and educate (and prototypically also with instruct) [NP] has to be +animate and thus, in this case, BENREC. The restrictedness with respect to ÆFFECTED in this slot is also reflected in another pattern which has two interpretations in connection with this verb group: [VC III] occurs with train and educate, whereas [VC IV] can only be found with teach.

[VC III] [SCU]_AGENT+ train/educate +[NP]_BENREC +[as_NP]_PREDICATIVE [VC IV] [SCU]_AGENT + teach +[NP]_ÆFFECTED +[as_NP]_PREDICATIVE

(20)BNC Mahmoud BENREC, for instance, had been educated as a lawyer in the French tradition PREDICATIVE. J10 1336 (21)VDE Two crew from each yacht BENREC have been trained as first aiders PREDICATIVE at the Royal Naval Medical School, Haslar. Phenomena identified in the analysis 133

(22)BNC …; English ÆFFECTED is taught as a subject PREDICATIVE from the first class. H88 273

Apart from this, there also appears to be a structural difference between these two patterns: the construction containing BENREC is typically found in passive patterns, while the construction with ÆFFECTED can also occur in active patterns. In verb group 14 (‘propose’), only recommend allows a change of roles in sentences such as the following: “recommend a patient/client to a spe- cialist” and “recommend a specialist to a patient/client” (Algeo 2006: 224). For propose and suggest this does not seem to be possible, although both have the relevant participants.

(23)QE And they in turn recommended me to Eric Gustavson, a long established consultant plastic surgeon. (Algeo 2006: 224) (24)VDE On leaving, she secured a job at Holborn Studios as a freelance hire assistant and, after only a year, the owner recommended her to David Bailey, who took her on as his assistant. (25a)VDE Listen, I’m going to propose something to you. ? (25b)NSA *Listen, I’m going to propose you to someone. (26a)VDE No one had ever suggested the possibility to him before, he said. (26b)NSA *No one had ever suggested him to a specialist before, he said.

Finally, one can allow, permit, and authorize [something] [to someone] (+[NP]_AFFECTED +[to_NP]_BENREC), and although entitle allows the same pattern and has the same participants, the participant-complement mapping differs, i.e. one can only entitle [someone] [to something] (+[NP]_ BENREC +[to_NP]_AFFECTED).

(27)VDE They have allowed no such luxury AFFECTED to anyone else BENREC. (28)BNC Under its terms, exchange of information and materials AFFECTED was permitted solely to those countries that had made “substantial pro- gress” in the development of nuclear weapons BENREC. ABA 836 (29)BNC He also confirmed that he had authorized US aid AFFECTED to Hondu- ras BENREC in return for Honduran assistance to the contras... HKR 412 (30)BNC I don't know. I'm BENREC not really entitled to anything AFFECTED.CCM 1377

These examples clearly show that although these verbs are undoubtedly se- mantically similar and are used in similar contexts with a similar pragmatic function, the coordination of formal pattern and semantic participant pat- tern is not always identical. This being the case, it is clearly problematic to argue in favor of a one-to-one correspondence between form and meaning. 134 Pattern choice and verb meaning

3.3.3 Same pattern restricted to a very specific context

All verb groups discussed below include verbs which have semantically similar uses (the “group” meaning) but which can express a different mean- ing when combined with a specific pattern. Thus a pattern which is com- patible with several (or even all) of the verbs in the groups has the expected “group”-meaning only with some of them. From the point of view of the pattern one could say that the pattern triggers a different verb meaning, with one (or some) but not all verbs in the group. This phenomenon was found in 10 of the 22 verb groups, in some of them even with several dif- ferent lexical units. This shows that semantic similarity is clearly confined to only part of the usage of the verbs under consideration. The constructions [SCU]_AGENT + verb +[NP]_ÆFFECTED can only be found in combination with gather in the sense ‘collect’ but not in the sense ‘assume’, while the pattern [SCU]_AGENT + verb +[NP]_ÆFFECTED +[from _NP]_SOURCE seems to allow both interpretations.

(1)VDE Don’t assume anything. (2)VDE She pauses, gathering her strength for the next sentence. = ‘collect’ (3)BNC He was not supposed to know whether the dealer was buying or selling on his client's behalf, but nonetheless he attempted to guess this from the runner's facial expression and comments. EUU 1876 (4)NSA He gathered this from the runner’s facial expression. (5)VDE Most prominent among the last category is the information gathered from the massive national survey of household resources and standards of living. = ‘collect’

Pronounce can be found in the pattern [SCU]+ ‘declare’ +[NP] +[to_NP] in the sense ‘pronounce a word in a certain way’, i.e. in the sense ‘speech’ but not in the sense ‘declare’. It might be argued that the pattern is blocked for the sense ‘declare’ because of this. However, the same does not apply to other patterns of pronounce which are available to both senses as illustrated in (8) and (9).

(6)VDE Failure to declare the site to the UN by the 18th of April is a violation of the ceasefire resolution. (7)BNC Okay so sometimes it helps if when you see a word you pronounce it to yourself. FMG 696 = ‘speech’ (8)VDE Anthony Spick pronounced the takeover as good news; ...= ‘declare’ (9)VDE Her voice was very chipped and British, firmly pronouncing ‘at’ as ‘et’. = ‘speech’ Phenomena identified in the analysis 135

The ‘intend’-group, and especially mean, are another case in point for the fact that the pattern a verb combines with has an effect on the meaning of that verb. Patterns which occur in combination with some of the other verbs in the sense ‘intend’ in part trigger a completely different meaning in com- bination with mean, even if in most cases this does not imply a difference in the participant inventories of these verbs (based on VDE: 528). Com- pare:

(10)VDE Did he mean that interest rates would come down rather slowly? = ‘WHAT I MEAN’ (11)VDE The opposition still intends that he should be removed from all decision making. (12)VDE I had been a serious smoker and it meant giving up a 30-year, 35-a-day habit. = ‘imply’/’CONSEQUENCE’ (13)VDE On another level, this shows that you intend doing things together and you will take decisions together.

With allow, the pattern [SCU]+ verb +[that_CL] triggers a meaning differ- ent from ‘permit’. As example (14) shows, the meaning is better described by means of ‘admit’ (cf. VDE) and clearly differs from (15) and (16).

(14)VDE I certainly would allow that things had taken an unfortunate turn. = ‘admit’ (15)BNC It is therefore recommended to authorize that the necessary steps be undertaken. HD0 734 (16)BNC This was “due entirely to the schoolmasters who will not permit that children in their schools use anything but their proper names”. J0P 268

The verb pick shows greater similarity to choose and select than to other verbs in the ‘choose’-group such as nominate or elect. Nonetheless there are several patterns of pick which trigger a reading different from the com- mon group-meaning ‘choose’, i.e. picking flowers, picking a fowl or picking on someone in the sense of harassing someone. Since the latter sense only occurs in the pattern with +[on_NP], it might also be seen as an idiomatic unit and is often analyzed as a phrasal verb (cf. LDOCE4).

(17)NSA An opposition leader in Zaire has confirmed that he has decided to de- cline an offer by president Mobutu to choose him prime minister. (18)VDE They want to make trouble asking the boys to pick them flowers. = ‘pick flowers’ (19)BNC He contends that economic prosperity has given the voters the opportu- nityto choose on issues like abortion and the environment. A3U 16 (20)VDE We’ve done nothing to them, yet all they do is pick on us. = ‘harrass’ 136 Pattern choice and verb meaning

In the ‘quarrel’-group, argue can be found with both +[NP] and +[that_ CL], which both realize the same participant. However, the meaning of argue is not the same as that of dispute, which also allows this form of complementation.

(21)BNC … it must be recognized that to make such claims was to dispute a wide- ly held belief, based on Aristotelian physiology, that women were by na- ture soft and therefore inconstant ... AN4 1516 (22)VDE The parties do not even argue the case in front of each other. = ‘dis- cuss’ (23)BNC No one can dispute that Sinead's behaviour of late -- particularly tear- ing up the Pope's picture -- has been outrageous but many feel that her protest statements are at least based on strong personal beliefs. CEK 381 (24)VDE Many farmers argue that the level of grants for planting woodland is too low. = ‘claim’

Teach, train, educate, and instruct can all occur in a monovalent pattern but they are not synonymous in this pattern. While educate, train, and teach express the meaning ‘teach’ as illustrated in (25)–(27), instruct is better interpreted as ‘order’ as in (28). The same applies to the pattern [SCU]+ ‘teach’ +[NP]+[to_INF] in examples (29) and (30). Moreover, in the case of train, the participant differs from the others in that the subject slot is filled by BENREC and not by AGENT (27).

(25)BNC This is a book which, like the BBC itself, informs, educates and enter- tains. BN9 1653 (26)VDE Her husband teaches at Portland State University. (27)VDE She trained at the Prue Leith School of Food & Wine in London, and also in Switzerland. (28)BNC Her mother talked and instructed and bustled until the minute she left the house. HJH 1550 = ‘order’ (29)VDE They were still trying to teach her to read and write. (30)BNC A note from Mellowes instructed me to cast my eye over the draft, pron- to, for inaccuracies. A0R 641 = ‘order’

Instruct, unlike the other verbs in this group, has the meaning ‘employ’ in contexts such as the one given in (32).

(31)VDE Two crew from each yacht have been trained as first aiders at the Royal Naval Medical School, Haslar. (32)BNC Michael Beloff QC and Richard Drabble have been instructed as coun- sel. HAJ 1377 = ‘employ’ Phenomena identified in the analysis 137

Recall +[NP]+[for_NP] means ‘call back’ rather than ‘remember’:

(33)VDE She is perhaps best remembered for her creation of the Belgian detec- tive, Mr Hercule Poirot. (34)BNC He recalled Congress for a special session and challenged it to enact the Republican programme. EWG 919 = ‘call back’

The verbs consider, think, reflect, ponder, contemplate, and judge can all be used more or less synonymously when complemented by +[(that_)CL] or +[Q/S]. Yet while think is also used in the pattern +[NP], this is not possible for judge and reflect in the same sense. Compare

(35)VDE Barge people are very often alone and have time to think grand thoughts. = ‘think’ (36)VDE Historically and socially, the system has reflected a male perspective, and men are still dominating the centres of power and influence. = ‘mir- ror’ (37)VDE It’ll be about six months before he can judge the success of the treat- ment, but there have been no problems so far. = ‘assess’

The pattern +[wh-CL] can have both interpretations with reflect but again only the ‘assess’-meaning in combination with judge.

(38)VDE I’ve never thought why I like something or why I don’t like something. = ‘think’ (39)VDE Above all, Steve’s house reflected how little time he had for a personal life. = ‘mirror’ (40)BNC He smiles to himself, reflecting how the roles of the Samaritan have gone back and forth in their relationship. BN1 905 = ‘think’ (41)VDE Mr Heath said it wasn’t for him to judge whether his mission had been a success. = ‘assess’

Also other patterns trigger the ‘assess’ meaning of judge or a further mean- ing which could be described by the template ‘competition’:

(42)VDE John Quayle said an extraordinary judiciary hearing would be called to judge on the incident at 8 o’clock today. = ‘assess’ (43)VDE Her presence makes me reassess my surroundings. My routine is infused with a consciousness of Ellen watching, judging.(only if clear from context) = ‘assess’ (44)VDE She judged at major British shows including the Royal Welsh and The Royal. = ‘competition’ (45)VDE Dave Wise judged the show and then explained what he looks for in a bird when judging. = ‘competition’ 138 Pattern choice and verb meaning

It is important to note that the “new meaning” is by no means just added by the pattern as suggested in frameworks such as construction grammar (e.g. Goldberg 1995, 2006a), i.e. it is not just the added complement type that needs to be taken into consideration, the lexical context plays an important role, too. In some cases, as with the ‘telephone’-group, the pattern can even be described as the same valency construction on a more abstract level. The verb call complemented by +[for_NP] can be interpreted as ‘demand some- thing’, not necessarily triggering a ‘telephone’ reading, however, not ex- cluding it either. In both cases, the complement can be seen as a realization of PURPOSE.

(46)VDE I asked the front desk to ring for a taxi. (47)BNC After a leisurely tea by a coal fire, I asked the clerk to phone for a taxi.ADM 1347 (48)BNC Mr Livingstone called for cuts in defence spending. A1P 78 = ‘demand’ (49)NSA I asked the front desk to call for a taxi.

In many of the cases listed above, a specific valency pattern seems to be reserved for one specific meaning in the sense that it blocks other interpre- tations. Then the different lexical units of a verb can clearly be distin- guished by the types of pattern they occur in (cf. 3.3.1). Yet this form of “blocking” is not necessarily always the case as examples (39) and (40) above show. Goldberg (2006a: 57–58) points out that some verbs are bi- ased towards certain patterns in some of their meanings, e.g. find in the sense “realize” favors a clausal complement. Such preferences, however, do not necessarily confine a verb to its preferred pattern, i.e. it might still occur in patterns which are more typically associated with other lexical units of the same verb. The Erlangen Valency Patternbank lists 1,247 lexi- cal units with the active pattern [SCU]+ verb +[NP], which are distributed over only 443 different verbs. Thus it is quite obviously the rule rather than the exception that a pattern is used for more than one lexical unit of a verb and that this does not result in any form of blocking. This does not only apply to frequent patterns such as [SCU]+ verb +[NP] but also to some much rarer ones. A similar result can, for instance, be found with the pat- tern [SCU]+ verb +[upon_NP], which is found in combination with 48 different lexical units but only with 34 different lexemes. What constitutes a different meaning of a verb, however, is by no means obvious. Herbst (2003: 69–70) points out the circularity of deciding whether such differ- ences are indeed a matter of the verb meaning or whether it is the meaning of the complement that highlights different aspects of the meaning of the Phenomena identified in the analysis 139 verb. Nonetheless it can be concluded that this kind of blocking does not seem to be a typical mechanism to account for restrictions: it is possible to have the same pattern in combination with the same verb expressing differ- ent lexical units. Examples can again be found in the ‘think’ group: reflect in a monovalent pattern can mean both ‘think’ and ‘mirror’.

(50)VDE After all, as any astrologer will avow, the moon reflects in much the same way as a mirror. (51)VDE The pilot reflected a second.

The same applies to [SCU]+ declare +[NP]+[against_NP], which means ‘declare war’ or ‘declare’. Here, a difference can be seen in the type of noun phrase available for the first slot. In case of ‘declare’, this appears to be a reflexive and in the case of ‘declare war’ it is quite obviously confined to the noun war.

(52)VDE But that title, of Earl of Falmouth, was granted in March 1665 as part of the honours bestowed when war was declared against the Dutch. = ‘declare war’ (53)VDE Poland’s Finance Minister has declared himself against giving large numbers of shares to people directly employed by privatised enterprises. = ‘declare’

Finally, [SCU]+ argue +[NP] means either ‘discuss’ or ‘claim’. Dispute, on the other hand, although it is not completely synonymous with quarrel either, as it could, for example, be replaced by deny in this context, is still closer to the group meaning than argue in examples (22) and (54).

(22)VDE The parties do not even argue the case in front of each other. = ‘dis- cuss’ (54)VDE I was amazed to read your leading article arguing the merits of a Scot- tish Super League. = ‘claim’ (21)BNC … it must be recognized that to make such claims was to dispute a wide- ly held belief, based on Aristotelian physiology, that women were by na- ture soft and therefore inconstant ...AN4 1516 (23)BNC No one can dispute that Sinead's behaviour of late -- particularly tear- ing up the Pope's picture -- has been outrageous …. CEK 381

Dispute differs from quarrel, argue, and bicker in that it has an additional participant ÆFFECTED, which explains a number of the syntactic differences between these verbs as well as its similarity to deny. Accordingly these restrictions were not counted as idiosyncratic. 140 Pattern choice and verb meaning

Thus synonymy is extremely dependent on the context in which a verb form occurs, syntactic valency pattern as well as semantic context – par- ticipants as well as lexical co-text. One could indeed identify judge and consider as synonyms but only confined to specific situations as in (55) and (56) as they can differ completely in others, e.g. (57) and (37).

(55)VDE Newcomers, living in homes built within the past 20 years, are often the most vociferous objectors and rarely seem to consider that they them- selves have benefited from recent planning permission. (56)VDE “I judge that Mrs. Plover was cooperative,” Mr.Macmillan said to him. (57)VDE He said he would take time to consider the matter. (37)VDE It l be about six months before he can judge the success of the treatment, but there have been no problems so far.

What these cases show is that the multiple use of one and the same pattern for different lexical units of the same verb is clearly far from unusual. Finally, it does not come as a surprise that several of the examples are monovalent uses, which obviously reduces the number of participants in- volved and consequently also the specifications these can bring with them. This applies to reflect (‘mirror’ or ‘think’) and judge (‘assess’ or ‘competi- tion’), but again this does not lead to any form of blocking either. Many more such examples can be identified and the number clearly ex- ceeds those cases specifically pointed out. Since the question addressed was only whether the verb can express the group meaning in this pattern or not, additional meanings were not pointed out in the overview presented in chapter 3.2. To put this into a wider perspective: what these examples indi- cate is that the correlation between form and meaning is specific to every single verb. Whether a verb allows a certain pattern or not has no predictive power as regards the use of a semantically similar verb in this pattern. De- spite the fact that some verbs can be used synonymously in certain patterns, this does not give a safe indication as to whether they can be used in other patterns as well. Moreover, it is not enough to check whether patterns can be transferred from one verb to another semantically similar verb, it also has to be made sure that this does not lead to a change in the meaning of the verb to which the pattern is transferred. Hoey (2005: 82) explains this phenomenon in terms of different lexical primings for different senses of one and the same word in the sense that “the more common meaning … drives the rarer meaning into a grammatical corner”. Phenomena identified in the analysis 141

Where it can be shown that a common sense of a polysemous word is primed to favour certain collocations, semantic associations and/or colliga- tions, the rarer sense of that word will be primed to avoid those colloca- tions, semantic associations and colligations. The more common use of the word will make use of the collocations, semantic associations and colliga- tions of the rarer word but, proportionately less, frequently. (Hoey 2005: 82)

How this could be explained in a model which takes a direct relationship between form and meaning for granted appears problematic. It is difficult to imagine a form of regulation that could account for such item-specific phenomena with a general, not lexically-driven mechanism.

3.3.4 Different formal realizations of the same participant

The phenomenon most frequently identified is a restriction of the formal realizations of a participant. Despite the fact that many of the verbs can be shown to have the relevant participants for the expression of a specific par- ticipant pattern, they cannot be found in the same range of possible formal realizations of this pattern. This can be due to a difference in optionality or the choice of a different complement type for the expression of the same participant.

3.3.4.1 Differences in optionality

Table 37 lists cases where semantically similar verbs exhibit differences as regards the optionality of certain participants or complements realizing these participants. Not expressing the respective patterns leads to ungram- matical sentences in the sense of Helbig and Schenkel’s (1973: 37) defini- tion of obligatoriness on the basis of non-deletability, i.e. if an obligatory complement is deleted, the sentence becomes ungrammatical (see also Heringer 1996: 161–166). If a participant is obligatory with part of the group but not with all of its members, this automatically leads to incompatibilities with all those pat- terns in which the respective participant is not expressed. One possible interpretation of such cases is to say that a difference in a verb’s minimum valency entails a difference in meaning. If a participant is obligatory with one verb but not with another (semantically similar) verb, this could be described as a difference in the participant inventory or as a difference in profiling. Goldberg, based on Langacker (1987), defines profiled roles as “entities in the frame semantics associated with the verb that are obligato- 142 Pattern choice and verb meaning rily accessed and function as focal points within the scene, achieving a spe- cial degree of prominence” (Goldberg 1995: 44).

Table 37. Verbs with obligatory complements which are (contextually) optional with other verbs in the same group

verb group differences in optionality of complements ‘intend’ mean demands ÆFFECTED, while intend, plan, and aim do not

‘like’ adore typically demands AFFECTED, while love and like do not

‘permit’ entitle demands [NP]_BENREC, while allow, permit, and author- ize do not; authorize and entitle demand ÆFFECTED, while permit and allow do not

‘remember’ bear in mind demands ÆFFECTED, while it is (contextually) op- tional with remember, recall, recollect, and reminisce

‘teach’ train demands BENREC (either as complement in the predicate or as a complement functioning as subject [ergative use]), while it is contextually optional with educate and instruct, and freely op- tional with teach.

Thus if a participant is obligatory, it might well have a different status in the scene described by the verb in the sense that it either receives the main focus or it is at least inseparably tied to the type of activity expressed. The following sections will present cases identified among the 22 verb groups examined here. The verb mean differs from intend, plan, and aim in that patterns with- out ÆFFECTED do not seem to be possible with mean in this sense.

(1)VDE But nothing would happen before they reached the lonely plains of West- ern Tibet, so at least he had time to plan/*mean. (2)VDE That’s precisely what Labour should aim/*mean at: the replacement of a market mechanism with planning mechanisms. (3)VDE Aim/*mean for clarity and precision; remove words that obscure your directions. (4)BNC And I think she intends/*means on having quite a few drinks. KD3 528

Note, however, that the underlined complements in (2), (3), and (4) are all realizations of AIM, a participant which is never expressed with mean. This could of course be an alternative explanation for the unacceptability of these sentences, which obviously reduces the amount of evidence for an obligatoriness of ÆFFECTED in combination with mean. Phenomena identified in the analysis 143

In the ‘like’-group, only love and like can be used in a monovalent pat- tern. The slot following the verb appears to be freely optional with love and contextually optional with like but obligatoy with adore. It has to be taken into account, however, that like in (5) is clearly a highly idiomatic expres- sion and cannot freely occur in monovalent patterns. Nonetheless it is not possible to use adore in analogy to like and love in either of the two con- texts.

(5)VDE Get a glass, if you like. ? (6)VDE He is no longer able to love/* adore and hate, to invest himself in life.

In the ‘permit’-group, entitle – other than allow, permit, and authorize – always seems to need a BENREC participant to be realized. None of the pat- terns without this participant is acceptable:

(7)VDE The parliament passed a resolution urging the Algerian government not to allow/*entitle the introduction of what it called an authoritarian po- litical order ÆFFECTED. (8)NSA *They entitled him. (9)BNC The weighting screen presents twenty options including on-line help, and saving of changes, while others allow/*entitle sorting of data, changing of titles, reversing of the XY ranges and so on ÆFFECTED. A19 1063 (10)VDE Originally the scheme allowed/*entitled for pensions to be calculated on the best twenty years of earnings ÆFFECTED. (11)BNC This was “due entirely to the schoolmasters who will not permit/*entitle that children in their schools use any but the proper names” ÆFFECTED. J0P 268

Authorize, on the other hand, obligatorily demands the realization of ÆFFECTED, i.e. it cannot occur in a monovalent pattern but it does not need to express BENREC.

(12)VDE India would be willing to take out other nationals if resources per- mit/*entitle/*authorize.

This can also account for the fact of why complementation with +[NP] +[into_NP] is not acceptable with authorize, yet it does not explain the restriction with entitle as the pattern contains BENREC. Thus it seems that entitle demands both ÆFFECTED and BENREC to be realized.

(13)VDE Burma had not allowed/*authorized/*entitled foreign journalists and other observers into the country. (14)BNC They were permitted into it only when giving evidence and not more than one at a time. CCC 1231 144 Pattern choice and verb meaning

In the ‘remember’-group, bear in mind always needs to express ÆFFECTED, while it is at least contextually optional with remember, recall, and recol- lect.

(15)VDE I remember the visit we made to the gliding championships in Epernay. Do you recall? (16)VDE I can’t tell you what happened, because I don’t remember/*bear in mind. (17)BNC …, Leonard found his peace; his “sitting-down time,” as he called it, where he could -- as all poets must -- recollect in tranquillity. A0P 1597 (18)BNC She could see that if the circumstances of their meeting had been hap- pier they would have had a good drink of her mother's secret stash of Bushmill's and reminisced all night. FAB 3757 (19)BNC Back at the car park we had a well earned cuppa and reminisced/*bore in mind over another hot day back in 1933. ECG 907 (20)VDE “Don’t you remember about Anna?” I shook my head. (21)NSA *Let’s bear in mind about this.

While (17) and (18) are clearly cases of contextual optionality, ÆFFECTED appears to be completely optional with recollect and reminisce in (19) and (20). Differences with respect to the optionality of BENREC accounts for the majority of restrictions in the ‘teach’-group. Train does not occur in any of the patterns without it.

(22)VDE In a 16-session programme, Fitness Director Mel Hillsdon teaches/ *trains how dieting can often be detrimental. (23)VDE Nutritionist Patricia Becker teaches/*trains how to improve your health with grain, noodle and vegetable salads. (24)VDE And in teaching about Auden, he taught/*trained about how to write, how to live and how living and writing are linked. (25)VDE The fact that schools are to be judged on the pupils’ performance puts pressure on teachers to teach/*train to the tests. (26)VDE The Scriptures teach/*train that God cannot lie.

Train can also be used in a divalent pattern where the subject complement unit is realized by BENREC and the post-verbal complement by the partici- pant PREDICATIVE. With instruct and educate it does not seem to be possi- ble to fill the subject slot with BENREC without passivization. Teach was found in the same pattern but no BENREC is implied in sentence (28) as opposed to (27). Phenomena identified in the analysis 145

(27)VDE The idea of specializing in sports medicine came naturally to Harris, who trained as a pediatrician. (28)NSA US Jesuit forbidden by Vatican to teach as Catholic theologian.

Thus only train can be used in an ergative way, which makes the following patterns possible. In both cases, she is BENREC rather than AGENT, while the subject position attributes agent-like characteristics to the participant. As discussed above, this can be described by means of an additional level of semantic roles, i.e. clausal roles (cf. Herbst and Schüller 2008: 158–163).

(29)VDE She started off wanting to train to be a nurse and that didn’t last very long. (30)VDE She trained at the Prue Leith School of Food & Wine in London, and also in Switzerland.

The cases presented above show two things. On the one hand, obligatori- ness can be seen as dependend on the meaning of the verb and thus corre- lates with formal restrictions as regards verb complementation. At the same time, this is an aspect of verb meaning which is extremely difficult to pre- dict, which stresses the arbitrary nature of pinning down verb meaning in the first place. From a cognitive point of view this clearly means that such aspects need to be memorized and might well help to ease memorization of the patterns associated with the respective verb. However, the information is still item-specific and it seems difficult to reconcile this with a view of general semantic rules.

3.3.4.2 Different formal realizations of the same participant – an overview

Cases where semantically similar verbs differ with respect to the range of possible complement types that can be used for the realization of one and the same particpant can be found in 16 of the 22 groups. Table 38 provides an overview of all the groups which exhibit this phenomenon and their various restrictions in the realization of certain common participants (verbs in brackets do generally not have the recepective participant or another participant of the pattern). The different cases will be discussed below, following a general discussion of the theoretical issues at stake. The cases demonstrate that the same meaning – in the sense that the verbs describe a similar activity and the same participants are involved – can be expressed by means of different formal complement types. 146 Pattern choice and verb meaning

Table 38. Different complements realizing the same participant

[SCU]+ PREDICATIVE +[NP]+[to_INF]: (*gather) +[NP]+[AdjP]: *guess, (*gather) ‘assume’

PREDICATIVE +[NP]+[as_NP] ?nominate, ?designate +[NP]+[NP] *choose,*pick, ?select, ?nominate, ?designate SOURCE +[NP]+[from_NP] all +[NP]+[among_NP] ?designate,?select ‘choose’ +[NP]+[out of_NP] ?designate +[from_NP] ?nominate, *?designate,(*name) +[among_NP] *pick, *elect, *appoint, *designate,(*nominate,*name) +[between_NP and NP] *elect, *appoint, *designate, (*nominate, *name) ÆFFECTED and BENREC +[NP]+[for_NP] all +[to_NP]+[that_CL] all +[to_NP]+[wh-CL] ?pronounce, ?proclaim +[to_NP]+[Q/S] *pronounce PREDICATIVE +[NP]+[as_NP] / +[AdjP] / [+NP] all ‘declare’ +[NP]+[as_AdjP] *pronounce “PREFERENCE” +[for_NP] *pronounce, *proclaim, announce = ‘candidacy’ +[in favor of/against_NP] *proclaim, *state, (*announce) ÆFFECTED + BENREC +[NP]+[for_NP] all ‘get’ ‘get’ +[NP]+[NP] *acquire, (*receive) ÆFFECTED +[NP] *aim +[V-ing] *aim, mean = ‘imply’ +[that_CL] mean = ‘imply’ AIM ‘intend’ +[at_X] *intend, *plan, (*mean) +[for_NP] *?intend,(*mean) +[on_X] *aim, (*mean) Phenomena identified in the analysis 147

ÆFFECTED +[for_NP] *authorize, *permit, (*entitle) +[of_NP] *authorize, (*entitle) +[to_INF] *?permit, *authorize, (*entitle) +[V-ing] (*entitle)

‘permit’ +[for_NP V-ing] *authorize, *permit, (*entitle) +[NP ([+]) +to_INF] all +[for_NP to_INF] *entitle,*authorize +[that_CL] (*entitle), allow = ‘admit’ ÆFFECTED +[NP]/[to_INF] all +[V-ing] *train

‘practice’ +[wh-to_INF] ?train

EFFECTED +[NP]/[wh-CL]/[Q/S]/[that_CL]/[V-ing]/[so/…] suggest, recommend, propose +[NP V-ing] *?propose

‘propose’ +[to_INF] *suggest

TOPIC +[about_NP]/+[over_NP] all +[with_NP]+[about_NP] ?dispute +[with_NP]+[over_NP] ?dispute ‘quarrel’ +[with_NP]+[on_NP] *bicker, *?dispute TOPIC +[over_NP] *remember, *recall, *recollect, (*bear in mind) +[about_NP] *recall, *recollect,(*bear in mind) +[what/sth/…]+[about_NP] ?recollect, *reminisce ÆFFECTED +[that_CL]/+[if/wh–CL] all ‘remember’ +[NP] / +[wh-to_INF] *reminisce +[QUOTE] *bear in mind +[V-ing] / +[NP V-ing] *bear in mind, *reminisce BENREC and ÆFFECTED +[to_NP]_BENREC +[that_CL]_ÆFFECTED all +[to_NP]_BENREC +[wh-CL]/[wh-to_INF]_ÆFFECTED *show +[NP]_BENREC +[wh-CL]_ÆFFECTED *demonstrate ? ? +[NP]_BENREC +[NP]_ÆFFECTED * indicate, * demonstrate +[NP]_BENREC +[that_CL]/[wh-to_INF]_ÆFF. *indicate,*demonstrate ‘show’ PREDICATIVE +[NP]+[as_NP] all +[NP]+[AdjP] *indicate, *demonstrate +[NP]+[as_AdjP] ?indicate, *demonstrate 148 Pattern choice and verb meaning

BENREC and ÆFFECTED +[NP]_BENREC +[that_CL]_ÆFFECTED *train +[NP]_BENREC +[to_INF]_ÆFFECTED instruct = ‘order’ ? +[NP]_BENREC +[wh-to_INF]_ÆFFECTED *educate, * train +[NP]_BENREC +[NP]_ÆFFECTED *educate, *train, *instruct ? +[NP]_ÆFFECTED +[to_NP]_BENREC *educate, *train, * instruct +[NP]_BENREC +[NP]_ÆFFECTED+[from_NP]_SOURCE *instruct (*edu- cate,*train)

‘teach’ ? +[NP]_BENREC +[more/sth…]_ÆFFECTED +[about_NP]_TOPIC * educate, *instruct, (*?train) TOPIC +[NP]+[about_NP] train = ‘direct’ +[NP]+[on_NP/wh-to_INF] *teach +[NP]+[more/sth/..]+[about_X] *instruct, *teach, ?*educate, (*train) ÆFFECTED +TOPIC +[what/sth/etc.]+[about_NP] ?think, *ponder, *contemplate, *reflect +[NP/what/sth/etc]+[of_NP] ?judge, *reflect, *ponder, *contemplate +[NP]+[on/upon_NP] *think, *reflect, *ponder, *?contemplate, judge = ‘assess’ +[Q/S]+[on_NP] *?judge, *consider, *think, *ponder, *contemplate TOPIC +[on_X] judge = ‘assess’ +[about_NP] *contemplate, *judge +[of_NP] *reflect,*ponder, *contemplate, judge = ‘assess’ +[upon_X] *consider, *judge +[over_NP] *reflect, *contemplate, *judge, think = phr. verb EFFECTED +[that_CL] all +[Q/S] ?contemplate

‘think’ +[NP] reflect = ‘mirror’, consider = ‘consideration’, judge =‘assess’ +[wh-CL] judge= ‘assess’ +[wh-to_INF] *reflect +[V-ing], *think, *?judge, consider = ‘ consideration’, reflect = ‘mirror’ +[to_INF] *consider, *ponder, *contemplate, *reflect +[so/not/otherwise] ?consider,*reflect,*ponder ÆFFECTED + PREDICATIVE +[NP]+[as_NP] think = ‘imagine’, reflect = ‘mirror’, (*ponder) +[NP]+[NP] / +[NP]+[AdjP] / +[NP]+[to_INF] *contemplate, (*ponder, *reflect) +[NP]+[as_AdjP] *think, *contemplate, (*ponder, *reflect) +[of_NP]+[as_NP/AdjP] *consider,*judge,*contemplate,(*ponder, *reflect) Phenomena identified in the analysis 149

PREDICATIVE +[NP]+[as_NP] all +[NP]+[NP] *respect, *admire ‘think highly’ ‘think

AFFECTED +[NP] / +[V-ing] / +[NP V-ing] all +[to_INF] *?tolerate

‘tolerate’ (+[with_NP] *tolerate, *endure) AIM +[to_INF] / +[for_NP] all +[NP]+[to_INF] / +[NP]+[for_NP] / +[NP]+[into_NP/V-ing] all +[that_CL] *?push

‘urge’ +[QUOTE] *push +[NP]+[QUOTE] *press,*push +[upon_NP]+[that_CL] *?press, *push

This argument obviously rests on the question of whether she, the keys, and him/to him in sentences such as (1a) and (1b) really represent the same semantic participants, a question which deserves a more detailed discus- sion.

(1a)INV She gave him the keys. (1b)INV She gave the keys to him.

Emons (1978: 30) describes such sentences as “bedeutungsgleich” (‘seman- tically identical’), and also Wasow and Arnold (2003: 120) argue with re- spect to a similar example that “[t]here is little, if any semantic difference …; they certainly do not differ in truth conditions.” Yet this question has also been answered differently in other frameworks. Ramchand (2008: 7–8) summarizes three strategies to deal with this discrepancy:

[1] Linking principles are not deterministic, i.e. exceptions do not pose a major problem (cf. Dowty 199021; Bresnan 2001). [2] The two sentences are derivations of a “common underlying configu- ration” (Larson 1988) [cf. also Chomsky 1975: 492–495]. [3] The participant roles in the two sentences are simply not identical.

21. Ramchand (2008: 7) is most likely referring to Dowty (1991) in this context. 150 Pattern choice and verb meaning

Claim [1] does not really conflict with the valency-theoretic approach pre- sented here, as it emphasizes that although there are obviously certain re- gularities as regards the use of semantically similar verbs, they do not hold 100%. The analysis of the 22 verb groups clearly shows that if there were general linking principles, they would not be deterministic as too many exceptions remain which cannot satisfactorily be accounted for on the basis of meaning. Consequently this line of thinking calls for a means of explain- ing the acceptability of a verb in a specific pattern other than general link- ing principles. If this is the case, then the question arises as to what extent such principles are necessary for creating the link in the first place, or whether the same result can be achieved by other means, i.e. grammatical information being stored in the form of valency properties within each lexi- cal entry. If principles are taken to be generalizations based on a frequent co-occurrence of certain structures with certain meanings, this does not contradict the idea of storage: “regular” combinations are supported by the generalizations, which could clearly facilitate storage. Langacker (1987: 59) describes this phenomenon as entrenchment so that “every use of a structure has a positive impact in its degree of entrenchment, whereas ex- tended periods of disuse have a negative impact”. The claim that these sentences go back to one and the same basic struc- ture (claim [2]), appears to be rooted in the transformationalist tradition. It assumes operations such as derivations of an underlying structure. In this context a surface-oriented approach is preferred. If “common underlying configuration” is interpreted in a purely semantic way, however – i.e. in the sense of participant pattern – claim [2] is analogous to the current interpre- tation of the data, although this does of course not provide an explanation, why this is the case (cf. Wasow 2002: 116–131 and 61–82). Claim [3] is obviously problematic since all the restrictions listed in Ta- ble 38 could simply be done away with by claiming that the complement types express different participants, even if the verb in general allows very similar participants. According to Halliday (1970: 164) “[t]he members of such a pair have the same ideational meaning but differ in information”. Thus, with respect to examples (1a) and (1b) above, [to_NP]_BENEFICIARY is associated with the function ‘new’, the other form with the function ‘giv- en’. Similarly, Pinker (1989) takes the difference to be one of perspective, i.e. whether the focus is on him or the keys (see also Groefsema 2001: 527). Such differences in meaning cannot be accounted for by means of partici- pant roles, since whether they are given or new, their relation to the verb remains the same, and in that sense they do not contradict the analysis pre- sented here. Moreover, such differences can be captured by other means, Phenomena identified in the analysis 151 such as, for instance, an additional level of clausal roles (Herbst and Schüller 2008: 158–163) (cf. 4.1). Construction grammarians typically argue that a change in form is ne- cessarily accompanied by a change in meaning (Principle of No Syn- onymy) (e.g. Goldberg 1995: 67; Wulff 2008: 189–190). This is a general claim which follows from the stipulation that meaning is inseparably tied to form. This is evident in the definition of constructions as form-and-mean- ing-pairs. In that sense, the examples listed above must differ semantically as a matter of principle. Thus the pattern [SCU]+ verb +[NP]+[to_NP] is sometimes also analyzed as an instance of the caused-motion-construction with [to_NP] representing GOAL and not BENEFICIARY (e.g. Goldberg 1995: 33–35; Gries and Stefanowitsch 2004). This way [NP] and [to_NP] are not realizations of the same participant. However, the semantic similarity be- tween the two patterns is obvious and there is a huge body of publications (compare also the discussion above) on the two serving as alternatives, which is why an analysis as the same participant pattern with possible dif- ferences in the information structure is at least as legitimate. Moreover, only one of the two – either BENEFICIARY or GOAL – can be realized in the same sentence, i.e. they exclude each other. This would favor an analysis of BENEFICIARY and GOAL as subtypes of one and the same superordinate par- ticipant. Construction grammarians such as Goldberg (1995, 2006a) hold that the licensing of a verb in a specific construction depends on the meaning of the verb as it rests on a successful fusion of the verb’s semantic participants and the construction’s arguments. The similarity between the parallel pat- terns listed above is undoubtedly very high since both the general verb meanings as well as the participants relevant to the scene are very similar. Other construction grammarians see the difference not so much in a differ- ent choice of participants but also rather “in terms of their information- structural properties” (Stefanowitsch 2008: 517). Gries (2003: 9), for in- stance, lists as cues for a prototypical realization of +[NP]+[NP] that the RECIPIENT is more likely to be animate or human, short, pronominal and new in terms of information structure and the AFFECTED participant inani- mate or non-human, long, lexical, and old, and vice versa for +[NP] +[to_ NP]. Concerning the question of whether there are truly alternative construc- tions, i.e. formally differing patterns with the same meaning, Croft (2001: 127) argues that changes of the functions of a pattern or adoptions of new functions can be regarded as analogous to the process of grammaticaliza- tion. If a “new” construction comes to be used for a function which has 152 Pattern choice and verb meaning previously been expressed by another construction, “the new construction at least partially imposes the conceptualization of its original structure and function. This is likely to be true (if at all) at the initial stage of the process, when the new construction is not yet conventionalized” (Croft 2001: 127). Thus according to this line of thinking, a semantic difference associated with different formal realizations does not necessarily need to remain con- stant, which makes alternative constructions at least possible. According to Croft (2001: 127) the “old construction” is then lost or marginalized in the function, leading to a conventionalization of the new construction and con- sequently polysemy. At the same time, it does not seem to be the case that the old construction is necessarily always lost (Hopper and Closs Traugott 2003: 124). What is more, the different preference or choice of a pattern does not appear to follow any obvious scheme in the sense that each type of different choice of pattern is accompanied by a specific type of meaning difference. Hunston and Francis conclude in the Pattern Grammar (cf. 4.2.2) that “two utterances with different patterns may exemplify the same roles, though in a different order” (2000: 125). While a difference in word order could in- deed indicate a difference in the information structure of the sentence (i.e. the stress a participant receives in a specific construal), word order is not even affected in a large number of the cases identified. Accordingly, phe- nomena such as these have also been taken as an argument against a direct correspondence between meaning and form. Helbig (1992: 15), for in- stance, concludes in reference to examples such as He considers the book good and He considers that the book is good, that semantic and syntactic valency do not necessarily correspond. Moreover, even if there were se- mantic differences, it would be necessary to relate them systematically to the syntactic differences between the constructions. This does not seem to be possible in a principled way, either. In the field of foreign language learning and teaching, the use of par- ticles is known to be a common source of mistakes, on the one hand, be- cause there are obvious cross-linguistic differences, e.g. between English and German at the moment vs. im Moment (‘in the moment’) or compare sth to sth else as well as compare sth with sth else which can only be trans- lated as etwas mit etwas anderem vergleichen (‘compare something with something else’), which makes a direct transfer difficult. On the other hand, the meaning of the particle which serves as complement does not seem to provide a clue transparent enough to allow a prediction as to which combi- nation is possible and which not (cf. chapter 2). The idiosyncrasies to be found here are also the reason why in many linguistic frameworks which Phenomena identified in the analysis 153 try to relate semantic content with grammatical form particles do not play a very central role in the first place. They are often excluded from the claim for predictability, e.g. by constituting a difference between direct and indi- rect arguments (e.g. Marantz 1984), where direct arguments are noun phrase arguments which receive their semantic role directly from the verb and indirect arguments in English “are governed by a preposition ... [and] receive their thematic roles from the preposition ... or the preposition ... in conjunction with the verb” (Tenny 1994: 9). If the choice of particle is commented upon, this is usually done in connection with alternations such as the so-called dative alternation (2), the benefactive alternation (3), the locative alternation (4) or alternations concerned with more specific par- ticles such as the so-called with/against alternation (5):

(2a)QE Bill sold a car to Tom (2b)QE Bill sold Tom a car. (Levin 1993: 46) (3a)QE Martha carved a toy for the baby. (3b)QE Martha carved the baby a toy. (Levin 1993: 49) (4a)QE Jack sprayed paint on the wall. (locative variant) (4b)QE Jack sprayed the wall with paint. (with-variant) (Levin 1993: 51) (5a)QE Brian hit the stick against the fence. (5b)QE Brian hit the fence with the stick. (Levin 1993: 67)

The particle in sentence (2a) is often taken as having a function similar to the expression of morphological case in case languages, i.e. the particle phrase is seen as the realization of a participant of the verb (e.g. Jolly 1991: 104). In (4) and (5) there appears to be a bigger contribution of the seman- tics of the particle to the whole clause, while (3a) seems to take an interme- diate state between (2) and (4)/(5) (e.g. Jolly 1991: 107). Possible alterna- tive realizations of the same complement slot (cf. 2.2) are rarely com- mented upon or are simply accepted as idiosyncratic, although this se- riously calls into question the idea that meaning is responsible for gram- matical form. Even if Rohdenburg (2003: 206) argues that such gram- maticalized uses of prepositions “still preserve a variety of subtle semantic constrasts”, the discussion in chapter 2 has already shown that it is unlikely that this can serve as a basis for predictions of pattern choices. Accor- dingly, particle complements are usually regarded as lexically specified, i.e. other than non-particle complements, they are analyzed analogous to the valency theoretic approach to complementation where all complement types are stored in the lexical entry (e.g. Herbst 2009; Klotz 2007, 2000; Heringer 1967). Yet the status of particle phrases as complements is not uncontroversial in valency approaches, either, as pointed out in chapter 2, 154 Pattern choice and verb meaning e.g. Welke (2005: 56–57), who considers prepositional phrases as less typi- cal arguments because they are “autonomously coded”, i.e. the semantic relation of such phrases to the rest of the sentence can be understood by looking at the phrase itself. However, the meanings of particles are not necessarily stable and the choice is not always predictable either, which is why the approach adopted in VDE is taken as the basis of this analysis. The fact that particles are often regarded as lexically specified, even in theories which presuppose a semantically regular correspondence between verb and pattern choice, is remarkable. Thus Jolly (1993: 275) comments:

A more interesting and systematic analysis of English would be one in which prepositions did not have to be idiosyncratically specified with each verbal lexical entry but rather would be predicted by some general prin- ciples related to the semantics of verbs and the assignment of arguments to morphosyntactic positions within the clause.

Also Jackendoff (1972) points at the problem without providing a solution in his own framework:

The other mechanism needed is a consistent way of handling prepositions … More specifically, one would sometimes like to avoid stating the per- mitted preposition in the subcategorization of the verb, allowing the mean- ing of the prepositions itself to specify whether the phrase in question is a Source or a Goal or an Instrument ... The formalism I have given does not have a way to take into account the meaning of the preposition. (Jackendoff 1972: 42)

Note that while Jolly (1993) prefers a solution based on the meaning of the verb, Jackendoff (1972) puts emphasis on the meaning of the particle and its possible combination with the meaning of the verb. In both cases par- ticles are seen as part of the subcategorization frame of the verb which, according to Jolly (1991), applies to the major theories concerned with the problem of linking, such as Transformational Grammar, Fillmore’s , Bresnan’s Lexical Functional Grammar, and Jackendoff’s Se- mantic Theory (Jolly 1991: 15). Also Pinker (1989: 95) claims that “most of the language-particular properties of prepositions and oblique case markers can be factored out of any linking rule for oblique arguments and localized in the lexical entries for those individual morphemes”. As has been pointed out above, this is very much in line with valency theoretic assumptions, which take the type of complement to be associated with a verb as a valency property of that verb, irrespective of whether it is a noun phrase, a clausal or a particle complement and without having to refer to Phenomena identified in the analysis 155 such occurrences as exceptions in the first place (cf. for instance Helbig 1971: 35; Helbig and Schenkel 21973: 34; Emons 1978: 23 and 28; Heringer 21972: 68). Excluding particle complements means excluding a considera- ble amount of a verb’s patterns which obviously weakens the explanatory force of such a theory. The idiosyncratic component is simply transferred from verb complementation to the lexical item itself. The inclusion of par- ticle and clausal complement types in the analysis results in a more ade- quate picture of actual restrictions when it comes to verb complementation. Thus such cases present a valid point in this discussion. In what follows, different participants identified in the study as showing variation in their realization within one single group will be discussed sepa- rately, starting with the above mentioned variation in the expression of the participant BENREC.

3.3.4.3 Different formal realizations of BENREC

This participant can typically be realized in three different ways, either as +[NP] or as +[to_NP] as has already been shown with respect to examples (1a) and (1b) above or as [for_NP] (see below). While some verbs accept two alternatives, others only accept one of the two. In the sample there are several verb groups which show a different distribution within the same group. One third of all restrictions in the ‘show’-group can be accounted for by the fact that show can express BENREC as a post-verbal noun phrase while demonstrate and indicate usually need a to-particle phrase instead. Interes- tingly, one example of indicate in the pattern [SCU]+ verb +[NP]_BENREC +[wh-CL]_ÆFFECTED was found in the BNC, given here as (6). Obviosuly, the use with the particle phrase appears to be much more frequent.

? ? (1)VDE Anne asked him to show/* demonstrate/* indicate her BENREC the type- script ÆFFECTED. (2)VDE The things that indicated success ÆFFECTED to them BENREC were: be- ing satisfied with life (80 percent); having a good marriage (73 per cent); being in control of life (72 per cent) (3)BNC This operation ÆFFECTED needs to be explained and demonstrated to every student BENREC. A0H 376 (4)VDE If you have childhood photographs, look at them. Show them ÆFFECTED to someone else BENREC. (5)VDE She didn’t want to show/*demonstrate me BENREC how sad she was ÆFFECTED. 156 Pattern choice and verb meaning

(6)BNC To be of maximum value to the user, an index should be included, to indicate the reader BENREC where each grammatical structure and pho- nological feature is discussed… ÆFFECTED H0J 1504

Groefsema (2001) argues that verbs which only allow the double object pattern ([SCU]+ verb +[NP]+[NP]) cannot present the entity which changes possession (ÆFFECTED in our description) as actually affected, e.g. in the case of ask “… John does not have a particular effect on the question by asking it, and so the verb does not appear with the [NP_NP to_NP] form” (Groefsema 2001: 544). Yet in the case of show and indicate the object which changes possession is basically the image of something in both cases. There is clearly no such difference that could explain the restriction. As pointed out above, such a difference can be expressed by means of clausal roles which are tied to the sentence structure. Thus an [NP] comple- ment directly following the predicator receives a special kind of focus, irrespective of the verb meaning, which can be described by the clausal role “FOCUSSED AFFECTEDNESS”, just as an [NP] functioning as subject receives an “AGENTIVITY” role, even if the participant role is strictly speaking not that of an AGENT (Herbst and Schüller 2008: 160–163) (cf. 4.1). In the ‘teach’-group, only teach allows the two alternative realizations of BENREC as +[NP] and +[to_NP] as illustrated in (7) and (8), while the choice strongly depends on the form ÆFFECTED takes: teach, train, and educate (but not instruct) allow +[NP]_BENREC +[to_INF]_ÆFFECTED, teach and instruct (but neither educate nor train) allow +[NP]_BENREC +[wh-to_INF]_ÆFFECTED as in examples (9) and (10), and finally teach (11), instruct (12), and educate (13) but not train allow +[NP]_BENREC +[that_CL]_ÆFFECTED.

(7)VDE This experience taught/*instructed/*educated/*trained me a valuable lesson in legal humility. ? (8)VDE In Greek mythology, Chiron, the centaur who taught/* instructed/ *educated/*trained medicine to Asklepios, suffered an incurable wound at the hands of Hercules. (9)VDE Your family doctor or the gynaecologist will instruct you how to inter- pret and use this data. (10)VDE Richard taught/*?trained/*educated me how to cook. (11)VDE This early experience taught/*trained me that people do want their boss to appraise them. (12)VDE The conservative establishment of scholars and educators have instruct- ed the British people that great poetry is universal poetry. Phenomena identified in the analysis 157

(13)VDE The committee sees the centenary as a good opportunity to educate the British that Sikhs have not just landed on their doorsteps but have a long history as allies.

While get, buy, purchase, and obtain can be used in the construction [SCU]+ ‘get’ +[NP]_BENREC +[NP]_ÆFFECTED, this is not possible with acquire, although it has the respective participants as illustrated in (22a). Note that obtain also appears to be rather rare in this construction: Goldberg (2006b: 405) lists it as one of the verbs which cannot be complemented with two noun phrases and which serves as one example where Gropen et al.’s (1989) morphophonological rules apply, according to which verbs containing certain Latinate affixes resist this form of complementation. However, example (21) given below shows that speaking of rules in this sense seems to be misguided, even if only few examples of this kind can be found.

(14)VDE Either of you ready for a refill while I get a drink for myself. (15)VDE I should love to get our children a proper Wendy house. (16)VDE At Christmas, you always buy presents for other people. (17a)VDE His family have bought him a new hi-fi for Christmas. (18)BNC Land gives power, hence Sher’s identification of the need for credit, mainly to purchase land for the community. ALE 289 (19)BNC “Then you can make it up by purchasing me a change of clothes.”HWN 4223 (20)BNC … five years later a council in the Kamiah valley resolved to obtain firearms for the tribe. ALX 72 (21)BNC “I believe,” he said, “that I can obtain you a position as reader in a publishing house in which I have a certain interest.” CD2 1516 (22a)BNC We have also acquired for the Society a vast range of photo-copied material, which is also open for viewing. HHN 78 (17b)NSA *His family have acquired him a new hi-fi for Christmas. (22b)NSA *We have also acquired the Society a vast range of photo-copied ma- terial, which is also open for viewing. Note that receive cannot occur in this construction, either, although it has both BENREC and ÆFFECTED as part of its participant inventory. However, receive does not have an AGENT which is not at the same time also the RE- CIPIENT of ÆFFECTED.

(23)BNC In 1982-83, 40 per cent of divorced wome (who had not remarried) were receiveing some maintenance; 23 pre cent reived it for their children. AP5 723 (24)NSA *23 per cent of divorced women received them maintenance. 158 Pattern choice and verb meaning

According to VDE, all three verbs rent, hire, and lease can be comple- mented with two post-verbal noun phrases to express the participants BEN- REC and ÆFFECTED. However, lease was not unanimously accepted in this pattern by native speakers, although it cleary has a participant BENREC as can be seen in sentence (29). Thus, while one cannot really speak of a re- striction, there is a clear difference in preference. Rent and lease can also be complemented with [to_NP] to express BENREC both in in divalent as well as in trivalent patterns. Hire was found acceptable in the trivalent version but native speakers did not agree with respect to the acceptability of the divalent pattern.

(25)VDE Then he found the nearest car rental company, rented himself the smal- lest car he could and drove towards Washington. (26)BNC He also hired himself a cook-housekeeper-secretary, Cathy Allen, the wife of an actor friend. C9U 829 (27)VDE Perhaps La Prade could lease him a few acres. (28a)NSA ?He leased himself a car. (28b)NSA ?He leased his daughter a car. (29)BNC The new rector appointed in his place had leased a house to the plaintiff Philips, who had been evicted by Bury. FE3 203

According to Gries and Stefanowitsch (2004), a semantic difference be- tween those verbs typically occurring in the constructions [SCU]_AGENT + verb +[NP]_BENREC +[NP]_ÆFFECTED and [SCU]_AGENT + verb +[NP]_ ÆFFECTED +[to_NP]_BENREC can be made out in the sense that the former express proximal transfer and the latter transfer over a greater conceptual distance (Gries and Stefanowitsch 2004: 105–106). Yet the examples they provide are to a certain degree again a matter of subjective assessment and cannot serve as error-free proof of this theory. While pass, for instance, is classified as indicating distal transfer in accordance with its preference to occur in the +[NP]+[to_NP] pattern by Gries and Stefanowitsch (2004: 105–106), Groefsema (2001: 537) emphasizes that pass demands physical presence. This again shows how problematic such subsequent semantic classifications can actually be. Wasow and Arnold (2003: 133) conclude in a comparison of alternating verbs in search of a correlation between se- mantic difference and preference to occur in one of the two constructions that “these differences are subtle and seem unlikely to account for all the lexical variation in ordering preferences.” As discussed in the preceding chapter, semantic differences between the constructions as well as the verbs occurring in them can be identified on a case to case basis but are not sys- Phenomena identified in the analysis 159 tematic enough to allow for a generalization and consequently a prediction of the patterns the verbs can occur in.

3.3.4.4 Different formal realizations of PREDICATIVE

The following examples illustrate cases where verbs allow alternative valency patterns with and without the particle as, i.e. verbs which demand to be complemented with an [as_X]_PREDICATIVE and verbs which must be complemented with bare adjective or noun phrases. While elect and appoint can express the participant pattern AGENT + ‘choose’ + AFFECTED + PREDICATIVE with both patterns, choose is clearly confined to the [SCU]+ ‘choose’ +[NP]_AFFECTED+[as_NP]_PREDICATIVE construction as illustrated in (5), since the other pattern can only be used to express a different construction, i.e. [SCU]+ ‘choose’ +[NP]_BENREC +[NP]_AFFECTED (6a). According to Algeo (2006: 226), appoint is unlikely to be used in the pattern without as in American English. Such a regional difference of course also emphasizes the role conventionalization plays for complementation.

(1)VDE The Nigerian Bar Association has criticised President Babangida for appointing a military man as vice-president in a cabinet reshuffle last week. (2a)VDE An opposition leader in Zaire has confirmed that he has decided to de- cline an offer by President Mobutu to appoint him Prime Minister. (3)VDE That was the official announcement late last night that Labour’s Peter Kilfoyle had been elected as Member of Parliament for the Liverpool- Walton constituency. (4)VDE His supporters elected him president in June. (5)VDE Why choose weight training AFFECTED as the activity upon which to base an exercise programme PREDICATIVE? (2b)NSA *An opposition leader in Zaire has confirmed that he has decided to decline an offer by President Mobutu to choose him Prime Minister. (6a)NSA He chose her RECIPIENT a present AFFECTED.

With respect to select, native speakers did not agree on whether a sentence such as (8) is possible or not. Similarly, there was no conclusive assessment with respect to nominate and designate in either pattern, even if semanti- cally such a scenario is by all means plausible.

(7)BNC A survey of graduate employers has selected Cambridg AFFECTED as the University producing the most employable graduates PREDICATIVE. BPE 20 (8)NSA ?They selected him AFFECTED their leader PREDICATIVE. 160 Pattern choice and verb meaning

(6b)NSA He selected her RECIPIENT a present AFFECTED. (2c)NSA ?An opposition leader in Zaire has confirmed that he has decided to decline an offer by President Mobutu to designate/nominate him (as) Prime Minister.

Name can also be found in both patterns but the meaning ‘choose’ is only expressed with the pattern containing the particle (9). Similarly, pick only seems to allow the particle pattern when it is employed to express ‘choose’ (11a):

(9)VDE Shortly before the end of last season Barnet became the first team to name a black man AFFECTED as their manager PREDICATIVE. (10)VDE In 1906 Tommy Sopwith (then only 18 years old) and Philip Paddon, bought a rather untidy secondhand balloon which AFFECTED they named Padsop PREDICATIVE. = ‘call’ (11a)VDE And I shouldn’t pick him AFFECTED as the murderer PREDICATIVE, even though that reduces by one the number of those in the line-up. (11b)NSA *They picked him the murderer. (12)VDE They want to make trouble, asking the boys to pick them RECIPIENT flowers AFFECTED.= ‘…flowers’

Gries, Hampe, and Schönefeld (2005: 641), in a comparison of the as-pre- dicative and the construction with two noun phrases, suggest that “con- structions with unmarked object complements can indeed realize a wider range of meanings than as-predicatives”. To them, the constructional mean- ing of the as-predicative construction boils down to the basic sense of re- garding or seeing or (metaphorically extended) understanding someone or something as something (Gries, Hampe, and Schönefeld 2005: 651). This assumption is based on the fact that regard, describe, and see have the highest collostructional strength, i.e. their association with this construction as opposed to other constructions and as opposed to other verbs is the high- est (cf. Stefanowitsch and Gries 2003: 214). Unmarked object con- structions, on the other hand, are regarded as resultative. In case of appoint and nominate, however, they argue (in a footnote):

The scenarios denoted [appoint, nominate, adopt, and establish] may impli- cate a partial change on the side of the object referent, and thus comprise a truly resultative dimension, which can be captured by paraphrases with ‘make’: ‘make so. their agent’, ‘make so. the leader of the party’, ‘make the county a unit of mapping’. This resultative aspect, however, is downplayed by the verbs appearing in the as-predicative, which emphasize the partial and temporary nature of a change of role and status. (Gries, Hampe and Schönefeld 2005: 672) [emphasis mine] Phenomena identified in the analysis 161

Arguing that the meaning of an actual sentence deviates from the assumed construction meaning and at the same time claiming that the meaning is nonetheless influenced by this assumed construction meaning (making it only a partial or temporary result) appears to be rather circular. Further- more this line of reasoning is necessitated exclusively by the assumption that the construction must have a fixed meaning which differs from the alternative pattern with two bare noun phrases. At the same time it does not seem to be possible to explain the different acceptability on the basis of differences between the respective verbs. Similar restrictions can also be found among other groups examined here. Despite the fact that declare, pronounce, announce, proclaim, and state all have the relevant participants, they do not have the same alter- native valency patterns to express them. Declare, announce, proclaim, and state were identified in all four patterns respectively (illustrated with pronounce below), even if announce appears to be rather rare in the pattern with the particle. State shows a clear preference for [SCU]+ ‘declare’ +[NP]+[AdjP] and [SCU]+ ‘declare’ +[NP]+[as_NP]. Pronounce was not considered acceptable in the latter pattern as illustrated in (15b).

(13)VDE Anthony Spick pronounced the takeover as good news; other dealers were scandalised. (14)VDE I collected the material upon my return, and the training course was pronounced a great success. (15a)BNC We were thus free to take the Cup to lunch, where, feeling better, the former rugby hooker, Jeffrey Bernard, pronounced himself happy that “if England can't have the Cup, then at least it's good that the Australi- ans have it ...”AKE 947 (15b)NSA *…, the former rugby hooker, Jeffrey Bernard, pronounced himself as happy that “if England can't have the Cup, then at least it's good that the Australians have it …”.

Strictly speaking, the pattern exemplified by (15a) differs from the one in (13) and (14) in that the noun phrase is realized by a . If such differences are taken into account – and there are several differences in this group with respect to this aspect – the number of idiosyncrasies must be considered even higher. With respect to the participant pattern AGENT + ‘show’ + AFFECTED + PREDICATIVE, demonstrate allows neither complementation with a bare adjective phrase nor with an adjective or noun phrase embedded in an as- particle phrase, while both of these patterns are acceptable with show as exemplified in (16), (19), and (20) below. 162 Pattern choice and verb meaning

(16)VDE The picture shows him as a man in his thirties with a receding hairline. (17)VDE The marks on his face and neck indicated strangulation as a possible cause of death. (18)BNC But the nineteenth century demonstrates it as a scientific fact. CCE 1063 (19)NSA The latest picture shows/?indicates/*demonstrates him as beardless and without glasses, looking quite different to the earlier pictures. (20)VDE The latest picture shows/*indicates/*demonstrates him beardless and without glasses, looking quite different to the earlier pictures. [without glasses is another complement of show which is coordinated with beard- less] (21)BNC He saw 2 cars in front of him and said one car, which was driven by Anthony Gallagher, had indicated left but pulled out in front of him. K21 2722 = ‘signal direction’

Indicate can be found with a post-verbal adjective phrase only in a very limited type of context (21). This justifies an analysis as a different lexical unit, i.e. indicate ‘signal direction’. All three verbs in the ‘think highly’-group can be found in the valency constructions [SCU]+ ‘think highly’+[NP]_ÆFFECTED+[as_NP]_PREDICA- TIVE but only esteem can occur in a pattern without the particle (25), even if the pattern appears to be rather rare.

(22)VDE I still respected him as a friend. (23)BNC Rotha Lintorn Orman admired Mussolini as a man who had dealt firmly with the socialist menace. CS6 364 (24)BNC … those who had already committed themselves, together with those who esteemed him as a scholar, voted for E. K. Chambers. A7C 1149 (25)BNC There was scarcely a divinity student in Cambridge, says a contempo- rary, who “made not himself a disciple of Mr Andrewes by resorting to his lectures and transcribing his notes, and ever since they have in many hundreds of copies passed from hand to hand and have been esteemed a very library to young divines”. CFF 455 (26)NSA *They admired/respected him a great man.

The verb guess has a PREDICATIVE participant which is typically realized in a valency construction which has not been taken into consideration so far. The pattern +[NP (]+[) to_INF] expresses the participant ÆFFECTED which incorporates a PREDICATIVE element in the form of a to-infinitve clause as illustrated in (27). A realization as [AdjP] – the prototypical realization of this participant in combination with assume, suppose, and presume – was rated unacceptable (28).

(27)VDE He guessed her age to be no more than twenty-two or three. Phenomena identified in the analysis 163

(28)NSA *He guessed her older than she actually was. (29)VDE One could assume the prototype for humanity to be female rather than male. (30)VDE Bail is extremely difficult to obtain and the accused is assumed guilty until proven innocent.

Note that the realization of PREDICATIVE was also considered acceptable by native speakers in the following form: He guessed her age as no more than twenty-three. Also the ‘think’-group exhibits variation with respect to the realization of this participant. While consider and judge appear to have identical pat- tern inventories with respect to the participant pattern AGENT + ‘think’ + ÆFFECTED + PREDICATIVE (cf. also Herbst 2009), both only being restricted with respect to [SCU]+ ‘think’ +[of_NP]+[as_NP/AdjP], think and contem- plate are less flexible. Think normally seems to need the particle of when the particle as is used, e.g. in (32a) below. CGEL analyzes such sequences as prepositional verbs (16.3–16.8). This has the clear advantage that this apparent irregularity is resolved. Yet while this reduces complexity in the description of the complementation patterns of this group, it increases the number of verbs (cf. also VDE: xxvi). Thus while both approaches are legi- timate in themselves, the irregularity cannot be avoided. Opting for a pre- positional verb think of sets the challenge to explaining why we do not have analogous forms such as consider of or judge of. Even if an example could be identified in the BNC for [SCU]+ think +[NP]+[as_NP], this pattern appears to have a different meaning, i.e. ‘im- agine’ (38). Contemplate occurs in only one of these patterns (44). Accord- ing to Algeo (2006: 226), the pattern illustrated by example sentence (37) is unlikely in American English, which again points out the possible regional variability with respect to complementation.

[SCU]+ ‘think’ +[NP]+[AdjP] vs. +[NP]+[as_AdjP] vs. +[of_NP]+[as_AdjP]

(31a)VDE I thought him charming. (31b)NSA *I thought him as charming. (32a)VDE “Many people think of angels as benign, pleasant and helping,” says University of Wisconsin psychiatrist Richard Thurrell. (33)VDE Mr Advani’s enemies consider him dangerous. (34)VDE They are most appropriately considered as helpful rather than a hin- drance. (32b)NSA *“Many people consider of angels as benign, pleasant and helping,” says University of Wisconsin psychiatrist Richard Thurrell. (35)VDE I judged it safe to proceed. 164 Pattern choice and verb meaning

(36)VDE While his supporters hail his work as ingenious, novel, witty, avant- garde and refreshing, his critics judge it as excessive, crude and super- ficial. (32c)NSA *“Many people judge of angels as benign, pleasant and helping,” says University of Wisconsin psychiatrist Richard Thurrell.

[SCU]+ ‘think’ +[NP]+[NP] vs. +[NP]+[as_NP] vs. +[of_NP]+[as_NP]

(37)VDE Those who think him a poet rather than a philosopher do so because of his styles. (38)BNC Sometimes I think it as a transcendent sound which speaks of unknown powers, of cosmic storms and sun winds sighing in the brain. B1F 541 =imagine (39a)VDE One always thinks of George Orwell as a great polemicist. (40)VDE The telephone has come to be considered a necessity, but is this in fact true? (41)VDE They simply considered him as a promising player. (42)VDE Any success it might have runs the risk of being judged a merely formal solution. (43)VDE What he once took as boldness he will now judge as foolhardiness. (44)BNC The test is not what the defendants contemplated as a likely or even an inevitable consequence of their conduct; ... FSS 1282 (39b)NSA *One always considers/judges/contemplates of George Orwell as a great polemicist.

[SCU]+ ‘think’ +[NP to_INF]

(45)BNC They report that nationally 70 per cent of the population thought/ *contemplated Woonerven to be attractive or highly attractive. C8F 323 (46)VDE Under Chinese law if a drug-related offence is judged to be particularly serious the defendant can be sentenced to death. (47)VDE Until the late eighteenth and early nineteenth centuries insanity was considered to be largely incurable.

If verbs like regard and deem are also taken into consideration, the emerg- ing picture is even more complex.

(48a)NSA ?Many people consider angels as benign, pleasant and helping. (49)VDE They are most appropriately considered as helpful rather than a hin- drance. (48b)NSA Many people regard angels as benign, pleasant and helping. (32c)NSA *Many people think angels as benign, pleasant and helping. (50)BNC Now that we have considered the type of information which the law deems as illegitimate for stock market trading, let us move on and look Phenomena identified in the analysis 165

at some of the circumstances when trading on the basis of such infor- mation is illegal. ECD 1187 (32d)VDE Many people think/*consider/?regard/*deem of angels as benign, plea- sant and helping. (51)VDE I began to seriously consider that she should not be attending school this year at all. (32e)NSA Many people think/*regard/?deem that angels are benign, pleasant and helping. (52)BNC Then his father returned to England, sought out his son and deemed that he, too, should study law. F9U 1090 (32f)NSA Many people consider/regard/think angels benign, pleasant and help- ing. (53)BNC We have usually deemed it strategically unwise not to employ a judi- cious mix of the two.ABA 63 (32g)NSA Many people deem angels benign, pleasant and helping.

The pattern illustrated in (49) might also be acceptable in the active voice, although the use without the particle seems to be strongly preferred by some native speakers.

3.3.4.5 Different formal realizations of TOPIC

There are four groups which exhibit a variation in the realization of the participant TOPIC, illustrated in Table 39 below. In the case of ‘quarrel’, the formal realization of TOPIC is also tied to the rest of the pattern, i.e. whether the complement +[with_NP] is realized or not. This strongly emphasizes the role of the pattern as opposed to the sin- gle participants or complements of a verb. One possible meaning distinc- tion in this context is a difference in formality. It has been argued that +[on_X] is reserved for a rather formal TOPIC and is thus not a possible realization in combination with quarrel (cf. CGEL: 9.60). Yet the pattern was judged acceptable in combination with a CO-AGENT. Moreover, for- mality does not seem to be a general feature of +[on_X] which again rend- ers any explanation based on this argument rather weak.

(1a)VDE I beg you, do not argue with me on this. (1b)NSA I beg you, do not quarrel with me on this.

While dispute is neither found in the pattern illustrated in (1), nor in com- bination with [upon_NP] or [of_NP], bicker only seems to be restricted with respect to +[on_NP]. 166 Pattern choice and verb meaning ) +/–[with_NP] +/–[with_NP] +/–[with_NP] consider quarrel reminisce argue bicker dispute ( ponder .] .] .] also with ) also with what/sth/etc what/sth/etc consider ( +[NP/ +[NP/ think think reflect ponder contemplate

also with +[NP] also with +[NP] ) +[with_NP] only with +[Q/S] +[with_NP] including possible further conditions the of for the use pattern (+/–) consider quarrel ( argue think reflect instruct educate TOPIC ] ] .etc] what/sth what/sth/etc. what/sth./etc. only with +[ only with also with +[ +/–[with_NP] +/–[with_NP] only with +[ +/–[with_NP] only in trivalent pattern only in trivalent argue bicker dispute mind in bear think reflect ponder quarrel remember reminisce recall consider teach instruct educate train +[about_X] +[on_X] +[upon_X] +[of_X] +[over_X] +[over_X] +[of_X] +[upon_X] +[on_X] +[about_X] Different formal realizations of realizations formal Different  Table 39. 39. Table ‘quarrel’ ‘remember’ ‘think’ ‘teach’ verb- verb- groups Phenomena identified in the analysis 167

Although argue is clearly more formal than both quarrel and bicker, only quarrel was accepted by native speakers in this pattern but not bicker, even if the complementation with [on_NP] in part contradicts the analysis of quarrel in CGEL (cf. discussion in 2.1). Thus it is impossible to predict when formality really “demands” a certain realization. All five verbs in the ‘remember’-group have a participant TOPIC, but they vary as to its formal realization as can be seen in examples (2)–(8) below. If all verbs in the ‘remember’-group are ranked according to their semantic similarity, reminisce is least similar as it highlights the process ra- ther than the result, which is highlighted by the other verbs in the group. Whether this is responsible for its compatibility with [over_NP] – which is not compatible with any of the other verbs in this group – in the sense that the focus on the process is especially compatible with the meaning of over seems to be questionable.

(2)BNC Back at the car park we had a well earned cuppa and reminisced/*re- called/*remembered/*recollected over another hot day back in 1933. ECG 907 (3)VDE “Don’t you remember/*recollect about Anna?” I shook my head. (4)NSA He reminisced about Anna. (5)NSA *He recalled about his maths homework. (6)VDE I remember/?recollect/*reminisce things about her that I found very intelligent.22 (7)BNC I cannot recall anything about Sleetburn. G39 390 (8)BNC One of the basic factors we have to bear in mind about fungi is that, …, fungal attack in one form or another is about and ready to take advan- tage of any weakness for much longer periods …. CMM 1175

The third example of this category can be found in context of the ‘teach’- group. It is possible to teach someone about something (TOPIC) but not on something, although one can instruct and educate both on and about some- thing.

(9)VDE He taught me about enjoying life itself. (10)BNC You will be instructed about essential fire precautions including exits and how to get residents outside. B32 538 (11)VDE We must educate ourselves about our past and work to enhance our prospects for the future.

22. The passage that I found very intelligent is as a postmodifying relative clause part of the first [NP] complement. There might also be a case for analyzing [about_NP] in such a sentence as a postmodification of the first [NP]-comple- ment (cf. 4.1). 168 Pattern choice and verb meaning

(12)VDE Instruct/*teach them on how to make a reverse charge call. (13)VDE Informed Consent is a means to educate yourself on risks, options, com- plications and alternatives to surgery.

In the ‘think’ group, not one single verb in the group exhibits the exact same pattern inventory as any of the others. Five of the six verbs can in one way or another express the meaning ‘think’ + TOPIC (with judge expressing the meaning ‘assess’ in this pattern): contemplate is limited to particle complements including on and upon and reflect, in addition to that, also allows particle complements with about.

(14)BNC I begin to contemplate on how rail travel has been a source of artistic inspiration to passengers for more than a century. ABS 743 (15)BNC PCT … is based upon a concept of the individual as an active meaning- maker and participant in the world, able both to contemplate upon and extend his/her understanding of it. H83 1605 (16)VDE We had both had an opportunity to reflect on the recent turbulent events. (17)VDE As we reflect upon the past, we must address ourselves to the present and the future. (18)BNC Theorists who reflect about the ways in which things might have been different are bound to consider the question of what might have made events take another course. CMN 974

Ponder can be complemented with all four alternative complements. The same applies to think, although the combination with the particle over is only possible in the form of the phrasal verb think over:

(19)BNC I pondered also on the question as to why it matters so much… CES 1457 (20)BNC And all the time I pondered upon the enigma of Mrs Rumney. G3S 948 (21)BNC They would be moving, and he pondered about that move, thinking about the gun. H85 731 (22)BNC I pondered hard over the reply to my friend's question. CD6 1347 (23)BNC She began with praise: -- the more I think on it Ellen the more beholden I am to you for how you have cared for Oreste and the more fortunate. ADS 1241 (24)BNC Our Lady will think upon your problem. AD9 2444 (25)VDE I was just thinking about that. (26)VDE Say you want to think it over and don’t sign anything. = ‘think about it before deciding’

Think can also be complemented with +[of_NP] as in example (27). This pattern is similar to, but not synonymous with, the valency constructions above as it rather expresses EFFECTED than TOPIC. Phenomena identified in the analysis 169

(27)VDE He thought of how he’d kissed her.

If consider is used to express the meaning ‘think’ + TOPIC, the typical reali- zation seems to be with +[about_NP]. It was neither identified in a context with +[upon_NP], nor did native speakers accept a construed sentence with it. For all other constructions, however, exactly one example sentence each could be identified in the BNC, which obviously renders these as rather untypical cases. They were not accepted by native speakers, either, which is why they are here marked by a question mark.

(28)BNC ?This is an important area for school librarians and teachers to con- sider about the design and layout of information held on the microcom- puter in relation to the identification of curriculum-related keywords and the development of information skills. JXK 1015 (29)BNC ?<“telephone rings, phone conversation starts”> hello, hello, right, no, yeah well I, yes I I think it all went great, unfortu- nately erm, so I really can't really consider on, on that. F7N 30 (30)BNC ?Mr. Gillespie wanted time to consider over it, and promised to let the Sec- retary know by the end of the week whether he would accept or not.” FTT 1201

Judge did not occur in any of the divalent patterns expressing a TOPIC, ex- cept in the sense ‘assess’, as for instance in (31), while it was considered acceptable in a trivalent pattern with TOPIC (see below).

(31)VDE John Quayle said an extraordinary judiciary hearing would be called to judge on the incident at 8 o’clock today. = assess

Further irregularity can be found in trivalent patterns containing TOPIC or REFERENCE. Ponder and contemplate do not occur in any of the patterns expressing the combination of the participants ÆFFECTED and TOPIC or REFERENCE (TOPREF), although both verbs have both of these participants. Nonetheless a combination does not seem to be possible as exemplified in sentences (34), (36a and b), (37b), and (38).

(32)BNC All this has been, so to speak, a mere preliminary to the central act of a murder story, the murder, and there are a number of things we should consider about this murder and the murderer. FF0 238 (33)NSA …, there is the issue of what we can judge/?think/*contemplate/ *reflect/*ponder about the functions of its components from knowing the effects of removing one of them. ? (34)BNC Each case has to be considered/*reflected/*pondered/*thought/* con- templated on its merits where an ambiguous answer emerges from the stable isotope data. AC9 567 170 Pattern choice and verb meaning

(35)BNC Now those other products are finally in production and for the first time Excel is going to be judged upon its merits rather than its laurels. CTX 1550 = assess. (36a)NSA *What do they contemplate/reflect/ponder on/upon the matter? (37a)BNC MI5 maintains a large registry of files on individuals and organisations it considers of “interest”. AN0 89 (37b)NSA MI5 maintains a large registry of files on individuals and organisations it thinks/*reflects/*ponders/*contemplates/?judges of “interest”. (36b)NSA *What do they contemplate/reflect/ponder of the matter? (38)VDE At the age of eighty she reflected/*pondered/*considered/*contem- plated/*thought/*?judged on this situation: “My satisfaction in living there was double.”

The occurrence of all other verbs in at least one valency construction in- cluding a TOPIC and an ÆFFECTED – but again not the whole range of theo- retically plausible constructions – could be confirmed: compare (32), (34), and (37a) for consider, (33) and (35) for judge, (37b) for think, and (38) for reflect. Despite the fact that all verbs except for contemplate can be com- plemented with +[Q/S], only reflect can be found in a trivalent pattern with this complement and the only possible realization appears to be +[on_NP] as illustrated in (38). The diverging choice of particles to express one and the same partici- pant pattern and especially the nontransparent choice of which particles can combine with which verbs, despite the fact that in this type of context they undoubtedly express the same meaning, is difficult to reconcile with the notion of predictability of pattern choice on the basis of verb meaning. Moreover, even if found acceptable as such, the complements cannot nec- essarily occur in all semantically plausible combinations of the respective participants (for similar cases see also chapter 4.1).

3.3.4.6 Different formal realizations of ÆFFECTED

The same phenomenon so far described can also be observed with different types of clausal complements, whether introduced by a particle, as in +[that_CL] or +[to_INF], or not, i.e. +[INF] or +[V-ing] (for a discussion of the status of to and that as particles see Herbst and Schüller [2008: 61– 68]). A participant typically realized by this kind of complement type is ÆFFECTED, which is also one of the participants showing most variation in its formal realization. One can argue that the different complements realiz- ing ÆFFECTED can be divided semantically into different subtypes, yet they usually represent one and the same participant as the complements cannot Phenomena identified in the analysis 171 co-occur. Whether there are indeed semantic differences on a subordinate level which is reflected in the acceptability or restriction of certain com- plement types will have to be discussed for each restriction separately. Note that ÆFFECTED can very often also be realized by noun phrases, i.e. what is prototypically regarded as an object. Whether such clausal complements are subsumed under the label object, is, however, not always clear. At any rate, many approaches simply seem to focus on noun phrases without either commenting on or dealing with other complements with the same semantic function in detail. With respect to the distinction between +[to_INF] and +[V-ing], Quirk, Greenbaum, Leech, and Svartvik argue in the Grammar of Contemporary English (GCE) (1972) that while infinitive constructions are more likely to express “potentiality”, participle constructions are more likely to express “fulfillment” (GCE: 835), which is also repeated in the successor of GCE:

Where both constructions … are admitted, there is usually felt to be a dif- ference of aspect or mood which influences the choice. As a rule, the infini- tive gives a sense of mere ‘potentiality’ for action, as in She hoped to learn French, while the participle gives a sense of the actual ‘performance’ of the action itself, as in She enjoyed learning French. (CGEL: 16.40)

However, CGEL (16.40) also accounts for the fact that “[with] other verbs, the difference is more subtle, and may be overruled or neutralized by the meaning of the verb of the main clause”. Closely connected to this, Kiparsky and Kiparsky (1970: 146) claim to show that factive verbs cannot be combined with to-infinitives and non-factive verbs at least not freely with present participle clauses.

(1)QE *He comprehends himself to be an expert in pottery. (2)QE *Everyone supposes Joan's being completely drunk.

This is taken as proof for the direct dependence of the choice of clausal complement type on the meaning of the verb, i.e. whether it is factive (such as comprehend) or non-factive (such as suppose) (Kiparsky and Kiparsky 1970: 147). Frequency data by Mindt (2000: 475) however already indi- cates that such a classification might not even describe a real tendency, since only about a quarter of the examples Mindt took into account for to- infinive clauses can be counted as expressing what he calls a purposive meaning, whereas factual meaning is more than twice as frequent. Ana- logous to this, Kleinke (2002) shows that to-infinitives refer to current, po- tential, and hypothetical events – a hypothesis which is based on classifica- 172 Pattern choice and verb meaning tions made by Bolinger (1968) and Dixon (1991) – while also [V-ing] complements can express the same three features “depending on the seman- tics of the matrix verb” (Kleinke 2002: 237–238):

(3)QE I appreciate your not wanting to maintain his having been to prison before. [current event] (4)QE I suggest leaving on Tuesday. [potential event] (5)QE She avoided meeting students in her office. [hypothetical event]

Herbst (2009) demonstrates that a specialization as described above does not hold beyond specific verbs. Compare (6) and (7), “where it can hardly be argued that [6] expresses ‘potentiality’ and [7] ‘fulfillment’”, as sug- gested by GCE and that “it does not seem to be possible, at least not in this case, to establish a fixed meaning for a particular type of complement” (Herbst 2009: 58):

(6)QE The FIAC has managed in 16 years to become one of the world's lead- ing contemporary art fairs. (7)QE The Council members suggest appointing a task force to study the prob- lem.

Generally speaking, it indeed appears that [to_INF] often entails some kind of purposive meaning. [V-ing], however, does not show such a homogen- ous picture. Taking the two examples quoted from Herbst (2009) into con- sideration as well, a comparison of the pattern choice of the two verbs yields an interesting picture. What is striking is the distribution of the pat- terns with respect to the two verbs manage and suggest, in that the former is very frequently used in combination with [to_INF] (6,130 examples in the BNC) but extremely rarely with [V-ing]. The only two examples with [V-ing] functioning as a complement are given below as (8) and (9). The opposite applies to suggest with two example sentences for [to_INF] given in (10) and (11) vs. 325 example sentences for [V-ing].

(8)BNC If she can manage A levels and a baby she can manage doing this work. FU1 1821 (9)BNC “How do you think you will manage working with a large number of school-leavers …” BNA 1695 (10)BNC …, it's tempting to go on I think about or personal expe- rience of floods, but this is not the time or the place I would suggest to do that . J41 71 (11)BNC I asked children to think of a witches' stew and this is what they sug- gested to put on a sign on the wall.C8P 164 Phenomena identified in the analysis 173

The examples above do not seem to differ semantically from their counter- parts with [to_INF] and [V_ing] respectively. Thus the question that has to be asked is why such a rare form is chosen in the first place, if there is no motive such as the expression of a different nuance of meaning. One typical psychological explanation for syntactic gaps is preemption, i.e. if there is an alternative pattern this blocks a different construction of the same con- tent (Tomasello 2003: 178–181). Examples like the ones above raise the question as to when preemption plays a role and when it does not: such rare uses at least imply that preemption is not an all-or-nothing mechanism but that there is a gradient between frequent uses, rare uses, and impossible uses. Fanego (1996: 31) points out with respect to emotive verbs and their complementation with +[V-ing] and +[to_INF] that “it is not difficult to come across counterexamples showing the inadequacy of practically every criterion proposed in the literature”. Giving a diachronic account of the development of these two alternative complement types, she notes that “the verbal gerund was slow to develop as object, so that by late Modern Eng- lish it was not yet regarded as fully acceptable after a number of matrix predicates with which it has now become established usage” (Fanego 1996: 33). If semantic determinist theories are taken seriously, this would have to lead us to the assumption that this development was accompanied or even instigated by a semantic change. Rudanko (1989) suggests that

… Equi [= to_INF and V-ing] is characteristically governed by verbs ex- pressing , and, further, that infinitival Equi constructions typically co-occur with verbs of positive volition while the ing pattern ... displays a tendency to favor verbs expressing negative volition. The former tendency is more marked than the latter. (Rudanko 1989: 150)

Yet this explanation cannot account for the differences identified here ei- ther. In the context of cognitive approaches (e.g. Langacker 1987), a major distinction is made between finite clauses, which are considered as proces- sual or temporal, and non-finite clauses, which are taken to be atemporal (Hamawand 2002: 5). Consequently, [that_CL] and [V-ing]

… share the same conceptual content, but the participant coded as subject of the main clause construes it differently … [which] results in a difference in meaning. In a processual construal … the complement clause will in all like- lihood happen or be true. … In an atemporal construal, … there is a pos- sibility that the complement clause might happen or be true. The comple- 174 Pattern choice and verb meaning

ment clause represents a particular situation that exists only in the mental space of the main clause subject. (Hamawand 2002: 5)23

Dixon (1991: 218) claims that the difference between +[V-ing] and +[that_ CL] is to be found in the fact that the former describes “an activity or state extended in time, perhaps noting the way in which it unfolds” whereas the latter describes “some activity or state as a single unit”. It follows from this that “some verbs are, by virtue of their meaning, restricted to only one of these complement constructions”. Similarly, Verspoor (1990) argues that differences in the selection of such clausal complements are due to the fact that verbs belong to different semantic groups such as epistemic and deon- tic (based on Palmer’s 1986 discussion of modality) in combination with notions such as causation and the immediacy of this causation.

… I will argue that once ... notions such as ‘causally relevant relationships’ and ‘non-causally relevant relationships’ and ‘immediate’ versus ‘non-im- mediate’ causation are distinguished, one can account for the fact that ma- trix verbs belonging to the same semantic category may select different syn- tactic structures. ... For example the presence or absence of a causal re- lationship might explain the exclusive occurrence of a finite clause after a verb like announce and the exclusive occurrence of a non-finite clause after a verb like compel ... the compelling is causally relevant to the action. (Verspoor 1990: 19 and 27–28) [emphasis in the original]

(12a)QE I announced that he was leaving. (12b)QE *I announced him to leave. (13a)QE *I compelled him that he should leave. (13b)QE I compelled him to leave. (Verspoor 1990: 27–28)

A BNC search, however, does yield examples of the asterisked type (12b), e.g. examples (14) and (15), which show that her criteria do not really stand the test of a corpus-based analysis (cf. also Noël 2003: 355–356).

(14)BNC The explosion was in fact caused by members of the Guandong Army, but the Japanese announced it to be the work of non-uniformed mem- bers of the Chinese army. EE2 880 (15)BNC That night a curfew was announced to start at 11 p.m. and Parma was occupied by the SS Panzer. G3B 903

23. Note that this contradicts Verspoor’s (1990) analysis. For a detailed cognitive analysis of non-finite complement clauses see Hamawand (2002), who dis- cusses various difference in construal reflected in the choice of the type of complement, e.g. the construal of self-relatedness. Phenomena identified in the analysis 175

Moreover, despite the fact that the distinction into epistemic and deontic seems to account for certain similarities in complementation for several such pairs (e.g. with respect to her group “epistemic verbs reporting a dec- laration” including on the one hand the verbs certify, declare, proclaim,and vote and on the other hand appoint, call, christen, crown, declare, elect, name, pronounce, and select), certain differences remain, e.g. in that only the first group allows complementation with [that_CL]: “If the matrix verb directly and immediately causes the resultant state expressed in the com- plement, the finite that-clause is not possible” (Verspoor 1990: 80). How- ever, the distinction into direct vs. non-direct causation cannot account for other differences observed with respect to appoint and elect in the present analysis (cf. 3.2) such as:

(16a)BNC ...they can, if they wish, elect for a voluntary audit, if this is needed for shareholders, lenders or creditors. CBY 3184 (16b)NSA *...they can, if they wish, appoint for a voluntary audit, if this is needed for shareholders, lenders or creditors (17)VDE That morning, they decided they would wait till after lunch when the roads might be quieter, and elected to pass the time by going into Birley to seek out Jarvis’s bookshop. (18)NSA *You may appoint to believe me, or not.

Wierzbicka (1988: 50–51) also claims that there are clear differences in meaning between +[to_INF], +[V-ing], and +[that_CL]. [to_INF], for in- stance, expresses personal knowledge, which is why, according to her, (19a) sounds more natural than (19b). Yet the passive of the pattern which is given in (19c) is quite natural, a point which she does not comment on.

(19a)QE John knows that Mary is a Mormon. (19b)QE ?John knows Mary to be a Mormon. (Wierzbicka 1988: 51) (19c)NSA Mary is known to be a Mormon.

Noël (2003: 352), criticizing her approach of presenting such judgments as empirical facts, demonstrates with the help of a corpus analysis that con- trary to Wierzbicka’s assertions, the structure in (19b) has a first person subject in only 20% of the cases which is practically identical with the re- sult for the structure in (19a) with 19%. Noël (2003: 356–357) also shows that the interpretations provided by Verspoor (1990) and Wierzbicka (1988), who both believe to be able to account for syntactic difference on the basis of a semantic difference, actually contradict each other. Verspoor (1990: 43), other than Wierzbicka, sees the difference in choice between the patterns in (19a) and (19b) in that +[NP to_INF] is chosen “when there 176 Pattern choice and verb meaning is some direct evidence for the belief, and therefore expresses a stronger de- gree of commitment to the truth of the proposition expressed [than with +[that_CL]]”.

It is rather unfortunate that when we try to relate this to Wierzbicka’s (1988) notions of “personal” versus “objective knowledge”, we seem to end up with a correspondence that is exactly the opposite. … For one the that- clause expresses objective knowledge and the infinitive subjective know- ledge, and for the other it is the infinitive which is objectively grounded whereas the finite complement lacks objective grounds. Neither suggestion is backed up by usage data, however. (Noël 2003: 357 and 369)

Riddle (1975: 467–469) argues in favor of controllability and the time se- quence of the event described in the complement and the matrix clause as a distinguishing criterion between the two complement types. According to her the verb can be complemented by [to_INF] only if the action expressed is controllable, which – according to her – applies to activities but not to states. Rudanko (1989: 71–73), however, shows that in sentences such as (20) the verb is not controllable according to Riddle’s definition and com- plementation with [to_INF] is nonetheless possible.

(20)QE John desires to own a car. (21)QE He decided that he would be a teacher. (Rudanko 1989: 73)

Riddle (1975: 469) argues that a verb is complemented with [to_INF] when the event in the complement follows the event described in the matrix clause (which makes it controllable as opposed to a state which is uncon- trollable). This is again proved wrong by sentences such as (21), where the predication of the [that_CL]-complement clearly follows the one of the matrix clause but complementation with [to_INF] is not possible (Rudanko 1989: 73). Kleinke (2002: 237–251) gives a detailed analysis of various attempts to provide a semantic classification of [to_INF] and [V-ing].24 She concludes

24. Kleinke (2002: 237–251) quotes the following semantic features: “real vs. non-real existence of the state of affairs” (referring to Dixon 1990 [most like- ly 1991]; Dirven 1989), “temporality” (e.g. Wierzbicka 1988; Verspoor 1990), “referent” (e.g. abstract vs. concrete in Dirven 1989), “volitionality” (e.g. Wierzbicka 1988; Dirven 1989; Dixon 1991; Verspoor 1990), “ma- nipulation” (e.g. Dirven 1989), “evaluation through the speaker” (e.g. Wierzbicka 1988; Dixon 1991; Verspoor 1990), “integration of the event” (Dixon 1991), “prototypical causal relationship” (e.g. Verspoor 1990), “pre- Phenomena identified in the analysis 177 that many of the various aspects of meaning can be verified in combination with different matrix verbs “without having the status of schematic mean- ing as partially assumed” (Kleinke 2002: 236). Instead she assumes these aspects to play a role in the form of semantic attributes which occur togeth- er in various different bundles depending on the semantics of the verb to be complemented. This way, however, she cannot account for differences in acceptability or frequency either (e.g. the two complement types are attri- buted the same bundle in combination with start and begin which differ as regards their preferred complementation [cf. 3.3.6]). Another recent work which focuses on the differences between comple- mentation with non-finite complements is Egan (2008), a usage-based ap- proach which focuses explicitly on various types of infinitive and participle complements and which provides a detailed overview on various ap- proaches. Similarly to Kleinke (2002), Egan (2008: 87) concludes on his review of various approaches that “while all of these approaches account satisfactorily for some types of to infinitive and -ing clauses, few of them do so in all cases” [emphasis in the original].25 According to CGEL, verbs complemented with +[that_CL] fall into four groups: factual verbs – with an indicative verb in the complement clause –, suasive verbs – where typically a putative should can be introduced into the clause complementing that –, emotive verbs – which can have both an in- dicative verb or putative should in the complement clause – and finally hypothesis verbs which are combined with hypothetical past or were-sub- junctive (CGEL: 16.30). The sample examined here contains a group which semantically falls into the group emotive verbs and which allows both +[to_INF] and +[V-ing] but in general not +[that_CL]. Only one example was identified in the BNC for each love +[that_CL] and like +[that_CL] but not for adore.

(22)BNC Fabia loved that he seemed as befuddle-headed as she. JYF 2621 (23)BNC She was so enthusiastic about what she saw and liked that she could trans- mit her ideas even if the appeal was not immediately obvious to us. GU9 746

supposition” (e.g. Kiparsky and Kiparsky 1970), and “illocutionary role” (e.g. Verspoor 1990). 25. Egan (2008), unlike many of the other approaches reviewed here, takes a usage-based perspective and rests his analyses on a larger number of authen- tic examples sentences. For this reason, this approach seems promising, even if the many distinctions he makes appear to be difficult to reconcile with a no- tion of straightforward semantic rules. 178 Pattern choice and verb meaning

Despite the fact that emotive verbs can generally be complemented with +[that_CL] this complement is neither listed in VDE for love and like nor are these verbs included in the CGEL verb-list for complementation with +[that_CL]. Hence, the fact that a verb belongs to a certain group does not necessarily indicate whether a certain form of complementation which is generally associated with this type of verb can actually occur with all of these verbs. Moreover, there is no indication as to whether the form of complementation is the usual choice or rather exceptional. Rudanko (1989: 82) concludes in his comparison of these complement types that …[a] verb expressing intention or intention and endeavour but not commu- nication, characteristically may take Equi [= V-ing and/or to_INF] or for to complement clauses [= for NP to_INF] only, but no that complement claus- es, except if Verb1 [= the matrix verb] means ‘come to a decision’ in which case a that complement is also possible.

In that sense adore serves as a good example, but love and like do not ex- press communication or ‘coming to a decision’, either and appear to be more likely to accept +[that_CL] than adore. Train, practice, and rehearse can all be complemented with the partici- pant ÆFFECTED but again not all verbs allow all possible realizations.

(24)VDE You can also stop the play at random and then practice/*train respond- ing spontaneously. (25)BNC At the start she would rehearse arranging them with the nun so that she would do it well when she got to the Hogans, but as the weeks went by she grew in confidence. CCM 638 (26a)VDE Preparation for tests should consist in learning the subject being tested, not practicing how to answer this or that type of question. (26b)NSA Preparation for tests should consist in learning the subject being tested, not rehearsing/?training how to answer this or that type of question.

The latter realization of ÆFFECTED can also be found in combination with train, given that the pattern expresses BENREC. Some native speakers also accepted the sentence when the particle for is inserted (28).

(27)BNC Utilizing the instructional capabilities of the PLATO computer-based education system, the course trains users BENREC how to carry out searches ÆFFECTED...B3D 240 (28)NSA ?It’s really hard to train for how to answer this or that type of question.

Finally, [on_NP], which is also listed in VDE as a complement of practice, fulfills the function of ÆFFECTED, while there seems to be a difference to Phenomena identified in the analysis 179 other realizations of this participant on a subtype level. Rehearse and train cannot occur in this pattern.

? ? (29)VDE Try practising/* rehearsing/* training on a cheap fish such as mackerel.

Although train can express a participant ÆFFECTED, it appears to be a ra- ther untypical participant in combination with this verb. There appear to be only two possible realizations of ÆFFECTED: [NP], where the lexical reali- zation of the noun phrase is rather limited, and [to_INF].

(30)BNC The short-term need is to deal with the roots of the doubt -- in this case a matter of incompleteness of the picture of God -- while the long-term need is to train a habit of faith which will close the door to this type of doubt in the future. C8V 1400 ? (31)NSA *Science teachers included those training physics, chemistry, biology and mathematics. (32)VDE We train to play 15-man rugby. (24)VDE *You can also stop the play at random and then train responding spon- taneously.

Despite the fact that neither practice, rehearse, nor train refer to a potential action, all of them can be complemented with +[to_INF] and despite the fact that all of them can refer to an actual performance, train cannot be complemented by +[V-ing]. An alternative explanation might be found in the fact that [to_INF] is more result-oriented whereas [V-ing] is rather process-oriented. [to_INF] has both similarities to [V-ing], with which it commutes and which was described as a realization of ÆFFECTED above, as well as with [for_NP], which is best described as AIM (a construction all three verbs can be found in). In many cases it does not seem to be possible to draw a distinct line between the two, since [to_INF] often appears to entail some kind of purpose, which makes it similar to AIM. Yet this inter- pretation does not hold across all groups. All four verbs mean, intend, plan, and aim can be complemented with +[to_INF] (33)–(36) and mean, intend, and plan allow complementation with [NP] (37)–(39).

(33)VDE If we had meant to stage a protest, we could … (34)VDE She intended to stand in the state elections in October. (35)VDE You should always aim to stay in control. (36)VDE I’m planning to have my baby at home, but I’m worried in case some- thing goes wrong. (37)VDE Magda didn’t mean any harm. 180 Pattern choice and verb meaning

(38)VDE I never intended/*aimed any harm. (39)VDE Plan something you can look forward to.

However, only intend and plan are conventionally complemented with a [that_CL] as illustrated in (40)–(43) and [V_ing] in (45)–(46).

(40)VDE I intend that you shall enjoy your stay on my island. (41)VDE Mr Ridley is the fourth cabinet minister to resign since October and Mrs Thatcher had planned that her summer reshuffle would involve only the lower ranks of government. (42)NSA *Mrs Thatcher had aimed that her summer reshuffle would involve only the lower ranks of government. (43)BNC ?We do not need here to rehearse the contentious issues which surround integration, but it is sometimes aimed that the placement of a child in a mainstream school constitutes integration, …. CMU 43 [not accepted by native speakers] (44)VDE Did he mean that interest rates would come down rather slowly? = ‘imply’ (45)VDE If you intend/*aim buying a pet Amazon parrot then you should obtain a hand-reared specimen. (46)VDE The week after I will plan/*aim decorating the house. (47)VDE I had been a serious smoker and it meant giving up a 30-year, 35-a-day habit. = ‘CONSEQUENCE’

This might be explained by the fact that aim expresses potentiality rather than performance, i.e. describing something that will potentially happen in the future. Indeed, the meaning of mean changes slightly when comple- mented with +[V-ing] and +[that_CL], emphasizing the consequence im- plied by something as exemplified by (44) and (47). However, also intend and plan both express predominantly potentiality and nonetheless allow complementation with +[V-ing] and +[that_CL]. Also in the ‘permit’-group there are irregularities with respect to com- plementation with clausal complement types. While allow, permit, and authorize can be complemented with +[V_ing] as in (48)–(50) and with +[NP]+[to_INF] as in (51)–(53), only allow can be complemented with +[to_INF] (54).

(48)BNC 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 (49)BNC Both the Lithuanian and Latvian leaderships had renewed calls during November for a Red Army withdrawal, and on Nov. 14 the Latvian Su- Phenomena identified in the analysis 181

preme Council (formerly Supreme Soviet) had authorized cutting off food and power supplies to Soviet military bases. HL3 563 (50)VDE Congressional negotiators have imposed limits which may permit build- ing only two more of the bombers. (51)VDE In 1986, I got joint custody of the children, but my wife has never al- lowed me to see them. (52)BNC Chair, it was one officer whom we've got on staff at the moment Karen Wheeler who's away nursing her new baby, and it's to au- thorize that officer to carry out certain roles under the health and safety. KS1 206 (53)VDE Today, it seems, we are no longer permitted to be Tories. ? (54)VDE Gently pour the butter into the sieve and allow/*authorize/* permit to drain through undisturbed.

This can be accounted for on the basis that when no subject is expressed in the infinitive clause, the implied subject – or understood subject as Rudan- ko (1989: 3) refers to it – is co-referential with either the subject or another complement in the superordinate clause. Thus, in (54) it is the butter which is the implied subject of drain. The fact that also permit and authorize can be complemented with +[NP to_INF] further indicates that the restriction here is not connected to the form of the complement but rather to a differ- ence in the optionality of the post-verbal noun phrase. This difference in optionality is, however, difficult to relate directly to a difference in mean- ing, especially between allow and permit. Only allow, permit,and author- ize can be complemented with +[for_NP to_INF], whereas +[for_NP V- ing] is again only possible with allow.

(55)VDE Originally the scheme allowed/*entitled for pensions to be calculated on the best twenty years of earnings. (56)BNC “Is it permitted for me to call you Cara?” he enquired. JYF 763 (57)NSA “Is it authorized for me to enter the premises?” he enquired. (58)VDE Hugh finds the 35mm format is better for his high speed flash work, as film stock is cheap and comes in 36 exposures, allowing/*permit- ting/*authorizing for things going wrong such as flash sync problems.

Also [of_NP] and [for_NP] could be seen as alternative realizations of ÆFFECTED. Authorize, other than for instance entitle, can be used without BENREC analogous to allow and permit but it is limited in the realization of this participant. Moreover, it is not clear why something can permit of something but not for something (but for something to happen and not for something happening, as in [56] and [57]).

(59)VDE The facts allow of only one explanation. 182 Pattern choice and verb meaning

(60)VDE Psychiatry is not yet well enough developed to permit/*authorize of an easy, agreed classification. (61)VDE The clock tower has gone to allow/*permit/*authorize for a central one- way traffic system that neither helps pedestrians nor speeds traffic.

Among the verbs in the ‘remember’-group the realization of ÆFFECTED also differs: while all of them allow complementation with +[that_CL] and +[if/wh-CL], there are also restrictions, especially with respect to bear in mind and reminisce.

(62)VDE I remember/*reminisce her well. (63)VDE I can’t remember/*reminisce how to pronounce it. (64)VDE “The last few weeks before term were horrendous,” Chris recalls/ *bears in mind. (65)VDE I remember/*reminisce/*bear in mind sitting on bundles and suitcases, waiting for the train that would take us we didn’t know where. (66)VDE I am old enough and lived close enough to remember/*reminisce/*bear in mind the factory being built.

Note that the meaning which is expressed by remember differs, depending on the type of complementation as exemplified in (67) as opposed to (68):

(67)VDE I remember sitting on bundles and suitcases, waiting for the train that would take us we didn’t know where. ? (68)VDE You should remember/* bear in mind to replace your child’s toothbrush every three months or so. = TASK (69)VDE I can’t recall/*bear in mind reading a book of over 600 pages so avidly. (70)NSA *Recall to leave your name, age and address. (71a)BNC “I don't recollect writing poems for anyone else for twenty year.” A0L 3006 (71b)NSA *“I don't recollect to write poems for anyone else for twenty year.” (72a)NSA *I reminisce writing poems. (72b)NSA *I reminisce to write poems.

Compare CGEL:

The infinitive construction indicates that the action or event takes place af- ter (and as a result of) the mental process denoted by the verb has begun, while the reverse is true for the participle construction, which refers to a preceding event or occasion coming to mind at the time indicated by the main verb: (CGEL: 16.40)

(73a)QE I remembered to fill out the form. [‘I remembered that I was to fill out the form and then did so’] Phenomena identified in the analysis 183

(73b)QE I remembered filling out the form. [‘I remembered that I had filled out the form’]

Dixon (1991: 221) points out a similar difference for remember when com- plemented with [to_INF] and a [that_CL] complement containing a modal in that a sentence such as (74b) expresses at least the intention of the sub- ject to actually “lock the door” if not its successful completion, which does not apply to (74a).

(74a)QE I remembered that I should lock the door. (74b)QE I remembered to lock the door. (Dixon 1991: 221)

Recall, reminisce, and recollect cannot express the potentiality of an event. With respect to remember and recall, which Verspoor treats as “epistemic verbs of recollection”, she points out that both can be complemented with either +[that_CL] or +[V-ing], depending on whether “the actual event, not circumstantial evidence, causes whatever is remembered” (Verspoor 1990: 87). However, only remember can be complemented with the two patterns +[NP]+[AdjP] and +[NP]+[NP] (one pattern according to Verspoor [1990: 72 and 81], i.e. a small clause), whereas recall cannot, which Verspoor does not comment on, nor does she address the differences in particle com- plementation presented in the comparison of these same verbs in this present analysis. Reminisce and bear in mind can be complemented with +[that_CL] but not with +[V-ing], which would then imply that these verbs focus on results more than on the course of events. This might indeed serve as an explanation of the restriction with respect to bear in mind but less so for reminisce which can express this meaning when a particle is added:

(75)NSA I reminisce about sitting on bundles and suitcases, waiting for the train that would take us we didn’t know where.

A similar situation is to be found with think, which can be complemented with +[that_CL] but not with +[V-ing]. Similar to reminisce, a particle must be added to express this progress meaning, i.e. think +[about_V-ing]. All other verbs in the ‘think’-group can also be complemented with [that_ CL] but not necessarily with other clause complements. Since all six verbs in this group occur in the valency construction [SCU]_AGENT + ‘think’ +[(that_)CL]_EFFECTED, the various restrictions cannot be deferred to poss- ible differences in the participant inventories. Except for this construction, there is no further valency construction in which all verbs in the sense ‘think’ can occur. For instance, not all of the verbs can equally well be used 184 Pattern choice and verb meaning as speech act verbs: judge and consider seem to be very rare in a pattern with +[QUOTE], and consider is more likely to be complemented with +[SENTENCE] than with +[QUOTE]. With respect to contemplate, native speakers disagreed. Moreover, it appears to be impossible to reflect how to do something, although ponder and contemplate – the verbs most similar to reflect in this group – have been found in the pattern [SCU]+ ‘think’ +[wh- to_INF].

(76)BNC Dorothy, 65, of Hinckley, Lincs, is putting the £10,000 in a safe place -- the bank, not behind a cushion -- while she ponders/*reflects how to spend it. CH1 7501 (77)BNC I have a streak of economy in me, even when contemplating how to épater les bourgeois, and recipes that make their sauce out of a stock made at an earlier stage tend to appeal most. G2E 2497

Ponder and reflect cannot be complemented with +[so/not/otherwise]. This might indeed be connected to the fact that this complement type usually serves as a proform for something such as an opinion, i.e. it is connected to the meaning think ‘opinion’ rather than to think ‘thought’ which is the pro- totypical meaning of ponder and reflect. Nonetheless, +[so/not/otherwise] typically substitutes [that_CL] (CGEL: 12.27), which is available with all of these verbs, e.g.:

(78)VDE I don’t think that that is entirely true. (79a)VDE Confused customers might think otherwise. (80)BNC The fact that the Court of Appeal contemplated that an appeal might lie to the House of Lords on this matter arguably indicates a confusion concerning the nature of the jurisdiction it was exercising. FBK 595 (81)BNC A trade must normally be exposed to the exchange's established pricing mechanism, such as open outcry or an electronic matching system, un- less the rules contemplate otherwise. J73 380 (82)VDE “I judge Mrs. Plover was cooperative,” Mr. Macmillan said to him. ? (79b)NSA Confused customers might judge/ consider/*reflect/*ponder otherwise. (83)VDE Newcomers, living in homes built within the past 20 years, are often the most vociferous objectors and rarely seem to consider that they them- selves have benefited from recent planning permission. (84)VDE He further reflects that by the time all the participants have adjusted, the rules of the game will change again. (85)BNC The Right has prompted the Left to ponder that perhaps capitalism and democracy have come to the parting of ways; … G3L 1315

The complement type +[to_INF] can only be found with think and judge, which cannot be complemented with [V-ing]. Phenomena identified in the analysis 185

(86a)VDE They never thought to give it a different name. (86b)NSA *They never thought giving it a different name. (87)BNC 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 ? (88)NSA *You should judge doing this.

Contrarily, ponder and contemplate, as well as consider and reflect – though with a slightly different meaning – can be complemented with [V- ing] but not with [to_INF].

(89a)BNC “I am contemplating abandoning my quest for silver to go for gold.” CH7 2664 (90)BNC Carole King, the sharpest female lyricist of all time, ponders going all the way and whether her lover would still respect her in the morning. CHA 2494 (91)BNC Overall the average K/U and U/Pb ratios define a broad hyperbolic trend that may reflect mixing, but is more likely to indicate a process that fractionates both K and Pb relative to U (Fig. 3b). CRM 585 = ‘mir- ror’ (89b)NSA I am *contemplating/*pondering/*reflecting to abandon my quest for silver to go for gold. (92)VDE In this case, you should consider seeking professional help. = ‘take into consideration’ (93)NSA *You should consider to seek professional help.

Also [of_NP] is used to express EFFECTED, while it is restricted to think, consider (rare), and judge (with a slightly different meaning).

(94a)VDE He thought of how he’d kissed her. (95)BNC “As the period of the completion of the undertaking is not now far dis- tant, your Committee have considered of the most eligible mode of ap- pointing medical attendants to the institution; CMG 1180 (96)BNC From the Books which I have sent, you will be able to judge of his hand- writing and perhaps also of his progress in arithmetic. B3H 808 = ‘assess’ (94b)NSA *He reflected/pondered/contemplated of how he’d kissed her.

In the ‘assume’-group, guess and assume can be complemented with +[wh- CL]_ÆFFECTED. Although all other verbs in this group also have the par- ticipant ÆFFECTED (they can for instance be complemented with [(that_) CL] or [Q/S]), example sentences with this complement type appear to be unacceptable (cf. Dixon 1991: 135).

(97)NSA I can only assume how he got a hold of it. ? ? (98)BNC And guess/*gather/* suppose/* presume how she heard of us? JYE 895 186 Pattern choice and verb meaning

Similarly, suggest, recommend, and propose, which all have the participant EFFECTED, exhibit restrictions as regards the formal realization of this par- ticipant, however, in different ways:

(99a)VDE You might get a good counsellor that might suggest you doing a college course or something but that’s as far as it really does go. (100)BNC “True psychologists do not recommend siblings competing directly because they could become jealous and envious, and might even come to hate each other,” says Polgar. A89 442 ? (99b)NSA *You might get a good counsellor that might propose you doing a col- lege course or something but that’s as far as it really does go. (99c)NSA He might propose your doing a college course. (99d)NSA ?He might propose you do a college course.

(99c) and in part (99d) were accepted by native speakers. The similarity is obviously striking but nonetheless (99b) was not accepted despite the pa- rallel to suggest and recommend and despite the similarity especially to (99c). One possible explanation might be that the existence of (99c) preempts a use such as (99b) (e.g. Tomasello 2003: 178–181). Moreover, [to_INF] as a realization of EFFECTED is only possible with propose and recommend but not with suggest.

(101)VDE So what I propose/*suggest to do is just generally have a look at infla- tion this morning. (102)BNC Committee recommended to approve a scheme which is a requirement of the planning permission for the neighbourhood development. J44 204

Besides these restrictions, there also seem to be regional differences: Algeo (2006: 246) describes the combination of recommend and [NP V-ing] – as illustrated in (100) – as acceptable in British English but unlikely in Ameri- can English. It appears to be rather difficult to give a semantic explanation for this kind of difference. Teach, train, instruct, and educate do not allow the same range of forms of complementation to express the participant ÆFFECTED, either.

(103)VDE The conservative establishment of scholars and educators have in- structed the British people BENREC that great poetry is universal poetry ÆFFECTED. (104)VDE The committee sees the centenary as a good opportunity to educate the British BENREC that Sikhs have not just landed on their doorsteps but have a long history as allies ÆFFECTED. (105)VDE This early experience taught/*trained me BENREC that people do want their boss to appraise them ÆFFECTED. Phenomena identified in the analysis 187

(106)VDE Your family doctor or the gynaecologist will instruct you BENREC how to interpret and use this data ÆFFECTED. ? (107)VDE Richard taught/*educated/* trained me BENREC how to cook ÆFFECTED. (108)VDE This early experience taught/*instructed/*educated/*trained me BEN- REC a valuable lesson in legal humility ÆFFECTED. (109)VDE In Greek mythology, Chiron, the centaur who taught/*educated/ ? *trained/* instructed medicine ÆFFECTED to Asklepios BENREC, suffered an incurable wound at the hands of Hercules. (110)VDE He taught/*instructed/*trained/*educated himself BENREC carpentry ÆFFECTED from library books SOURCE. (111)VDE His early career was also helpful in that it taught/*educated/ *?instructed/*?trained him BENREC more “AMOUNT” about how not to run a business than how to run one TOPIC.

Note that CGEL classifies the pattern exemplified by (105) as “unusual” in combination with train yet no example could be identified and native speakers did not accept train in this sentence. In case of (111), the pronoun which constitutes the second predicate complement unit (“AMOUNT”) could also be interpreted as a subtype of ÆFFECTED. In the ‘tolerate’-group, bear and endure can be complemented with +[to_INF], which does not seem to be acceptable with tolerate although it can realize ÆFFECTED with [NP] (114) and [NP V-ing] (115).

(112)VDE I had a little money of my own coming in, which was very important to me because I could never bear to ask my husband for it. (113)BNC But Carteret's wife, who frequented health spas, could not endure/ ? * tolerate to live with him or he with her: there were no children. CD2 1061 (114)BNC One young man, unable to tolerate the thought, burnt himself alive in a public square. A08 1361 (115)BNC I’m not going to tolerate you working on this team. HGM 474

Respect and admire, finally, allow more flexibility as regards the formal realization of ÆFFECTED than esteem:

(116a)VDE I admire/*esteem what you’re doing and I’ll help. (116b)NSA I respect what you’re doing and I’ll help. (117)BNC The more I see of you, the more I esteem you. FNT 1068 (118)VDE The bottom line is that you gain respect from others only when you have learned to respect yourself. (119)VDE Davis said that he had always admired Paul McCartney’s musical ta- lent. 188 Pattern choice and verb meaning

These analyses clearly show that whether a participant is generally accept- able with a word or not can by no means be taken as the basis for predicting possible complementation patterns. Moreover, the semantic differences appealed to when trying to account for the differences in complementation do often not hold across different verbs or are so verb-specific that they clearly do not constitute an argument against a model in which storage plays an important role. Two patterns which are also often discussed in connection with clausal complement types are [SCU]+ verb +[NP]+[to_INF] and [SCU]+ verb +[NP to_INF], which differ in that the former consists of three separate complements forming a trivalent pattern, whereas the latter is a divalent pattern, where the noun phrase functions as subject complement unit within the subordinate infinitive clause. The distinction between the two is, how- ever, not always clear-cut. Both CGEL and CamG provide several criteria for distinguishing between divalent (in their terminology monotransitive) and trivalent (ditransitive) complementation (CGEL: 16.36, 16.41, 16.43, 16.66; CamG: 1201–1204), even if these criteria are not necessarily uncon- troversial (for a discussion see, for instance, Standop 2000). The divalent (or monotransitive) pattern with [NP to_INF] is, according to CGEL, “re- stricted to a small number [of verbs] chiefly denoting (not) liking or want- ing: (can’t) bear, desire, hate, like, love, prefer, want, and wish” (CGEL: 16.41). Nonetheless, many of the verbs included in this sample were listed in CGEL as verbs allowing this form of complementation, and in many cases verbs were indicated to accept complementation with both +[NP to_INF] and +[NP]+[to_INF] in VDE. Semantically, the patterns could be said to express ÆFFECTED and some form of PREDICATIVE or AIM, either in a merged form (in the divalent pattern) or as two separate complements (in the trivalent pattern). Matthews (1981: 184) refers to this type of construc- tion as a fused construction “in which a single element is a complement of both a controlling and a dependent predication”, and in terms of earlier versions of generative grammar this phenomenon is described as subject-to- object-raising (Haegeman 1991: 401). Again, it appears that semantically similar verbs do not necessarily “behave” in the same way.

(120a)VDE Today, it seems, we are no longer permitted to be Tories. (121a)VDE In 1986, I got joint custody of the children, but my wife has never al- lowed me to see them. (122a)NSA Section 2 authorizes the Home Secretary to issue warrants.

The underlined passage in (122a) with authorize is better analyzed as a single complement [NP to_INF] – other than in (120a) and (121a) – be- Phenomena identified in the analysis 189 cause authorize is, for example, not completely acceptable in a pseudocleft sentence, unlike allow and permit.

(122b)NSA ?What section 2 authorizes the Home Secretary is to issue warrants. (121b)NSA What my wife did not allow me was to see them. (120b)NSA What they do not permit us is to be Tories.

However, neither the tests which are typically applied to differentiate be- tween these two patterns nor the native speaker assessments in this respect appeared watertight enough to make any claims about this variability.26 For this reason, the pattern was indicated as +[NP (]+ [)to_INF] in the tables given in chapter 3.2.2. However, examples such as the one quoted above indicate that semantically similar verbs do not necessarily exhibit the same structural properties in this field either.

3.3.4.7 Different formal realizations of PREFERENCE and AIM

A similar situation can be found with the participant pattern AGENT + verb + PREFERENCE. Declare, for instance, can be found in both valency patterns expressing this participant pattern [SCU]+ ‘declare’ +[for_NP] and [SCU]+ ‘declare’ +[in favor of/against_NP], whereas pronounce and state only occur in one of these patterns, however, in different ones. Announce can occur in the first pattern only in one specific context given in (3), while proclaim was not considered acceptable at all.

(1)VDE One of the first to declare/*pronounce/*proclaim for Mr Heseltine, having wavered over the weekend, was Peter Rost. (2)BNC While her husband threw up violently into the lap of the Japanese Prime Minister and the world took Valium at the thought of Sandpit Dan mov- ing into the White House, she stated for pudding. BNP 312 (3)VDE I have to decide whether it is good or better for the people of my state or not good at all for me to announce for the presidency. = ‘candidacy’ (4)BNC The turnout, across the Soviet Union as a whole, was 80 per cent, … and of those who voted, 76.4 percent declare/*stated/*proclaimed/*an- nounced in favour of the question that had been asked of them. FYT 432 (5)VDE The Bishop is said to have declared against adultery but not adulterers. (6)VDE The plaintiff asked the court to pronounce in favour of that 1986 will. (7)VDE Russian nationalists have also raised the subject and pronounced/ *stated/*proclaimed*/announced vehemently against concessions.

26. See htttp://dx.doi.org/10.1515/9783110240788_appendix3 for the sentences assessed by native speakers. 190 Pattern choice and verb meaning

VDE lists two trivalent patterns for declare which have here been sub- sumed under one single pattern, which also expresses a certain preference: [SCU]+ ‘declare’ +[REFLPRON]+[in favor of/against_NP]. Announce, proclaim, and state cannot be complemented in this way as indicated in (9a) and (8b). Pronounce in (8b) and (9b) was considered acceptable by native speakers, although no such examples were identified in the BNC.

(8a)VDE A number of naval officers urged Michelsen to recognize this and to declare himself openly in favor of the constitutional regime. (9a)VDE Poland’s Finance Minister has declared/*announced/*stated/ *?proclaimed himself against giving large numbers of shares to people directly employed by privatised enterprises. (8b)NSA He pronounced/*proclaimed/*announced himself in favour of a consti- tutional regime. (9b)NSA Poland’s Finance Minister has pronounced himself against giving large numbers of shares to people directly employed by privatised enterprises. (10)NSA *After a long debate, he stated himself in favour of the lowest of these sums.

As discussed above, mean cannot express a participant AIM. The other members of the ‘intend’-group differ as regards the range of possible reali- zations of this participant, e.g.

(11)VDE Aim/*intend for clarity and precision; remove words that obscure your directions. (12)VDE We must begin now to plan for a time when people will ultimately be able to work for the cheque they get. (13)BNC And I think she intends/*aims on having quite a few drinks. KD3 528 (14)VDE The Beaton family are planning on a trip to Guyana next month. (15)VDE That’s precisely what Labour should aim/*plan/*intend at: the replace- ment of a market mechanism with planning mechanisms.

A correct paraphrase of (11) with plan would include either [on_NP] as in (14) above, [for_NP] as in (12) or [NP], i.e. in the latter case AFFECTED in- stead of AIM. The similarity between the two is very obvious, so there might even be a case for arguing in favor of a blocking of the +[at_NP]- pattern due to the similarity in its communicative function with +[NP]. This however does not explain why it is possible to plan on something or for something. Parallel to this, aim can only be complemented with +[at_NP] but not with +[NP] (16).

(16)NSA *I never aimed any harm. Phenomena identified in the analysis 191

Push, which has both AIM and AFFECTED as indicated in sentences (20) and (21), does not occur in the valency construction [SCU]_AGENT + ‘urge’ + [NP]_AFFECTED + [on_NP]_AIM, i.e. [on_NP], which also expresses what is desired by the AGENT, is not possible. Compare:

(17)VDE I mean I never pressed him AFFECTED on it AIM. (18)BNC “Don't press/*push it AIM on him AFFECTED,” said Lady Grubb. FSP 1133 ? (19)NSA *Every night she conferred with John Cullen, trying to push speed on him, instead of the perfection he desired. (20)VDE The kids AFFECTED are encouraged but not pushed. (21)NSA Opposition Labour MPs are again pushing for a public inquiry AIM.

Several restricted patterns in this group include realizations of the partici- pant AIM, which all three verbs have as part of their participant inventory, e.g. [that-CL], [QUOTE], and combinations of these with [upon_NP] and [NP] respectively:

? (22)VDE We write again to urge/* push that the government act against rape within marriage by changing the law. (23)BNC Will he press that where commitments are made to open markets for textile products, a proper verification procedure must be established? HHV 11793 (24)VDE “Come on,” Jarvis urged. (25)BNC “You love me?” he pressed/*pushed, “really love me? HHA 4373 (26)BNC “Get a ticket, not a criminal record,” the ads urge/*press/*push you; but this is easier said than done. AAV 1025 ? (27)VDE Will he urge/*push/* press upon Tony Blair that the returned-to-local democracy, new foundation schools be forced to become real centres of subject excellence in, say, music, sports, or even drama?

3.3.4.8 Conclusion

The cases discussed above show that, even if it is difficult to provide an exact semantic role distinction, the semantic similarities are obvious. Al- though there are clearly cases where a semantic difference is expressed when more than one complement type can fill a certain slot, there are also cases where this does not seem to be the case. For this reason the exact range of possible formal realizations of a verb cannot be predictable on the basis of general semantic rules. Moreover, there are several studies which indicate that factors other than verb meaning are responsible for the choice of one pattern over another. Wasow (2002: 36–41), for instance, found that in the case of the dative alternation (+[NP]+[NP] vs. +[NP]+[to/for_NP]), the complexity as well as the length of the object has an influence on the 192 Pattern choice and verb meaning preferred order in the sense that the heavier and more complex an object the more likely it is expressed at the end of a sentence (cf. GCE: 14.8; see also Gries 2003: 9). Thus when the heavy complement represents BENREC, a particle complement is more likely, which also appears to be linked to con- straints on speech planning and production (Wasow and Arnold 2003: 129). At the same time, however, they acknowledge that “[w]eight and informa- tion structure do not suffice to account for constituent ordering” (Wasow and Arnold 2003: 130; see also Hawkins 1994). Bybee (2007: 288) – based on experiments conducted by Bock (1986) – reports on priming effects with respect to the dative alternation showing that the choice of construction is highly influenced by the occurrence of constructions in the preceding sentences. Such syntactic priming effects have also been confirmed in corpus-based studies. Gries (2005: 374), for instance, showed that “the corpus-based analysis of syntactic priming re- vealed significant priming effects for ditransitives and prepositional da- tives”. What he also showed, however, is that “a verb strongly associated with a particular construction resists priming and rather sticks to its asso- ciated constructions” (Gries 2005: 379). Whether the stronger collostruc- tional association can be seen as based on the semantics of the verb is not completely obvious. It is equally likely that the results point in direction of a storage phenomenon. Gries (2003: 16) concludes that “it is highly un- likely that we will ever be able to predict native speakers’ behavior com- pletely flawlessly irrespective of the number of variables we might still want to include in the analysis”. Another non-semantic factor in the choice of alternative patterns ap- pears to be what has been called the principle of horror aequi which “in- volves the widespread (and presumably universal) tendency to avoid the repetition of identical and adjacent grammatical elements or structures” (Rohdenburg 2003: 205). Thus if the matrix verb is already embedded in a [to_INF] pattern, it is more likely to be complemented with +[V-ing] or +[that_CL], if these alternatives exist, than with another +[to_INF] pattern. This in part contradicts the idea of priming. The fact that both can be shown to have an effect again emphasizes the complexity of the processes in- volved. Also cognitive complexity itself has been considered to play a role in pattern choice. Vosberg (2003) found that in cases of verbs allowing complementation with both +[to_INF] and +[V-ing], the former is pre- ferred, even if this leads to a repetition of structure (two to-infinitives), when “sentential complements [are] separated from the superordinate clause by any intervening material. … This phenomenon is most likely due to an increase in cognitive complexity provided by these discontinuous Phenomena identified in the analysis 193 constructions” (Vosberg 2003: 321). These principles have also been em- ployed to explain the changes in the complementation preferences or possi- bilities diachronically (Vosberg 2003: 322). Again it needs to be stressed that all these phenomena show that the de- cision in favor of or against a certain complement is not necessarily based on semantic aspects at all. The fact that aspects such as the ones mentioned above can override possible semantic differences also proves the point that the meaning in the sense of pragmatic function of different constructions – even if formally different – can be very similar or indistinguishable from one another depending on the actual situation of usage.

3.3.5 Different flexibility in the combination of participants

Some verbs were analyzed as having the same participants while exhibiting restrictions as to how they can be combined. In the ‘propose’-group and the ‘think’-group some verbs are restricted as regards the combination of par- ticipants which can occur separately although both a separate and a com- bined occurrence of the participants is possible with semantically similar verbs of the same groups (for a discussion see also chapter 4.1.4). Contrari- ly, in the ‘choose’ group, for instance, some verbs only allow certain parti- cipants in combination but not separately, i.e. only in trivalent and not in divalent patterns, although the participants can occur in divalent patterns with other semantically similar verbs in the same group. Despite the fact that the complement type +[among_NP] is accepted in a trivalent pattern by most of the verbs in the ‘choose’-group, choose and select are the only ones to also occur with +[among_NP] in a divalent pat- tern, even if for select only one single example was identified in the BNC.

(1)VDE Therefore it is especially important to know the various options and to choose among them with the fullest possible information. (2)BNC At the same time teachers are being treated more and more like workers and less and less like professionals, so that their sense of power and freedom to evaluate and select among these opportunities feels dimi- nished. CKS 274 (3)NSA He can *appoint/*elect/*designate/*pick among several options. (4)BNC Not least in America, where else, where companies know they can pick and choose among executives desperate to work again. K5M 6400

Example (4) is the only example of pick complemented by [among_NP] identified in the BNC, however, pick is coordinated with choose in that case. With some verbs (e.g. nominate and name) [NP] is generally ob- 194 Pattern choice and verb meaning ligatory, however, this is not the case with designate and elect, which can occur in divalent patterns without +[NP], and pick and appoint in addition to that in a monovalent pattern (cf. 3.2). Thus their unacceptability in (3) is not due to the missing of an obligatory complement. At the same time, only elect and appoint can occur in the divalent construction [SCU]_AGENT + verb [from_NP]_SOURCE and only appoint in the divalent construction [SCU]_AGENT + verb +[for_NP]_ GOAL as illustrated in (5)–(7), despite the fact that appoint, nominate, and designate can be found in trivalent con- structions with an additional [NP]-complement as in (8)–(11).

(5)BNC It is certain that the vice-chancellor in the chair gained the impression from what was said at the meeting that they were forced to elect/ ?no- minate/*name/*?designate from a weak field of candidates. A68 1687 (6)BNC Keith Cox says the directors are up at Bolton tonight when they will look at the list of people who have already applied for the job and decide whether to appoint from that or advertise the post. K1Y 1000 (7)BNC ...they can, if they wish, elect/*name/*appoint/*pick/*nominate/ *designate for a voluntary audit, if this is needed for shareholders, lenders or creditors. CBY 3184 (8a)BNC If his successor at British Coal were appointed from within the corpo- ration he considers that it would be impertinent to tell him how the job should be done. A6L 1267 (8b)NSA If his successor at British Coal were nominated/designated from within the corporation he considers that it would be impertinent to tell him how the job should be done. (9)BNC So in 1972 a new secretary was appointed for the task, Anthony Gray, who worked away for some years and helped set it on its feet again. A7H 745 (10)BNC Prior to writing the script, I had made a list of the cast, designated local actors for the various roles and had them called for the 2pm “wood- shed” run-through. B11 1383 (11)NSA The main opposition leader could nominate a person for the post of Vice-President.

Note that pick can occur in sentences such as He picks for his sister. How- ever, for his sister does clearly not have the same role as for a voluntary audit in (7), i.e. he does not choose his sister but instead of her. [For_NP] does not seem to be a realization of the same participant in the trivalent combinations with appoint (9) and designate (10) and the divalent pattern with choose and select. While in combination with choose, select, and elect +[for_NP] is best interpreted as ÆFFECTED, i.e. “the thing elected” as in (12), (13), and (14) below, trivalent patterns have an [NP] complement Phenomena identified in the analysis 195 realizing ÆFFECTED, which only allows for an interpretation of [for_NP] as PURPOSE in such patterns as illustrated in (9)–(11) above.

(12)BNC What natural selection actually selects for, we now know, is the repro- ductive success of individual genes and in fact this explains all these figures. HUM 468 (13)BNC Choose for espaliers, fans, larger bushes and cordons on poor soil. A0G 1180 (14)BNC ...they can, if they wish, elect for a voluntary audit, if this is needed for shareholders, lenders or creditors. CBY 3184

Given that all verbs can be complemented with [NP]_ÆFFECTED, the fact that choose and select allow an alternative realization by [for_NP], cannot be accounted for by differences in the participant inventories of these verbs, either. Further cases of this phenomenon are presented below, first of all in the ‘intend’-group. Despite the fact that aim, intend,and plan have the par- ticipants ÆFFECTED and AIM, a combination of the two is not possible with plan. With respect to intend, no examples could be identified in the BNC for this combination, and native speakers did not agree about possible com- plementation patterns.

(15)VDE It’s ÆFFECTED not aimed/?intended/*planned at one particular party AIM, but rather at politicians as a breed.

Secondly, although recommend can be complemented with both [QUOTE] _EFFECTED and [to_NP]_RECIPIENT, a combination was neither identified in VDE or the BNC nor was it accepted in native speaker interviews. While recommend was classified as wrong right away, the native speaker judg- ments with respect to propose were inconclusive.

(16)VDE Interviewer Owen suggested/*recommended to Nesta: “The fact that you are here speaks volumes, I think.” (17)NSA ?Bishop Empey proposed to the synod: “Somehow we have to nail the lie that permissiveness flows from the Church of Ireland.”

Finally, quarrel, argue, and bicker allow combinations of TOPIC and CO- AGENT, two participants all four verbs have in common, whereas dispute is clearly not conventionally complemented in this way:

(18)VDE I beg you, do not argue/?dispute with me about this. (19)BNC In their first scene together, … , they dispute about the iconography of the goddess Fortune, Pistol in verse, Fluellen in prose, more coherently. CRV 166 196 Pattern choice and verb meaning

(20)VDE Credit for this trade was hard to get, since publishers argued over terms with distributors. (21)NSA ?He disputes with his colleague over everything. (22)BNC She heard the drone of some bees in the garden and the angry chatter of birds disputing over their seeds. H9H 1007 (23)VDE I beg you, do not argue/*dispute with me on this.

These cases clearly show that despite a general semantic similarity of the verbs and striking parallels as regards their participants, the concept of single complements as realizations of participants is not sufficient for a prediction of how they can be combined.

3.3.6 Frequency-based differences in the choice of complements

Another factor which obviously plays an important role and which is rarely taken into consideration when the problem of linking is discussed (except in usage-based approaches) is frequency. The question is often dealt with as if it were a matter of yes or no, even if in many cases the answer can only be something like “theoretically yes, but practically no” in the sense that there is a clear difference with respect to the frequency of alternative pat- terns. Select, for instance, can be found with the complement +[among_NP], which is also available to other verbs in the ‘choose’-group, yet this pattern appears to be very restricted with select. Instead, there seems to be a clear preference for the pattern +[from among_NP], which obviously expresses the same participant, i.e. one selects someone or something from among a group of people/things rather than among it. With choose, on the other hand, the latter pattern is completely conventional and it was also accepted in combination with elect.

(1a)VDE She was chosen/?selected among eighty highly qualified women who applied for £ 330,000 a-year part-time post. (1b)NSA She was selected from among eighty highly qualified women who ap- plied for £ 330,000 a-year part-time post. (2)NSA A further 100 seats were set aside for members to be elected propor- tionally among political parties winning at least 5 per cent of the vote, including 20 reserved for overseas Chinese. (3)BNC It also established a 150-member National Assembly elected from among the members of eight directly elected regional councils. HL4 2737 Phenomena identified in the analysis 197

Similarly, it appears to be much more acceptable or at least frequent to call someone on a number than at a number. While phone and ring can easily be combined with at, no such example could be identified with call. The fact that the verbs differ as regards the acceptability of the type of particle phrase speaks in favor of an analysis as complements – as opposed to ad- juncts – since the form of the complement appears to be determined by the valency carriers phone, call, and ring respectively.

(4)VDE She left a message for John McLachlan, telling him to phone her at that number at a certain time. (5)VDE Please phone Age Concern, on 081-965 7711. (6)BNC Anyone who can rekindle a friendship should write to him at Flat 11, Grove Court, Cooden Drive, Bexhill, Sussex or ring him at 0424 - 212456. A67 941 (7)BNC Anyone who has any information should ring Sandy on: (0767) 680551 and ask for the Gill Net Hot-line…ARE 657 (8)BNC He rejected the temptation to call Celia at the office. GUU 1381 = ‘LOCA- TION’ (9)BNC For more information about ACET's Home Care Service please call us on 081 840 7879 and ask for Jackie Sears. A00 291

The preference of one pattern over another pattern as opposed to a com- plete unacceptability of a pattern is a phenomenon that can be observed throughout almost all groups in the sample. Wasow (2002) points out the close connection between the restriction of a pattern and frequency:

What is the basis for distinguishing so sharply between possibility of cooc- currence and probability of cooccurrence? After all, if we were to encode probability of cooccurrence, then obligatoriness would just be one limiting case, and impossibility would be the other. And there is ample evidence that people know a great deal about the frequencies with which words appear in various environments. (Wasow 2002: 133)

Hoey (2005: 8) takes such frequency phenomena as motivation for arguing in favor of lexical priming, where “priming is seen as the property of the word and what is primed to occur is seen as shedding light upon the prim- ing item rather than the other way around”. Hoey also argues in favor of a strong idiosyncratic component because his corpus findings could not con- firm the hypothesis that “priming descriptions were to centre on the seman- tic set, [since] it would need to be the case that members of a semantic set should share the great majority of primings” (Hoey 2005: 63). Thus the 198 Pattern choice and verb meaning cases listed below clearly strike at the heart of the matter even if we are not dealing with restrictions to be accounted for by meaning. In chapter 2 the focus was on verbs which allow the same alternative complements, irrespective of the meaning of the verb. Thus the verbs ima- gine and intend, which can be complemented with both +[V-ing] and +[to_INF] to express a certain prospective activity, exhibited differences when it comes to pattern preferences.

Number of occurrences of IMAGINE and INTEND in the BNC

73 imagine [to_INF] 2,084 intend

514 [V-ing] 288

Figure 19. Frequency of complementation with [to_INF] and [V-ing] for the lem- mata imagine and intend

There is only a very limited number of sentences in the BNC with the pat- tern [SCU]+ imagine +[NP] +[to-INF] (73 occurrences) as opposed to [SCU]+ imagine +[NP]+[V-ing] (514). Intend, on the other hand, shows a clear preference for the pattern including [to-INF], with 7,084 example sentences as opposed to 288 for the pattern with [V-ing]. Now if the choice of a specific complement type were due to a correspondence between the meaning of the participant and the meaning of the verb one would expect similar graphs for semantically similar verbs.27 With respect to start and begin, the total frequency is fairly similar (start: 40,194 / begin: 42,142 [based on a lemma search]), however, the lemma begin is much more frequently complemented with one of these two complement types than the lemma start (total frequency of +[to_INF]/+[V- ing] with start: 13,726 / begin: 22,034). Start exhibits a fairly even distribu- tion with respect to these two complement types (cf. Figure 20). Compared to the semantically similar begin (both exhibit a high correlation with re- spect to pattern choice in general), an interesting distribution emerges. Not only is the choice of [to_INF] over [V-ing] opposite to the preference to be found with start (which is, as has been mentioned before, fairly balanced),

27. Similar analyses have been done under the keyword subcategorization prefe- rences of verbs. See for instance Hunston and Francis (2000). Phenomena identified in the analysis 199 the combination of begin complemented with a clausal complement type is far more frequent than the combination with start. For begin a clear prefe- rence of the pattern [SCU]+ begin +[to_INF] can be found (for similar ap- proaches see Biber, Conrad, and Reppen [1998]; Biber, Johansson, Leech, Conrad, and Finegan [1999], and Mindt [2000]). Number of occurrences of START and BEGIN in the BNC

6,207 start [to_INF] begin 19,204

7,519 [V-ing] 2,830

Figure 20. Frequency of complementation with [to_INF] and [V-ing] for the lem- mata start and begin

It appears difficult to relate this to the meaning of the two verbs. Looking at the lexical filling within the complements there seems to be a great deal of overlap with respect to the kinds of verbs to be found with begin and start in the two different patterns respectively. In the case of start more than half of the verbs which occur in the [V-ing]-construction as a complemtent of start are also among the most frequent collocates in the [to_INF]-con- struction (cf. get(ting), cry(ing), look(ing), make(-ing), go(-ing), laugh(-ing), run(-ing), take(-ing), work(-ing), come(-ing), and talk(-ing)), i.e. the choice of pattern does not coincide with a specific choice of lexical item describ- ing the initiated activity. The same applies to only three verbs with respect to begin (think(-ing), make(-ing), and walk(-ing)), i.e. here the lexical fil- lings seem to play a much more important role than in combination with start. What is of much greater interest, however, is the comparison of col- locates across the two verbs since this might mirror a difference in mean- ing: do the verbs choose different lexical fillings? Such a cross-comparison yields interesting insights: with respect to +[to_INF], start and begin share 8 out of 20 collocates (look, move, take, cry, fall, make, walk, and grow), with respect to +[V-ing] 10 out of 20 (taking, thinking, making, working, playing, writing, taking, shouting, shipping, building). This makes the cor- relation between pattern and lexical filling higher than that between valen- cy carrier (start vs. begin) and the lexical filling of the clausal complement, which might be taken as an argument against a difference in meaning, i.e. 200 Pattern choice and verb meaning the choice is more likely to depend (if at all) on the construction than on the meaning of the matrix verb. Also Ramchand (2008: 14) pointed towards the relevance of the lexical filling as regards the acceptability of a pattern. In reference to examples such as (10) and (11), she concludes that “[there] is no evidence that differ- ences in inference properties at this level are linguistic at all. Rather, it seems more as if language allows different structures, but the real world determines felicity and detailed inferential patterns.”

(10)QE John painted me a picture. vs. ? John painted me a wall. (11)QE John painted a wall red. vs. ? John painted a picture red.

Some of the lexical fillings (e.g. make and work) occurred in all patterns with both verbs.Yet while begin to make is the most frequent combination following the general trend of the overall higher frequency of begin as compared to start, this is not the case across all lexical fillings.

Number of occurrences in the BNC start to 131 make start making 185

begin to make 298

begin making 55

Total: 669

Figure 21. Make/making as collocates of START/BEGIN +[V-ing]/[to_INF]

When one wants to express that some kind of work commences,start work- ing is clearly more frequent than any of the other possible combinations, including begin to work. This distribution is not an anomaly, a very similar picture can be found for instance with start/begin + talk (see Figure 23 below). At the same time, there are of course numerous lexical fillers which show a distribution as expected from the general distribution, for instance ‘commence’ + think/look/cry/take, all of which prefer the combination with as indicated in the illustrations below. They do differ however, in their preferences with respect to pattern choice in combination with start. While the choice for begin is fairly stable, start +[V-ing] and Phenomena identified in the analysis 201 start +[to_INF] show an even distribution with respect to cry and take, whereas look – and even more so think – clearly prefer complementation with +[V-ing]. Number of occurrences in the BNC

start to work 84

start working 151

begin to work 84

begin working 70

Total: 389

Figure 22. Work/working as collocates of START/BEGIN +[V-ing]/[to_INF]

Number of occurrences in the BNC

start to talk 52

start talking 256

begin to talk 136

begin talking 52

Total: 496

Figure 23. Talk/talking as collocates of START/BEGIN +[V-ing]/[to_INF]

Number of occurrences in the BNC

start to think 79

start thinking 195

begin to think 375

begin thinking 29

Total: 678

Figure 24. Think/thinking as collocates of START/BEGIN +[V-ing]/[to_INF] 202 Pattern choice and verb meaning

Number of occurrences in the BNC

start to look 127

start looking 192

begin to look 346

begin looking 26

Total: 691

Figure 25. Look/looking as collocates of START/BEGIN +[V-ing]/[to_INF]

Number of occurrences in the BNC

start to cry 85

start crying 83

begin to cry 145

begin crying 9

Total: 322

Figure 26. Cry/crying as collocates of START/BEGIN +[V-ing]/[to_INF]

Number of occurrences in the BNC

start to take 111

start taking 112

begin to take 379

begin taking 40

Total: 642

Figure 27. Take/taking as collocates of START/BEGIN +[V-ing]/[to_INF]

Thus the likelihood of expressing the meaning ‘commence’ + LOOK from most likely to least likely is Phenomena identified in the analysis 203

BEGIN to look Æ START looking Æ START to look Æ BEGIN looking while for ‘commence’ + WORK it is

START working Æ BEGIN to work Æ START to work Æ BEGIN working

A comparison of the two verbs by Mindt (2000) shows that the choice of pattern also differs from genre to genre, with start +[V-ing] being clearly most frequently used in conversation (71%), while only 4% of the begin +[V-ing] combinations are to be found in this genre. As a semantic difference in connection with these two verbs, Rericha (1987: 130) suggests that +[to_INF] focuses on the not-completed activity, whereas +[V-ing] “implies the initial phase of a repeated activity or an activity which is fully developed in its initial stage.” Schmid (1993: 267– 269), on the other hand, sees the difference between start and begin – rather than between the two complement types – in that start is more oriented to- wards agentive and dynamic situations and begin towards the stative and cognitive domain. Yet such explanations cannot account for the distribution presented above either. The apparent arbitrariness is also reflected in the quote by Quirk (1974: 166–167): “[t]here ought to be a big award for any- one who can describe exactly what makes him say ‘I started to work’ on one occasion and ‘I started working’ on another.” Mair (2003: 329), in a comprehensive overview on this alternation in combination with start and begin, describes this as a phenomenon being “in a state of flux diachronically, with -ing-forms gradually encroaching on the infinitive …”. His analysis yielded interesting regional differences in that American English, as opposed to British English, exhibits a clear increase in the use of +[V-ing] in written language and a decrease in +[to_INF]: in written English “gerund complements after begin are a statistical Ameri- canism” (Mair 2003: 338). As regards semantic determinants Mair (2003: 340) concludes: “Any semantically based account thus seriously overde- termines the observed distribution.” Thus even if there were slight semantic differences, they could not be taken as the basis for predicting pattern choice as long as they remain so subtle as in the case of this example.28 What this comparison shows is that it is certainly extremely difficult to account for such differences in complementation on the basis of the mean- ing of the verb functioning as valency carrier. This is also the result of a study conducted by Klotz (2007) on the distribution of what he calls propo-

28. For a similar concept which focuses on different preferential choices more generally cf. the notion of probabemes in Herbst and Klotz (2003: 145–153). 204 Pattern choice and verb meaning sitional arguments. Grouping verbs together according to their qualitative valency with respect to the valency slot expressing the participant PROPOSI- TION and, independently of this, grouping these verbs into semantic groups, he correlates the type of alternation a verb allows with possible semantic groups. With respect to the two groups (1) permitting [that_CL] and [NP to_INF] and (2) permitting only [that_CL], Klotz (2007: 126) concludes that “[it] is immediately apparent that both complementation classes co- occur with all [but one] semantic groups … Apart from this there is very little to suggest that the realizational possibilities of a propositional argu- ment could be predicted from the meaning of the verb to which it belongs”. Comparing what he calls communication verbs and opinion verbs, he finds a slight preference for class (1) yet his overall conclusion clearly empha- sizes the general tendencies also found here:

Assuming that this distribution is not just coincidence, we are still nowhere near anything that might be called a regularity … In sum it seems fair to say that the statistical analysis of complementation data from the Valency Dic- tionary of English does not lend any support to the view that the valency of a verb can be deduced from its meaning. The storage view of valency which sees it as an irregular lexical rather than semantically rule-based phenome- non is strengthened further. (Klotz 2007: 127)

Wasow (2002) analyzed data from the New York Times to find out about a possible lexical bias in the dative alternation. He found that some verbs are more likely to occur in one of the two constructions than in the other and does not exclude that there is a semantic explanation for a general prefe- rence for one pattern over the other in connection with specific verbs. However Wasow (2002: 88) concludes that

…[a]lthough I would like to be able to provide such an explanation, I do not see how to do it. In the absence of a deeper explanation, I conjecture that speakers’ mental lexicons include information about the tendency of par- ticular verbs to appear more or less frequently in various syntactic con- texts”.

If he is right, semantics would not really facilitate the choice but the prefe- rences would have to be stored alongside other syntactic information. Mair (1995: 260) examined pattern preferences with the help of a cor- pus-based study with the Brown Corpus (American English from 1961) and LOB Corpus (British English from 1961) and FLOB (British English from 1991). In an analysis of the verb help and its occurrences in the two parallel patterns +[to_INF] and +[INF], he found that over the past thirty years help Phenomena identified in the analysis 205 complemented with a bare infinitive has become more frequent in British newspaper language than it used to be, although it had been more frequent in American English than in British English (Mair 1995: 261–264). Even if there might be some reservations as far as these results are concerned due to the size of the corpora, both tendencies, the dialectal difference and the change in frequency, do not speak in favor of a choice of the complement on the basis of the meaning of the verb. It appears rather unlikely that help has a different meaning in American than in British English. Equally, to argue that the meaning of the verb help changed in between 1961 and 1991 so that this could account for a change in pattern choice is not very con- vincing. Wierzbicka (1996: 391) argues that cross-linguistic comparisons are not valid for judging the predictability of form on the basis of meaning since “[i]t is the meaning that is predictive not the denotation”. Meaning can thus still differ due to different conceptualizations, which “may be ex- plainable in terms of history, culture, living-conditions, religions and so on” (Wierzbicka 1996: 391). Thus meaning according to Wierzbicka (1996: 391) includes those aspects which “underlie certain ‘semantic rules’”, i.e. which are predictable. However, if meaning is considered to predict form but only those aspects which predict form are acceptable as “meaning”, the definition becomes obviously rather circular (cf. Palmer 1990 for a discus- sion). If a larger time span is taken into consideration, it becomes quite clear that strictly speaking what is described as the pattern inventory or the pat- tern choices of a verb is actually in flux. Fanego (1996: 55), for instance, shows that the use of +[V-ing] as a complement to subject-control verbs became “an established feature of English usage” “only from the second half of the 17th century onwards”. This makes identifying criteria for the occurrence in one but not in another pattern much more difficult. Fanego (1996: 57–59) lists as possible “gerund-promoting properties” the “type of verb” (e.g. verbs of avoiding appear to be more “hospitable to the gerund”), the “informality of the text” and the “synsemantic environment” (e.g. “neg- ative or non-assertive contexts containing the modals can/could”). Howev- er, she concludes that

… it must be acknowledged that, in certain cases, one gets the impression that the two constructions were probably felt to be very much alike, so that the ultimate choice of either one or the other was prompted by considera- tions of rhythm, style, or perhaps even individual preference. To, conclude, then, it seems to be that the grammatically regulated use of the gerund and the infinitive, if this indeed can be proved to exist in Present-day English, must be a later development. (Fanego 1996: 59–60) 206 Pattern choice and verb meaning

There are clearly preferred patterns as far as frequency is concerned, which is interesting from the point of view of storage: one could assume that a more frequent pattern is stored whereas a less frequent one is created, e.g. in an ad hoc fashion in analogy to other verbs. Frequency might also play a role in the sense that a verb which is rather infrequent in a certain meaning as compared to other members of the same group could “behave” differently than other verbs, e.g. by allowing a smaller range of patterns. With respect to the present sample a direct com- parison is not possible, since what has been compared so far is not the number of all possible patterns compatible with a lexical item but only those patterns which are associated with the lexical units with the respec- tive group meaning. While the BNC provides information on the raw fre- quency of lexical items, it is obviously not possible to limit this to the lexi- cal units of interest by an automatic lemma query, especially considering the fuzzy nature of these categories. Moreover, tagging mistakes in connec- tion with words such as state which can be both a verb and a noun must be taken into consideration. Nonetheless, there are some tentative statements that can be made in this respect: there are cases where frequency plays a role in the sense that a verb with a high token frequency also exhibits a higher type frequency with respect to complementation patterns. Remember which occurs 25,635 times in the BNC (lemma search) has been found in fifteen different valency patterns, whereas recollect, with only 214 oc- currences in the BNC has only been found acceptable in ten. Similarly, al- low (33,577 hits) can be found in thirteen patterns and permit (4,440 hits) only in eleven, and love (14,219 hits) is acceptable in nine patterns as op- posed to adore (502 hits) with six patterns. At the same time, cases where such a connection between token and type frequency does not seem to play a role, have also been identified: presume, which is much rarer than sup- pose (1,156 as opposed to 11,493 hits), can be found in the same number of patterns as suppose and the same applies to recommend (5,882 hits) and suggest (28, 246 hits) (pattern number always in reference to the group meaning). All in all, the examples in this chapter clearly show that neither parallels nor differences in meaning can account for differences in preference of certain patterns and other frequency phenomena in a satisfactory way. As regards the frequency of verbs, it is not possible to confirm a tendency for a correlation between type and token frequency for this sample. Phenomena identified in the analysis 207

3.3.7 Formal similarity between verbs of opposite meaning

Another area of interest when comparing the patterns of semantically simi- lar verbs are verbs of opposite meanings since they are of course also se- mantically related. Goldberg (1995) argues that the participants of verbs with opposite meaning can still practically be the same and that the only difference is to be found in the fact that the goal, instead of being achieved, is simply not achieved. Thus, deny and refuse are acceptable just as give and pass in the notion of transfer since they simply express “X CAUSES Y not to RECEIVE Z” (Goldberg 1995: 38–39). If the participants remain the same, they can, according to Goldberg, still fuse with the argument struc- ture slots in the respective argument structure constructions. Indeed, verbs with opposite meanings can sometimes be used in the exact same patterns:

(1)VDE You should remember to replace your child’s toothbrush every three months or so. (2)VDE Don’t forget to leave your name, age and address. (3a)NSA Let’s remember him as he really was. (3b)NSA Let’s not forget him as he really was. (4a)VDE I like it when it snows. (4b)NSA I don’t mind it when it snows. (5a)BNC What I'd really like for us to have, and this is some time in the future, is one nice sort of saloon type car for best. KD4 798 (5b)NSA What I really wouldn’t mind for us to have, and this is some time in the future, is one nice sort of saloon type car for best.

This has also been seen as one of the great advantages of construction grammar since it enables us to link verbs like give and deny:

… the ditransitive construction is regarded as a prototypical category which also includes verbs that cannot be easily mapped onto – but are still taken to be related to – the semantic core of ditransitivity, e.g. deny. This prototypi- cality approach makes it possible to also link verbs such as envy, (be)grudge and spare, which according to Gropen et al. (1989: 241) could be subsumed under ‘future not having’, to the ditransitive transfer event. (Mukherjee 2005: 51)

However, it has to be doubted whether this really holds across all possible patterns antonymic verbs have. Compare:

(6a)VDE She is perhaps best remembered for her creation of the Belgian detec- tive, Mr Hercule Poirot. 208 Pattern choice and verb meaning

? (6b)NSA * She is not forgotten for her creation of the Belgian detective, Mr Her- cule Poirot.

One might argue that it is world knowledge that tells us that when we for- get something we usually do not go through this process consciously and thus usually do not have a specific reason to be connected with what we forget. Thus this is not part of the meaning of forget. However, such an explanation is much more difficult in case of the examples below. One could easily imagine a CAUSE for not minding something

(7a)VDE She used to love it for us to go and take the children out. ? (7b)NSA *She didn’t use to mind it for us to go and take the children out. (8a)VDE I would love it if my family could be more comfortable with themselves. (8b)NSA ?I wouldn’t mind it if my family would be less intrusive.

Another interesting case can be found with respect to start and stop: while start can be complemented with both +[V-ing] and +[to_INF], expressing ÆFFECTED, stop only seems to allow +[V-ing]. A corpus search for the string stop + to_INF yields quite a considerable number of hits, yet the to- infinitive does not express the same participant:

(9)VDE Consequently, bright children can get bored or start to hate school. (10)BNC Birdwatchers should also stop to look at The Stack on Scarlett Point -- a favourite with seabirds. A65 539

While in (9), the complement could be replaced by +[V-ing], this is not the case in (10), which rather means stop in order to look at, i.e. quite the op- posite of stop looking. Although the examples presented here do not provide a basis broad enough to thoroughly discuss the overlap in patterning between verbs of opposite meanings, the examples show that, even if there is a certain cor- respondence, other factors which might not be as predictable have to be taken into consideration as well.

3.4 Different aspects of verb meaning and pattern choice

The previous chapter mainly discussed different phenomena which emerged in the course of the comparison of semantic verb groups. These indicate that there are a considerable number of factors which have to be taken into account when the syntactic potential of a verb is to be explained. They also show that this potential cannot be accounted for by assuming a Different aspects of verb meaning 209 direct relationship between the meaning of a verb and its pattern inventory. To do justice to the many theories taking different semantic aspects as the shaping force when it comes to pattern choice this chapter will also syste- matically address these aspects with respect to the verb groups and care- fully examine each of them in order to see to what extent they play a role for the verbs analyzed here.

3.4.1 The participant inventory

The notion that the meaning of a verb is (more or less) determined by the number and type of its participants is part of various semantic theories. Fillmore (1970), for instance ascribes a crucial role to the presence or ab- sence of certain participants by claiming that

…two verbs can differ in that one manifests an n-place predicate and the other manifests and m-place predicate, the roles of the arguments that are present in the one and absent in the other accounting for differences in the semantic interpretation of the sentences which contain them. … The role by which KILL differs from DIE, and that by which PERSUADE differs from BE- LIEVE is that of the individual that is ‘agentively’ involved in the events named by these verbs. Apart from this difference, we are dealing here with pairs of synonyms. (Fillmore 1970: 259) [emphasis mine]

Welke (2005: 50) argues that semantic differences between verbs can be described in this respect: “Verbs differ according to the number of their arguments. Thus they can be differentiated semantically already by the mere number of arguments which they are laid out for [translation mine]”. In a frame-semantics approach (e.g. Fillmore’s FrameNet project), semanti- cally similar verbs have the same frame, i.e. the same list of participants, but they can choose the perspective of different participants. However, the perspective of which participant is expressed in combination with which verb is considered to be idiosyncratic (Fried and Östman 2004: 43). More- over, even if the participant inventory of a verb provides for a number of slots and a number of semantic roles, it is not clearly predictable how they can be formalized. While there are roles which are fairly limited as regards their structural function in a clause, as for instance AGENT which is typi- cally realized as grammatical subject of an active declarative clause or by a by-phrase in a passive clause, this does not apply to other roles quite so easily. While most approaches focus on the linking of semantic role to grammatical function, such as subject and object, constituents having other functions or being realized by phrases other than noun phrases are rarely 210 Pattern choice and verb meaning taken into consideration. Even in the analysis of subjects and objects a great deal of work exclusively focuses on noun phrases. The possible formal variability even within these two functions is rarely taken into considera- tion or dismissed as peripheral phenomena. The analysis in the preceding chapters reveals that the verbs are indeed very similar with respect to the participants they have, even if their partici- pant inventory was not directly taken into account in the constitution of the different groups (cf. 3.1). They also exhibit considerable similarity with respect to the kinds of patterns they can occur in and a fair number of the restrictions which were nonetheless identified can indeed be accounted for by the fact that the verbs have similar but not necessarily identical partici- pant inventories. However, the majority of the restrictions identified (55%) cannot be explained by means of differences in the participant inventory or other semantic features of the respective verbs (cf. 3.2.3). There is only one single group in this sample, where all restrictions can be accounted for by such systematic differences in the participant inventory (the ‘not lose’- group). This shows that the participant inventory of a verb is far from reli- able when it comes to predicting the patterns available with it. Again this can only lead to the conclusion that storage must play an immensely important role here. Moreover, the mere existence of formal valency patterns as realizations of quite different participant patterns – de- spite an obvious similarity in the overall meaning of the verbs –, seems to indicate that accounting for the connection between a verb’s valency and the meaning of the verb is problematic. Croft (1998: 82) argues that se- mantic roles “are simply not helpful for predicting the semantics of gram- matical relations: Almost any semantic role can occur in almost any gram- matical relation. Yet, intuitively, the choice of subject, object, and oblique in English (and in other languages) is not chaotic”. Thus explaining the patterning of a verb exclusively on the basis of the semantic roles the verb is typically associated with does not appear to be very promising. Also Levin (1993), for instance, argued with respect to her study of verbs and their patterning behavior that “the complex pattern of behavior manifested by verbs with respect to diathesis alternations cannot be ex- plained with a lexical semantic representation that takes the form of seman- tic roles” (Levin 1993: 16). Another argument against an explanation resting solely on a verb’s par- ticipants is the realization that additional elements can be added which are not necessarily assumed to be part of the participant inventory of a verb. Different aspects of verb meaning 211

Sometimes participants are added ... [which] cannot be projected by the lex- ical predicate (i.e., they are not even in the inventory of frame elements as potential additions) and ... the addition does not the meaning or inter- pretation of the lexical predicate ... These patterns pose a serious challenge to syntactic theories that assume the clause structure to be always a direct projection of verbal argument structure. (Fried and Östman 2004: 52–53)

It is for this reason that construction grammar along the lines of Fillmore and Kay “does not subscribe to the notion of a universal hierarchy of se- mantic roles that would automatically predict for any given valence in any given language which semantic role is the privileged one” (Fried and Östman 2004: 44). Moreover, the presence of one role is often dependent on the presence of another role, i.e. it cannot be regarded in isolation as occupying a certain position in a hierarchy:

[A] hierarchy of isolated roles takes no account of their relational na- ture…The idea of an event notionally parsed into, say, a location and an ex- periencer is incoherent because “experiencer” makes no sense without the presence the cause or content of an experience and “goal” makes no sense outside of the kind of scene in which some patient or theme (potentially) moves toward some intended destination. (Fillmore and Kay 1993: 8.21 as quoted in Croft 1998: 29)

Goldberg (1995: 226) takes this insight as the justification for a separate level of construction meaning: “It has been suggested that the degree of regularity in the relation between semantic role types and overt syntactic expression is sometimes exaggerated, and that many linking generalizations are construction specific”. According to this form of construction grammar, profiled participant roles must be fused with argument roles that are rea- lized as grammatical functions. Profiled refers to the focus put on certain participants, e.g. steal profiles the stolen goods whereas rob profiles the victim (Goldberg 1995: 45–48), making them obligatory participants. Ac- cording to Goldberg (1995: 53–54), only profiled participants are a contri- bution from the verb meaning to the overall meaning of verb and construc- tion, while the construction can add additional arguments for which there are no profiled participants, e.g. Joe kicked Bill the ball, where Bill is not profiled in the verb meaning. Profiled participants of a verb can fuse with the arguments in the argument structure construction, while this fusion can be seen as the licensing of a complement within a pattern in combination with a specific verb (Goldberg 1995: 50–52). Yet such differences as re- gards the focus a participant receives does not always make the participant obligatory, especially since there are cases where two different participants 212 Pattern choice and verb meaning can be highlighted, as in the case of teach which can put the focus on the material taught and the learner, whereas instruct can only highlight the learner. For Goldberg, it is a combination of lexical semantics, i.e. a verb’s par- ticipant roles, and constructional semantics, i.e. the argument roles of the construction, which explains the licensing of a pattern for a verb. However, the lexical specifications which could hinder or allow the fusion of the par- ticipants with the argument roles are not defined or specified in order to account for restrictions (Boas 2003: 106). Boas (2003: 107), who put her assumptions to the test in a study of resultative constructions, concludes that “Goldberg’s amount of frame semantic information is not sufficient to describe the full range of attested resultative constructions while ruling out unattested resultative constructions”. As has been shown in detail in 3.3.4, many of the restricted patterns analyzed seem to express participants all verbs have but the verbs differ with respect to the range of possible patterns they allow for the expression of these participants. The following chapter will take into account a semantic notion which is often closely connected to the description of participant roles, i.e. selection restrictions, in order to check to what extent differences with respect to selection restrictions match with pattern restrictions and sameness of selec- tion restrictions with pattern overlap.

3.4.2 Selection restrictions

Selection restrictions have been referred to in various linguistic models – from generative grammar to valency grammar – for various purposes, while the general notion implies that “argument terms underlie semantic restric- tions due to logical conditions imposed on possible arguments terms” (Löbner 2002: 114). Thus verbs do not only specify participants, they also seem to establish restrictions on the possible semantic “cast” of such par- ticipants. Hence a verb such as murder needs both a living ACTOR and a (previously) living EXPERIENCER, thus excluding plants and objects. More- over, murder is most likely to have a human being as EXPERIENCER; an animal might be acceptable under certain circumstances but is rather un- usual (cf. Löbner 2002: 116). Thus the restrictions for the EXPERIENCER of the action murder are [+alive at the outset, –alive afterwards, +human]. However, a brief look at corpus material provides us with several examples where inanimate objects can be murdered, e.g. thing, music, and hope.

(1)BNC Rhetoric was a thing he would gladly have murdered; A1B 976 Different aspects of verb meaning 213

(2)BNC Some will argue that it was the decade when “real” music was mur- dered by Stock, Aitken & Waterman or rap, but on the other hand the “real” musician of mature years had never had it so good... A8F 128 (3)BNC She murdered his hopes, didn't she? ECU 2352

This already demonstrates that it is far from easy to establish what consti- tutes a real selection restriction and not just prototypical realizations. Löbner (2002: 117–118) explains this by means of “meaning shifts”, for instance by metonymy (“Moscow declares the Chechen rebels defeated” where a geographical term becomes an AGENT) or metaphor (“His courage evaporated”, where the verb meaning is shifted from “vanishing of a phys- ical substance” to “vanish completely”), thus accepting complements that fulfill the selection restriction of this metaphorically created verb meaning. Chomsky describes selection restrictions as distinct from subcategoriza- tion principles (1965: 95, 113–120) and Helbig (1971: 38–39) covers such semantic aspects by what he calls the third step or layer in his valency model besides a first quantitative layer and a second layer covering the formal realization of complements, and selection restrictions still assumes a separate level in his extended six-layer model (Helbig 1992: 154–155). Selection restrictions are usually seen as part of the meaning of a verb and are to some extent inseparably tied to a verb’s participant roles (see above). Fillmore (1968: 24) defines AGENTIVE on the basis of the selection criterion +animate, i.e. it is part of the role definition, even if exceptions like “‘inanimate nouns’ like robot or ‘human institutions’ like nation” are acknowledged (Fillmore 1968: footnote 31). Classifications of selection re- strictions range from rather general restrictions similar to those used for example by Fillmore, i.e. +ANIM or +HUM, to more specific elements, e.g. the fact that the complement of drink must be fluid (Helbig 1971: 38– 39). Also Pinker (1989), based on Talmy’s ([1985] 2007) list of semantic elements of verbs in different languages, includes various properties of a possible [NP]-complement in his list of “syntactically relevant features of meaning” such as “animate vs. inanimate, human vs. nonhuman, dimensio- nality (0D vs. 1D vs. 2D vs. 3D extendedness), count/mass, rigid/flexible, substance/aggregate, liquid/semisolid” (Pinker 1989: 185–187, 208–209).29

29. Groefsema (2001: 528–529) points out that Pinker (1989) cannot account for all alternations in this way and thus has to introduce additional morphophono- logical rules. Such rules, however, are rejected, for instance, by Goldberg (2006b: 404) as they “largely coincide with distinctions between Latinate and native vocabulary”, so that this would imply that children must have recourse to etymological information in the process of language acquisition. 214 Pattern choice and verb meaning

While the list appears rather extensive, it does not really serve the purpose of sufficiently describing inanimate, abstract entities, such as respect or honor, which constitute a considerable part of typical post-verbal noun phrases for the verbs under consideration. Selection criteria listed in CGEL include: plurality/singularity (applicable to both subject and predicate com- plement units), concrete/abstract, animate/ inanimate, human/nonhuman (CGEL: 10.51). As mentioned above, only 3% of all restrictions in the sample can be accounted for on the basis of a difference with respect to selection restrictions. Since the role definitions of the participant roles used in the analysis above do not directly include any selection restrictions, a separate analysis of selection restrictions appears necessary in order to show to what extent the verbs differ in this respect in the first place. In order to show what kind of role selection restrictions play, it is necessary to examine those groups which actually differ in this respect, in order to see to what extent syntactic restrictions might still be related to this aspect of meaning. For a comparison of the verbs in this study, the verb groups are analyzed as to possible selection restrictions of the post-verbal noun phrase in the pattern [SCU]+ verb +[NP]. On the one hand, this pattern was chosen be- cause it is one of the most frequent patterns, and, on the other hand, [NP] is a complement where selection restrictions are likely to occur.30 The selec- tion restrictions used are those listed in CGEL (see above). Animate is tak- en as a superordinate term for human unless there are clear restrictions with respect to nonhuman participants. Table 40 lists all verbs in their respective groups which allow complementation with [NP] and a semantic classifica- tion of this complement based on examples from VDE and the BNC.31

30. For other restrictions see also Rudanko (1989: 34), who found that in order to allow complementation with +[to_INF] the subject of the matrix clause is typ- ical +animate and the action volitional but there are also counterexamples. 31. Several cases were not taken into consideration, e.g. when the restriction is possible with a different meaning of the same verb (e.g. recollect +animate as in recollect oneself, consider +animate [= ‘take into consideration’] or judge +animate [= ‘assess’]) or when the restriction applies to other non-divalent patterns, as in the case of allow and authorize and mean, intend, and plan which can be complemented with a +animate noun phrase in the pattern +[NP] +[as_NP]. Thus this classification is first and foremost concerned with divalent patterns ([SCU]+ verb +[NP]) and not with trivalent patterns ([SCU]+ verb +[NP]+[X]). Likewise, sentences such as ‘Guess who’ are left aside since who does not refer to the person but rather an action it is involved in. Different aspects of verb meaning 215

Table 40. Selection restrictions of post verbal [NP]-complements

selection restrictions of meaning verbs [NP] in the pattern [SCU]+ verb +[NP] ‘answer’ answer +/–animate reply, respond ––– ‘assume’ assume, suppose, presume, guess –animate gather ––– ‘choose’ appoint +animate choose, elect, pick, select, name, +/–animate designate nominate +/(–)animate ‘declare’ declare, pronounce, proclaim, –animate state announce +/–animate ‘get’ obtain, acquire (+)/–animate receive, get, buy, purchase +/–animate ‘hire’ hire +/–animate rent, lease –animate ‘intend’ mean, intend, plan –animate aim –––

‘like’ love, like, adore +/–animate

‘not lose’ keep, preserve, sustain +/–animate, +/– abstract maintain –animate, +abstract

‘permit’ allow, authorize –animate permit –/(+)animate32 entitle –––

‘practice’ train +animate practice –animate rehearse –/(+)animate

32. +Animate is only possible for BENREC as in Children under 14 are not per- mitted in bars except when passing…BNC: HC4 724. 216 Pattern choice and verb meaning

selection restrictions of meaning verbs [NP] in the pattern [SCU]+ verb +[NP] ‘propose’ propose, suggest –animate recommend +/–animate

‘quarrel’ quarrel, argue, bicker ––– dispute –animate ‘remember’ recall, recollect –animate remember +/–animate reminisce ––– bear in mind +/–animate

‘show’ indicate, demonstrate –animate show +/–animate

‘start’ start, commence –animate, +/–abstract begin –animate, +abstract

‘teach’ instruct +/–animate teach, train, educate +animate

‘telephone’ phone, ring, call +animate (or institution)

‘think’ think, consider, ponder, judge –animate reflect ––– contemplate +/–animate

‘think highly’ respect, admire, esteem +/–animate

‘tolerate’ bear, tolerate +/–animate endure –animate ‘urge’ urge +/–human press –animate push +/–animate

First of all, verbs in the sample which exhibited an absolute overlap of pat- terns compatible with the verbs were compared. There are verbs which allow the exact same range of patterns and also have the same selection restrictions. This applies to: suppose and presume; buy, purchase, and get, (and possibly obtain); love and like; phone, ring, and call; respect and ad- Different aspects of verb meaning 217 mire; rent and lease. In case of buy, purchase, get, and obtain, all four verbs can be complemented with +/–animate [NP]. However, in combina- tion with obtain, the selection restriction +animate is confined to certain idiomatic expressions such as a male obtains a female or someone obtains someone as a bride. Generally, such cases seem to confirm the correlation between pattern choice and selection restrictions. However, it has to be taken into account that in three of the five cases there are other verbs in the group which also have the same selection restrictions but do not allow the same range of patterns, e.g. guess and assume, adore, and esteem. Thus there must be further aspects which play a role. However, most verb groups are not completely identical as far as their selection restrictions are concerned, either. However, there are many sub- groups which are, i.e. very often it is a single verb in the group that seems to differ from the rest. Comparing all groups and subgroups which seem to have the same selection restrictions with respect to [NP] in divalent pat- terns, the following categories can be formed: (1) verbs with the same se- lection restrictions also select the same patterns (see above); (2) subgroups of verbs with the same selection restrictions – within a larger group of verbs with different selection restrictions – exhibit a greater degree of pat- tern overlap than compared to the rest of the group, i.e. verbs with the same selection restrictions are more similar as regards their pattern choices than verbs with different selection restrictions in the same group; and (3) wheth- er the verbs have the same selection restrictions or not has no effect on the number of pattern restrictions. Consider, judge, think, and ponder (without reflect), for example, have the same selection restrictions and as opposed to other verbs in this group two additional patterns, i.e. [SCU]+ verb +[wh-to_INF] and [SCU]+ verb +[Q/S]. However, contemplate, which also differs from the rest with re- spect to selection restrictions, can be found in the first pattern as well.

(4)VDE Mrs Thatcher also said NATO must consider how to extend its role from one of preventing war to one of building peace. (5)BNC In practice, judging when to close is difficult. K94 780 (6)BNC But initially she couldn't think how to raise the money for it -- until she hit on the idea of Greener Gifts. A70 133 (7)BNC Dorothy, 65, of Hinckley, Lincs, is putting the £10,000 in a safe place -- the bank, not behind a cushion -- while she ponders/*reflects how to spend it. CH1 7501 (8)BNC I have a streak of economy in me, even when contemplating how to épater les bourgeois, .... G2E 2497 218 Pattern choice and verb meaning

Teach, train, and educate (without instruct) have one additional pattern in the sense ‘teach’. Instruct allows the same pattern but in the sense ‘order’. That this is connected to the fact that instruct allows + and –animate [NP]- participants and the other members of the group only +animate appears highly unlikely.

(9)VDE They were still trying to teach her to read and write. (10)BNC A note from Mellowes instructed me to cast my eye over the draft, pron- to, for inaccuracies. A0R 641 = ‘order’

Keep can be used to describe the holding on to a concrete object (animate or not) or something abstract, while maintain has the meaning of holding on to something abstract only (I would never keep a vicious dog!VDE and I was able to keep control of my emotions.VDE as opposed to The government will maintain tight control over the money supply.VDE). Keep, preserve, and sustain (without maintain) have one additional pattern, but this was already accounted for by a difference in the participant inventory of maintain, which does not have any ways of expressing an AIM or GOAL. Again, there is no connection to selection restrictions.

(11)NSA *Please read these notes carefully and maintain them for future refer- ence.

Assume, suppose, presume, and guess (without gather) have one additional pattern; the overlap is highest, however, when only assume, suppose, and presume are compared, i.e. guess, which does not differ with respect to selection restrictions, is still more restricted in its formal realization than these other three verbs. The additional restrictions of gather can be ac- counted for in reference to a difference in its participant inventory.

(12)NSA *One could gather the prototype for humanity to be female rather than male. [assume in the original VDE sentence]

Recall and recollect (without remember, bear in mind,and reminisce) have eight additional patterns. Yet when remember, which has different selection restrictions in that it can be complemented with a +animate [NP], is also taken into account, the overlap is even higher between remember and recall than between recall and recollect. Similarly, demonstrate and indicate (without show) have one additional pattern. Yet when indicate and show are compared they have even more patterns in common. Different aspects of verb meaning 219

Cases like these clearly show that a reduction of verbs in a group often leads to a higher overlap and this is obviously not directly connected to the selection restrictions. The more verbs are compared, the more restrictions typically have to be accounted for. The biggest group, however, consists of those cases where whether the verbs have the same selection restrictions or not has no effect on the number of pattern restrictions. For example, start differs from commence and begin in that it is not li- mited to +abstract [NP], i.e. one can, for instance, also start an engine which does not seem to be possible with begin and commence. However, this does not reflect in the patterns these verbs allow: the overlap between start and begin is higher than that between begin and commence. Propose and suggest (vs. recommend) have the same selection restrictions but still unrelated pattern restrictions. The same applies to choose, elect, pick, se- lect, name, and designate (vs. appoint and nominate), and bear and tolerate (vs. endure). In the case of appoint, the selection restriction must be seen as part of its meaning which implies that a person is chosen “for a job, esp. for an official position” (VDE: 43), and this cannot apply to a non-human complement. There is basically one exception to this, i.e. the appointing of a time or place. Here one could either argue for a different lexical unit as has been done in VDE, or for an additional selection restriction. Allow and authorize, which have the same selection restrictions, have fewer patterns in common than allow and permit, which have different selection restric- tions. Finally, all members in the ‘like’ and the ‘think highly’ group have the same selection restrictions but formal restrictions can still be identified. The fact that the groups for which no difference can be found is the big- gest already shows that the effect of selection restrictions cannot be deci- sive: there does not seem to be a convincing link between a general simi- larity in meaning, similarity with respect to selection restrictions, and pat- tern choice for this sample. Moreover, in none of the groups where the pat- tern overlap was higher when a “deviating” verb was discarded, could the restrictions really be shown to be connected to the selection restrictions in the first place. If selection restrictions could help to predict whether a verb can occur in a pattern or not, semantically similar verbs which in general exhibit a great- er formal overlap, should also exhibit greater similarity in the selection restrictions than those which formally differ. The other way around seman- tically similar verbs allowing the same selection restrictions should not differ with respect to the patterns they allow. The data above shows that neither of these claims could be verified. Formally and semantically similar verbs do not necessarily have the same selection restrictions. Moreover, 220 Pattern choice and verb meaning semantically similar verbs with the same selection restrictions do not ne- cessarily occur in the same patterns, even if it has to be taken into account that in the analysis above the focus was on a pattern which does actually occur with the verbs under consideration. Nonetheless, this might throw light on possible differences between otherwise semantically similar verbs which should have an influence on their overall pattern inventories. Moreover, further restrictions identified in the previous chapter are also taken into consideration. There is only one verb group, where selection restrictions might really be seen as responsible for pattern-verb mis- matches. Name and nominate need a +human AFFECTED, which could ex- plain their unacceptability in clausal patterns as opposed to elect, choose, select, and pick which allow + and –human. The pattern [SCU]+ ‘choose’ +[to_INF] cannot express a +human AFFECTED in its entirety and instead “refer[s] to such abstractions as events, facts, dates, and ideas rather than to perceptible objects.” (CGEL: 15.2). The same applies to +[wh-CL] and +[wh-to_INF], while, with respect to the latter, native speakers disagreed about the acceptability. However, appoint, which did not exclusively occur with +human AFFECTED, was not necessarily acceptable in this pattern ei- ther.

(13)VDE You may choose/*name/*nominate/*appoint to believe me, or not. ? (14)VDE How do we choose/* elect what we teach in the classroom? (15a)BNC It is up to you to select how much you want to invest in your TESSA each year or each month, without putting any undue strain on your finances. B27 405 (15b)NSA It is up to you to *name/*nominate/*appoint how it should be done. (16)VDE In recent years women have begun a fight to gain back what they have lost: the right to choose where to have their babies. (17)BNC ..., the perceived ability of an institution to get students through courses successfully and into good jobs is very important in selec- ting/*nominating where to study. HC6 151 (18)NSA *Women have the right to appoint where to have their babies. (19)NSA ?You have to name where to go.

Also with respect to the ‘teach’ group, differences could be identified: instruct in examples (20) and (21) does not seem to be acceptable since the AGENT is realized by a –human [NP]. At the same time, teach in example (21) appears to be a fixed phrase. Finally, example (22) shows that al- though [SCU] is usually realized by a +human AGENT this is not neces- sarily the case in the sense ‘order’.

? (20)VDE His early career was also helpful, in that it taught/* instructed him more about how not to run a business than how to run one. Different aspects of verb meaning 221

(21)VDE This experience taught/*instructed me a valuable lesson in legal hu- mility. (22)BNC The resolution also instructed the politburo to give all necessary sup- port to the rump of the Lithuanian CP which had remained loyal to the CPSU. HKR 179 = ‘order’

Moreover, other restrictions in this group cannot be accounted for on this basis: examples (23) and (24) both have a +human AGENT and still the pat- terns are not possible in combination with instruct:

? (23)VDE In Greek mythology, Chiron, the centaur who taught/* instructed medi- cine to Asklepios, suffered an incurable wound at the hands of Hercules. (24)VDE He taught/*instructed himself carpentry from library books.

Train receives a different reading when complemented with –human [NP], however, the complement as such is only possible in trivalent patterns, where the meaning is described in VDE as “[s]omething such as a light or a weaponII can be trained on someone or somethingIII, i.e. directed towards them.” (VDE: 876). In that sense, there is a very clear semantic restriction for this pattern, i.e. –human +weapon/light/etc. Practice, on the other hand, can only be complemented with –human [NP], while this does not lead to any pattern restrictions. Again, the difference in selection restrictions mir- rors a necessary distinction into different lexical units.

(25)VDE Bright lights are trained on the crowd. (26a)VDE His first interest in music came when he listened to his brother and sis- ter practicing/*training the piano. (26b)NSA *His first interest in music came when he listened to his brother prac- tising his sister.

Similarly, indicate and show differ in that a +human [NP] in a divalent pat- tern is only possible with indicate. Another difference is that an AGENT is implied (as opposed to a THEME), which is not the case in the pattern [SCU]+ show +[NP]. For this reason this lexical unit of indicate has been described as ‘point at’ as opposed to ‘show’ in VDE.

(27)VDE Voters in the local government elections will not have to show a voter’s card at the polling station before they are allowed to vote. (28)VDE A profusion of white water lily indicates good water quality. = ‘show’ (29)VDE “You two report directly to me,” he said, indicating the Duke and me. = ‘point at’ 222 Pattern choice and verb meaning

The same pattern [SCU]+ verb +[NP] realizes different participant patterns with press and urge and consequently also mirrors their different selection restrictions.

(30)VDE He did not press her AFFECTED. (31)VDE A statement today in Brussels urges an end to customs duties in both directions EFFECTED.

Another rather unsystematic restriction can be found with respect to an- swer, which does not exhibit any selection restrictions with respect to ani- macy in the pattern [SCU]+ answer +[NP]. However, +animate [NP] can be seen as a different participant, i.e. BENREC (32a), rather than –animate [NP], which is classified as ÆFFECTED in VDE (33a). This choice has also formal relevance as far as passivization is concerned. (33b) appears to be much more natural and also more frequent than (32b), although both were found to be acceptable by native speakers.

(32a)VDE Gwen jutted her chin forward, her nose in the air, and did not bother to answer Sister Esmee. (32b)NSA Sister Esmee was not answered by Gwen. (33a)VDE After his speech, Mr Gorbachev answered a series of questions sub- mitted by parliamentary deputies. (33b)NSA After his speech, a series of questions submitted by parliamentary dep- uties were answered by Mr Gorbachev.

In that sense selection restrictions might be grammatically relevant in that they influence the formal realization of patterns. The question to what ex- tent this is predictable, however, is highly questionable, since this cannot be taken as a general rule: VDE provides several examples for BENREC com- plements appearing as the subject of a passivized clause in the same pat- tern, e.g.

(34)VDE Lovesey wanted to be able to advise customers on the basis of practical experience. (35)VDE To help in the prevention of heart disease we are advised to eat fish three or four times a week. (36)VDE I’ve no idea. You must ask my wife. (only if clear from context) (37)BNC I couldn't, I was asked and I just couldn't think of it. KE3 6212

Despite such differences, most of the cases discussed above do not neces- sarily lead to a difference in the pattern inventories of the verbs. Different aspects of verb meaning 223

Besides frameworks relying on abstract selection restrictions, there has also been criticism of such notions for being presented as too general to be actually relevant in complementation. Ickler (1985: 366–370), for instance, considers selection restrictions to be extra-linguistic factors, which are in- dependent of language, and Heringer (1996: 64) sees them as “completely idiosyncratic”. Ágel (2000: 187) concludes that “[t]he more general a se- mantic feature the less likely is it that it is relevant in a specific language, i.e. that one can find oppositions between verbal valency carriers which rest exclusively on this feature [translation mine].” Even if one considers such criteria to be relevant, it is hard to imagine a straightforward relationship between selection restrictions and pattern selection when the different phe- nomena presented above are taken into consideration. That it is not possible to predict a verb’s formal realizations from the participant roles and selection restrictions of a verb does, however, not completely refute the notion of semantic determinism. Restrictions could still be due to other differences in the verb meaning beyond its participants and their specific selection restrictions. While it is undoubtedly the case that the participants of a verb form the basis of its meaning, other aspects might play a role as well. The more general the semantic roles used in a description of participants, the more verbs can be found allowing the same kinds of participants. Thus there must obviously be other factors dis- tinguishing such verbs. It is for this reason that an explanation has been sought in a different aspect of the meaning of the verb. Gisborne (2001: 348) argues that representations of meaning simply focusing on the number and semantic role of a verb’s participant are usually regarded as insufficient for explaining the linking between semantics and syntax. Instead theories addressing this issue “generally include information such as causation ac- tion, resultant states”. The following chapter will therefore focus on aspects of meaning which are not directly connected to the notion of participants.

3.4.3 The situation type of the verb

Semantic differences between verbs can also be identified in their aspectual nature, i.e. their lexical aspect or the situation type they express (also Ak- tionsart). Also the term event semantics often refers to the notion that be- sides semantic roles there is a meaning component in the verb describing it for instance as bounded or unbounded, punctual or expanding in time, or implying a cause or not doing so, i.e. verbs can be said to have an inherent aspectual meaning (e.g. Dahl 1985: 26–27; Welke 2005: 171). Geuder and Butt (1998: 7) argue that 224 Pattern choice and verb meaning

…there are at least two reasons which speak for a concentration on event semantics in order to explain the variability and compositional effects ob- served with regard to argument projection. For one, events have internal structure, which means that event description can be complex. For another, events can be conceptualized in different ways, thus allowing different va- riants of predicates that share a common core meaning.33

Reference to a verb’s lexical aspect has been made in different attempts to relate the meaning of verbs and their syntactic realizations. Mukherjee (2005: 46–47), for instance – in what he calls a lexico-grammatical ap- proach for the linkage between words, pattern, and semantics – identifies three levels of analysis, of which situation types together with semantic roles constitute one level – besides functional categories and lexico-gram- matical patterns such as give + V n to n. Others take lexical aspect as the decisive grammatically relevant semantic feature of a verb: Tenny (1994) puts forward what she calls the “Aspectual Interface Hypothesis”:

The universal principles of mapping between thematic structure and syntac- tic argument structure are governed by aspectual properties. Constraints on aspectual properties associated with direct internal arguments, indirect in- ternal arguments, and external arguments in syntactic structure constrain the kinds of event participants that can occupy these positions. Only the aspec- tual part of thematic structure is visible to the universal linking principles. (Tenny 1994: 2)34

Along these lines, particle complement types can function as “event termi- nus” or “path objects”, which delimit situations and thus affect the aspec- tuality of a sentence (Tenny 1994: 68–69 and 72). Pinker (1989: 207), how- ever argues against this, claiming that more is needed to predict verb pat- terns: “For example, there are no obvious aspectual differences between roll and bleed, own and have, choose and win or load with and fill with, though they contrast in their willingness to undergo alternations.”

33. Note that argument selection or projection is mainly concerned with the ques- tion of which participants realize subject and object in a clause. While the discussion is typically confined to noun phrases, the choice of pattern goes further than that. Patterns explicitly include information on the formal realiza- tion of these functions and also account for formal realizations such as clauses or particle phrases. 34. The distinction into internal and external arguments goes back to the notion of theta roles in the context of Government and Binding theory (Chomsky 1984), where internal arguments are part of the verb phrase and the external argument is realized as subject. For an overview see Welke (2005: 100–101). Different aspects of verb meaning 225

The situation types a verb can express can be regarded as a combination of a number of features describing the different characteristics a situation can have. Differences in the composition of these features lead to different situation types. The features that define a situation type are, for example, duration, , and the distinction, whether a situation is stative or dy- namic (e.g. Brinton 1988: 23–27; Comrie 1976: 41–51; Welke 2005: 170– 178). The most widely used classification of situation types goes back to Vendler (1967), who proposed a four-way distinction into states, activities, accomplishments, and achievements (Brinton 1988: 27–28). Aspectual differences which have an effect on the structure of sentences are, for instance, seen in the distinction between accusative languages (such as English or German) and ergative languages (such as Dyirbal). The ques- tion whether this distinction can also help to predict structures within the English language will have to be addressed. In order to check to what extent aspectual distinctions can account for the restrictions found in the verb groups under consideration, the lexical units were analyzed according to their aspectual profile. The list of situa- tion type differentiations used for this analysis is illustrated in Table 41, including eleven situation types (in small capitals and bold type).They are based on CGEL, which are again mostly based on Vendler’s (1967) four- fold distinction, supplemented with additional distinctions, e.g. Smith’s (1991: 223, 236–237) category of semelfactives which denote telic achievements (grayed areas corresponding to Vendler’s original distinc- tions). Ramchand (2008: 19–20) argues that “the original division into states, activities, achievements and accomplishments cannot correspond directly to what is specified in the lexicon” which is why “many theories attempt to use lower-level aspectual features that are derived from the larg- er natural classes”. The reason for supplementing the original list in CGEL basically rests on the attempt to provide explanations for the use of the progressive form in English, which makes them grammatically relevant categories (CGEL: 4.33). The situation type of a verb is not necessarily fixed: Welke (2005: 171) suggests an “original aspect” for each verb which can be altered – in Eng- lish for instance by the choice of morphological aspect, i.e. progressive form vs. simple form. Van Valin and Lapolla (1997: 112–113) distinguish between a “basic lexical meaning of a verb” and its meaning in context, while “[the] former would be its representation in its lexical entry in the lexicon”, while they also accept that a verb can be “ambiguous between two related meanings”. 226 Pattern choice and verb meaning focusing on initial stages indicating non-conclusion non-conclusion indicating multiple event view ------/ pretending purpose on acting situation unbounded unbounded situation situation unbounded indicating non-conclusion non-conclusion indicating situation unbounded focusing on initial stages multiple event view sit down down sit fill up fire a gun be tall be angry sit example example aspect progressive effect of drink drink ripen ripen rain drop sneeze .) ) ) ) ) ) AGEN ( ) ) ON - AGENTIVE AGENTIVE AGENTIVE AGENTIVE AGENTIVE ( - - - - NON AGENTIVE NON NON NON AGENTIVE ( ACCOMPLISHMENTS ( ( ACTIVITIES ( TRANSITIONAL ACTS ( MOMENTARY ACTS ( PROCESSES PROCESSES GOINGS TRANSITIONAL EVENTS MOMENTARY EVENTS = =

QUALITIES STATES conclusive = conclusive = ACCOMPLISHMENTS ACTIVITIES ACHIEVEMENTS ACHIEVEMENTS non- stative durative situations stative punctual situations Situation types based on CGEL (4.27) and Vendler (1967) (1967) Vendler (4.27) and CGEL on based types Situation STANCE SITUATIONS non-conclusive situations situations Table 41. 41. Table 1 stative-durative- type situation 2 3 4 dynamic 5 6 non-conclusive 7 8 non- 9 non-conclusive10 11  Different aspects of verb meaning 227

To what extent lexical aspect actually figures in restricting patterns and thus making such restrictions predictable and consequently making situa- tion types relevant to the grammatical realization of participants will be shown below.What the typology provided in CGEL includes in addition to Vendler’s original classification is the feature agentivity, e.g. the distinction into going-ons (like rain) and activities (like drink). This is similar to a further dimension which is also often considered as important, i.e. the dis- tinction – non-causative. This dimension does not focus on the inherent temporal structure of verbs but is also often regarded as aspectual. According to Van Valin and Lapolla (1997: 107), each of the four situation types can be either causative or non-causative. Welke (2005: 171–179) argues in favor of a distinction into Handlung (‘act’) and Vorgang (‘event’), emphasizing the role of causality/intentionality. Such distinctions have been formalized by means of different primitives such as DO, CAUSE, BECOME or +/– control (cf. Dik [1978] 1979: 32–34). Dik (1979: 33) defines control as follows: “A state of affairs is controlled if one of the entities involved in it, the controller, has the power to determine whether or not that state of affairs will obtain.” Alternatively, features such as +dynamic and +control are employed (Welke 2005: 182). Since this is already part of the CGEL distinction, a difference as regards causativity will only be indicated where necessary. The verbs in the verb groups were classified on the basis of different tests. An often proposed test for conclusiveness is the question: “if one was verbing but was interrupted while verbing, has one verbed?” (Brinton 1988: 26): a conclusive situation interrupted before its completion cannot be con- sidered as having taken place, and the question must be answered negative- ly. Another test for determining situation types in English is the compatibil- ity with the progressive aspect form and the effect achieved by this (Comrie 1976: 35) (for an overview of different test criteria see Van Valin and La- polla 1997: 94–101; see also Table 41). Those verbs which did not exhibit any differences as regards their lexi- cal aspect are portrayed in Table 42. Those verb groups which exhibited difference are discussed below. If aspectual distinctions cause differences in patterning, verbs with the same aspectual profile should also be more similar with respect to their complementation potential than verbs with different aspectual profiles. Among the verb groups exhibiting no differences in complementation, in- cluding subgroups within the 22 groups, this is clearly the case. 228 Pattern choice and verb meaning

Table 42. Verb groups with verbs which have the same situation type

meaning verbs situation type ‘answer’ answer, reply, respond accomplishments ‘choose’ appoint, choose, elect, pick, select, name, accomplishment nominate, designate ‘declare’ declare, pronounce, announce, proclaim, accomplishment state ‘get’ obtain, receive, get, acquire, buy, purchase accomplishment ‘hire’ hire, rent, lease accomplishment ‘like’ love, like, adore state ‘not lose’ keep, maintain, preserve, sustain activity ‘permit’ allow, permit, authorize, entitle accomplishment ‘practice’ train35, practice, rehearse activity ‘quarrel’ quarrel, argue, fight, dispute, bicker activity ‘start’ start, begin,commence transitional act ‘propose’ propose, suggest, recommend activity ‘teach’ instruct, teach, train, educate activity ‘telephone’ phone, ring, call activity ‘urge’ urge, press, push activity

There are indeed no differences in complementation between the following pairs/groups: suppose and presume can both be described as accomplish- ments; the same applies to lease and rent, and get, obtain, buy, and pur- chase,and reply and respond. Phone, ring, and call are activities, and love and like, finally, express stative situation types. However, respect and ad- mire also occur in the same valency patterns, although they differ in that respect rather describes a state and admire an activity. Considering the number of restrictions identified, one would expect clearer discrepancies between the situation types and the verbs if aspect is to play a major role. This is simply not the case. Among the 22 verb groups analyzed here, only seven exhibit differences as regards the potential situa-

35. Van Valin and Lapolla (1997: 100) would probably classify train as an active accomplishment similar to march in The seargent marched the troops to the barracks as opposed to The troops marched. (Cf. When we train professional athletes, we get them into the lab and monitor their progress very careful- ly.VDE and “It’s really hard to train on your own,” she admitted.VDE). This nicely accounts for the difference between train as opposed to practice and rehearse with respect to this alternation. In the present analysis train was di- vided into two different lexical units (‘practice’ and ‘teach’). Train ‘practice’ is best taken to be a causative activity. Different aspects of verb meaning 229 tion types the verbs within the groups can have. This by no means corre- lates with the general rate of discrepancies described in the preceding chap- ters. Only 94 of all 193 restrictions which could not be accounted for on the basis of the participant inventory or selection restrictions are associated with those eight groups which exhibit differences in their aspectual profile in the first place (42 of these restrictions are to be found in the ‘think’- group alone). Even if this is a considerable number, there is not necessarily a direct relationship between the situation type and these restrictions. The eight groups will be discussed below. assume, suppose, presume accomplishment guess, gather activity/accomplishment

It has already been noted above that suppose and presume have parallel valency constructions and the same situation types. It could be argued that guess is less conclusive than suppose and presume. The pattern in which guess does not occur ([SCU]+ verb +[NP]+[AdjP]) is indeed not really compatible with an activity reading.

(1)NSA *He guessed her older than she actually was.

However, other patterns guess is compatible with do not necessarily ex- press activities either, e.g. (2), including the pattern which was exclusively found for guess given in (3).

(2)VDE He guessed her age to be no more than twenty-two or three. (3)VDE In fact, he had awakened some while ago, and had at once guessed where he was and what had happened.

Gather, on the other hand, clearly has an activity reading when used in the sense ‘collect’ but not in the sense ‘assume’. Yet whether this can interfere with the acceptability of patterns in the sense ‘assume’ is not at all obvious. Moreover, assume expresses the same situation type as suppose and pre- sume but cannot occur in the pattern containing [wh-CL]. mean, intend state plan, aim state/activity

Mean and intend, which do not seem to differ with respect to their situation types, indeed do not exhibit any restrictions which cannot be accounted for by other semantic aspects. Plan and aim, on the other hand, which can have 230 Pattern choice and verb meaning both a more stative but also a more dynamic reading, differ much more clearly as regards complementation, which should not be the case if restric- tions are to be explained on the basis of situation type differences. There is only one difference between the two subgroups – mean/intend, on the one hand, and aim/plan, on the other – in the sense that a pattern is acceptable to the second but not to the first subgroup.

(4a)VDE Aim for clarity and precision; remove words that obscure your direc- tions. (5)VDE We must begin now to plan for a time when people will ultimately be able to work for the cheque they get. ? (4b)NSA *Mean/* intend for clarity and precision.

Consequently, this would be the only case available for an explanation with respect to this semantic difference but the examples do not provide such an explanation at all.

recall, remember, recollect, bear in mind activity/accomplishment reminisce activity

The fact that reminisce differs from recall, remember, recollect, and bear in mind in that it focuses on the activity of remembering and less so on the thing remembered, can indeed explain certain restrictions to be found in this group. Moreover, reminisce, does not need to express ÆFFECTED, a participant which is more central for the other verbs, since it could be seen as embodying the endpoint of the accomplishment, even if it is not obliga- tory. This might account for the patterns without this participant, which are exclusively compatible with reminisce, given in (6)–(9).

(6)BNC At first, as he worked, he had reminisced pleasantly to himself about the smells and the texture of paper and ink…H84 485 (7)BNC Back at the car park we had a well earned cuppa and reminisced over another hot day back in 1933. ECG 907 (8)BNC The entertainment is the pleasure of reminiscing with one's friends and making a couple of witty speeches. FSN 358 (9)BNC The soprano Birgit Nilsson reminisces with John Higgins about her distinguished career. ED6 767

The restrictions of remember in examples (8) and (9) were taken to be se- mantic differences since remember never expresses a CO-AGENT. Yet since there are patterns where remember can be complemented with about which is very similar to over and since it can also be complemented with a Different aspects of verb meaning 231

[to_NP], the restrictions exemplified by (6) and (7) are not taken to be di- rectly related to meaning. All patterns in which reminisce could not be found include ÆFFECTED. There is one further pattern, however, which includes ÆFFECTED and is exclusively compatible with reminisce. This shows that this participant is not excluded simply because reminisce represents an activity.

(10)BNC “Conceive what I must have been at fourteen,” ÆFFECTED he remi- nisced to James Gillman, his first biographer. B0R 128

Thus a straightforward explanation of the restrictions on the basis of the situation type difference is not completely satisfactory either, although a certain interrelationship must be acknowledged. indicate activity/transitional act show activity demonstrate activity/accomplishment

In the case of indicate, show, and demonstrate, one might argue that there is a difference in the sense that demonstrate is more concerned with the completion of the event than show. One of the tests to distinguish between activities and accomplishments is the question used in (A) and (B), where it appears that (A) can more easily be answered with yes than (B) (based on Brinton 1988: 26).

(A) If one was showing the solution to the audience but was interrupted while showing, has one shown? (B) If one was demonstrating what would be the best solution but was in- terrupted while demonstrating, has one demonstrated?

Moreover, it appears that indicate is more punctual than both show and demonstrate.

(11a)NSA She showed him her pictures. (12)BNC This operation needs to be explained and demonstrated to every stu- dent.A0H 376 (11b)NSA She indicated her pictures to him.

However, looking at real language data in the BNC, it is much more diffi- cult to find such clear-cut examples since again the use of the progressive form is widely found with all three verbs presenting them as extended in time and in very similar contexts. This turns the distinction into something 232 Pattern choice and verb meaning much too unreliable to use it as the basis for making any kinds of predic- tions of pattern-verb incompatibilities, even more so, since there is no sys- tematic relation to the choice of valency patterns.

think, consider, reflect, ponder, contemplate activity judge transitional act

A similar distinction can be made in the case of think, consider, reflect, ponder, contemplate, and judge, where the latter, as opposed to the other verbs in the group, does not really have an inherent time structure and is rather punctual, which is also reflected in the fact that judge is not fre- quently used in the progressive form, even if there are such examples in the BNC. Judge cannot occur in the following patterns, which all express a TOPIC: *[SCU]+ verb +[about_NP]/+[upon_NP]/+[over_NP]. It could be argued that these patterns fail to express a necessary endpoint, i.e. the judgment itself, and that they are consequently not compatible with the situation type of judge. However, there are restrictions which cannot be explained in this way, e.g. *[SCU]+ judge +[Q/S]+[on_NP] and *[SCU]+ judge +[of_NP]+[as_NP] which all express ÆFFECTED in combination with other verbs in the group. Again, the semantic difference is by no means directly related to the observed restrictions. Note that despite this, such differences were counted as differences explainable on the basis of a differ- ence in the participant inventory in that judge in the sense ‘think’ never expresses TOPIC.

respect, esteem state admire activity

In the case of respect, admire, and esteem, it seems that admire more easily portrays an activity than for example respect or esteem.

(13)BNC “We were admiring your garden,” says Denis, the candidate's minder, to the lady on the step. CAF 1177 (14)BNC Moreover, Cézanne had felt he was respecting the laws of traditional, scientific perspective. GUJ 705

While LDOCE4 describes both respect and admire as not compatible with the progressive aspect, the BNC provides several examples for uses of ad- mire in the progressive, clearly expressing an activity, but only 4 examples with respect and none for esteem. Again this clearly shows that while cer- tain preferences are indisputable, it seems that at least respect can adopt a Different aspects of verb meaning 233 more activity-like event type. However, none of the restrictions in this group correlate with the distinction drawn on the basis of the situation type difference. bear state tolerate, endure state/activity

Finally a difference between bear on the one hand and tolerate and endure on the other might be found in the fact that bear does not typically express an activity and is therefore not typically used in the progressive. Nonethe- less, a BNC search also revealed a number of examples such as sentence (15).

(15)BNC Practitioners are bearing most of the costs of regulation themselves. CBT 2248

Moreover, the restrictions within this group again do not seem to be related to this difference and they are difficult to explain in reference to a differen- tiation between state vs. activity. Consequently, the role of situation types for the verb groups in this sample has to be regarded as not sufficiently relevant. Boas (2003: 59) also argues that event structure alone is not the key to argument realization: “based on a comparison of verbs that are closely re- lated in meaning and exhibit the same type of event structure I argue that it is extremely difficult to capture the distribution of resultatives on the basis of event structure alone”. In order to account for the fact that there are cases where the aspectual quality of a verb plays a role and others where it clear- ly does not, attempts have been made by dividing verbs in two groups, weak predicates, which show greater variation, and strong predicates (cf. also Ritter and Rosen [1998: 143–145]) which

… actually encode some of the relevant aspectual information lexically: Their interpretation [of strong predicates] is fixed and may not vary with the context, they have strong semantic requirements, and their event classi- fication, as well as their adicity and case properties are fixed. (Geuder and Butt 1998: 13)

This, however, is rather a circumscription of the fact that aspectual qualities do not generally help to predict the form of a verb but give such indications in only a limited number of cases, rather than providing a real explanation of this variability. 234 Pattern choice and verb meaning

The fact that several of the verbs do not seem to be fixed as regards their situation type, i.e. several of the more stative verbs can be used in the pro- gressive and can consequently express a more activity-like event, already exhibits another problem with the hypothesis that a verb’s aspectual nature determines the choice of patterns it can occur in: the situation type of a verb is very much dependent on the context the verb actually occurs in. There- fore situation types are descriptive features for a verb in use rather than for the lexeme as such. One lexeme can be used in different contexts and con- sequently have different situation types. Hence, the event type a verb ex- presses in a specific sentence can be seen as a composition of the verb’s situation type, the choice of aspect form such as progressive form vs. sim- ple form, and even the type of pattern: the use of a particle in drink vs. drink up or also certain types of noun phrases – as in play tennis vs. play a game of tennis – or particle phrases – as in She's swimming vs. She's swim- ming across the canal – can change the event type of a verb (Brinton 1988: 4; Kabakþiev 2000: 69–90 and 241–262). With respect to the latter it be- comes obvious that the more or less artificial division into differences re- lated to the participant inventory and differences related to the situation type does not really hold. These features obviously interact and the division cuts across a wider semantic classification system.36 This is also evident in the fact that some of the restrictions presented here, i.e. restrictions which might be due to the aspectual nature of the verbs in question, have been “counted” as semantic differences based on a difference in the participant inventory. This is the case, for example, with judge in the sense ‘think’, which never expresses a participant TOPIC. This might also explain why judge is less likely to express an activitiy – extended in time – but rather a punctual transitional act – or the other way around. Thus whether it is the participant or the lexical aspect which causes this difference is difficult to say. As both are counted as a semantic difference explaining a structural

36. Cf. also Dik’s (1979: 32–36) use of what he calls nuclear predications such as (+/– control) to explain (in)compatibilities with certain participant roles. Another semantic classification of verbs, which e.g. Jackendoff ([1990] 1991) proposed as relevant for event type description and possibly influential for ar- gument selection, is based on a differentiation between “motion and location events”. Levin and Rappaport Hovav (2005: 79) refer to this as the “localist approach” as in Jackendoff (1991: 43–44) who classifies verbs on the basis of semantic primitives such as GO, CAUSE, or BE. However, according to Le- vin and Rappaport Hovav (2005: 83), this theory clearly fails to account for a considerable number of “activity verbs” which cannot be analyzed as motion- al in any way, e.g. cry or play. Different aspects of verb meaning 235 difference in the quantitative evaluation of the results of this analysis, this does not affect the overall result that meaning is not sufficient for explain- ing pattern choices. Methodologically the division is justified as it allows for looking at these phenomena from as many different angles as possible. It must be very clear, however, that by this no assertions are made about the make-up of the semantic structure of verbs (cf., for instance, Leech 1971) or a decision taken which of the two is the actual cause for the difference. A problem which seems to be inherent in all aspectual approaches ac- cording to Levin and Rappaport Hovav (2005: 111) is the fact that they have so far not been very successful in describing which participant is rea- lized in which position other than the direct object and the realization of prepositional phrases. Moreover, the focus in such approaches is clearly not on the exact formal realization of functional units such as objects, which obviously delimits what they actually try to predict.

The notions of telicity, measure and incremental theme appear to be impli- cated in certain facets of argument realization, especially in the choice and expression of direct objects. … Nevertheless, … not all facets of argument realization can be reduced to these [aspectual] notions. (Levin and Rappa- port Hovav 2005: 112)

The approach to take the whole event and consequently the complete par- ticipant pattern into consideration for its formal expression is by all means an approach with greater potential than projectionist approaches which take as their sole starting point the semantic role of single participants. Nonethe- less, also in the analyses above, the distinctions made do not seem to play a role. The influence of the selected criteria on the connection between mean- ing and form seems to be doubtful.

3.5 Implications

With the data analyzed, it is certainly unproblematic to agree with Sinclair (1991: 65) that there is “a strong tendency for sense and syntax to be asso- ciated”. To define the exact nature of this association, however, is clearly much more problematic. To put the results of the comparison of 22 groups with altogether 87 semantically similar lexical units into perspective, differ- ent approaches will briefly be reviewed. Hunston and Francis (2000) dif- ferentiate between what they call a weak, a strong, and a medium statement to the connection between pattern and meaning: 236 Pattern choice and verb meaning

Thus a weak statement … would be that a list of verbs frequently having a particular pattern is not totally random with respect to meaning. A strong statement would be that a verb has a particular pattern because it has a par- ticular meaning. A medium view would be that, given a list of words occur- ring with a particular pattern, the majority will be divisible by most observ- ers into reasonably coherent meaning groups ….The strong statement is a theoretical claim, but is as yet insufficiently substantiated in our view. (Hunston and Francis 2000: 86)

The strong claim is often referred to as projectionist approaches, which assert that formal properties are determined by lexical properties and, more specifically, semantic lexical properties (cf. Levin and Rappaport Hovav 2005: 18). The term projection goes back to Chomsky’s conception of the projection principle but has also been adopted in other models such as Lex- ical Functional Grammar in the form of the completeness and coherence condition (Bresnan 2001: 63) or by VanValin and LaPolla in what they call a completeness constraint (1997: 325). In the context of Government & Binding, Chomsky proposed a determination of the syntactic form of the arguments through theta-roles (1984: 36–37), thus clearly strengthening the lexical component of this syntactic model. Every lexicon entry specifies, according to Chomsky, its phonological form and its semantic properties. The semantic roles of possible complements of a verb are therefore deter- mined by the lexicon, i.e. the so-called semantic selection or s-selection. This information is accessible from the syntactic side via the projection principle. According to the projection principle, the form of a complement is based on the theta roles the verb assigns or the arguments it demands, so that specific semantic roles are realized by complements of a specific form (Chomsky 1984: 29). This means that for each a number of so- called canonical realizations are expected.

… the Projection Principle expresses the idea that D-structure is a “pure” representation of thematically relevant GFs [= grammatical functions]. … [T]he complements of the Head need not be specified since they are deter- mined by the lexicon, under the Projection Principle. (Chomsky 1982: 9) [bold type mine]

This notion does not in itself refer to the semantics of the verb but assumes the formal properties of a sentence to be projected by the lexical head, while the source is not exclusively the meaning of the lexical head. Besides s-selection, a second important concept is c-celection, i.e. categorical se- lection. However, in more recent models, Chomsky (1995: 31) puts even more emphasis on the role of semantics arguing that subcategorization “fol- Implications 237 lows almost entirely from theta-role specification”, thus assuming a more direct relationship between semantic role and syntactic function (cf. Welke 2005: 101). Yet the analysis of verb groups has clearly shown that a verb’s participants are by no means sufficient for explaining the valency patterns of a verb, since there is still a considerable degree of variation in their spe- cific formal realization. This is in line with the fact that the strong state- ment has also previously caused much debate and even generativists seem to have doubts whether it can be said to be applicable to such a strong de- gree.

The variation in meaning that accompanies many argument alternations has led a number of researchers to conclude that the program of deriving the syntactic contexts a verb is found in directly from its meaning – what we re- fer to as the projectionist approach – is misguided. (Levin and Rappaport Hovav 2005: 18)

Nonetheless, the response so far does not necessarily take the form of at- tempts to disprove the claim but rather to search for evidence in favor of it – or at least a modified version of it – by widening the scope of what is taken to be a relevant semantic feature or delimiting the list as is evident, for instance, in Pinker’s Grammatically Relevant Subset Hypothesis (1989: 166). Levin and Rappaport Hovav (2005: 15) speak of “a real challenge” when it comes to pinning down those factors which are actually relevant for argument realization: “The most obvious components of meaning may not be the actual semantic determinants of syntactic behavior. A failure to iden- tify such elements may impede the formulation of a perspicuous theory of argument realization”. This clearly implies that it is not the complete mean- ing of the verb that is relevant to argument realization but only a subset of its meaning, e.g. not what is typically understood as its encyclopedic mean- ing (cf. for instance Ramchand 2008: 2). In the framework of valency theory, such a distinction has also been made: Bondzio (1971: 92–93) dif- ferentiates between meaning components relevant to valency and meaning components which are not. In that sense Bouchard (1995), for instance, argues that syntax is indeed still dependent on semantics but on a different kind of semantics than Chomsky had envisaged:

... of the linguistically relevant aspects of meaning, some affect syntactic form (I call them Grammar Semantics …) while others do not (I call them Linguistic Semantics …) Some aspects of the compositional semantics of a sentence project fairly directly onto its syntactic structure so that syntax is dependent on semantics in some of its formal aspects. (Bouchard 1995: 17– 18) 238 Pattern choice and verb meaning

There are, however, also linking theories which take encyclopedic infor- mation as sufficient for the establishment of a link between meaning and form. Ramchand (2008) provides an overview of the major trends in link- ing, which, according to her can be divided into mainly two approaches: the lexical-thematic approach and the generative-constructivist approach. The lexical-thematic approach, as Bouchard (1995) above describes it, takes verbs to have syntactically relevant semantic information which is pro- jected onto the verb’s syntax, i.e. which restricts it. The link is often taken to depend on a hierarchy of participant roles or, as Croft claims, on “force- dynamic relationships”:

... the force-dynamic relationships constitute the event frame, and only part of the event in the event frame is profiled by the verb, combined with a par- ticular assignment of subject, object, and oblique. ... Whereas full affected- ness appears to be the salient semantic feature for assignment as object, full control appears to be the salient semantic feature for assignment to the sub- ject position. (Croft 1998: 88–89)

These approaches often do not cover any formal realizations as defined in this study but rather functional realizations, and if formal realizations are taken into consideration, the focus is clearly on noun phrases (see also Croft 1998: 21). The second approach, the generative-constructivist ap- proach, differs mainly in that it “allows free building of syntactic terminals, but allows general encyclopedic knowledge to mediate whether a particular lexical item may be inserted in those terminals or not” (Ramchand 2008: 4– 5). What both theories have in common is the objective of finding syntacti- cally relevant features such as causal or aspectual features (event structure) and/or participant roles, while the latter are not considered as “theoretical primitives but derived from event structure” (Croft 1998: 21), following the general assertion that a “thematic role hierarchy is insufficient for deter- mining argument linking, and [that] another layer of representation is ne- cessary”. This additional layer can then be either a syntactic layer (e.g. Rappaport and Levin 1988; Bresnan 1994) operating with additional syn- tactic arguments or a semantic one, basing the distinction on semantic “su- per-roles”, which override the thematic roles as in Dowty`s (1991) proto- roles (Croft 1998: 21).37 Croft (1998: 21–22) argues that there “are rules specifying how participants in participant roles are linked to the appropriate

37. For a detailed overview on different theories of argument realization see Levin and Rappaport Hovav (2005). Implications 239 casemarked NPs using the concepts in the preceding three layers [event structure, thematic roles, and syntactic arguments/super roles]”. The general appeal of models which take syntax to be systematically re- lated to meaning is that they can obviously reduce the idiosyncratic compo- nent of their grammar. If complementation is based on semantic rules which apply to more than just a single verb and can account for the bigger part of syntactic configurations, the complexity of single lexical entries can be reduced (cf. for instance Levin 1993: 11). The question yet to be ans- wered, as has been show above, is what form such semantic notions could take to allow for such a wide generalization without generating sentences which do not occur in actual language. In the analysis presented in the previous chapters, the focus was clearly on what is typically considered to be grammatically relevant semantic in- formation, e.g. aspectual features, selection restrictions and the participants of a verb. Given that the verbs in the groups are clearly not semantically identical, one could assume that differences between them result in com- plementation differences. The hypothesis was that it should be possible to provide explanations for such differences between the verbs on the basis of their meaning, i.e. by examining various semantic aspects such as the ones mentioned above. Since there is by no means agreement about the exact location of the dividing line between what is syntactically relevant and what not, encyclopedic information was also taken into consideration in the discussion of separate restrictions where it appeared to be promising. This is in line with Ramchand (2008: 14) who argues that “[t]he decisions about what kind of meaning fall on which side of this divide is of course a subtle and empirical question, and should not be prejudged.” If semantics is the decisive factor, verbs which are semantically more similar, e.g. verbs which do not differ with respect to their aspectual profile or their selection restric- tions, should also be syntactically more similar. The hypotheses could not be verified. A considerable number of restric- tions could not be accounted for in a satisfactory way and not all differ- ences in the field of supposedly grammatically relevant semantic features had an effect on the range of patterns of the different lexical units. While this does not refute the hypothesis that syntax is determined by meaning, it clearly casts doubt on the notion that meaning can be captured by means of general rules. A more promising attempt is clearly that of construction grammar, which allows for an additional level on which meaning plays a role for complementation, i.e. not just the lexical entry but also constructions. However, also in the framework of construction grammar, verb meaning 240 Pattern choice and verb meaning plays a crucial role in explaining whether a verb can occur in a construction in the first place, i.e. whether verbal participants can fuse with the argu- ments of the argument structure construction. The verb hand can occur in the ditransitive construction because the hander can fuse with the argument role agent, the handee with the argument role recipient and the handed with the argument role patient (Goldberg 1995: 51). According to Goldberg (1998) constructional meaning largely builds on basic verb meanings, such as give for the ditransitive ([SCU]+ verb +[NP]+[NP]). Goldberg (1995: 133) describes verb classes as being “implicitly represented speaker- internally as generalizations over learned instances” and “general categori- zation processes”, while accepting “scattered positive exceptions and vary- ing degrees of acceptability”. Construction grammarians reject the idea that syntax is predictable from semantic roles, i.e. that the syntactic form and the meaning of a sentence are exclusively projected from specifications of the main verb (Goldberg 1995: 225–226). This is based on the realization that the construction can provide semantic aspects which go beyond the meaning of the verb. The presence of profiled participants is not necessarily a prerequisite for the acceptability of a verb in a construction: non-profiled participants – as the mailee – can become obligatory in certain construc- tions as Paul mailed her a letter, i.e. “the construction poses a profiled status on the … role” (Goldberg 1995: 53). Contrary to the mailee, the han- dee in Paul handed her a letter is always obligatory (cf. Paul mailed a let- ter vs. *Paul handed a letter). Similarly, the caused-motion construction can provide a GOAL element for the verb sneeze in He sneezed the napkin off the table (Goldberg 1995: 54–55). Goldberg (1995: 225–226) concludes “that an entirely lexically based approach to grammar is inadequate” in that it cannot account for such cases and that constructions are needed instead as an additional layer. At the same time, Goldberg (1998: 209) argues that “semantically similar verbs show a strong tendency to appear in the same argument structure construc- tions”. According to her view, the verb is related to one or a few basic senses which must be integrated into the meaning of the construction if the verb is to be used in it.38 This means that the verb does not adopt a new meaning in a new syntactic configuration but what is new is added from the construction. This way, the verb sense is not used to explain the whole syn- tactic configuration (Goldberg 1995: 9–12). Whether a verb can occur in a

38. Goldberg (1998: 209) implies that these basic verb meanings are the first ones acquired in the respective constructions. This was not confirmed in psycho- linguistic studies (Campbell and Tomasello 2001; Tomasello 2003: 121–122). Implications 241 construction or not is regarded as a matter of compatibility of the meaning of the verb and the meaning of the construction (Gries and Stefanowitsch 2004: 99). Thus one difference between projectionist approaches and con- struction grammar in this sense is that the pattern itself is considered to have meaning independently of the verb, and that it is not the verb meaning alone but the combination of verb meaning and construction meaning that determines the formal patterns a verb can occur in. As evidence for the fact that the verb meaning remains constant and cannot be taken to account for restrictions, Goldberg points out the different degrees of acceptability of sentences (1) and (2) where the meaning of the verb obviously does not change.

(1)QE This room was slept in by George Washington. (2)QE *This room was slept in by Mary. (Rice 1987 in Goldberg 1995: 21)

However, it has to be taken into consideration that a sentence such as “This bed has not been slept in by anybody” is clearly more acceptable. This shows that it is information structure that plays an important role here: the construction is only possible, when the room or the bed are really an appro- priate theme which can be expanded in what follows, which is less likely with sentence (2) above (see also Halliday 1970: 161). Croft (1998: 91) argues that

…to some degree, any word can in principle be used in any construction. Hence, distributional patterns do not establish grammatical categories in the strict sense. What matters is the semantic interpretation of a word in a par- ticular grammatical construction. By examining the meanings of verbs in constructions, we can establish semantic classes of events and conceptual meanings of constructions.

Such an additional level as constructions is clearly problematic for any theory which takes meaning associated with lexical items as having predic- tive powers, as suggested in projectionist approaches. Hunston and Francis (2000: 145), for instance, conclude in this context that “if true this would throw doubt on Levin’s assertion ... that the syntactic behavior of a verb can be uncontroversially determined by a description of its semantic prop- erties” (cf. 4.2.2). A different form of criticism of projectionist approaches comes from the field of conceptualist semantics. Langacker (1999: 71), for instance, takes “the question of whether grammar is predictable on the basis of meaning 242 Pattern choice and verb meaning

[itself as] misguided, for it presupposes prior and independent knowledge of what an expression means”.

Perhaps ironically, a conceptualist semantics implies the absence of such knowledge, the reason being that meaning is largely a matter of construal, and construal is largely invisible. (Langacker 1999: 71)

This could easily be taken as an argument in favor of storage and an item- based view on language, which is also common in context of valency theo- retic frameworks, in which there is not necessarily a need for rules to ac- count for the connection between meaning and form. The type of comple- ments a verb takes and consequently the patterns it can occur in is consi- dered as specific to the lexical unit and could thus be considered to be stored together with the lexical unit. This is also evident in the fact that valency phenomena are often covered in the form of dictionaries. If this is the case, there is no need to account for idiosyncratic behavior since it can simply be accepted as such. One might argue that this means storing a great deal of redundant information. However, this does not have to be a knock- out criterion for this theory. Baker (1999: 234) argues that “[a]lthough models of the lexicon which minimize the number of senses have a certain theoretical appeal, the preponderance of evidence suggests that people do store quite a number of relatively specific senses for highly polysemous words”. If we can store different senses in combination with one lexeme and do not have to evolve them from a core meaning, then why should it not be possible to store a certain amount of idiosyncratic pattern information to- gether with the separate senses? Moreover, the principle of economy with respect to storage is not as necessary as it has been envisaged by linguists in the linking tradition. Slobin (1997: 311) notes that “[i]t is unsettling to realize how many of our theories are aimed at the simplistic criterion of ‘economy’ when we have no rational measure of that economy”. Croft (2001: 121) also points out that “[i]f anything, psycholinguistic evidence tends to favor redundancy and computing parsimony” (see also Barsalou 1992: 440). The findings of the previous chapters clearly show that when in addition to noun phrases particle complements and clausal complements are also taken into consideration, typically chosen semantic “explanations” do not apply and fail to account for syntactic gaps or different pattern preferences. This is also supported by various other corpus-based studies. Hudson, Rosta, Holmes, and Gisborne (1996: 439–440), for instance, conclude in what might be described as a small-scale test of what was intended here at a Implications 243 much larger scale: “we also believe that there is an irreducible residue of cases which can never be explained”. They also argue against the construc- tionist notion that different patterns cannot express the same meaning, i.e. the claim that syntactic difference equals semantic difference, especially on the grounds that a coherent relationship between the syntactic difference and the (supposed) semantic difference is often not available:

However so long as the relations between syntax and semantics are a matter of empirical debate this argument works only as a heuristic for drawing at- tention to semantic differences which remain to be discovered and which may or may not exist. Not only do the putative differences need to be de- monstrated independently, but even more importantly, they must also be such as to explain the syntactic differences …. It is not enough to show that one of the words can be used in a way which is not possible for the other, unless this helps to explain the syntactic facts. (Hudson et al. 1996: 440)

This was also found with respect to the sample above: many of the restric- tions could not be satisfactorily explained and even if a general difference in meaning could be identified and described, there was simply no direct relation to the restriction itself. Hudson et al. (1996: 442) conclude that “in any case, it is hard to see how the alleged meaning differences could be learned except on the basis of the syntactic differences that they are meant to explain, so the arguments are dangerously circular.” Boas (2003) presented similar results in a comparison of the distribution of the semantically similar phrases to death and dead in combination with different verbs expressing the basic sense ‘make + [– alive]’. Why one shoots or strikes someone dead but stabs or crushes someone to death (see Tables 43 and 44 below) cannot really be explained on the basis of the meaning of these verbs which all describe activities with the exact same participants (AGENT, AFFECTED, WEAPON, GOAL) and the same selection restrictions. One might argue in favor of a difference in the lexical aspect of shoot and stab in that shoot dead is more punctual than stab. However, such a difference does not seem to exist between strike and crush. Boas refers to the mechanism responsible for this correlation of pattern and verb as conventionalization. He concludes that “[t]he data thus suggest that a verb’s ability to occur in resultative constructions is a matter of con- ventionalization, i.e. it cannot be explained by more general constraints” (Boas 2003: 126). 244 Pattern choice and verb meaning

Table 43. Distribution of to death in resultative constructions in the BNC (Boas 2003: 131)

Verb Occur. stab 114 beat 74 put 44 batter 39 frighten 34 crush 25 scare 24 burn 18 torture 16 drink, starve 15 bludgeon, hack 12 shoot, kick 11 club 9 bore, knife, choke 8 blast, trample, work, worry 7 love 6 strangle 4 dash, poison, kiss 3 ax, bayonet, boil, bring, clap, suffocate, kick, freeze, spear, spray, 2 stone, suck, gun, hammer, hug, knock, nag, peck, play, rape, shag, sting annoy, eat, bleed, blend, bug, bully, stab, flog, frit, cudgel, curse, 1 dance, feed, gas, flog, jog, laugh, pitchfork, pound, run, schmaltz, scorch, scratch, seduce, shock, sing, smother, squash, squeeze, stamp, strike, sweat, whip

Table 44. Distribution of dead in resultative constructions in the BNC (Boas 2003: 130)

Verb Occur. shoot 408 cut 11 kill 9 strike 8 stop 6 make, knock 3 make the ball 2 flatten, kick, smite 1 Implications 245

Besides such conventional constructions, there are what Boas calls transparent compositional constructions which should be explicable on the basis of general semantic principles and a border area between convention- al and compositional which he describes as motivated (2003: 139–140). Due to this “in-between” status such motivated constructions are easily interpreted or “decoded”, using the distinction made by Fillmore, Kay, and O’Connor (1988: 504–505), but it is impossible to “encode” them, i.e. to know how to use them if the speaker is not already familiar with them (Boas 2003: 141). Further arguments in favor of conventionalization are differences to be found in a contrastive analysis of different varieties of a language or even different languages. Thus Boas names the following examples in order to illustrate the point that although the meaning of the verb send and drive are similar in both varieties, British send has an additional use which can only be accounted for as a conventionalized use absent in American English (Boas 2003: 142):

(3)QE This problem drives me mad. Æ British and American English 39 (4)QE This problem sends me mad. Æ only British English

In a corpus-based study of the into-causative, Wulff, Stefanowitsch, and Gries (2007: 279) also found that there are “differences in the way British and American speakers make use of the same construction”, e.g. there are more “movement-initializing cause predicates” (such as trigger) in British English and “movement-restricting cause predicates” (such as coerce) in American English. Thus the choice of pattern is the same but the verbs in the pattern differ. Hoey (2005: 8) describes structural differences between synonyms as a differences in priming, which according to him is “a property of the word and what is primed to occur is seen as shedding light upon the priming item”. Words can thus be primed for collocations, colligations, and seman- tic associations (Hoey 2005: 79). He concludes that

…cumulatively, with all caveats and cautions in place, the evidence sug- gests that synonyms are not typically identically primed. There are indeed shared primings, and in so far as there are, they reflect the close similarity

39. Sentence (4) was judged unacceptable by a British native speaker and there is no such use in the BNC, either. However, the unacceptability seems to be connected to the idiomatic status of ‘verb + me + mad’. The invented sen- tence The chemicals send him mad was accepted. 246 Pattern choice and verb meaning

of sense. But they also differ in important ways, the difference marking var- iations in use and context and providing a reason for the existence of syn- onyms in the first place. (Hoey 2005: 73)

The notion of conventionalization or priming differences describe the phe- nomenon but do not really provide an explanation, i.e. the question remains why a certain combination of pattern and verb becomes conventionalized while others are constructed fairly freely and transparently. Note that this also has a historical perspective which cannot be dis- cussed here in detail. As pointed out above, when talking about the valency pattern inventory of a verb, it is important to bear in mind that its conven- tionally used patterns are not necessarily fixed. Conventions change and so do complementation patterns of lexical items, i.e. a verb can come to be used in new patterns over time, while other patterns might be lost (cf. 3.3.2). Jacob (2004: 115) even argues that a discussion of linking is only possible taking a diachronic perspective “which explains the preference of a certain morphosyntactic position for a certain communicative function in reference to the etymological origin of the respective position and its syn- tactic properties resulting from it” [translation mine].40 Another important dimension, which can only be alluded to briefly in this context, is language acquisition. From such a point of view, the use of a verb in a novel pattern can be described as a gradual emergence of structure (see, for instance, MacWhinney 2001). According to Tomasello (2003: 163–169), the process of analogy plays a crucial role for the gradual growth of a verb’s pattern inventory (and patterns in general) in the process of first language acquisition:

When an analogy is made, the objects involved are effaced; the only iden- tity they retain is their role in the relational structure. … [T]he learner makes an analogy between utterances (or constructions) by aligning the ar- guments one to one, and in making this alignment she is guided by the func- tional roles these elements play in the larger structure. (Tomasello 2006: 274–275)

At the same time, semantic analogy was shown to have no effect in a study of training 2;6-year-olds the transitive construction (Abbot-Smith, Lieven, and Tomasello 2004). While this does not mean that semantic analogy does not play a role at all, it is difficult to prove experimentally, and there might also be other factors that play a role. In a further study Kidd, Lieven, and

40. For an overview on diachronic changes with respect to valency in the German language, see Habermann (2007). Implications 247

Tomasello (2006: 101) conclude that “[t]he results suggest that from an early age children keep track of the distributional environments in which individual verbs appear, and annotate this frequency information onto their lexical entries for verbs”. As important factors supporting the child in this, Tomasello (2003: 165–166) lists “type variability in the slots” as well as “consistency of the items in the slot”, and potentially something like a “crit- ical mass” of exemplars. Thus isolated meaning components are clearly not decisive factors in the acquisition of verb-pattern correlations. However, highly frequent verbs, such as think and give, might have facilitatory effects on learning and abstracting finite complement clause constructions or the ditransitive construction respectively (Kidd, Lieven, and Tomasello 2006: 102; Goldberg 1998). Hunston and Francis (2000: 96) also point out that the pattern inventory of a verb can be extended by means of analogy which means that “what words belong to a list is in a state of flux”. Pinker also argues along a similar line, however, he transfers this to his notion of lexi- cal rules which then determine the choice of possible structures:

… the historical processes that cause lexical rules to be defined over some subclasses but not others seem to favor the addition and retention of verbs whose own meanings exemplify or echo the semantic structures created by the rule. ... The full motivation for a subclass may come from the psycholo- gy of the first speakers creative enough or liberal enough to extend a lin- guistic process to a new item, as such speakers are unlikely to make such extensions at random. (Pinker 1989: 109)

If Tomasello’s criteria also apply to language change and not only to the acquisition of language, frequent patterns which can be used with many different verbs are more likely to be used in combination with a semanti- cally similar verb with which it has not been used previously. There were several cases discussed above where only an extremely limited number of examples were found in the BNC, which were partially accepted by native speakers and partially not. These might indeed be considered as sponta- neously formed in analogy to the use of a semantically similar verb in this pattern. However, this does not explain why others were still rejected by native speakers. Moreover, what leads to the establishment of such uses is still to be determined. In any case, it is extremely unlikely that this is rule- governed in any way. MacWhinney (1998: 215) argues in the context of “irregularities” that “the existence of differences between regular and irre- gular processing does not, in itself, provide strong evidence for the exis- tence of rules”. Referring to Bybee’s (1995) “Regular Morphology and the Lexicon” he states that “these attempts to preserve a role for rules in human 248 Pattern choice and verb meaning cognition have run into problems with the fact that even the most regular patterns or ‘rules’ display phonological conditioning and patterns of gra- dience” (MacWhinney 1998: 215; see also MacWhinney 1989: 197). If it is not rules which account for the mapping of meaning onto form, i.e. in the sense that a verb’s valency patterns cannot be inferred directly from its meaning in a rule-governed way, the lexical item itself must adopt a more important role in the localization of this piece of information. Lamb (2001: 177) argues:

We don’t have to look very far to see that … every verb of any other sub- category, has its own distribution. … And so we can conclude that every lexeme has its own syntax. … According to such a view the acquisition of syntactic knowledge is the acquisition of lexical knowledge.

Goldberg concludes in reference to complementation irregularities between the semantically similar verbs help and aid that

[e]ven though this sort of example may be rare, a learner cannot possibly know which patterns will turn out to be productive and which will not be on initial encounter. Thus it is clear that all early-learned instances must be stored. (Goldberg 2006a: 57)

Langacker (1987: 494) points out that the two notions “productive genera- lizations” and “elaborated lists” might not be in competition at all, which can be captured in usage-based models of language which he describes as “a non-reductive approach to linguistic structure that employs fully articu- lated schematic networks and emphasizes the importance of low-level schemas”. Thus, this does not mean that there are no generalizations at all. However what can be concluded from this is that storage of valency infor- mation in combination with the valency carrier must play a crucial role, since generalizations on the basis of meaning are simply not always possi- ble and the restrictions clearly amount to more than just some random ex- ceptions.41

41. For evidence of storage of whole patterns including the verb, see Bybee (1995). 4 The meaning of patterns

4.1 The pattern as an additional entity

4.1.1 Complement type-independent pattern restrictions

Even if it is possible to correlate all participants of a verb to specific com- plement types, it is still not necessarily possible to predict which patterns are actually available to that verb. Even if the combination of different par- ticipants is semantically plausible, this does not mean that all possible rea- lizations of these participants can be combined in one single pattern. This can be demonstrated with the verb force. The participants of force can be described as AGENT/THEME (the person or circumstances exerting force or creating pressure), AFFECTED (the person or thing at which the action is directed), and AIM (the desired effect on AFFECTED). The difficulty lies in the fact that one single form – [NP] – can realize all of the three partici- pants – yet never in the same sentence, i.e. in trivalent patterns either AF- FECTED or AIM must be realized by a different complement type as illu- strated in Table 45 and in the examples provided below.

Table 45. Valency patterns of force (based on VDE: 325)

SCU PHU PCU1 PCU2 e.g. *[NP]_AGENT/THEME force [NP]_AIM [NP]_AFFECTED (3) [NP]_AGENT/THEME force [NP]_AFF. [into_NP/V-ing]_AIM (4) [to_NP]_AIM (5) [to_INF]_AIM (6)

[NP]_AGENT/THEME force [NP]_AIM [on_NP]_AFFECTED (7a) [upon_NP]_AFFECTED (8)

(1)VDE And now Socialists in turn have decided to force a referendum AIM. (2)VDE The best kind of progress she can make will be slow and steady; try not to force her AFFECTED. 42 (3)NSA *They tried to force her a referendum. (4)VDE But would the Queen force her youngest into a Windsor wedding?

42. See Appendix 4 at htttp://dx.doi.org/10.1515/9783110240788_appendix4 for example sentences in chapter 4 which were judged by native speakers. 250 The meaning of patterns

(5)VDE One is forced to the conclusion that she wasn’t very good at whatever it was she did. (6)VDE The rise of the Nazis forced the family to move to London. (7a)VDE Quotas are now being forced on the Bahamian people. (8)BNC The council forced it upon us. CH6 1758 (7b)NSA *They force quotas the Bahamian people. (7c)NSA *They force the Bahamian people quotas. (7d)NSA *The Bahamian people are now being forced quotas. (7e)NSA *Quotas were forced the Bahamian people.

As the participants can co-occur in one and the same sentence as illustrated by examples (4) to (8) above, they are not merely subtypes of one and the same participant, which could explain such restrictions.

Valency Slot I [AGENT/THEME] Complement [NP] Adjunct force

Valency Slot II [AFFECTED] 1 Complement [NP] 2 Complement [on_NP] Valency Slot III [AIM] 3Complement [upon_NP] 1 Complement [NP] 2 Complement [to_NP] Interdependent 3 Complement [into_NP] pattern choice 4 Complement [to_INF]

Figure 28. Valency structure of force (based on VDE: 325)

Even if [NP] is considered polysemous in combination with force (ex- pressing AIM and AFFECTED), this alone cannot explain why not both par- ticipants can be realized by [NP] at the same time. Neither the order of participants nor whether the sentence is active or passive plays a role here. Obviously, the pattern [NP]+ verb +[NP]+[NP] is not generally blocked (e.g. They gave Peter the book.). The prototypical meaning of this pattern, – the so-called ditransitive construction – is that of transfer. According to Goldberg (1995, 2006a), this meaning is independent of the verb, i.e. the construction can be used with verbs which do not generally express transfer or have the participants BENREC and AFFECTED, since they can be contri- buted by the construction itself (e.g. Goldberg 1995: 54–55). Despite the fact that one could easily imagine a metaphorical transfer in the case of force, i.e. with her as a kind of BENREC and the referendum being forcefully The pattern as an additional entity 251 transferred to her, sentence (3) is not possible. Moreover, even if force can occur with the same participants in various patterns with a particle phrase, there are further differences between the patterns. In the case of the pattern in which AFFECTED is realized by [NP], there are numerous example sen- tences where [NP] is lexically much more restricted than would be ex- pected from the description as AFFECTED, e.g. in the form of the very com- mon phrase [NP]+ force + [(possessive pronoun) + way] +[into_X], illu- strated by example (9) and often discussed as the “way-construction” (e.g. Jackendoff 1991; Goldberg 1995).

(9)BNC He AGENT forced his way AFFECTED into our 3rd Division promotion team AIM. B2H 1306

With respect to AIM, it is possible to make a further distinction (i.e. on a subtype level): an alternative to AIM is GOAL which implies that the verb describes a movement towards a certain destination. In such cases AF- FECTED is followed by some kind of locative phrase. As can be seen from example (10a), this is independent of whether GOAL is inanimate or not, i.e. there are also examples for an animate GOAL as in (11).

(10a)BNC ... Austria, Britain and France signed a treaty on the basis of the Four Points which was designed to force Russia AFFECTED to the conference table GOAL by the end of the year. HY7 545 (10b)NSA *... Austria, Britain and France signed a treaty on the basis of the Four Points which was designed to force the conference table GOAL on the Russians AFFECTED by the end of the year. (11)BNC “Not 'ungry,” said Claudia, trying to force potato purée AFFECTED into little Paolo GOAL. CA0 1753

Idiosyncrasies, which are not directly connected to the participant types ex- pressed but rather to how they are combined, clearly show that the pattern itself must be considered as an additional level of analysis. Treating these valency patterns as realizations of the same participant patterns does not mean that the resulting valency constructions are completely synonymous.

(12a)NSA They forced them into slavery. (12b)NSA They forced slavery on them.

One might argue that semantically there is a difference between (12a) and (12b) in that there is a difference in the focus taken on AFFECTED in (12a), and on AIM in (12b), even if the participants expressed in the pattern remain the same. In terms of clausal roles (Herbst and Schüller 2008: 158–163) 252 The meaning of patterns one could argue that slavery in (12b) receives an additional role FOCUSSED AFFECTEDNESS because it is expressed as a noun phrase directly following the verb. Such clausal roles provide an additional level of semantic roles which do not function on the level of the complement but on the whole pattern. AGENTIVITY for instance is a role that elements receive – irrespec- tive of their participant role – when they fill the subject valency slot of a verb expressing an activity (Herbst and Schüller 2008: 158–163). This ac- counts for differences between pairs of sentences such as (13a) and (13b), where the factory is still AFFECTED by the activity but also clearly has an agentive component in (13b).

(13a)INV They closed down the factory. (13b)INV The factory closed down.

Likewise, the first complement in the predicate receives a specific kind of AFFECTEDNESS which is added to the participant role.

(14)QE Pat loaded the wagon with hay. 43 (15)QE Pat loaded the hay onto the wagon. (Goldberg 2006a: 34)

While wagon and hay are “CONTAINER” and “LOADED ITEM” in both pat- terns, it is the wagon in (14) and hay in (15) which are put into focus (Herbst and Schüller 2008: 159). Thus the pattern is clearly more than just the sum of the participants involved. Another example which underlines the relevance of the concept of pat- tern is the verb hear, which is discussed in Herbst (2007: 21–22). Hear specifies – among other participants – AFFECTED (the message), TOPIC (what the message is about and which can replace AFFECTED), and SOURCE (where AFFECTED – the message – comes from). According to VDE (381– 382), the following participant patterns are possible:

(16)VDE Some Liberal Democrats seem eager to hear from me SOURCE again on the subject of their party TOPIC. (17)VDE I love to hear about people who do things like that TOPIC. (18)VDE We want to hear from you SOURCE.

43. For a discussion of these examples see Goldberg (2006a: 34–37), Dowty (1991), Palmer ([1976] 1981: 148), and Fillmore (1968: 48). For an overview of the literature on this alternation see Levin (1993: 49–50). The pattern as an additional entity 253

The prototypical realization of SOURCE is [from_X] and for TOPIC, three parallel complement types are available (see below): [about_X], [of_X], and [on_X]. Yet SOURCE and TOPIC cannot freely be combined:

AGENT + hear +[from_NP]_SOURCE +[about_X]_TOPIC AGENT + hear +[from_NP]_SOURCE +[on_NP]_TOPIC *AGENT + hear +[from_NP]_SOURCE +[of_X]_TOPIC AGENT + hear +[about_NP]_TOPIC AGENT + hear +[of_NP]_TOPIC *AGENT + hear +[on_NP]_TOPIC

[About_NP], [of_NP], and [on_NP] can all function as realizations of TOP- IC. The difference between them is not semantic in nature and consequently it is not necessary to specify different subtypes. Instead, the difference here appears to be purely formal: [from_NP] is incompatible with [of_NP], and [on_NP] needs support from an additional complement, whereas [about_ NP] does not show any restriction in this context, yet still seems to be very rare in combination with SOURCE (only one example in the BNC).

(19)BNC Time and again we have heard tonight -- especially from the hon. Mem- ber for Withington -- about shortfall and about the measures that we should take to meet it. HHX 8640

Valency Slot I [AGENT] Complement [NP] hear Valency Slot II [AFFECTED] 1 Complement [NP] 2 Complement [that_CL] 3 Complement [wh-CL] Valency Slot IV [SOURCE] Valency Slot III [TOPIC] Complement [from_NP] 1 Complement [about_NP] 2 Complement [of_NP] Interdependent Adjunct 3 Complement [on_NP] pattern choice

Figure 29. Valency structure of hear (based on VDE: 380)

VDE lists only one single pattern in which AFFECTED and TOPIC can co- occur, i.e. +[NP]+[of_NP]. 254 The meaning of patterns

(20)VDE Often we hear more of birds than we see. [more than we see = discon- tinuous complement]

TOPIC must be realized by [about_NP] or [on_NP] when combined with SOURCE. The conclusion Herbst (2007: 22) draws is that “[t]his situation is difficult to describe in terms of a complement inventory since [from N] is optional in patterns with [about N] but obligatory in patterns with [on N]”. The semantic value of the participants alone does not provide a basis on which the form of the pattern can be predicted either: only knowledge of the complete list of participants activated for a specific scene (SOURCE or TOPIC vs. SOURCE and TOPIC) sheds light on the form the participants ac- tually take. Consequently, interdependency has to be taken into consider- ation between, on the one hand, the meaning of the complement type and, on the other hand, the complex structure of participants which co-occur in an event (cf. Herbst 2007: 22–23 for a similar example). Knowing the par- ticipants of a verb does not enable speakers to predict the patterns available to that verb, even if the complement types realizing the participants are known to them.

4.1.2 Participant mergers

Certain participants can occur separately but also in a joined form. This clearly speaks against a one-to-one relationship between participants and complement types, even with respect to the complements of a specific verb. There are several verbs which allow a participant CO-AGENT in addition to an AGENT, while these can be realized in different forms, e.g. A cooperates with B or A and B cooperate (Heringer 1996: 68; Herbst and Klotz 2002; see also Fillmore 1972 on symmetric predicates).44 This phenomenon is also taken into consideration in VDE by portraying these participants in a com- bined slot I+II with the different formal realizations they can take: as a single entity [Npl/group] (They met.) or [N and N] (Jim and Jon met.), or as two separate entities which semantically represent the same participants, i.e. [N]A and [N]P (Jim met Jon.) or [N] and [with_N] (Jim met with Jon.).

44. The fact that two elements in these sentences can be assigned a very similar participant role was taken to be counter-evidence for the theta-criterion, which demands a one-to-one assignment of roles to form (Frawley 1992: 233). To resolve this problem Jackendoff (1985, 1987, 1991) argues in favor of a model in which participants are assigned roles from two different tiers, which can lead to two different participants overlapping in one of their two roles. The pattern as an additional entity 255

This is not to say that such patterns are completely synonymous: there is at least a difference in the information structure in that the CO-AGENT is either part of the theme or the rheme (see, for instance, Halliday 1967b and Brown and Yule 1983: 126–133). Again, the difference in meaning could also be described by means of clausal roles (Herbst and Schüller 2008: 158–163): the participant functioning as subject usually receives an agen- tive reading even if the participant role is not necessarily a prototypical agent and the participant following the verb in the form of an [NP]-com- plement appears to be more affected than if it fills the subject slot (see also Dowty 1991). In that sense the CO-AGENT can appear more or less agentive and more or less affected. Also Dixon (1991: 11) argues that the element which is not the transitive subject “is regarded as most salient for the ac- tivity (often, the role which is most affected by the activity) which is put into O-relation [=transitive object].” Nonetheless, this interpretation ap- pears to be on a different level than the role specification, and it still makes sense to regard them as the same participants. Yet even if one accepts a slight semantic difference, it is striking that this variation among comple- ment types is not possible with all verbs specifying an AGENT and a CO- AGENT. Compare:

(1a)NSA Angelina kissed Brad. (1b)NSA They kissed. (1c)NSA *Angelina kissed with Brad/to Brad.

(2a)NSA Angelina loves Brad. (2b)NSA They love each other. (2c)NSA *They love. (2d)NSA *Angelina loves with Brad/toBrad.

(3a)NSA *Peter agreed/argued Tom. (3b)NSA Peter and Tom agreed/argued. (3c)NSA Peter agreed/argued with Tom. (3d)NSA *Peter agreed/argued to Tom.

(4)NSA *Jamila chatted Zahira. (5)BNC Taff and I chatted while he sat on the floor of the trench smoking his cigarette. A61 515 (6)BNC They chatted for a while longer, then Mrs Lennox and Susan set about packing, getting ready for the next day's evacuation. AN7 2084 (7)VDE I went to his shop after a missed lunch-hour simply to chat to him. (8)BNC Most of the day Jamila and Zahira chatted to each other. A6V 798 (9)VDE Hadn’t got the time to chat with her. 256 The meaning of patterns

(10)VDE When calling a number, you hear a nice woman’s voice saying ‘We are trying to connect you’. (11)BNC “Now get off the phone, Robert, for Christ's sake.” / “When will we con- nect?” / “Are you on duty tonight?” / “Yep. We're still short-handed.” / “Well head for my place when you're finished. ...HTG 3580-3585

This clearly shows that the possibility of merging is not dependent on the meaning of the participants but is an idiosyncratic feature of the individual verbs with the respective participant inventory. Such cases clearly speak against a determination of the pattern inventory of a verb solely on the basis of its participant inventory. Moreover, participants cannot only merge among each other. It has also been pointed out that what constitutes a separate participant with respect to some verbs might already be incorporated into the verb meaning or merged with the verb. Helbig (1992: 15) refers to this phenomenon as “verdichtetes Verb” (‘concentrated verb’). Also Jackendoff (1991: 23) argues that “not every conceptual constituent in the meaning of a sentence corresponds to a syntactic constituent, because … many conceptual constituents of a sen- tence’s meaning are completely constrained within lexical items”. His ex- ample pierce includes the meaning ‘through’ and jump includes the notion ‘over’ and ‘across’. Leisi (1971: 80–110) discusses this matter under the label “complex word content” and includes spatial aspects (‘swim includes water’), vectors (e.g. sink, rise) but also resultative verbs (e.g. construct). Lyons (1977: 262) refers to this phenomenon as encapsulation. In that sense it is of course the meaning of the verb – in the sense of semantic components – that restricts the combination with certain participants: if a relation is already implied in the verb meaning, an expression in the form of an additional complement is redundant. This, however, shows that a generalization across participant patterns is not possible. The fact that a participant pattern exists and is semantically compatible with the activity or event described by the verb does not necessarily mean that it is indeed ac- ceptable.

4.1.3 Instability of complement-participant correlation

The relationship between the participant pattern of a verb and its realization as valency pattern in many cases is a relationship of one-to-many and clear- ly not a one-to-one relationship. As has been shown in chapter 3.3.4, there are cases where variation is possible in the realization of one and the same participant. Yet even beyond this clearly defined space of a valency slot, The pattern as an additional entity 257 there is the possibility of conveying the same semantic idea in different ways, i.e. in the form of verb-dependent complements but also in the form of verb-independent adjuncts. This, of course, depends heavily on the defi- nition of these terms and the distinction between complement and adjunct status. Hence, the argument partly rests on whether a particle phrase such as on that point in (1) is regarded as a complement or as an adjunct. In a valency context it makes sense to define this phrase as a complement since it is clearly determined in its form by the verb concede, which cannot be complemented by any other particle phrases but [on_X] to express the par- ticipant TOPIC.

(1)BNC She was persuaded to concede on that point, then proposed to share everything with the production team -- …. BN6 581 (2a)BNC On finance, Heseltine conceded that the public “had not been per- suaded that the [community] charge is fair”. HL5 1751 (2b)NSA Heseltine conceded - on finance - that the public “had not been per- suaded that the [community] charge is fair”. (3)BNC On Cambodia, the two sides agreed that the UN “framework docu- ment”… was a “good foundation” for the achievement of an all-em- bracing political settlement. HL7 843

If sentence (1) is compared to sentence (2a), it seems to be quite clear that the fronted particle phrase on finance represents the same participant role. Also the verb agree demands to be complemented with [on_X] or [about_ X] if a REFERENCE is to be expressed and example (3) can be interpreted as a realization of the participant pattern REFERENCE + AGENT + verb + ÆFFECTED, where REFERENCE is thematically fronted. From a purely for- mal point of view, however, example (2a) and (2b) as well as (3) constitute at least borderline cases and one might argue that it is more sensible to classify them as adjuncts as they are completely optional, mobile – at least to a certain degree, and they could easily be replaced with such phrases as for instance “with respect to finance/Cambodia”, “when it comes to finance/Cambodia”.45 Welke (2005: 47) argues that semantic participants can be both complements and adjuncts, pointing out the notorious difficulty of differentiating between these entities:

Complements and adjuncts (arguments and modifiers) are obviously diffe- rentiated between in a vague and prototypical way. Thus, they are well dis-

45. For different syntactic-operational tests for the determination of complement status versus adjunct status and their evaluation see for instance Helbig (1992: 78–87). See also Somers (1987). 258 The meaning of patterns

tinguished in a core area. There is, however, a broad border and transition area. This renders the differentiation a notorious problem. (Welke 2005: 44) [translation mine]

Despite the lexical property of verbs to accept certain formal complement types and deny others to express certain participants, the speakers’ options are not necessarily completely limited by such conventions. Moreover, whether an element is classified as a complement or as an adjunct might also be connected to the frequency of a certain form occurring with the verb and thus the degree of entrenchment of such a combination. In the case of concede and agree, which can only be complemented with a particle complement containing on, the classification as complement is, due to this limitation, rather uncontroversial. There are, however, also verbs which can be complemented with a variety of particle phrases where the choice of particle depends on the following noun. While this is generally an indicator for a classification as an adjunct as the form is not determined by the verb and it is not obligatory, there are still borderline cases. This was also taken into consideration in the compilation of VDE, while here se- mantic valency, i.e. the link to a participant of the verb, was taken as an important factor.

Whether adverbials in such cases [i.e. when they are not obligatory] are to be considered complements or adjuncts largely depends on how strongly one would want to argue that the semantic roles expressed by them form an integral part of a verb’s semantic valency. (VDE: xxxiii)

Thus the pattern illustrated with examples (4a) and (5) below is described as +[NP]+[ADV], where [ADV] is not obligatory as can be seen in (4b) and (6).

(4a)VDE Subsequently, other measures have been adopted to liberalise the econ- omy and persuade the rest of the world that Vietnam should be accepted back into the global economic community. (4b)NSA Subsequently, other measures have been adopted to liberalise the econ- omy and persuade the rest of the world that Vietnam should be accepted. (5)VDE Burma had not allowed foreign journalists and other observers into the country. (6)VDE No persons over 16 allowed. (only if clear from context)

This is not to say that the distinction between complement and adjunct is not considered sensible as such – an issue raised for instance by Jackendoff (1991: 176), who argues that “[t]he argument-adjunct distinction, while it The pattern as an additional entity 259 has been useful as a rough-and-ready-criterion, has on the whole simply been assumed. If it should turn out that a more precise treatment of the distinction reveals intermediate cases, so what?” While valency theory developed criteria and tests which show that it is useful to make such a distinction (cf. 1.2.2), it clearly has to be acknowledged that there is a gra- dient. From a purely semantic point of view, however, the dividing lines between participants do not necessarily coincide with complement bounda- ries. To predict the form of a sentence on the basis of the participants is thus even more imprecise since it is not simply a choice of complement type but of a combination of elements – including complements and ad- juncts – that would then have to be taken into account for the determination of form through meaning. Similarly to the examples above, participants which are usually ex- pressed as verb-dependent complements can also be realized as noun-de- pendent complements, i.e. integrated into another verb-dependent comple- ment. The verb unite can be complemented either by +[in_NP] realizing the AIM of the action or by +[against_NP] realizing the OPPONENT. Compare (7a) with (7b):

(7a) Every radical revolt of mid-nineteenth-century Spain could unite the artisans and workers AFFECTED in a protest AIM. FB7 584

Figure 30. Stemma for example sentence (7a)

(7b) Every radical revolt of mid-nineteenth-century Spain could unite the artisans and workers AFFECTED against the consumos OPPONENT.

Figure 31. Stemma for example sentence (7b) 260 The meaning of patterns

However, it is also possible to combine these elements, which results in a different structural hierarchy: the two complements have formally fused into one, while the two participants can still be identified.

(8)BNC Every radical revolt of mid-nineteenth-century Spain could unite the arti- sans and workers AFFECTED in a protest against the consumos. FB7 584

Figure 32. Stemma for example sentence (8)

This clearly shows that participants, when verbalized, are neither neces- sarily tied to specific complement types nor to specific grammatical con- stituents. Such cases also show that the different levels and hierarchies established by means of such an analysis are not necessarily equally rele- vant from the point of view of language processing. Yet even from the point of view of a descriptive analysis it has to be taken into account that it is possible to establish different hierarchies (cf. Herbst and Schüller 2008: 6–12). Similar examples can be found with many verbs which can be comple- mented by noun phrases and particle phrases, as for instance the verb hear, which has already been discussed above. Strictly speaking, the two patterns below represent two different participant patterns.

[1] hear +[from_NP]_SOURCE +[on_NP]_TOPIC [2] hear +[from_NP]_SOURCE +[NP]_ÆFFECTED

Yet it depends crucially on the type of noun heading the noun phrase in pattern [2] whether ÆFFECTED is an appropriate label or whether TOPIC might not be a better choice, as for instance in example (9) below, due to the complementation of the noun information by [on_NP]: the comple- ments of hear and information are typically headed by on. Thus it is ambi- guous to a certain degree on which of the two [on_NP] depends.

(9)BNC We now, I think er, going to hear from members of the Management Team, more detailed information on working progress and priority is- sues. JNK 149 The pattern as an additional entity 261

In that sense, the underlying participant pattern seems to be quite flexible: the lexical meanings of the complements and their own valency conditions can influence the overall realization of the participants of the verb.

4.1.4 Realization of participants dependent on the overall pattern

The analyses presented in chapter 3 also produced several cases where one and the same participant, formally realized by the same complement type(s), is generally compatible with semantically similar verbs. However, it occurs in different patterns depending on which of the verbs it combines with. In some cases, it can occur as the only predicate complement unit and not in combination with other complements, i.e. in a divalent pattern and not in a trivalent pattern (despite the availability of such a pattern with a semantically similar verb). In others, only a pattern with additional comple- ments appears possible (while uses with it as the only complement in the predicate have been identified with semantically similar verbs). Suggest and recommend can be used as speech act verbs (cf. Leech 1983: 203 and 206), i.e. they can be complemented with a participant EF- FECTED, which can either be realized by +[QUOTE] or by +[that_CL]. Moreover, both verbs have a participant BENREC, realized by [to_NP].

(1)VDE “They make an excellent reindeer stew here,” he recommended. (2)VDE “They might be doing the dishes,” David suggested after a moment. (3)VDE The Commission recommends that rape within marriage be made a criminal offence. (4)VDE Reports from West Germany suggest that the government in Bonn is likely to offer an enormous package of financial assistance.

Yet they are not freely compatible: one can suggest and recommend [to someone] (BENREC) [that something should be done] (EFFECTED), i.e. com- plement them with +[to_NP]_BENREC +[that_CL]_EFFECTED as in exam- ples (5) and (6). However, only suggest can freely combine BENREC and the other formal realization of EFFECTED, i.e. [QUOTE], whereas recommend was rated as unacceptable:

(5)VDE Senior officers in his force have recommended to him in a report that the force should not contest an appeal being made by the group of six Irish people, known as the Birmingham Six. (6)BNC McLeish decided that it was probably not the moment to suggest to her that the real reason any civil servant disliked lobbyists must be that the 262 The meaning of patterns

chaps were paid to make sure Ministers got a view other than the De- partmental one. AB9 630 (7)VDE Interviewer Owen suggested to Nesta: “The fact that you are here speaks volumes, I think.” (8)NSA *She recommended to John: “Go and see a doctor.”

Similarly, it is possible to ponder and contemplate on something (+[on _NP]_TOPIC), as in (9) and (10), as well as to ponder and contemplate something (+[NP]_ÆFFECTED), as in (11) and (12), but it is not possible to combine the two in the case of ponder or contemplate, although +[NP] +[on_NP] is an acceptable pattern with other verbs in the group, cf. (13).

(9)BNC I pondered also on the question as to why it matters so much… CES 1457 (10)BNC Needless to say, the sight of such impressive architecture stimulates me, and I begin to contemplate on how rail travel has been a source of ar- tistic inspiration to passengers for more than a century. ABS 743 (11)BNC Only the previous day Louisa had been pondering a passage from the Aurelia occulta in which …. H82 190 (12)BNC If we are contemplating a two-tier service, so is the Labour party. HHX 2336 (13)VDE At the age of eighty she reflected/*pondered/*contemplated on this situation: “My satisfaction in living there was double.”

Train can be used without any predicate complement units (+[0]) as in (14), one can train someone (+[NP]) as in (15) or train someone how to do some- thing (+[NP]+[wh-to_INF]) as in (16), but no example could be identified for train how to do something (*+[wh-to_INF]), although one can practice how to do something (17).

(14)VDE “It’s really hard to train on your own,” she admitted. (15)VDE When we train professional athletes, we get them into the lab and mon- itor their progress very carefully. (16)BNC Utilizing the instructional capabilities of the PLATO computer-based education system, the course trains users how to carry out searches on the library shelf-list database. B3D 240 (17)VDE Preparation for tests should consist in learning the subject being tested, not practicing/?training how to answer this or that type of question.

It is possible to bear something in mind about something (+[what/sth/etc.] +[about_NP]) (18), to bear something in mind (+[NP]) (20), but not to bear in mind about something (*+[about_NP]) (22), although it is possible to remember about something (23). The pattern as an additional entity 263

(18)BNC One of the basic factors we have to bear in mind about fungi is that, …, fungal attack in one form or another is about and ready to take advan- tage of any weakness for much longer periods …. CMM 1175 (19)VDE What do you remember about her voice? (20)BNC …, you should bear in mind a few good rules. BNA 1157 (21)VDE I remember her well. (22)NSA *Let’s bear in mind about this. (23)VDE “Don’t you remember about Anna?” I shook my head.

These examples clearly indicate that even if one knows the semantic par- ticipants of a verb and how they can be formally expressed, it is not possi- ble to predict the actual uses, i.e. patterns a verb can occur in. Moreover, the similarity is not just due to a similar set of participants, also the activi- ties are rather similar and also the complement types acceptable are iden- tical. Nonetheless an explanation based on meaning – of the participant, the complement type or the verb – is problematic. Construction meaning along the lines of Goldberg (1995: 54–55) adds an additional dimension, since it can provide a role which is not part of the participant inventory of a verb. To what extent one can really say what is part of a verb’s participant inventory and what not on the basis of such sentences appears to be problematic. The reason for a classification of off the table as a goal contributed by the construction and not the verb can only be based on the fact that there is not much evidence for such a construction, i.e. it is at least highly infrequent, but the sentence is still considered ac- ceptable by native speakers. In that sense, establishing the participant struc- ture can never be anything but the search for prototypical and rather fre- quent complementation patterns. Cases like these and the examples discussed above clearly show that the notion of verbs being equipped with complement inventories from which they freely choose, only coerced by the need to provide for a certain se- mantic plausibility is misguided (see also Herbst 2007: 19–20, 27).

4.1.5 Conclusion

The fact that participants do not always directly match up with comple- ments which are generally taken to be their formal correspondents is prob- lematic when the choice of complement is considered to be determined by meaning. Different participants can merge into one single complement or they can be expressed in separate complements, while – as has been shown above – which different forms are available to a verb for this separate ex- 264 The meaning of patterns pression is again item-specific. Similarly, the fact that a participant (e.g. TOPIC) can also merge into another complement adopting the form of a postmodification shows that this clear pairing of complement and partici- pant is problematic. The same applies to the fact that the same semantic content can sometimes be expressed in the form of verb-dependent com- plements but also in the form of verb-independent adjuncts. Moreover, as the examples above show, different participants can be expressed by the same complement types, although there is interdependence between them as regards the choice of pattern, i.e. knowledge of possible patterns goes beyond the knowledge of a verb’s participants and their formal realizations. All this makes a prediction of valency patterns on the basis of meaning very difficult and probably even impossible.

4.2 The relationship between pattern and meaning

4.2.1 Same valency pattern – same participant pattern?

4.2.1.1 Theoretical background: assumptions of construction grammar

As shown in the preceding paragraphs, the valency pattern must be taken into consideration as an additional level of analysis – at any rate from a de- scriptive point of view but also very likely from a cognitive point of view. This is also reflected in one of the major developments in current linguis- tics which has already been mentioned here in context of different aspects but which clearly demands a more detailed discussion, i.e. constructions. In the framework of construction grammar of the Goldbergian type (1995, 2006a) argument structure constructions are not just formal patterns – as valency patterns have been defined – but per definition form-meaning pair- ings. Here, the term valency construction is employed to refer to combina- tions of formal valency patterns and semantic participant patterns, yet with- out raising any claims as regards a one-to-one relationship for pattern form and pattern meaning (cf. Herbst and Schüller 2008: 139–141; cf. chapter 1.2). As already partly discussed in chapters 2 and 3 Goldberg (1995: 67) as- sumes the following “psychological principles of language organization” as relevant for the notion of constructions: (1) Principle of Maximized Motiva- tion46, (2) Principle of No Synonymy, (3) Principle of Maximized Expres-

46. Cf. also Baker’s (1988: 46) Uniformity of Theta Assignment Hypothesis (UTAH), according to which “[i]dentical thematic relationships between The relationship between pattern and meaning 265 sive Power, and (4) Principle of Maximized Economy. According to the first principle, syntactically related constructions are also semantically re- lated. The second principle assumes that formally distinct constructions must also be semantically distinct. According to the third principle, “[the] inventory of constructions is maximized for communicative purposes”, while the overall number of constructions, according to the fourth principle, is minimized (Goldberg 1995: 67–68). The last two principles emphasize two important aspects which play a role in human language, i.e. creativity and processing constraints. On the one hand, speakers of a language evi- dentially make use of innovations in order to express one and the same proposition in a different way, thus drawing specific attention to what is being said and/or adding additional facets like irony. This phenomenon is, for instance, seen as one of the driving forces in the process of grammati- calization: although speakers have available to them certain means of ex- pressing a certain function, for instance future time reference, a new means becomes established (e.g. referring to future time by means of be going to). As a consequence, the meaning of formal patterns can change over time and they either lose their original meaning or, alternatively, maintain the original meaning and become polysemous (cf. Hopper and Closs Traugott 2003: 124–125 for the concept of layering). This creativity – and thus also the process of grammaticalization – is constrained by pragmatic principles such as the cooperative principle which is to ensure that comprehension is facilitated as much as possible and by not placing too dramatic strains on speech processing (Hopper and Closs Traugott 2003: 71–73). This is ex- pressed by Goldberg’s principle of maximized economy (Goldberg 1995: 67–68). However, the fact that according to grammaticalization theory con- structions can be polysemous – as an intermediate state or possibly even as a stable situation – contradicts Goldberg’s principle of no synonymy. This issue has already been discussed to some extent in chapter 3: the comparison of semantically similar verbs has led to the conclusion that there are indeed patterns that are at least semantically similar enough to preclude a satisfactory prediction in which of such alternative patterns a verb can occur. Thus the assumption that different patterns must express different meanings could not be supported to a satisfactory degree in the preceding chapter. As this chapter is concerned with the pattern as the pri- mary unit, the same question will be approached from the opposite direc-

items are represented by identical structural relationships between those items at the level of D-structure.” 266 The meaning of patterns tion, starting out with valency patterns and trying to relate them to their underlying participant patterns and the verbs associated with them. Goldberg’s principle of maximized motivation (see above) assumes that formally identical or at least similar patterns also share common semantic ground. This chapter will examine this maxim in reference to VDE data. One obvious counter-example is the trivalent pattern [NP]+ verb +[NP] +[NP] which can either realize the participant pattern AGENT + verb + BEN- REC + ÆFFECTED as in (1) or AGENT + verb + ÆFFECTED + PREDICATIVE as in (2).

(1)INV She called him a taxi. (2)INV She called him a fool.

According to Goldberg (2006a: 21), these two constructions are not related and constitute different constructions, since PREDICATIVE (PRED in her pa- radigm) can also be realized by an adjective phrase. Consequently she ar- gues in favor of a classification of the construction by means of the rela- tional term PREDICATIVE as opposed to a such as NP because this way the two constructions can be distinguished (Goldberg 2006a: 21). At the same time, this clearly contradicts the notion of con- structions as form-meaning pairings: the two constructions are formally identical in case of examples (1) and (2). This appears to be a clear contra- diction to the maximized motivation hypothesis. Similarly, in chapter 2, where the focus was on alternative complement types, the close analysis of example sentences revealed that there are cases where one and the same complement type can express different semantic nuances in the same pat- tern. The pattern [SCU]+ cheat +[on_NP], for instance, has been analyzed as expressing clearly more than one meaning (cf. 2.2.2), i.e.

(3)BNC If Iran is cheating on nuclear weapons REFERENCE, the International Atomic Energy Agency has yet to catch it at it. CRC 1468 (4)BNC “How did you feel about cheating on Celeste VICTIM?” JY9 3218

The question which clearly needs to be asked here is what establishes a “different” meaning. In the examples above, the differences can clearly not go back to differences in verb meaning because the verbs used in the pat- terns remain stable. Goldberg’s paradigm does not presuppose semantic identity within the same formal constructions as long as they are still related or retraceable to a common basic constructional meaning. While she does “not deny the exis- tence of constructional homonymy”, this qualification is only mentioned in The relationship between pattern and meaning 267 a footnote where she also states that “it is important not to assume massive ambiguity without seeking out broader surface generalizations” (Goldberg 2006a: 36 footnote 14). Thus the construction is taken to carry meaning independently of verb meaning, which goes back to some basic verb mean- ings such as give for the ditransitive construction (Goldberg 1995: 39–43; 1998). Consequently, this construction meaning ought to be recognizable, when instances of the same formal pattern are compared. Such a compari- son is facilitated by the Erlangen Valency Patternbank, an electronic data- base which, according to its introductory page on www.patternbank.uni- erlangen.de:

…lists all valency patterns that were identified for the verbs, adjectives and nouns contained in the Valency Dictionary of English. In its present format, it allows researchers to find − a complete list of the valency patterns of English verbs (active and passive), adjectives and nouns on the basis of the VDE − for each valency pattern a list of all lexical units given in the VDE as occurring in that pattern − for each word all valency patterns indicated in VDE.

Since the patterns can be sorted according to the number of lexemes as well as the number of lexical units with which they combine (in reference to VDE), it is possible to identify patterns which are open to many verbs as well as patterns which are more restricted in their use, i.e. patterns with a high and a low type frequency. The correlation between pattern form and pattern meaning as well as the question of verb meaning(s) related to a specific pattern are two questions for which the Patternbank constitutes a prime research tool. Regarding its size with patterns of more than 500 verbs and an overall number of more than 3,500 patterns, these questions will only be examined here in an exemplary way. Yet in order to provide a pic- ture of the relationship between form and meaning as thoroughly as possi- ble, formal pattern meaning is included as another important aspect.

4.2.1.2 The pattern [NP + verb + NP + NP]

Returning to examples (1) and (2) above, the trivalent active verb pattern [NP + VHCact + NP1 + NP2] (consisting of a noun phrase subject, an active verbal head complex47 , i.e. the main verb and potential auxiliary verbs, and

47. The term verbal head complex refers to the combination of auxiliary verb (or prehead) and main verb (or verbal head) without any further complements of 268 The meaning of patterns two noun phrase complements) occurs with 88 of the verbs covered in VDE, yet since this pattern can in part occur in combination with more than one meaning of these 88 verbs, the lexical unit count amounts to 121.48 The prototypical participant patterns realized by this valency pattern are AGENT + verb + BENREC + ÆFFECTED (I call him a taxi) and AGENT + verb + ÆF- FECTED + PREDICATIVE (I call him a fool) as pointed out above. Roughly one fifth of all occurrences of the pattern in VDE appear to realize the par- ticipant pattern AGENT + verb + ÆFFECTED + PREDICATIVE, while the rest are for the most part realizations of AGENT + verb + BENREC + ÆFFECTED.49 Besides a semantic difference between the two constructions, a typical ar- gument for referring to this pattern as two different constructions – also in non-constructionist frameworks, e.g. in CGEL – is the fact that the par- ticipant PREDICATIVE cannot become the subject of an equivalent passive construction, i.e. it is not a grammatical object, when object is defined in this way, while both BENREC and ÆFFECTED, which are typically analyzed as two objects, can.50 The actual number of patterns in which both noun phrases can actually become the subject of an equivalent passive construc- tion, however, reveals that this claim does not seem to hold true for every so-called ditransitive construction.51 Although only about one fifth of the constructions include a PREDICA- TIVE, which cannot be passivized, the number of patterns where the second post-verbal noun phrase in the pattern cannot be passivized amounts to more than a third of all patterns (see Figure 33 where a subscript pass indi- cates whether the noun phrase can become the subject of a corresponding passive clause). This means that there are patterns which are structurally identical to the pattern expressing a PREDICATIVE with respect to passiviza- bility and different from the prototypical ditransitive construction, although semantically they are clear cases of the participant pattern + BENREC +

the verb (cf. Herbst and Schüller 2008: 43–45), avoiding the ambiguity of the term verb phrase. 48. Based on www.patternbank.uni-erlangen.de (as of May 26, 2009). 49. See Appendix 5 at htttp://dx.doi.org/10.1515/9783110240788_appendix5 for the example sentences on which this classification is based. 50. For the problem of defining object see Herbst and Schüller (2008: 167–172) and Herbst and Faulhaber (forthcoming). 51. The information on passivizability of complements in the Patternbank and the VDE rests on native speaker assessment in cases where no passive examples were identified in the Cobuild Corpus, which served as the basis for VDE. Additional native speaker tests were made for the examples discussed here. The relationship between pattern and meaning 269

ÆFFECTED. Compare examples (5) and (6), where none of the noun phrases can become the subject of a corresponding passive clause.

(x) number of patterns x proportion in percent

NP+NPpass(2) 2%

NPpass+NP (36) 30%

NPpass+NPpass (73)

60% 8% NP+NP (10)

Total: 100% (121)

Figure 33. Formal differences for [NP]+ verb +[NP]+[NP] for lexical units in VDE52

(5a)VDE Can you AGENT answer me BENREC one question ÆFFECTED, John? (5b)NSA *The question can be answered (to) him. (5c)NSA *He was answered the question. (6a)VDE Little did she know that the portrait of her mother so carefully and faith- fully made up of exact memories would be later published and would help to earn her BENREC the Nobel Prize ÆFFECTED. (6b)NSA *She was earned the Nobel Prize. (6c)NSA *The Nobel Prize was earned (to) her.

The construction including PREDICATIVE also has two variants in this re- spect: patterns, where the first noun phrase can become the subject of a corresponding passive clause and patterns where this is not possible. Com- pare examples (7) and (8):

(7a)VDE I AGENT’ll call you ÆFFECTED Den PREDICATIVE.

52. Note that some of the lexical units also occur in patterns where, “extraposi- tion with a dummy subject it is possible” or even obligatory (VDE: xii). 270 The meaning of patterns

(7b)NSA He is called Den. (7c)NSA *Den is called him. (8a)VDE Thea said she AGENT had not imagined you ÆFFECTED a poet PRE- DICATIVE. (8b)NSA *You were not imagined a poet. (8c)NSA *A poet was not imagined you.

Structurally, example (8) and examples (5) and (6) are identical. Similarly, example (9) below represents a case where only the first but not the second post-verbal noun phrase can become a subject (+[NPpass]+[NP]), just as in examples (7) and (10). With respect to example (10c), native speakers did not agree, which makes the pattern at least unconventional:

(9a)VDE An opposition leader in Zaire has confirmed that he has decided to de- cline an offer by President Mobutu to appoint him ÆFFECTED Prime Mi- nister PREDICATIVE. (9b)NSA He was appointed Prime Minister. (9c)NSA *Prime Minister was appointed (to) him. (10a)VDE Perhaps she AGENT would allow him BENREC the pleasure of taking her to lunch ÆFFECTED afterwards? (10b)NSA He was allowed the pleasure of taking her to lunch afterwards. (10c)NSA ?The pleasure of taking her to lunch afterwards was allowed (to) him.

These inconsistencies clearly show that meaning alone is very problematic for determining syntactic behavior. Moreover, there are also semantic dif- ferences, i.e. constructions which neither fall into the prototypical category of the ditransitive construction nor into the “predicative construction”. In the case of example (11a), not only the post-verbal part of the construction differs structurally from what could be called a “prototypical” ditransitive construction, i.e. in that passivizability is not possible. Semantically, also the noun phrase functioning as subject complement unit is less AGENT-like than Peter in a sentence such as Peter called him a taxi and cost does not really express any kind of transfer, either. In that sense a third construction would have to be assumed (cf. the corresponding FrameNet roles in brack- ets).

(11a)VDE The Government estimates that roadworks THEME (GOODS) cost road users BENREC (PAYER) at least £55m a year ÆFFECTED (ASSET). (11b)NSA *Road users were cost at least £55m a year. (11c)NSA *£55m were cost (to) road users. The relationship between pattern and meaning 271

There are further examples where the classification as AGENT + verb + BENREC + ÆFFECTED must be doubted. In (12) I is clearly more BENREC than her.

(12)VDE Susan was so generous with her time and her information that I owe her pretty well everything I know about publishing.

Finally, some verbs can express two different participant patterns with the same valency pattern:

(13)VDE The shopping took me two hours. (invented example) (14)VDE During the chase, the gunmen AGENT took four children ÆFFECTED hostage PREDICATIVE. (15)VDE The union AGENT voted itself BENREC larger welfare benefits ÆFFECTED. (16)VDE She ÆFFECTED was voted Most Promising Actress PREDICATIVE by the London theatre critics AGENT.

Similarly as in example (11), the shopping in (13) is not really AGENT-like, and two hours are not really ÆFFECTED. This use of take is very idiomatic, which makes a classification problematic. The same applies to take some- one hostage or prisoner, where the second post-verbal noun phrase is re- stricted to a very clearly defined set of lexemes. A classification as PRE- DICATIVE appears possible here, since four children and hostage are refe- rentially identical. In the case of vote, sentences (15) and (16) can indeed be analyzed as representing the two prototypical participant patterns with AGENT, BENREC, and, ÆFFECTED and AGENT, ÆFFECTED, and PREDICATIVE respectively. The same applies to the following examples:

(17)VDE We AGENT had to step in and find her BENREC a home ÆFFECTED. (18)VDE I AGENT personally find them ÆFFECTED a nuisance PREDICATIVE and disagreeable in appearance. (19)VDE Radice is also returning to his favourite Sardinian restaurant, where the boss, old friend Giovanni of the Hotel Gallia AGENT, has kept him BE- NREC a table ÆFFECTED. (20)VDE The expense of it THEME will keep me ÆFFECTED a very poor man PRE- DICATIVE.

Make can even be used in three different participant patterns, the prototypi- cal ditransitive construction with AGENT, BENREC, and ÆFFECTED (21) and two different types of predicative constructions: the prototypical con- struction discussed so far in which PREDICATIVE refers to ÆFFECTED (22) 272 The meaning of patterns but also a second construction which includes a BENREC and in which PRE- DICATIVE refers to THEME as in (23) (see also CGEL 16.46):

(21)VDE I AGENT let her in and made her BENREC a cup of coffee ÆFFECTED. (22)VDE The superhuman effort he brought to his work THEME made him ÆFFECTED a superstar PREDICATIVE. (23)VDE There can be little doubt she THEME would make him BENREC a happier wife PREDICATIVE.

These cases clearly show that even with respect to such a well-attested construction, there is still a considerable amount of idiosyncrasy which simply cannot be captured by means of semantic generalizations.

4.2.1.3 The pattern [NP + verb + for_NP]

The same applies to other constructions as well, e.g. constructions in- cluding particle phrases. Leaving completely idiomatic uses aside, the ac- tive verb pattern [NP + VHCact +for_NP] (consisting of a noun phrase as subject, an active verbal head complex and a particle phrase headed by for and complemented by a noun phrase) occurs with 136 different lexical units (= type frequency in the Patternbank). The range of participants to be found in this pattern is much wider than with the pattern consisting exclusively of noun phrases discussed above. While noun phrases are not determined in their meaning by any morphological elements which are an intrinsic part of the pattern as is the case with particle complements, the latter seem to be semantically more flexible. Contrary to what might be expected, the “neu- tral” noun phrases are much more restricted in constructions than particle complements, as it is their configuration which determines the meaning of the construction. This indeed points towards an independent construction meaning as suggested in the context of construction grammar, while the examples above clearly show that there are also exceptions. Particle phrases do not have one stable meaning which shapes the mean- ing of the pattern but are already inherently polysemous. [NP + verb + for_NP] can realize the following participant patterns (in reference to the data included in VDE) as exemplified in sentences (24)–(30). The preverbal complement typically expresses AGENT, although there is also one partici- pant pattern where an interpretation as THEME appears to be more plausible (cf. chapter 1 for a role description). The relationship between pattern and meaning 273

AGENT +verb + AIM (24)VDE At the moment, it is the social forces and opposition forces who have to push for something concrete.

AGENT + verb + BENREC (25)VDE I collect for the hospital charity.

AGENT + verb + GOAL (26)VDE We then turned to the right and made for Trenton, a village on the De- troit River near its entrance into Lake Erie.

AGENT + verb + CAUSE (27)VDE ... “from a Marxist point of view it is unjustifiable to punish for theft, vandalism, rape, murder, fraud, etc.”

AGENT + verb + REFERENCE (28)VDE How many plants of a kind should I group to get the best effect? I have to calculate for this.

THEME + verb + PRICE (29)VDE Touts said yesterday that semi-final tickets were going for £250 and men’s final tickets for £ 1,000 a pair.

AGENT + verb + PRICE (30)VDE One bright morning you pass through St James Park and the cloth lawn chairs are set out by the pond, yours to rent for 50p.

The examples show that the pattern clearly has a fairly wide range of mean- ings, even if slightly more than half of all the occurrences in VDE of the pattern [NP+ verb +for_NP] are indeed best analyzed as realizations of the participant pattern AGENT + verb + AIM (see Figure 34 below). This is not to say that the distribution for participant roles or the choice of [for_NP] in a pattern is completely arbitrary. Goldberg (1995: 69–70) refers to the term motivation as an intermediate state between arbitrariness and predictability. Quoting Langacker (1987), she argues that “our inability to predict what pattern a language uses does not entail that the choice has no semantic ba- sis” (Goldberg 1995: 70). Referring to Lakoff’s (1987) definition of moti- vation, a construction is motivated when it “inherits” properties from another construction, if these are not in conflict with the original construc- tion specifications (quoted in Goldberg 1995: 70). According to this line of thinking, semantic differences between constructions are not random but in some way or other semantically explicable. The type of link that is of inter- est to the present question is the so-called polysemy link since it accounts 274 The meaning of patterns for semantic differences between identical constructions (for further types of link cf. Goldberg 1995: 75–81). Figure 34 gives an overview of the distribution of participant patterns. At this point, it has to be noted that identical argument structure construc- tions are not the same as identical valency constructions, since the Gold- bergian type of construction is in many cases defined functionally and not formally (despite their definition as form-meaning pairings). This means that for example in the case of the caused-motion construction, the con- struction label allows for all kinds of particle phrases specifying the loca- tion in direction of which the motion is caused (covered by the term ob- lique) (e.g. on the table, into the car, etc.).

(x) number of patterns x proportion in percent unclassified (14)

10% mixed forms (13) 10% AGENT + CAUSE (4) AGENT + AIM (72) 3% 53% THEME + "PRICE" (4) 3% 3% AGENT + REFERENCE (4) 5% AGENT + GOAL (7) 13% AGENT + BENEFICIARY (18)

Total: 100% (136)

Figure 34. Participants realized by [for_NP] for lexical units in VDE53

From a valency point of view, each different particle phrase in this con- struction constitutes a different valency pattern, while these patterns might still be realizations of one and the same participant pattern. What appears to be a contradiction of the principle of no synonymy (Goldberg 1995: 67) at first sight is thus only a matter of a difference in categorization: while in VDE formally defined patterns can be semantically similar or even identi- cal (i.e. possible realizations of the same participant pattern), Goldberg subsumes them under one single construction straight away (cf. also the

53. See Appendix 5 at htttp://dx.doi.org/10.1515/9783110240788_appendix5 for the example sentences on which this classification is based. The relationship between pattern and meaning 275 resultative construction where the element expressing the result is simply defined as resultative phrase allowing various formal realizations [Gold- berg 2006a: 73]). Consequently, different forms are treated as instances of the same construction, which obviously also stresses their semantic com- patibility. A similar “cover term” is used in VDE in the form of the label ADV for “an adverbial which can be realized in various ways” (VDE: xvi). Thus, in the case of the caused-motion construction mentioned above – e.g. Peter put the book into his bag – the complement realized by into his bag, could also be realized by other complement types, e.g. on the table. ADV is not determined in its form by the valency carrier but its realization is de- manded by the valency carrier and the complement thus dependens upon it. Particle complements which are determined in their form by the valency carrier are labelled separately. This distinction, i.e. between elements which are determined in their form and those which are not, is not sufficiently captured by the term oblique. Consequently, the principle of no synonymy only applies to parallel pat- terns which constitute different constructions such as the ditransitive con- struction and the “to-dative” construction (I gave him the book. / I gave the book to him): since these are not part of the same construction, they must, per definition, be semantically different (see above). The question that re- mains however, is, to what extent formally identical valency patterns are motivated, i.e. whether differences between them can be accounted for by polysemy links. The notion of a polysemy link is exemplified in Goldberg (1995: 75–77) with respect to the ditransitive construction, which is consi- dered to have as a central sense ‘X causes Y to receive Z’ (Joe gave Sally the ball) but motivates additional senses such as ‘X enables Y to receive Z’ (Joe permitted Chris an apple) or ‘X causes Y not to receive Z’ (Joe re- fused Bob a cookie) among others.54 Returning to the analysis of the pattern [NP + verb + for_NP], it might indeed be possible to relate the participants AIM and GOAL to one another, which then clearly cover the majority of all uses of the pattern. If unclear or undecided cases are left aside, this group covers almost three quarters of all uses of this pattern. However, it appears to be much more difficult to relate BENREC, REFERENCE, PRICE, and CAUSE in the same way, which means there is obviously homonymy, without a clear link. Moreover, in the case of PRICE, the different participant in the subject slot also has to be accounted for. This is illustrated in Figure 35.

54. Cf. Tomasello (1998a: 436) on metaphorical extensions for the caused motion construction. 276 The meaning of patterns

Among the original number of this type of pattern, so-called idiomatic uses, which have not been discussed so far, do indeed exhibit semantic similarities which might be motivated by a “prototypical pattern meaning”. Uses of [for_NP] in (31) and (32) clearly have the most frequent meaning AIM, although the verbs play and shoot do not necessarily have this par- ticipant as a central member of their participant inventory.

(31)VDE Playing for time, Hillsden looked down at his feet and traced patterns in the sand with a piece of driftwood. (32)VDE We’re shooting for a deficit that’s below $200 billion by the end of fiscal year 1998.

Shoot obviously has a GOAL but not necessarily an AIM as defined above. GOAL is typically realized by [at_NP] but not by [for_NP]. This is a good example to illustrate one of the main purposes and also advantages of a construction grammar approach: the construction itself is taken to carry meaning which can be added to the lexical verb meaning. This allows for a more stable verb meaning, since different nuances within different uses of the same verb result from the construction meaning (Goldberg 1995: 11).

(x) number of patterns x proportion in percent

THEME + "PRICE" (4) AGENT + CAUSE (4) AGENT + REFERENCE (4) 4% 4% 4% AGENT + BENEFICIARY (18) 17%

72% AGENT + AIM / GOAL (79)

Total: 101% (due to rounding) (109)

Figure 35. Idealized distribution of participants realized by [for_NP] for lexical units in VDE The relationship between pattern and meaning 277

Yet whether this is necessarily due to the construction [NP + verb + for_ NP] or just the predicate construction [verb + for_NP] or even just the complement type [for_NP] in cases such as the one above is not absolutely clear. Moreover, it is not necessarily always the most frequent or the most typical meaning which must be regarded as the constructional meaning added to the verb meaning. In sentence (33), another idiomatic use taken from VDE, [for_NP] clearly represents the participant BENREC and in (34), a classification as RESULT appears more fitting. In the latter case, there are also good reasons for arguing in favor of an idiomatic multi-word verb use, at least from a semantic point of view.

(33)VDE You know I’ve been pulling for you. [American English: supporting you] (34)VDE Top speed now was just over twenty-two knots. It made for a hard ride.

Moreover, there are also a number of uses, where more than one role seems to be necessary to provide a semantic pattern description, even in combina- tion with one and the same verb. Pay can, for instance, occur with a partici- pant which is similar to AFFECTED in (35) or CAUSE in (36), both realized by [for_NP].

(35)VDE The UN says it’s ready to intervene and has funds to pay for the evacua- tion. (36)VDE She’d got to pay for his silence.

In the case of cry, sentences (37) and (38) also seem to differ, even if it is possible to analyze both instances of [for_NP] as CAUSE. Nonetheless they are not identical.

(37)VDE He felt like crying for joy. (38)BNC She cried for Adam, cried because she was suddenly frightened he was dead, cried for the loss of something she had only just found. G0L 4080

While in (37) the crying is done because the person experiences joy, the crying in (38) because she is afraid that something could have happened to Adam. The difference is also evident in a German translation, where in (37) ‘(weinen) vor’ and in (38) ‘(weinen) um’ would be the appropriate choice of particle. What follows from this is that it is not possible to generalize pattern meanings and thus consistently predict the availability of a pattern in a certain sense. Neither is it possible to foresee the grammatical flexibil- ity of one and the same valency pattern, e.g. with respect to passivizability. The comparison of semantically similar verbs in chapter 3 also yielded 278 The meaning of patterns several examples where verbs can occur in the same valency construction but differ with respect to whether the pattern can or even has to be realized in a passive form. While the pattern meaning can indeed usually be abduced (cf. Hopper and Closs Traugott 2003: 42) from similar uses or from the meanings of components of the patterns, i.e. the outcome can be interpreted in a certain way, it is not possible to make any kinds of predictions. The fact that ab- duction is possible, i.e. that the combination of forms is not completely random but semantically motivated might indeed facilitate learning and storage of such patterns (cf. Goldberg 1995: 70). Yet it must again be stressed that it is item-based knowledge, especially about verbs, which tells the speaker how they can be used, i.e. in which patterns they can occur and what this pattern then means. This point of view has, for instance, also been expressed by Fillmore (2008: 1): “I entertain the common image that each lexical item carries with it instructions on how it fits into a larger semantic- syntactic structure, or, alternatively, on how semantic-syntactic structures are to be built around it.”55 It seems that working with large amounts of actual language data instead of generalized abstract constructions cannot but lead us in this direction.

4.2.2 Same valency pattern – similar verb meaning?

Another approach, when starting from the vantage point of the pattern, is to test to what extent verbs occurring in the same pattern are semantically alike. One of the prerequisites for a verb to occur in a specific construction is that the lexically specific participant roles can fuse with the argument structure construction roles. Although not all roles need to be provided by the verb, the verbal roles must not be in conflict with the ones provided by the construction (e.g. Goldberg 2006a: 39–40). Approaches which consider verb meaning as the determining force for verb complementation go a step further by assuming a stable relationship between pattern and verb mean- ing. Consequently, verbs allowing the same complementation pattern ought to be semantically similar (i.e. same pattern, similar verb meaning). Chap- ter 3 already examined this relationship from the opposite direction, i.e. semantically similar verb meanings were the starting points for testing possible similarities in patterning. As explained in detail in 3.1, this direc-

55. Quoted from an unpublished manuscript of a paper given at the Euralex con- ference in Barcelona, 2008: Fillmore, Charles (2008), Border Conflicts: FrameNet Meets Construction Grammar. The relationship between pattern and meaning 279 tion (from verb meaning to pattern) was preferred for the simple reason that starting from the pattern and subsequently evaluating verb meanings occur- ring in these patterns exlcudes verbs which do not allow the pattern from the start, thus concealing possible irregularities. Nonetheless, it will be worthwhile to take the view “from pattern to verb meaning” as an addition- al perspective when examining the relationship between meaning and form, which is the objective of this chapter. Again, taking into account the amount of data available in the Patternbank, this can only be done in an exemplary way.

4.2.2.1 Verb class studies – a brief review

As mentioned above, the association between verb meaning and pattern has previously been studied in the form of verb group comparisons with pattern choice as the starting point (cf. also Gropen et al. 1989).56 A highly recog- nized study in this field is Levin’s (1993) English Verb Classes and Alter- nations, which appears to be cited whenever there seems to be need for some “real” data in favor of linking rules and ideas along those lines, as it seems to be one of the very few empirical approaches to the question. The second major attempt to align formal and semantic similarity by means of patterns is to be found in the Collins Cobuild Grammar Patterns / 1 Verbs (Francis, Hunston, and Manning 1996) and the Pattern Grammar by Hunston and Francis (2000) (henceforth both publications will be referred to under the label Pattern Grammar). The motivation of these two works clearly differs from that of Levin (see below). These studies – their scope and implications – will be briefly reviewed, before the pattern groups estab- lished in the preceding chapter are examined more closely with respect to the verb meanings occurring in them. Levin’s focus is on a description of the mental lexicon with respect to verb meaning, and especially the question to what extent syntactic behavior of verbs can be explained with the help of generalizations across verb classes and thus need not be specified any further in the mental lexicon entry of single verbs (Levin 1993: 11–12).

This work pursues the hypothesis of semantic determination seriously to see just how far it can be taken, even if it does ultimately turn out to meet with

56. Cf. also Herbst (1986) who examined the correlation between the valency patterns and the meaning of adjectives, and correspondingly established se- mantic adjective groups on the basis of similarities in their complementation. 280 The meaning of patterns

limited success. … If the hypothesis ... is taken seriously, then the task is to determine, first, to what extent the meaning of a verb determines its syntac- tic behavior, and second, to the extent that syntactic behavior is predictable, what components in verb meaning figure in the relevant generalizations. (Levin 1993: 13–14)

Her approach presupposes that there is a considerable overlap between verbs with similar meaning and similar grammatical behavior. Conse- quently she holds on to the hypothesis that even if the semantics of a verb is not a 100% indication of its argument realization, it must at least be a very good one. Her verb groups are first of all based on the verbs’ potential to allow “alternations between different constructions”, while the reason for this “behavior” is that they “share semantic properties” which are consi- dered to be causally relevant for this potential in the first place (Levin 1993: 7–8). Thus, verbs allowing the same kinds of alternations are put to- gether and the claim is made that the resulting groups are semantically sim- ilar (Levin 1993: 14). There are not necessarily always compelling reasons for such groups: a prior sorting according to syntactic similarity easily leads to a certain bias in creating groups which would not be so “obvious” if there was no such external motivation. Even Levin (1993) herself admits that

…the verb classes that are identified in this book should be “handled with care,” since there is a sense in which the notion “verb class” is an artificial construct. Verb classes arise because a set of verbs with one or more shared meaning components show similar behavior. Some meaning components cut across the classes identified here, as attested by the existence of proper- ties common to several verb classes. (Levin 1993: 17)

In Levin’s analysis, for example, call and phone, end up in two different groups: call belongs to “verbs of manner of speaking” and phone to “verbs of instrument of communication”, while one might easily interpret call as another instance of the latter group, the instrument either being the tele- phone or simply the person’s speech organs. Her tell-group represents a further methodological problem: it consists only of the single verb tell, which makes it doubtful whether the term “group” is really applicable. If a single verb shows a specific kind of patterning it seems to be intuitively more reasonable to take this as a case in point for idiosyncrasy. What is also problematic is the distribution of some of the verbs to these classes on the basis of their alleged participation in certain specified diathe- sis alternations. For instance, for her class “verbs of manner of speaking”, a The relationship between pattern and meaning 281 defining property is the possible complementation with +[at_X] and +[about_X]. This results in sentences like (1) and (2) the grammaticality of which is at least dubitable.

? (1)NSA *Susan chattered at Rachel. ? (2)NSA * Susan hissed about the party.

For the verb chatter, the BNC offers all in all 339 matches in 233 different texts, not a single one of which is complemented by +[at_X]. The verb hiss has 786 matches in 324 different texts and again not a single match with +[about_X]. The sentences were also rejected by native speakers, with the exception of one. As mentioned above, the study is often quoted as empiri- cal evidence for the close relationship between semantics and syntax. It has to be taken into consideration, however, that the data is not corpus-based but mainly rests on native speaker intuition and a number of learners’ dic- tionaries (e.g. OALD and LDOCE) (Levin 1993: xviii).57 Experience in connection with the analyses conducted here clearly shows that there are indeed differences as to what is considered acceptable by a native speaker, the coverage of patterns in dictionaries and what is to be found in a corpus. Compare Greenbaum and Quirk (1970: 2–3):

We may have strong beliefs about the forms we habitually use and we may also have strong views about the forms that ought to be used; these may be in harmony or in rueful conflict, but – needless to say – our beliefs about our own usage in no way necessarily correspond to the facts of actual usage.

Levin’s restriction to alternations, even if the list as such is quite impres- sive, is nonetheless very limiting. Alternative realizations as regards the choice of particles are not sufficiently taken into consideration and “sen- tential complements are for the most part ignored” (Levin 1993: 18). In- cluding the whole valency potential of a verb, as has been done in chapter 3, makes finer distinctions possible, and, what is more, it allows for identi- fying many more restrictions which pose a problem to any strong or de- terministic statement about the association of form and meaning. Verbs such as demonstrate and show, declare, announce, proclaim, and state, and propose and suggest, which are part of the present sample and Levin’s group of “Say-verbs”, exhibit acceptance of a certain number of the same alternating patterns, yet this does not apply to the whole range of patterns

57. Levin (1993: 193) quotes as another source the lexicon project of the MIT, while the nature of this data is not made clear. 282 The meaning of patterns these verbs allow: there are quite obvious differences in the structural be- havior of these verbs, even if other senses of the verbs are not taken into consideration. The treatment of complements introduced by a particle needs some fur- ther comment. As Levin treats particles as parts of prepositional verbs she comes up with groups like “Appeal Verbs”, with niggle at, grate on, jar on, appeal to, and matter to, which are all complemented by a noun phrase (Levin 1993: 193). By including the idiosyncratic component into the verb, as is done in approaches which assume a category prepositional verb, the complementation automatically appears much more systematic and regular than if the particle is seen as part of the complement, as is done in most valency approaches. However, this way, the idiosyncrasies are simply transferred to a different part of grammar. From a usage-based point of view, the combination of verb and the kind of particle it is associated with (i.e. in this case at, on or to) is still an idiosyncratic part of language which most likely has to be learned and stored by the language user and cannot be construed on the basis of general semantic rules. The verb groups in the Pattern Grammar (Hunston and Francis 2000; Francis, Hunston, and Manning 1996) are compiled by grouping together verbs which can be used in a certain pattern and then sorting this rather heterogeneous group into semantic groups. Hunston and Francis (2000: 3) argue that “words which share a given pattern tend also to share an aspect of meaning”. There are at least three major differences between the Pattern Grammar approach and Levin’s verb classes, two of them concern the me- thod of assembling the verb groups and the third one is connected to the interpretation of the results. To address the methodological differences first, the patterns in the Pattern Grammar are based on the COBUILD corpus and thus have a much sounder empirical base than Levin’s groups. Secondly, Francis, Hunston, and Manning (1996) identify different semantic groups for every single verb pattern, i.e. one verb can be a member in as many groups as it has patterns (e.g. argue belongs to the following groups, some of which are even listed more than once as they apply to more than one pattern: ‘agree’; ‘agree’ and ‘disagree’; ‘argue’; ‘ask’; ‘decide’; ‘discuss’; ‘preach’; ‘remonstrate’; ‘report’; ‘say’, ‘describe’, and ‘call’; ‘say’; ‘speak’; ‘talk’), whereas Levin takes alternations, i.e. at least two patterns at a time into consideration. Hunston and Francis (2000: 145) note in refer- ence to Levin’s groups that “allowing all patterns of a verb to be consi- dered, instead of just a few, might militate against the consistent formation of semantic groups, i.e. it is more likely to find similarities if only verbs in one single pattern are compared”. Hunston and Francis try to present their The relationship between pattern and meaning 283 findings in a rather “theory neutral” way, which is why this does not have to be of great concern to them, yet it is certainly problematic for determi- nistic approaches to verb complementation: when the meaning of a verb is considered to determine its formal patterns, then this should not only reflect in one single pattern but in the whole range of its patterns. It is for this rea- son that Levin feels the need to address such “irregularities” by claiming that mismatches are no evidence for the “non-existence of correlations” but merely indicate that the wrong meaning components were chosen when grouping the verbs (1993: 13). The third major difference between Levin and the Pattern Grammar ap- proach lies in the conclusions they draw: Hunston and Francis (2000: 85) do not claim that form is determined by meaning: “We must also say that the association of pattern and meaning is not entirely predictive, in the sense that not every verb with a meaning similar to those given above [verbs in the pattern V of n] will share the pattern V of n.” This means that there are semantically similar groups to be found with different patterns, as well as cases where one pattern might be represented by several verb groups, which differ semantically to a considerable degree (for the exam- ples below cf. Francis, Hunston, and Manning 1996: 173, 143, and 230).

V by amount: ‘INCREASE’ & ‘DECREASE’ with decline, drop, rise V prep: ‘RISE’ & ‘DROP’ with change, drop, rise

V on to/onto n 1. ‘HOLD’ group with cling, fasten, hang, hold, hook, latch, slot 2. ‘BACK’ group with back, front, lead, open, open out 3. ‘GET’ group with come, get, move, get back 4. “rest-category”: burst, get, come, lead

At the same time this also leads to verb groups which are so small that their existence can hardly be seen as the result of or the basis for generalizations, such as the ‘BREAKFAST’ group, including solely the verbs breakfast, dine, and lunch (Francis, Hunston, and Manning 1996: 134) (cf. the ‘tell’-sub- group in Levin [1993] quoted above). The label “rest-category” (see above) implies that verbs which could not be integrated semantically into any of the other groups, did not constitute a number large enough in order to es- tablish new, specific groups for them. The fact that there is a category in- cluding verbs which do not systematically fall into any of the other seman- tically motivated groups can already be seen as an indication for the fact that there is simply no such one-to-one relationship. This is also part of Hunston’s and Francis’s (2000: 85–86) conclusion: 284 The meaning of patterns

There is no one-to-one correspondence – it is not the case that a single pat- tern occurs with verbs of a single meaning. It does not even mean that all the instances of a particular pattern can be covered by a set of meaning groups – we are left with a ‘ragbag’ of ... verbs that do not fit in any group ... The division into meaning groups ... is not achieved through anything other than the intuition of the person looking at the list.

Moreover, Hunston and Francis (2000: 96) observed that there is also rea- son to believe in the prototypicality of verb groups, where prototypical verbs in a group can lead to the adoption of the pattern by less prototypical members in a process described as analogy. Moreover, the fact that the meaning groups “are not necessarily consistent between patterns … sug- gests that the traditional semantic sets may not be as fixed as might have been thought, that there might be prototypical members of each set, but that other verbs might slide between one set and another” (Hunston and Francis 2000: 145). These suggestions can easily be aligned with the findings pre- sented so far, i.e. that any theory of language has to be able to account for item-specific properties of words.

4.2.2.2 Semantic verb groups based on pattern groups

The lexical units occurring in the pattern groups compared in chapter 4.2.1 ([NP + verb + for_NP] and [NP + verb +NP + NP]) have been scrutinized as to possible semantic similarities. To provide an objective division into verb groups, no other verb group studies were taken into consideration for a first sorting. They were only consulted in a second step. The methodology is thus practically identical with the approach in the Pattern Grammar ac- cording to which “given a list of words occurring with a particular pattern, the majority will be divisible by most observers into reasonably coherent meaning groups” (Hunston and Francis 2000: 86). Some of the verbs in the pattern [NP + verb + for_NP] indeed exhibit a striking similarity, which allows the establishment of semantic groups. In part they form pairs such as try and test or count and calculate, but also larger groups can be established. The biggest group is formed by lexical units which can be grouped together under a fairly wide notion of wanting. Figure 36 presents this group in the form of a mind map, where some lexi- cal units are semantically closer, which is illustrated by grouping them in one single field, in overlapping fields or grouping fields more closely to- gether, while others differ more, which is illustrated by an increased dis- tance between the fields. The relationship between pattern and meaning 285 pray cry cry ask ask beg beg ring whistle whistle distance between bubbles indicates semantic semantic indicates bubbles between distance verbs between distance send send apply apply in the Erlangen Valency Erlangen Patternbank) the in bid bid call shout claim claim + for_NP] + die die act live burn burn want vote [NP + VHC wish hope hope compete care care fight aim aim hunt hunt chase reach reach push push press press -group (based on the the pattern on (based -group want The The strike protest protest campaign campaign canvass canvass adverstise adverstise Figure 36: 36: Figure 286 The meaning of patterns

Want as a neutral term forms the center of the group. Closely connected to want are lexical units like care, hope or wish, and burn, live and die (for something). The other lexical units in this group express various activities which are applied in response to the want, i.e. activities to reach the desired goal. These can again be subdivided in various different types of activities: more assertive ones, like hunt, chase, press, push, and fight and also protest and strike, less “aggressive” ones like ask, beg, cry and activities which seem to be situated between these, such as call, claim, shout etc. Moreover, there are ways of expressing the want through a medium such as ring and whistle and there are numerous activities which are closely connected to a certain context: If people want something from God, they pray for it, if they want a job, they apply for it, if they want a specific party to be represented in their government they vote for it, etc. Accordingly the group could also be subdivided into several groups such as ask, want, fight etc. What is quite obvious here is that this is by no means a homogeneous group and the concept of group might be questioned altogether, considering the fact that several of the lexical units need explanation in order to be ac- ceptable as members of it in the first place. Secondly, compete appears to be connected to a subgroup within the overall group, i.e. fight, yet not to the central concept want. Similar pictures could be provided for other possible verb groups within this pattern group, e.g. (A) pass, rush, race, march, go, come, and move, (B) find, inquire, check, seek, look, and search, or (C) cry, grieve, suffer, stand (tolerate), tremble, fear, and feel. In case of the first group (A), activities describing the manner of move- ment (race) as well as movements indicating a direction (come) or a rela- tion to something else (pass) are joined. Searching, seeking, and looking for something form the center of group (B) which is matched with find (as the final stage of searching) and inquire and check as certain types of search- ing. Group (C) includes both immediate feelings like grieve and expres- sions of such feelings such as cry. What this test demonstrates is that most of these lexical units would clearly not have been grouped together intuitively if there had not been the intention of finding such groups in the first place.This also shows that what they share is not necessarily a similar verb meaning: in many cases it is indeed the pattern meaning (e.g. thriving for a goal) which creates a certain similarity between verbs such as protest and whistle: both activities can be performed in order to achieve something, which is then expressed by [for_NP]. Arguing that the verbs can occur in the pattern because they are semantically similar ignores the fact that the pattern can also be extended to various other verbs: an AIM can easily be added to a large number of verbs. The relationship between pattern and meaning 287

Yet while one can protest for something as well as protest against some- thing, whistle can only be complemented with [for_NP]:

(1)VDE Those that protest for animals’ rights don’t go in for much intellec-tual- ising. (2)VDE The bill was sent to Congress after a wave of strikes by public sector workers protesting against the government’s economic austerity pro- gramme. (3)VDE When Freddy came to the house he would whistle for my mother. (4)NSA *The protesters whisteled against the manager.

Hence, the occurrence in such a group does not provide any reliable clues as to how the lexical units can actually be complemented. Another argu- ment against the merit of such verb groups as indicators for the close re- lationship between meaning and form is the simple fact that in addition to the five groups illustrated above, there is still a considerable number of very small groups (e.g. pay, charge, and rent, or dress and change, or count and calculate), and there are verbs which do not fit into any of the estab- lished groups but rather appear to be fairly idiomatic.

(5)VDE Frana was serving for the match at 5–4 ... (6)VDE I had to sit for a little examination.

If the formation of semantic groups on the basis of occurrence in one pat- tern applies to only such a limited number of lexical units, these semantic groups cannot be considered as indicating semantic determinism. What they do show, however, is that some verbs might be used in a certain pat- tern in analogy to a semantically similar verb. The use of whistle in this pattern could be interpreted as an extension of the pattern from a more fre- quent and established pattern of calling for someone. In that sense one can again conclude that the occurrence of verbs in certain patterns is not neces- sarily random but also clearly far from predictable. Another argument against taking semantic similarities among verbs oc- curring in the same pattern as a strong indication for semantic determinism is the unreliability of semantic judgments of this sort. Above it has already been indicated that different groupings might be possible as well, and this is confirmed by comparing the groups here to the groups offered in Francis, Hunston, and Manning (1996: 177–184). They make a distinction between the following groups: ask, compete, work, deputize, volunteer, argue, opt, yearn, care, search, prepare, pay, plan, wait, last, sell, and a rest-cate- 288 The meaning of patterns gory.58 While there are clear similarities to the groups established above with respect to the search-group and the work-group and also if subgroups within the want-group are taken into consideration such as the ask-group, there are also clear differences. For instance, while Francis, Hunston, and Manning establish the so-called ‘care’ group with care, (not) care, fear, feel, grieve, mourn, and pine, the verbs (as far as covered in VDE) fall into different groups in the analysis above: grieve, feel, and fear are part of the feel-group, while care (as in Would you care for a drink or a cup of cof- fee?VDE) is part of the want-group. The want-group here also includes wish, which is part of the Pattern Grammar ask-group, and hope and die,which are both members of the Pattern Grammar yearn-group. This clearly shows that such semantic judgments are not necessarily very reliable or that there are various ways of grouping verbs semantically which might be equally acceptable. This being the case, it appears to be very dubitable whether the relationship between pattern choice and verb meaning is indeed such a strong indicator for a close relationship between the two as far as present- day English is concerned. For the pattern [NP + verb + NP + NP], VDE lists all in all 120 lexical units in VDE. As discussed in chapter 4.2.1 above, the pattern can be re- lated to predominantly two different valency constructions, AGENT + verb + BENREC + ÆFFECTED and AGENT + verb + ÆFFECTED + PREDICATIVE. Tables 46 and 47 show the groups which are again formed on an intuitive basis: five for the participant pattern AGENT + verb + ÆFFECTED + PREDI- CATIVE and ten for the participant pattern AGENT + verb + BENREC + ÆFFECTED (both including one rest-category with lexical units which do not fit into any of the established groups), resulting in a joined number of thirteen semantic verb groups and one larger rest group for this pattern. The verbs in the vote-group describe how a person is given something such as a post in a process of election. The verbs in the consider-group ascribe a certain property to a person, while in the name-group it is a classi- fication or a label rather than a property that is ascribed to someone. The verbs in the declare-group, finally, have a more official character than the ones in the consider-group in that the verdict usually has some direct con- sequences for the person to which it applies.

58. Note that Francis, Hunston, and Manning (1996) cover a wider range of verbs in the COBUILD Grammar Pattern Series than VDE does. For this reason, the groups are not entirely comparable because the basis out of which they are formed differs. The relationship between pattern and meaning 289

Table 46. Verb groups for the valency construction [NP]_AGENT + verb + [NP]_ ÆFFECTED + [NP]_PREDICATIVE

vote consider name declare other vote think name declare keep (‘remain’) elect imagine call pronounce make (‘be or be- appoint find label (‘state publicly’) come’) count pronounce judge make ‘cause’ consider (‘speech’) account prove

Again, these groups are clearly not uncontroversial and one might indeed subsume some of them under a larger, more broadly defined semantic verb group (as has been done, for instance, for some of these verbs in chapter 3, where they are subsumed under the label ‘choose’) or subdivide them into more specific ones. Francis, Hunston, and Manning (1996: 272–279), who also divide the pattern into different – in their terminology – “structures” correlating with the valency constructions presented above, establish only one single group for this construction. Thus, semantic distinctions which can be made with respect to [NP]_AGENT + verb + [NP]_ÆFFECTED + [NP] _PREDICATIVE are not made at all, which shows that the semantic distinc- tions cannot be so compelling after all. The groups established for the valency construction [NP]_AGENT + verb + [NP]_BENREC + [NP]_ÆFFECTED are illustrated in Table 47 below. The verbs in the work-group have in common that some kind of activity or work is done for someone, e.g. an employer. The promise-group includes verbs which assure a transfer of something in the future. The buy-group includes verbs describing the transfer of goods typically in exchange of money. The get-group describes a kind of transfer where the recipient is put into focus. The deny-group includes verbs blocking some kind of transfer (cf. also Goldberg 1995: 38). The prepare-group includes activities where some- thing is arranged for someone else. The activities in the allow-group em- phasize the permission of a transfer. The give-group includes all those lexi- cal units where the notion of transfer is the most obvious feature. The de- mand-group includes activities which express a request that a transfer takes place. Goldberg (1995: 141–151) claims that the verbs in this pattern must be able to express some kind of transfer (see also 4.2.1) and can be consi- dered as various types of extensions based on the prototypical verb in this pattern give. Indeed many of the groups are obviously connected to a notion of transfer. 290 The meaning of patterns ÆFFECTED +[NP]_ BENREC + verb +[NP]_ + verb (‘money’, ‘sacrifice’), owe (‘money’, ‘derive’), take ‘derive’), owe (‘money’, ‘sacrifice’), (‘money’, ant’), excuse, offer (‘kindness’), leave (‘not interfere’) leave (‘not (‘kindness’), offer excuse, ant’), AGENT ), pay, bid, bet bet pay, bid, ), Į ), design, play, build, sing, read, tell, quote (A /B), teach /B), (A quote tell, read, sing, build, play, design, ), pass, hand, lend (‘money’), bring (‘carry’, ‘effect’), send (‘letter’, ‘person’),deal ‘person’),deal (‘letter’, send ‘effect’), (‘carry’, bring (‘money’), lend hand, pass, ), ensure, guarantee, grant (‘concede’), award, insure, concede, bid (‘auction’) bid concede, insure, award, (‘concede’), grant guarantee, ensure, ), Į ), show (‘be noticeable’), make aware’), feed (‘food’, ‘join’), pour, serve, light, pay, leave leave pay, light, serve, pour, ‘join’), (‘food’, feed aware’), make noticeable’), (‘be show ), Į ȕ , gloss), (subclassifications based on VDE) on based (subclassifications , į Ȗ (+ (+ (‘hostage’), pick (‘will’, ‘deposit’) ‘deposit’) (‘will’, gain, get, earn, win, save, keep (‘retain’) keep save, win, earn, get, gain, (‘cards’, ‘strike’),slip, strike (‘hit’, ‘ball’), set, shoot, find (‘discover’), afford, rent (‘make available),offer available),offer (‘make rent afford, (‘discover’), find shoot, set, ‘ball’), (‘hit’, strike ‘strike’),slip, (‘cards’, ( vote (‘provide’), members rent (‘hire’), lease, buy, sell, sell ( sell buy, sell, lease, (‘hire’), rent  Verb groups for the valency construction [NP]_ demand demand other call answer, ‘request’), ask (‘inquire’, wish, order, cost (‘time’), take pay (‘visit’, cause, do, mean, give deny deny prepare allow pack fix, lose deny, refuse, mix, make (‘produce’), cut, prepare, lend (‘gr (‘allow’), grant forbid, permit, allow, get get buy buy Table 47. 47. Table promise( groups verb work promise, promise draw,write, write (  The relationship between pattern and meaning 291

However, this does not seem to apply unequivocally to all the lexical units: the members of the work-group and the members of the prepare- group, for instance, do not really describe a transfer. The notion of transfer might indeed derive from the frequent association of the pattern with verbs such as give. Yet this raises the question what coerces this construction, i.e. why there are not more verbs that are typically associated with it (cf. verbs which only allow the so-called to-dative such as indicate as opposed to show [cf. 3.3.4.3]). Francis, Hunston, and Manning (1996) establish the following groups for this construction: give, bring, tell and send, cost and save, envy, and a rest category. The bigger number of groups above cannot simply be mapped to the smaller number of Pattern-Grammar-groups by joining some of the groups above. The pattern-grammar ‘bring’-group, for instance is defined in reference to the criterion that “the verbs are concerned with doing something for someone, usually something which is beneficial to them” (Francis, Hunston, and Manning 1996: 274). It includes verbs like bring (which is part of the give-group above), play (work-group), guarantee (promise-group), make (prepare-group), etc. Thus the belief that verbs allowing the same pattern neatly fall into semantic groups appears to be misguided. 5 Conclusion

5.1 Different approaches towards the syntax-semantics interface – an assessment in the light of empirical findings

The objective of this work was to throw light on the relationship between meaning and form with respect to verb complementation patterns. The mo- tive for a further exploration of this much studied subject is the apparent lack of real empirical research which seriously tries to evaluate the actual scope of the influence of meaning on the verbalization of a communicative event. In order to put the results of this work in perspective and underline their relevance for an assessment of the relationship between meaning and form, this chapter will briefly summarize the main positions which are found in the literature concerning this question as discussed in the previous chapters. For a considerable time, semantic aspects were considered as both too unreliable and not central enough to play a role for syntax. Tomasello (1998b: x) comments on this in reference to Chomsky (1980):

If there is an analysis of a specific linguistic phenomenon in which seman- tics appears to affect syntax, it is assumed that either: (a) there is something wrong with the analysis ... or (b) the phenomenon is deemed to be outside of “core grammar” and placed into “the lexicon” or “the conceptual sys- tem”.

In the wake of the Chomskyan approach of marginalizing the relevance of semantic, lexical, and idiosyncratic aspects in favor of what was considered to be the core of language, i.e. syntax (at least in his early models), a coun- ter-movement can be observed which focusses especially on semantic as- pects. Research in the last twenty years has increasingly come to emphasize the role of meaning in accounting for how a verb can be used. The relation between grammar and meaning has been considered as “probably the most crucial issue in current linguistic theory” by Langacker (1991: 1), who also points out that “[even] in the generative tradition, which has long and loud- ly proclaimed the autonomy of grammatical structure, semantic consid- erations have not only intruded but taken on progressively greater signi- ficance”. There is growing agreement among linguists that “[a] functional analysis of syntactic structures should take into account such correspon- Approaches towards the syntax-semantics interface 293 dences between syntax and semantics whenever possible” (Mukherjee 2005: 9). This development is also closely connected to an ever stronger fo- cus on lexis as opposed to syntactic configuration. This manifests itself in an increased interest in lexis-oriented frameworks or approaches with at least a strong lexical component such as Word Grammar (Hudson 1984, 1990), valency theory (cf. 1.2), construction grammar (Fillmore, Kay, and O’Connor 1988; Goldberg 1995, 2006a; Croft 2001), Sinclair’s work on idiomaticity (e.g. 1991, 2004) but also HPSG (Pollard and Sag 1994) and Lexical Functional Grammar (e.g. Bresnan 2001). This again is closely connected to an increased interest in, and the availability of, corpora and the research possibilities these offer. The focus on lexical items has increa- singly revealed correlations between syntax and word meaning which need to be accounted for.

Recent corpus linguistic studies provide ample testimony of the fact that such correspondences between syntactic patterns and their semantic values (which may be more or less abstract in nature) can be found in many fields of language use, which calls into question the widely held, formalist view of an autonomous syntax. (Mukherjee 2005: 9)

Also from the point of view of language processing, the lexical item – in- stead of the syntactic structure – appears to be a better starting point.

…lexis is communicatively prior. As communicators we do not proceed by selecting syntactic structures and independently choosing lexical items to slot into them. Instead, we have concepts to convey and communicative choices to make which require central lexical items, and these choices find themselves syntactic structures in which they can be said comfortably and grammatically. (Francis 1993: 142)

Similarly, Lakoff (1987: 228) emphasizes the communicative function of language, which again suggests a stronger emphasis on meaning than on purely formal aspects:

…if language is a way of framing and expressing thought so that it can be communicated, then one would expect that many (not necessarily all) as- pects of natural language syntax would be dependent in at least some way on the thoughts expressed.

Currently, there seems to be a broad consensus that how speakers phrase what they want to express is not completely random but that verb comple- mentation is related to the meaning of the verb. Quite obviously, the num- 294 Conclusion ber and types of participants of a verb delimit the number and range of complements with which it can co-occur.

Syntactic information can provide cues to verb meaning just because the structural privileges of a verb (the number, type, and positioning of its asso- ciated phrases) derive, quirks and provisos aside, from its argument-taking properties. The number of argument positions lines up with the number of participants implied by the logic of the predicate. (Gillette et al. 1999: 164)

Thus a verb expressing transfer occurs with an entity receiving the trans- ferred good as well as the good itself and the initiator of the transfer. Uses with an AGENT participant only are not possible (other than for instance with a verb such as read).

(1)NSA I gave him a book. (2)NSA *I gave. (3)NSA He frequently gives to charities. (4)NSA ?I gave a book.(only if clear from context) (5)NSA *I gave him.

Both the transferred good and the recipient can remain unexpressed pro- vided they are retraceable from the context and that at least one of the two participants is expressed as in (3) and (4). However, as example (5) shows, there are also formal restrictions which have to be accounted for. A verb expressing a mental activity, on the other hand, can express the outcome of this process, e.g. a thought, and the person pursuing this activity. It does not normally occur with a beneficiary or a recipient.

(6)NSA I think that he is wrong. (7)NSA *I think him that he is wrong. (8)NSA ?I imagined us our dream holiday.

This tendency has increasingly led to the assumption that “grammar and meaning are indissociable” (Langacker 1999: 1) and that there is “a prin- cipled interaction between the meaning of a word and its grammatical properties” (Dixon 1991: 6). What exactly constitutes verb meaning and which semantic features are relevant to linking have been central questions in this context. A sentence such as (8) shows that the acceptability of a pat- tern does not depend exclusively on the semantic participants of a verb. Hence, the normal complementation characteristics of a verb can be over- ridden and it can adopt additional participants/complements which are con- tributed by the construction (e.g. Goldberg 1995, 2006a). Yet while useful Approaches towards the syntax-semantics interface 295 for such “creative” verb uses, this additional level does not sufficiently explain with which complementation patterns a verb is conventionally used. Thus the question remains, as to how the participants of a verb are conven- tionally realized in English and to what extent this is determined by mean- ing. Such a correlation can be seen in the fact that it is prototypically the person performing the activity that is encoded as subject and the entity most affected by this action that is encoded as a complement directly fol- lowing the verb in the predicate (i.e. an object in traditional terminology). 59 Gillette et al. (1999: 164) also argue that “the type of complement is also derivative of aspects of the verb’s meaning. …. For example, because one can forget things, this verb licenses a noun-phrase complement; and be- cause one can also forget events, it also licenses clausal complements” (al- though events can of course also be encoded as things as He forgot the meeting, which complicates the matter). In that sense there is quite ob- viously a certain dependency between meaning and form. The initial cautious notion that semantics must somehow play a role in how a verb can be used is increasingly being replaced by stronger state- ments which take a more and more deterministic stance. Dixon (1991: 6), for instance, claims that “once a learner knows the meaning and gram- matical behavior of most of the words in a language, then from the meaning of a new word he can infer its likely grammatical possibilities”. According to this line of thinking, it ought to be possible to predict the uses of a verb on the basis of similarity between semantically similar verbs. What makes this appealing for a theory of complementation is that if semantic informa- tion systematically predicts certain structures, these aspects do not have to be accounted for in any other way and they do not have to be built into a theory of grammar: the apparent idiosyncrasies would be reduced to a min- imum – ideally so slight that they are negligible – while conceptual aspects and meaning are taken as the basis of grammar, without grammar itself losing its elegant simplicity, cf. Langacker (1999: 1): “grammar reduces to the structuring and symbolization of conceptual content and thus has no autonomous existence at all”. Compare also Levin (1993: 11) who argues that “the ideal lexical entry for a word should minimize the information provided for that word. This goal can be achieved by factoring predictable information out of the lexical entries, leaving only idiosyncratic informa- tion”.

59. For a summary of possible explanations for exceptions see Levin and Rappa- port Hovav (2005: 32). 296 Conclusion

Thus within such approaches the complementation patterns of a verb are considered to be predictable from verb meaning, except for some (minor) idiosyncrasies. According to Pinker (1989), ideally universal linking rules control the mapping from meaning to syntactic form. Consequently, dif- ferent complementation patterns are the outcome of semantic differences. Similar positions are taken for instance by Levin (1985) and Gropen et al. (1989) (cf. Goldberg [1995: 8] and Goldberg [2006b] for a summary and a critique of Gropen et al.[1989]). Chomsky postulates that semantic theta roles are mapped onto syntactic forms (e.g. 1984, 1995). This is mirrored by various similar claims about the shaping force of meaning for the occur- rence of a verb in certain syntactic structures (e.g. the Aspectual Interface Hypothesis by Tenny [1994] and, with a different emphasis, also Goldberg’s [1995] Semantic Coherence Principle and Correspondence Principle). Semantic determinist theories – as Noël (2003) calls them – assume that the complementation pattern must be determined through meaning, for instance via semantic role hierarchies, event semantic properties of verbs or their selection restrictions. If the relevant semantic criteria have not been identified, it is often only considered to be a matter of time before this will be achieved (e.g. Levin 1993). However, there are also more cautious statements about meaning as just one important factor for the choice of formal patterns in which a verb can occur (Hunston and Francis 2000; Goldberg 1995, 2006a). Yet while such approaches often acknowledge that complementation patterns cannot be predicted on the basis of meaning, this is often no more than a side note leading up to a more central claim which again highlights the role of meaning in how a verb can be used. Goldberg (2006a: 57), for instance, mentions the complementation irregularities be- tween the semantically similar verbs help and aid, stressing that such cases – which according to her are rare – have to be stored. Yet at the same time she underlines that the knowledge of whether a verb can occur in a specific construction or not depends on whether the verbal participants are “seman- tically compatible” with the argument structure roles or at least not in con- flict with them (Goldberg 2006a: 40 and 1995: 70). It appears to be symptomatic for research on linking that the claims that are made are often based on a number of handpicked examples which clear- ly indicate a strong relationship between form and meaning but rarely on large-scale samples which are selected without reference to their syntactic properties as in this present analysis. There is also a tendency toward claim- ing that such semantic features are universal, and therefore do not have to be learned (e.g. Pinker 1989: 248). However, there seems to be no real em- pirical basis for this hypothesis either. Croft (1998: 25) claims that “[p]artly Approaches towards the syntax-semantics interface 297 this is because there is hardly any discussion of the cross-linguistic evi- dence. But partly this is because universal claims about linking are made on fine-grained analyses of English where we simply do not know what as- pects are universal and what are English-specific”. Welke (2005: 117) points out in reference to such cross-linguistic differences as Someone frightened John but not *John frightened vs. Jemand/Etwas ängstigte John and John ängstigte sich that axioms are too often just centered on the Eng- lish language. For this reason no claim is made here that the results have any kind of universal relevance.60 Moreover, the focus is usually on functional units such as subject or ob- ject and prototypically their verbalization by means of noun phrases. This clearly leads to an underestimation of the actual range of formal uses in which a verb can be found and the possible range of irregularities to be identified in this area. While this is not to say that there might not be a level of analysis at which such units can be useful in providing important in- sights into the relation between form and meaning, they are too broad when the purpose is to examine the range of possible idiosyncrasies. It is also for this reason that approaching the issue from a valency point of view, which allows for a fine-grained description of complements and thus a less gene- ralizing analysis, provides an important additional perspective: only by looking beyond elements such as noun phrase objects and instead com- paring them with the various patterns of present-day English including clausal and particle complements can it be attempted to approach the ques- tion of the actual extent to which form is predictable from meaning. More- over, defining the slots of each construction by separate formal and seman- tic labels, as we already suggested in Herbst and Schüller (2008), allows for more flexibility and thus accuracy than can be achieved by combined labels such as object, where it is often not clear whether they are based on seman- tic or on syntactic features or both. The same applies to describing com- plements with respect to the type of particle phrase or clause by which they are realized as opposed to labels such as oblique under which various phrases are subsumed.

60. According to Fabricius-Hansen (2004: 187, 194) the question as to the signi- ficance of sentence patterns and construction types is to be answered on a language specific level. When processing German, for instance, one needs to rely on constructions more than in English due to the occurrence of subject- object-verb sequences, i.e. the valency carrier comes fairly late during online processing, unlike in English with its subject-verb-object word order, where the reader can more successfully operate with valency information. 298 Conclusion

Taking all of these aspects into consideration, it can be shown that there is considerably more “irregularity” than is typically inferred from such statements as exemplified by the Goldberg quote above. The idiosyncratic component is clearly too substantial to simply be dismissed as peripheral. This is very much in line with valency theoretic approaches in their various forms and shapes (cf. chapter 1.2). They all have in common that comple- mentation is described as a predominantly lexical phenomenon. While this by no means excludes analogies between semantically similar verbs, the model has no difficulty dealing with idiosyncrasies in the first place and it is generally acknowledged that a correspondence between semantic valency and syntactic valency is not an obligation (e.g. Helbig 1992: 15). Thus, unlike construction grammar, which generally assumes form-meaning pair- ings, valency theory can more easily cope with formal discrepancies be- tween semantically similar patterns (cf. Herbst 2009). The acknowledge- ment that valency phenomena and verb complementation are therefore best dealt with in the form of a lexically based representation as opposed to general grammar rules is evidenced by the many valency dictionaries, such as Helbig and Schenkel (21973), VALBU (Schumacher et al. 2004), and, for English, VDE (Herbst et al. 2004). The latter two are corpus-based, which stresses another important factor in this context. Working with cor- pora and thus having to cope with large amounts of authentic language data and the idiosyncrasies to be found there leads to a very cautious stance towards making generalizations which might be too broad. The heightened awareness of such idiosyncrasies is also tied to the fact that valency theory is often applied in the context of foreign language teaching – especially German as a foreign language and with VDE and the Patternbank also for English foreign language linguistics. When examining verb complementa- tion from a contrastive point of view, one is confronted with numerous cases where the same semantic-pragmatic context is verbalized differently in different languages and this also affects the complementation patterns of verbs (e.g. Welke 2005: 117). Finally, the fact that valency approaches allow for a fine-grained formal analysis which includes a functional level and possibly also a semantic level – as is done in VDE and especially Herbst and Schüller (2008) by means of a semantic role description – makes valency theory a model that can successfully cope with the chal- lenges posited by the many restrictions identified here – as opposed to models which use predominantly functional elements or mixed categories such as object. Summary of the results 299

5.2 Summary of the results

The comparison of 87 verbs distributed over 22 semantic groups clearly shows that the valency patterns of verbs cannot simply be inferred from their meaning. In order to be able to examine the extent of idiosyncrasies, i.e. in order to arrive at a reliable list of restrictions, the starting point of the analysis had to be a semantic one despite the obvious pitfalls of any classi- fication not based on formal criteria. Comparing formally similar verbs to point out their semantic similarity – as was done by Levin (1993) and Hunston and Francis (2000) – would have thwarted an identification of formal differences with respect to complementation patterns from the start. Again it has to be stressed that the fact that there are hardly any true syn- onyms was also acknowledged and that no claim was made that the verbs in the semantic groups are semantically identical. However, they are semanti- cally similar enough to allow for a comparison, i.e. they are semantically equivalent in the sense that they can – at least in part – be used with the same participant patterns or can be used in the same kinds of context for the same kinds of pragmatic purpose. If semantic determinism reflected lan- guage reality then the syntactic differences between these verbs should be retraceable to their meaning. For a considerable number of the restrictions identified this was simply not possible. This must be seen as a clear argu- ment against semantic determinism and evidence for a strong idiosyncratic component.

5.2.1 Lexical aspect, selection restrictions, and participant inventories

The analyses of 22 groups of semantically similar verbs presented in chap- ter 3 has shown that semantic features which are typically considered cru- cial for determining the complementation possibilities of a verb are neither a reliable factor for predicting restrictions nor do they help in accounting for them. No significant correlation could be identified between restrictions and the situation type the verbs in the groups typically express (cf. 3.4.3). Most of the semantically similar verbs represent the same situation type, i.e. they have the same lexical aspectual quality in the first place but still exhibit differences in complementation. Answer, reply, and respond are, for instance, all accomplishments, i.e. they are dynamic, durative, conclusive, and agentive, yet they cannot be used in exactly the same patterns. All in all, more than half of the groups are completely homogeneous with respect to their aspectual profile. In the rest of the groups, too, there is often only a difference in the sense that one verb – or even just another lexical unit of 300 Conclusion one of the verbs in the group – can express an additional situation type (e.g. guess and gather can be both activities and accomplishments whereas as- sume, presume,and suppose are only accomplishment verbs). Nonetheless, with the exception of one group (phone, ring, and call) and a number of subgroups, the verbs in the homogeneous groups cannot be used in exactly the same way, i.e. 15 of the 22 groups are homogeneous but still exhibit differences in complementation. Obviously, complementation restrictions cannot be accounted for on the basis of a verb’s aspectual nature alone. This does not exclude it as a contributing factor in combination with other factors. However, all restrictions were closely examined and a causal con- nection to the situation type of the verb could not be established. Also the restrictions identified within the remaining seven groups with heterogene- ous aspectual profiles do not seem to be in any way connected to the aspec- tual difference, with one exception: reminisce differs from recall, remem- ber, recollect, and bear in mind in focussing on the activity of remembering and less so on the remembered than the accomplishment verbs in the group do. This might explain why it is the only verb in the group that can occur in a number of patterns without ÆFFECTED, a participant which plays a more central role for the other verbs since it could be seen as embodying the endpoint of the accomplishment (however, ÆFFECTED is not obligatory with the other verbs in the group, either). This shows that the situation type can have an effect on in which complementation patterns a verb occurs, however, there is no basis for predicting complementation or even provid- ing a systematic ex post explanation of restrictions. This clearly speaks against theories which consider the aspectual dimension as decisive for linking (e.g. Tenny 1994; see also Levin and Rappaport Hovav 2005: 96– 117). Differences between verbs within one group with respect to their selec- tion restrictions had an effect on the types of noun occuring in a pattern but not on the pattern type itself (cf. 3.4.2). Verbs with the selection restriction +animate (appoint and train) were less likely to be complemented with clausal complements, however, sentences with clausal complements were identified in the BNC or accepted by native speakers, so that no predictions are possible on this basis either. Only 3% of the restrictions in chapter 3.2 could be related to selection restrictions. The kinds of semantic restrictions Pinker (1989: 185–187 and 208–209) envisaged as “syntactically relevant features” did not apply to the sample examined here. The participant inventory of these verbs, finally, was indeed the best in- dicator of whether a pattern is available to a verb or not (cf. 3.4.1). When a verb does not have a certain participant, patterns expressing it are obviously Summary of the results 301 not available, either. Ponder, for instance, – other than consider and think – does not express a PREDICATIVE. Consequently it cannot be found in those patterns which specify this participant. A difference in the participant in- ventory can also be related to how central a participant is to the meaning of a verb, i.e. whether it always has to be expressed or whether the verb can also occur in patterns without it. Goldberg describes such participants as profiled participants, which have to be “obligatorily expressed, or, if unex- pressed, must receive a definite interpretation” (Goldberg 2006a: 39), while this typically only refers to active constructions, i.e. profiled elements are often optional in passive constructions (Herbst 2010). Other conceptions of this term also include a general prominence or salience of a profiled partic- ipant (Langacker 1987, 1991; Fillmore 1977). In valency terms, this often coincides with what is understood as an obligatory valency slot, which might have various fillings but cannot remain unexpressed. The verbs name and nominate cannot, for instance, be used without the chosen or named entity, i.e. AFFECTED, while choose, select, pick, elect, appoint, and desig- nate, which can as such be used in a very similar way, can be used without it. Yet here, too, a considerable number of irregularities remain (cf. 3.3.4.1; see also Fillmore 2007). Note that this argument is to some extent circular: the fact that ponder was not found in a constructions with a complement typically expressing a PREDICATIVE and that it was marked as incorrect by native speakers when used in such patterns was taken as evidence that it does not have such a participant. At the same time, the fact that it does not have such a partici- pant explains the restriction of these patterns. Similarly, the fact that name was never found in patterns not expressing AFFECTED was taken as evi- dence that the participant has a different status than the same participant for other verbs in the group. In that sense there is already arbitrariness as to which verbs have which participants out of the number of all conceivable participants for a specific scene described by the verbs. This leads to the core of the question as to what kind of categories languages establish in the first place, i.e. what kinds of activity or parts of an activity receive a sepa- rate lexical label in a specific language. This has not been of major concern here, since it clearly goes beyond the question of the relationship between verb meaning and complementation. Consequently, such cases have been classified as restrictions related to verb meaning, while the latter has simply been assumed to be an invariable factor for the purposes of this analysis. Nonetheless this shows unmistakably that these kinds of questions are not even exclusively tied to complementation but rather to more general processes of categorization, which can be seen as arbitrary to a certain de- 302 Conclusion gree in addition to the arbitrariness identified here for the formal realization of certain participants: the participant inventory of a verb cannot necessari- ly be predicted even if one knows what the verb means. Another factor which has to be taken account of and which was dis- cussed in chapter 3.4.3 is that the differentiation between what is caused by a difference in the participant inventory and what by a difference in lexical aspect is – in some cases – rather arbitrary. This might in part also account for the comparatively small number of restrictions attributed to the latter aspect, even if it does not alter the overall number of semantically explica- ble restrictions. Taking this for granted, 32% of all syntactic gaps could be accounted for on the basis that they contain a participant which has generally not been identified with the verb, or more specifically, the lexical unit under con- sideration. 10% of the restrictions are due to the pattern being associated with a different lexical unit of the verb which blocks the use of the pattern for further lexical units, while there are also numerous cases where the same pattern can occur with two different lexical units of the same verb. This leaves us with 193 restrictions amounting to 55%, which cannot be accounted for in a systematic semantic way. At the same time, the analysis also confirmed that there is a considerable overlap between semantically similar verbs: the overall percentage of overlap, i.e. where verbs accepted the same patterns amounts to 68% (as opposed to 16% of unsystematic restrictions). Since the objective of this work was to identify actual syntac- tic gaps, rare verb-pattern associations were also taken into consideration (e.g. where only one or two examples were identified in the corpus or con- strued examples which were accepted by some but not all native speakers). Hence, if such controversial uses are also regarded as restrictions, the num- ber is clearly higher. However, even 16% clearly appears too much to be marginalized in any theory which concerns itself with the relationship be- tween meaning and complementation.

5.2.2 Competition with other lexical units of the same lexeme

In some cases patterns are blocked since they automatically trigger a dif- ferent reading of the verb, e.g. it is possible to teach or train someone to do something, while instruct in instruct someone to do something means ‘or- der’ rather than ‘teach’, although instruct in many other patterns is seman- tically equivalent to teach and/or train. Teach was identified in the pattern teach someone on something, e.g. a specific topic, yet train in the same pattern means to ‘direct’, as in train a weapon on someone (cf. 3.3.3). In Summary of the results 303 part this also has an effect on whether a certain valency slot is obligatory or not (cf. 3.3.4.1); for instance the post-verbal noun phrase in [SCU]+ an- nounce +[NP]+[for_NP] is obligatory in the sense ‘declare’ (1), since the pattern [SCU]+ announce +[for_NP] is only found in the context of an- nouncing candidacy for something (2).

(1)BNC EC announces £420m for R&D -- but is it enough? A19 111 (2)VDE I have to decide whether it is good or better for the people of my state or not good at all for me to announce for the presidency.

This kind of blocking is not a general mechanism either. If a certain sense is expressed with a certain pattern, analogies to semantically similar verbs which are found in this pattern are not always preempted. There are numer- ous examples of one and the same pattern being used with different lexical units of one and the same lexeme, e.g. gather can be used in the sense ‘col- lect’ as well as in the sense ‘guess’ in the pattern [SCU]+ gather +[NP] +[from_NP] and assume can mean ‘suppose’ and ‘take’ in the pattern [SCU]+ assume +[NP].

(3)VDE Through the year feathers had been gathered from every bird trapped, netted, arrowed, tamed, hatched, or coaxed into reach.= ‘collect’ (4)NSA He gathered this from the runner’s facial expression. = ‘guess’ (5)VDE You shouldn’t mind, assuming a fair interest rate, keeping the account at that bank. = ‘suppose’ (6)VDE The Front assumed power in Romania during the revolution. = ‘take’

While the semantic difference between (3) and (4) can be explained on the basis of a metaphorical use of gather, this is much less obvious in case of examples (5) and (6). Thus the notion of metaphorical extension, as used by Goldberg (1995: 3–4) in accounting for the occurrence of “untypical” verbs in a certain construction gives valuable clues about the spread of patterns but cannot account for all cases where a pattern is available to different uses of a verb. If it were the form of the complement or the verb that triggered a certain reading as suggested in projectionist approaches, cases where one and the same verb accepts the same pattern but has several meanings in each should not be possible. Likewise the pattern in combination with the same verb should have a stable meaning, and this is not the case, either. Several cases were identified where a pattern represents two different valency construc- tions with one and the same verb (cf. 3.3.2) as in: 304 Conclusion

(7a)VDE A small book dealing with the history of insurance plaques will, I hope, answer all your questions AFFECTED. (8a)BNC When no one answered her BENREC, she came stamping into the room with a scowl on her face. AC4 647

At the same time this variability cannot be explained with reference to the meaning of these verbs either, since semantically similar verbs which can express the same participants do not allow this alternation.

(7b)NSA *A small book dealing with the history of insurance plaques will, I hope, reply all your questions AFFECTED. (8b)NSA *When no one replied her BENREC, she came stamping into the room with a scowl on her face. (9)VDE Pasternak replied to him BENREC in a personal letter. (10)VDE Mitterrand replied that the question of sanctions had to be decided by the European Community summit in June AFFECTED.

5.2.3 No stable complement type or pattern meaning

While the main focus of this investigation has been on verb meaning, spe- cific complement types as well as valency patterns have also been analyzed in order to provide a broad coverage of this phenomenon. As regards com- plement types, it is clearly possible to make certain semantic generaliza- tions (e.g. that particle phrases headed by on, about, or over can express a TOPIC). However, beyond that there is hardly anything such as a clearly discernable complement type meaning, which could – from a synchronic point of view – account for either the restriction of some of these types in certain contexts or the preference of other types in other contexts (cf. 2.3.1). Although they are morphologically marked, which contributes to the overall meaning of the pattern – as opposed to patterns which consist only of noun phrases and are thus only configurationally marked –, it is not pos- sible to take this as the basis for predictions as to which other verbs allow the same complement type and which do not. This does not only hold for particle complement types but also for clausal complement types, even if here, too, certain semantic tendencies can be identified. Moreover, even if there seem to be explanations for the choice of one over the other in combi- nation with a specific verb – in the sense that a certain specialization or se- mantic differentiation can be attested – these cannot serve as a basis for ge- neralizations beyond specific contexts. Such verb-specific specializations cannot form the basis of a lexical rule to predict the aptitude of a certain Summary of the results 305 form to serve as a complement to other verbs.61 This clearly contradicts ap- proaches like Wierzbicka (1988, 1996) and Verspoor (1990). The semantic criteria which it is claimed determine the choice of one complement type over another are often contradictory and can certainly not serve as a basis for predicting the complementation patterns of a verb. What is more, the specialization is by no means necessarily semantically motivated, i.e. it was not possible to identify stable semantic features which are connected with one rather than another semantic subspecification. The fact that [on_NP] is used for more formal TOPICs than [about_NP] applies to talk but not to inform or cheat. When complementing inform, the complement can express what has been labeled a “wrongdoer” as in (1) but also an absolutely “neu- tral” topic, which does not have to be formal, either, as in (2). However, the meaning exemplified in (1) indeed appears to be the prototypical meaning when AGENT and TOPIC are both +human.

(1)BNC He also was quickly released, but he informed on two other well-known villains, MacDonald and Collins. ANK 484 (2)BNC Among the many items the Group offers are ... training programmes and seminars and guidance on the best software to buy to inform on exercise and nutrition matters. CB4 899

In combination with cheat, which is not an information verb and therefore does not express a TOPIC, [on_NP] can express REFERENCE, i.e. the field in which the cheating takes place (3), an entity that is somehow manipulated (4) or a person which is cheated, typically in the context of relationships (5), while outside this context the “VICTIM” is expressed with [NP] (6).

(3)BNC If Iran is cheating on nuclear weapons, the International Atomic Energy Agency has yet to catch it at it. CRC 1468 (4)BNC In 1931 Andrew Mellon was accused by Congressman Louis McFadden of cheating on his 1931 income tax return ... EBV 1822 (5)BNC ‘How did you feel about cheating on Celeste?’ JY9 3218 (6)VDE Everybody tries to cheat her.

Examples (3)–(5) show that semantic differences are not necessarily related to a difference in form. One and the same complement – in combination with some verbs – can be considered as the realization of different semantic

61. Cf. Dürscheid (1999: 4) who comes to a similar conclusion for German case in that meaning of the case-marked noun phrase cannot be determined in an “absolute fashion” but only “relative to its context”. Cf. also Helbig (1992: 23). 306 Conclusion subtypes, i.e. there is a considerable degree of polysemy (cf. 2.2.2). Lan- guage must map a large number of semantic types onto a limited inventory of form types. This is also in line with Hopper and Closs Traugott (2003: 42) who relate one form having several meanings to memory and parsing limitations, as well as speakers’ inclination to metaphorical uses “to sustain interest in verbal interaction”. In this context they conclude that “‘[o]ne form – one meaning’ would in these circumstances be dysfunctional” (Hopper and Closs Traugott 2003: 42). With talk, which allows more than one complement type for the realiza- tion of a specific participant, each type is prototypically (but not exclu- sively) used for a specific semantic nuance as indicated above. Yet the ex- amples of cheat and inform show that some complement types can express various senses even in combination with the same verb, so there does not seem to be anything forcing them to express differences in a formally dis- tinct way. This contradicts Goldberg’s (1995: 67) principle of maximized motivation, according to which syntactically related constructions are also semantically related. At the same time, such semantic subtype-variations within one partici- pant can have an effect on the combinatorial potential of the participant with other participants. Argue, in the sense ‘dispute’, can, for instance, only be complemented with [on_NP]_TOPIC in combination with a [with_NP] complement, while TOPIC can be expressed without such restrictions with [about_NP] and [over_NP].

(7)BNC “I'm not going to argue with Jane Austen on that one,” Matthew said. GV8 3310 (8a)VDE They talked and argued about the money for about an hour and a half. (8b)NSA *They argued on the money for about an hour and a half. (9a)VDE We had been arguing a lot over little things. (9b)NSA *We had been arguing a lot on little things.

Examples like these also show that it is not enough to limit lexical infor- mation to semantic knowledge about complements since such configura- tional knowledge is hard to capture this way (cf. 4.1). The examples have also clearly shown that the relationship between complement and participant is one-to-many, i.e. one complement type can have several different semantic interpretations depending on the verb it is complementing. The same applies in the case of patterns, i.e. when com- plement types are combined. This has been shown for divalent [NP]-pat- terns, for instance by Schlesinger (1995) and Van Valin (2004). With the analysis of the patterns [SCU]+ verb +[for_NP] and [SCU]+ verb +[NP] Summary of the results 307

+[NP] (cf. 4.2.1) it could be demonstrated that clear predictions as to the meaning of the complements are not possible on the basis of their occur- rence in a fixed order, either. [For_NP] can express up to six different par- ticipants (AIM, BENREC, GOAL, CAUSE, REFERENCE, and PRICE) in this pat- tern. As has been shown in chapter 4, some of these (AIM and GOAL) might be subsumed under one super-role, yet clearly not all of them. The trivalent [NP]-pattern can express AGENT/THEME + verb + BENREC + AFFECTED (the often discussed ditransitive construction expressing transfer as with give), AGENT/THEME + verb + AFFECTED + PREDICATIVE (e.g. consider) or THEME + verb + BENREC + PREDICATIVE (as in the VDE-example There can be little doubt she would make him a happier wife.). This clearly shows that the semantics of the complement type itself can hardly be taken as fixed in any way: Patterns with particle phrases are clearly polysemous and not even ditransitive patterns are completely predictable. At the same time it is not possible to predict whether a verb can occur in this pattern on the basis of the participant roles realized by the pattern. A verb like explain can express AGENT, AFFECTED, and BENREC (He explained it to me) but cannot occur in the trivalent [NP]-pattern (*He explained me it) and verbs which a have an AIM participant cannot necessarily be complemented with [for_NP] (?He intended for it). Taking a diachronic point of view provides another strong argument against a projectionist theory: even if a particle was originally selected on the basis of its semantic properties, it cannot be assumed that a speaker today has access to this “original” meaning. Thus if constructions rested on a semantic basis from a diachronic point of view, this must not be regarded as an explanation for synchronic pattern choices as the pattern forms must be seen as conventionalized uses. The fact that it is semantic change that is made responsible in the Oxford English Dictionary for the difficulties in classifying on (cf. 2.1.2) demonstrates that such aspects can easily be ob- scured in the course of the development of the language and thus appears to be of no importance to present-day children acquiring English as their first language, learners of English as a foreign language, and finally to expe- rienced speakers of the language.62

62. Compare Croft (2001: 127) on the conventionalization of new functions and the loss of semantic distinctions in the course of grammaticalization. 308 Conclusion

5.2.4 Verb meaning not predictable from pattern choice

Moreover, the claim that verbs in the same pattern are generally similar (e.g. Goldberg 1995, 2006a; Hunston and Francis 2000) is highly doubtful. That there are certain similarities is unquestionable, indeed. Yet the com- parison of verbs occurring in the two patterns indicated above has clearly shown that the kinds of semantic group are too numerous to take them as evidence for the link between form and meaning in any deterministic way. Moreover, the verb meanings rather seem to form a network with various possible connections and groupings rather than coherent semantic groups (cf. 4.2.2). Thus there are indeed certain similarities. However, they cannot be taken as the basis of a prediction of what kinds of verbs can occur in a pattern. With respect to the similarities, Hunston and Francis (2000: 21) ask re- ferring to Sinclair (1991: 105) how this link between a group of semanti- cally similar verbs and a pattern can be accounted for, i.e. “does a pattern necessitate the selection of a particular sense of a word, or does the selec- tion of a particular sense, necessitate the use of a particular pattern”. From a construction grammar point of view, the answer must probably be: both. According to Goldberg (1998), the meaning of argument structure con- structions goes back to basic events that human beings experience and might have developed in connection with certain prototypical verbs such as give for the ditransitive construction or put for the caused-motion construc- tion. Through the use of the construction with more and more different but still semantically similar verbs the construction has adopted a meaning of its own, which it can attribute to other verbs occurring in it, provided the verbs do not directly contradict this construction meaning.63 From this it follows that a fairly wide range of verbs can – at least theoretically – be used in a wide range of constructions (cf. Croft 1998: 91). The notion of construction meaning indeed nicely explains which semantic interpretation nonsense verbs receive in a construction, e.g. She topamased him some- thing as a transfer verb, and it can also account for creative uses like He sneezed the napkin off the table (both examples from Goldberg 1995; see also Goldberg 2006a: 58–59).64 In an experimental study Gillette et al. (1999: 165) however found that while “mental verbs [see, look, want, know, like, think] are correctly identified 90% of the time via syntactic evidence

63. Compare Behrend (1998: 448) for a criticism of the notion of construction meaning relating to “basic events” on the basis of the failure of earlier such models to “[stand] the test of empirical time”. 64. For an experimental study of nonsense verb interpretation see Naigles (1990). Summary of the results 309

… all the other verbs are collectively identified by this evidence only 40% of the time [emphasis mine].” This means that the interpretation was more often incorrect than correct. A limitation to such constructions underesti- mates the amount of lexical properties of verbs which still have to be ac- counted for: not all verbs with a certain semantic profile can occur in all semantically compatible constructions as has clearly been shown above. At the same time not all patterns are associated with one specific construction- al meaning that they force on the verb: depending on the verb, one and the same pattern can be the representation of very different participant patterns, which can only be explained in an item-based approach (see also Goldberg 2006a: 94).

5.2.5 Accounting for alternative valency constructions

Some verbs can have more than one possible realization of a certain par- ticipant or participant pattern (cf. 3.3.4). However, other semantically simi- lar verbs are often more restricted by not being complemented with the whole range of theoretically available options. Many of the syntactic gaps cannot be related to the fact that a verb does not have a certain participant but that it only allows a restricted range for its realization as opposed to other semantically similar verbs. This provides a second important vantage point for an evaluation of the relationship between form and meaning. While no claim is made that such alternative constructions are completely synonymous, no systematic difference related to the form of the pattern or the complement type could be identified, either. That different comple- ments can express the same participant does not exclude the possibility that a certain specialization of each variant can be observed. This has already been shown above in connection with the realization of TOPIC in combina- tion with talk. This is also in line with the notion of economy in language, which claims that any extension in language which is not economic will be lost (Martinet [1955] 1981: 85; Zipf 1949). Several aspects can account for a difference between such alternative realizations. Often – but not always – these alternative constructions differ with respect to their information struc- ture: a BENREC which is realized as an [NP]-complement directly following the verb (1a) is possibly considered more affected than a BENREC realized by [to_NP], while in the second example it is informationally more high- lighted (1b).

(1a)INV I gave Sonya the book. (1b)INV I gave the book to Sonya. 310 Conclusion

In construction grammar terms (1a) and (1b) are realizations of different constructions, the ditransitive construction and the caused-motion con- struction (Goldberg 2006: 34). While the semantic difference can be cap- tured in this way, the semantic similarity which obviously exists between these two alternatives cannot (see below). The difference between such constructions can often better be described on an additional level of analy- sis. Clausal roles can explain the specific focus a noun phrase directly fol- lowing the verb receives. Semantic subtypes as an additional differentiation of participants can account for semantic differences, which do not lead to a change in the sequence of participants. The analysis as subtypes (cf. 2.1, see also below) has the clear advantage that the demand that each partici- pant can be expressed only once (cf. Fillmore 1968, Chomsky’s projection principle 1984) is fulfilled, while it is still possible to account for semantic differences. Moreover, the actual choice of one of these alternative constructions is influenced by a range of other factors as well, such as the length of the elements functioning as BENREC and AFFECTED. There is a clear tendency to shift the longer elements to the end of the sentence (cf. Gries 2003) (cf. 3.3.4).

(2a)INV I gave him the book I had told him about three weeks ago. (2b)INV ?I gave the book I had told him about three weeks ago to him. (3a)INV ?I gave John, an old friend of mine who I’ve known at least since we went to kindergarten, the book. (3b)INV I gave the book to John, an old friend of mine who I’ve known at least since we went to kindergarten.

Correspondingly, an element that is realized by a pronoun is more likely to be realized in the slot directly following the verb.

(4a)INV I gave her the book. I gave it to Anna. (4b)INV I gave the book to her. ?I gave Anna it.

Schlesinger (1995) lists as factors for a possible restriction of the alterna- tion between complementation with +[NP] and with a particle complement what he calls completion and feat:

(5a)QE know the boy shot the fox grasp the rope (5b)QE know of the boy shot at the fox grasp at the rope (6a)QE John swam the lake. John swam in the lake. (6b)QE ??John swam the pond. John swam in the pond. Summary of the results 311

While the examples in (5a) always imply that the action is completed, the alternatives with the particle always imply that this is not the case (Schlesinger 1995: 63–64). According to Schlesinger (1995: 64–65), the difference between (6a) and (6b) rests on the fact that a noun phrase rea- lization is only possible “when the activity leads to a not quite usual achievement, a ‘feat’, so to speak … When the activity does not have this property, the locative Noun Phrase will appear as a prepositional object, even when the activity has the property completion”. Here, the particle complements and noun phrase complements were tak- en to be different subtypes of a participant. This clearly depends on the role concept that is taken as the basis for the description of participants. Cate- gorical roles, other than relational roles, are static and thus refer to the enti- ties involved irrespective of what kind of function they have in a sentence (Tarvainen 1985: 13–15). In that sense the two sentences above exemplify alternative realizations of the same role. In order to capture the semantic difference as well as the fact that they exclude each other in the same con- text, they are here treated as subtypes of one participant. Again, one could also refer the fact that the lake in (6a) appears to be more affected when realized as a noun phrase directly following the the verb than when realized as a particle phrase to the level of clausal roles (Herbst and Schüller 2008: 158–163) (cf. 4.1). However, variation between [AdjP/NP] and [as_AdjP/NP] as PREDICA- TIVE, between [about_NP] and [on_NP] as TOPIC, or between [V-ing] and [to_INF] as ÆFFECTED, cannot be accounted for in this specific way. Prim- ing effects might play a role for all kinds of alternative realizations as these are not tied to a change in position of a certain participant in the pattern. These have been shown to play a role (e.g. by Gries 2005), while this again may be counterbalanced by horror aequi effects (e.g. Rohdenburg 2003) (cf. 3.3.4.8). Wasow and Arnold (2003: 134) also indicate that the frequen- cy of different alternatives plays a role in language processing:

These differences have been shown to influence both comprehension and production. Hence, in the absence of a demonstration that they can be de- rived from other factors, we tentatively conclude that lexical bias informa- tion is a distinct factor in determining constituent ordering.

These factors show one thing quite clearly: if it is possible to sidestep one pattern and choose another one for clearly non-semantic reasons in order to express the same content with the intention of avoiding an increased pars- ing load, then they must be true semantic alternatives. Describing them simply as different constructions which are not related in any way because 312 Conclusion of a difference in form fails to adequately account for the clear parallels in their usage. Moreover, pragmatic aspects need to be taken into considera- tion as well. Fine distinctions in utterance meaning are not necessarily con- nected to verb meaning in a strict sense but rather, for instance, to the in- tention of the speaker. Such factors can only account for the kind of variant chosen if more than one realization of a participant is available but not why certain verbs can occur with only one or a restricted number of alternatives in the first place. Since semantic factors fail to provide reliable predictions as to the possible occurrence of verb-pattern combinations, other factors must play a role. Schlesinger, for instance, argues that example (8) is not possible since it would inevitably lead to garden path effects, since John mentioned Paul is a plausible sentence in which Paul has the role AFFECTED.

(7)QE John told Paul the news. (8)QE *John mentioned Paul the news. (Schlesinger 1995: 68–69)

The linear processing of the sentence would therefore necessitate a reanaly- sis of Paul as BENEFICIARY, which is not the case with Paul in (7) which can only be interpreted as BENEFICIARY in the first place, also in the sen- tence John told Paul. The limitation to a particle complement for the ex- pression of BENEFICIARY in combination with mention thus facilitates processing (Schlesinger 1995: 69–71). This garden path restriction can, however, not be attributed general validity: a sentence such as (9) would then be excluded as well, which obviously is not the case.

(9) John called George irresponsible.65

This example indicates that the most likely solution is one, where fre- quency factors, processing constraints, semantic features and other features jointly account for how certain combinations have been conventionalized. However, what a discussion of such aspects obscures is the fact that there does not always seem to be a semantic difference at all. The compari- son of alternative complement types in chapter 2 showed that there are cases where native speakers were not able to identify a semantic difference, nor did a close analysis of the co-text and context in which these valency patterns actually occurred reveal any form of specialization: the alternative patterns were found in the same contexts with the same collocates, i.e. de facto no specialization has taken place (cf. 2.3.2). From a diachronic point

65. Example by courtesy of Michael Klotz (September 2009). Summary of the results 313 of view this might indeed be an intermediate state in the sense that a specia- lization might only be a matter of time. Differences in the frequency of such alternatives might also be an indicator in this respect. At the mean- time, there does not seem to be sufficient evidence for such a hypothesis and it could very well be the case that idiosyncrasies continue to exist. What is important, however, is that in the process of language acquisition, only synchronic data is available to the language learner and also in later stages language competence does not depend on historical developments in the language. Thus, language processing must be able to cope with such irregularities as have been identified in the previous chapters.

5.2.6 Idiomaticity in complementation

It was not possible to link the restrictions which were identified for the verbs examined here to the meaning of these verbs in any systematic way. What the data suggests is that conventionalization, and thus storage of enti- ties larger than single words, plays a very important role. Further evidence for this comes from the lexical properties of some of the patterns examined here, which have a clear effect on complementation. If a verb is strongly associated with a certain context or co-text, i.e. if it is part of an idiom or a collocation, this can also influence how strongly it is associated with a spe- cific pattern; for instance, of the 241 occurrences of the lemma keep +[NP/ REFLPRON] +[to_X], the element functioning as complement in the to- phrase is a minimum in 91 cases (e.g. keep something to a minimum)and a reflexive pronoun in 140 cases (e.g. keep it to yourself or she keeps herself to herself). Thus with keep the pattern is severely restricted to certain lexi- cal fillings, which is not the case with semantically similar verbs, for ex- ample sustain. Sustain only occurs 15 times in this pattern in the BNC, while six of these uses are related to a different sense of sustain (sustain injuries/damage to something), which also appear to be fairly fixed expres- sions. For the other occurrences no pattern could be identified. Such collo- cations can in part also explain why semantically similar verbs simply can- not fill the same slot: the pattern [SCU]+ ‘teach’ +[to_NP] was exclusively found with teach and only in a very limited lexical environment illustrated in (1), while educate was rated as unacceptable in this pattern and no con- sensus was reached among native speakers as regards the acceptability of instruct.

(1)VDE The fact that schools are to be judged on the pupils’ performance puts pressure on teachers to teach/?instruct/*educate to the tests. 314 Conclusion

Similar cases are listed in the Patternbank as lexically specified patterns. This is very much in line with a Sinclairian view of language as a system where the choice of one element can help to predict the choice of the fol- lowing elements in the sense that they are co-selected as suggested by the idiom principle (e.g. Sinclair 2008: 409). Yet this is not necessarily purely based on verb meaning but again indicates the important role of conventio- nalization. For instance the combination fight +[NP] +[on_NP] seems to be closely associated with the context of elections as illustrated in example (2): of the 89 examples of this pattern in the BNC, 66 belong to the domain of elections.

(2)BNC That general election will be fought not simply on fair rates, but on a fair Government. HHW 6894

Although fight someone/something about something is generally much more frequent, it is not found in this context and fight something + over something is found only three times. Hence, there seems to be a certain potential for predicting certain patterns in reference to specific contexts as is the case here with fight on. However, this can neither be located in the meaning of fight nor in the meaning of on. Similarly, support can be com- plemented with [on_NP], [over_NP], and [upon_NP] to express the partici- pant BASIS. Yet while the first two options are more frequent and do not seem to be restricted to a specific topic, the pattern [NP]+ support +[NP] +[upon_NP] occurs in the BNC only in the clearly confined context of ar- chitecture. This is certainly not predictable from the meaning of upon, and the semantic field architecture as such does certainly not block combi- nations with on, yet in all other areas upon seems to be impossible or at least extremely unusual.

(3)VDE If Tunisia did leave the League, the nine members which support it on this question might be tempted to follow suit. (4)VDE About sixty Conservative MPs failed to support the government over a motion by Labour to have the Poll Tax abolished. (5)BNC This, as at the Pantheon in Rome, is to support the hemisphere upon cylindrical walls. HWB 1057

Another factor for the choice of one pattern over another one might also be a stylistic one. A prime example here is the variation between complemen- tation with on- and upon-phrases. Even if the distribution can possibly not be completely explained by means of style, different aspects of style are at least a fairly good marker for most of the occurrences of upon. Play, for Summary of the results 315 instance, is complemented by +[on_NP] for any type of musical instrument, while +[upon_NP] is found in the BNC predominantly with old-fashioned instruments or music (6). The verb ride complemented with +[on_NP] can be found both with animate, e.g. horseback, as well as mechanical means of transportation, e.g. bus (see also Klotz 2000 for similar examples). The complement +[upon_NP] was only found with animate beings (7).

(6)BNC A shepherd who kept his flock upon the side of the mountain and knew the legend, began one day to play upon his bagpipes. ASW 1252 (7)BNC Then the Cid clapt spurs to the mule upon which he rode, and vaulted into a piece of ground which was his own inheritance, and answered, Sir, I am not in your land, but in my own. ASW 544

Note that in these cases it is not so much the style or the historical origin of the text (all BNC examples are from various age groups and the texts all stem from the second half of the 20th century) but rather the topic of the text that might influence the choice of pattern. Upon creates clear connotations such as higher degree of formality, old age or “the past” and a higher de- gree of prestige (“the ‘good’ old days”). Thus there are obviously many mechanisms at work influencing the pattern choice which are clearly not purely semantic. Thus pattern choices can also correlate with lexical choices within the complements. The phenomenon of collocation has regularly been observed as delimiting the semantic options which ought to be availa- ble but are not employed: “Many other perfectly grammatical ways of say- ing the same things are conceivable: but people just don’t say them” (Stubbs 2001: 58). While this emphasizes the notion of conventionalization and storage in the sense that such combinations are highly entrenched and thus possibly preempt other plausible uses, it also contradicts a rule-based view for meaning and form correspondences.

If we metaphorically suppose that a word can be written into the [mental] lexicon, then each time a word in processing is mapped onto its lexical re- presentation it is as though the representation was traced over again, etching it with deeper and darker lines each time. Each time a word is heard and produced it leaves a slight trace on the lexicon, it increases its lexical strength. (Bybee 1985: 117)

The types of example show that these co-textual indicators have no predict- able power in the sense that one could deduce the choice without knowing the collocation already. General rules cannot apply since the types of re- striction or preferences are extremely closely bound to the specific context in which they occur. It does not seem plausible to speak of rules under such 316 Conclusion circumstances since the number of rules or the number of exceptions to one more general rule could be expected to be as memory-intensive as storage of individual co-occurrences or patterns. While this does of course not prove that whole patterns are stored, it counterbalances the original advan- tage of rules, i.e. saving memory capacity. While a pattern might be blocked or preempted because an alternative pattern is closely associated with the verb, a less categorical case can be found in a great number of cases in the sense that one pattern is more close- ly associated and thus more frequent than an alternative one. Even if there are no restrictions as such, i.e. different semantically similar verbs can oc- cur in the same valency constructions, they often differ with respect to their preferred way of expressing the same semantic content. A well-studied example is the difference between start and begin (cf. 3.3.6) with respect to complementation with a to-infinitive or a participle clause. Further exam- ples can also be quoted here from the comparison of different complement types expressing TOPIC, e.g. agree on something is much more frequent than agree about or upon something (cf. 2.2.1.2), although for instance talk about something is much more frequent than talk on something. No seman- tic differences could be identified explaining this preference in the case of agree and start vs. begin. Since frequency can be considered one of the driving forces for gram- maticalization (e.g. Bybee 2003), one might assume that frequent verbs might also exhibit differences with respect to pattern compatibility when compared to less frequent but similar verbs. While this applies to some of the verbs examined here, there are also counter-examples (e.g. recommend with 5,882 occurrences in the BNC and suggest with 28,246 occurrences can be found in the same number of patterns). Thus the token frequency of verbs does not necessarily play a crucial role. However, the type frequency of valency patterns clearly seems to play a role (e.g. Bybee 2001: 119). The Patternbank demonstrates this with respect to the type frequency informa- tion about patterns: the pattern with the highest type frequency is [SCU]+ verb + [NP], followed by the so-called intransitive pattern [SCU]+ verb +[0], i.e. patterns consisting of few elements which are not morphologically marked (i.e. semantically most neutral). The likelihood that a verb can be found with one of those patterns or that such a pattern is extended to a new verb as a new pattern is higher than for other patterns. However, the pattern [SCU]+ verb +[that_CL] is the third most frequent pattern in the Pattern- bank, which does not mean that it can freely be extended to other verbs, as has been shown above. The same applies to the pattern [SCU]+ verb Summary of the results 317

+[for_NP], which is also among the most frequent patterns and which is by no means semantically neutral, as the analysis in chapter 4 has shown.

5.3 Implications for a theory of complementation

All in all, the summary has clearly shown that while a large percentage of verb complementation seems to be closely linked to different semantic aspects, this in itself is not sufficient to account for which patterns are con- ventionally used and which are not: on the one hand, the analyses in chap- ter 3 have shown that there are still restrictions which cannot be accounted for and that semantic features such as lexical aspect or selection restrictions do not sufficiently correlate with restrictions in general. Moreover, chapter 2 has shown that complements do not necessarily have such a straightfor- ward meaning that would make them candidates for complementation based on meaning. This is illustrated in Figure 37.

Results of empirical Factors considered relevant for Results of empirical research (chapter 2) complementation research (chapter 3)

• Too many semantically plausible patterns A Verb meaning which are not • Participants acceptable (>3.2) • Aspectual Profile • No sufficient • Selection correlation with • Examples with no Restrictions situation type stable semantic differences for this properties (>2) sample (>3.4.3) • Examples of • No sufficient synonymous B Meaning of correlation with complement types complements selection (>2) restrictions for this • Morphological sample (>3.4.2) • Examples of components polysemous • Configurational complement types components (>2)

Figure 37. Factors generally considered relevant for verb complementation as opposed to empirical evidence presented in chapters 2 and 3

Verb meaning (participant inventory, lexical aspect, selection restrictions) (A) and complement meaning (B) are not sufficient to account for how semantic concepts are formally realized in the sense of verb complementa- 318 Conclusion tion. There is much more problematic data (indicated in the arrows to the left and right of [A] and [B]) that needs to be accounted for, if the claim that form is determined through meaning and that this can be described in terms of general rules is to hold. On the one hand, this insight contributes to a further elaboration of a theoretic descriptive model of the English lan- guage. On the other hand, this question is also important from a psycholin- guistic point of view: if complementation is too irregular to be totally rule- governed, storage must play a considerably more important role in the processing of sentences, and this might also have pedagogical implications with respect to the learning and teaching of valency structures. The association between meaning and form is not random, which is also apparent in the detailed analyses of example sentences provided in chapters 2 and 3. However, a one-to-one correspondence between semantic partici- pant pattern and formal valency pattern is unusual at best for the verbs ana- lyzed. The form-meaning pairings or valency constructions which are the re- sult of this matching up of participants and valency complements are not necessarily synonymous in their entirety. They might differ semantically to a certain degree yet not with respect to the participant roles they express. This is apparent in the fact that complements expressing the same par- ticipant cannot occur in one and the same pattern, since they fulfil the same semantic function in the activity or a situation described by the verb, e.g. TOPIC in (1) and AFFECTED in (2) (cf. 2.1.3):

(1)INV *He talked on insects about butterflies. (2)INV *She loves to cook reading.

If a semantic difference can be attested that is not “different enough” to block a co-occurrence of the elements in one pattern, the introduction of a secondary level enables us to deal with co-occurrence restrictions coin- ciding with semantic similarity (cf. 2.1.3). A specification of valency slots into subtypes thus provides a model for coping with such cases. As illustrated in Figure 38 each valency slot of a verbal lexical unit is considered to be associated with a general participant role (X, Y) under which more specific semantic subtypes – if available – are subsumed (Y_1, Y_2, etc.). Likewise, associated with each valency slot are one or more formal complements which express the general participant role and the semantic subtype roles (if available). Thus a semantic differentiation is as- sumed on two levels: (1) that of the valency slot, and (2) – within the va- lency slot – that of subtypes. Thus a speaker of a language has to know which participants are associated with a specific lexical unit (participant Implications 319 inventory) and with what kinds of complements they can be expressed (complement inventory).

Valency Slot I Valency Slot II Particitpant Role [X] Particitpant Role [Y] Complement: [Į] Complement [ȕ] Subtype [Y_1] Complement [Ȗ] Subtype [Y_2] Complement [į] Subtype [Y_3] verb

Figure 38. Idealized valency structure with semantic subtype specifications

The illustration of the valency structure in Figure 38 is idealized to a certain extent since the analyses in chapter 2 produced several deviations from it. First, as indicated above, it is by no means always the case that a valency slot has subtypes in the first place – not even when the slot can be realized by different complement types. Thus a formal difference must not neces- sarily lead to a different semantic subspecification. This does, of course, not exclude differences in the field of pragmatics or information structure. Secondly, while this model can explain restrictions such as the ones in (1) and (2) above, there are further restrictions which simply fall outside the scope of this scheme and which can only be captured if the pattern as a unit is taken into consideration as well (cf. 2.2.2 and 4.1). The significant num- ber of irregularities identified with respect to the combination of comple- ments which are as such available to a verb also shows that it would not be sufficient if the participant inventory were stored alongside a complement inventory. While such a grammar would produce a considerable number of correct sentences, it would also generate unacceptable combinations. There are still cases where the whole pattern or construction must be stored. Va- lency structures as modeled here can capture a considerable number of cases where a participant can be realized by more than one complement type and consequently illustrate verb-specific specifications of complement types. However, they cannot (and were not intended to) serve as a fully- fletched explanation of form-meaning pairings on the level of verb valency. Consequently, an additional element of storage has to be introduced, i.e. a pattern inventory which includes the combinations of complements which are available in a specific language. This can be envisaged in the form of a 320 Conclusion valency pattern inventory and a participant pattern inventory (or a com- bined valency construction inventory) for each lexical unit. Table 48 illustrates the theoretic variability such a matrix of valency constructions and lexical units might permit. Theoretically speaking, the range of possible variability would even be higher than indicated here since already the association between valency pattern and participant pattern (dimensions 1 and 2) is not completely systematic.

Table 48. Variability of valency constructions for different lexical units which are semantically similar

Dimension 1 and 2 valency pattern + participant pattern VC I VC II VC III lexical unit 1 x x x lexical unit 2 x x lexical unit 3 x x lexical unit 4 x x lexical unit 5 x lexical unit Dimension 3 3 Dimension lexical unit 6 x lexical unit 7 x

With so many different variables, the range of possible combinations would be extensive, yet the associations that actually exist in the English language are of course not completely random, as has also been shown in the analys- es in chapter 3. Obviously, the semantics of the lexical unit, its participants, and that of its complements limits these options considerably. Each seman- tic participant – and consequently each participant pattern – can only be realized by a limited number of formal elements. Consequently, for each participant pattern one can assume a confined list of possible valency pat- terns, depending on the kinds of participants expressed. Following Herbst (2009) such lists of possible realizations of one specific activity or situation by means of a set of different valency constructions can be described as a constructeme, i.e. “the set of all valency constructions that share the same participant structures.”66 Such a constructeme is illustrated in Table 49.

66. This term has first been used in the introduction to the Erlangen Valency Pat- ternbank which was launched in July 2009 (www.patternbank.uni-erlangen. de). It has to be kept in mind that it was never the aim of this analysis to pro- vide a collection of synonymous constructions: instead the semantic com- Implications 321

Table 49. Constructeme AGENT + verb + AFFECTED + PREDICATIVE

AGENT +[SCU] +[SCU] +[SCU] +[SCU] AFFECTED +[NP] +[NP] +[NP] +[NP] PREDICATIVE +[NP] +[as_NP] +[AdjP] +[as_AdjP] assume 9 suppose 9 presume 9 choose 9 select ? 9 pick 9 nominate ? ? designate ? ? appoint 99 name 99 elect 99 declare 99 9 9 announce 99 9 9 proclaim 99 9 9 state 99 9 9 pronounce 99 9 judge 99 9 9 consider 99 9 9 think 99 9 deem 99 9 9 find 99 believe 99 contemplate 9 regard 99 recognize 99

Apparently, not all verbs which can be used to express certain configura- tions of participants can occur in all possible valency patterns which ex- press them. With respect to verbs occurring with a PREDICATIVE partici- pant, no clear system is observable. For instance, while appoint, name, and elect might indeed be more different from choose than select is from choose (the latter occurring in the same patterns), nominate and designate

parability of many of the constructions was taken as the basis for a search of semantic differences between them. As demonstrated above, it was not possi- ble to relate such differences systematically to the meaning of either verbs or complement types. 322 Conclusion appear to be more similar to appoint and elect but do not allow the same range of patterns. Despite the fact that the verbs are semantically similar and have the same participant patterns, they obviously differ and the differ- ences cannot directly be related to a difference in meaning. The argument that the meaning of lexical units and complements turns a vast amount of combinatorial possibilities into a predictable set of actual verb complemen- tation patterns obviously overestimates the extent to which irregularities can be explained on the basis of meaning. The choice of patterns cannot be deduced from the meaning of this lexical unit in a systematic way. Thus a considerable amount must be covered by the storage of conventionalized information. One can clearly say that semantically close verbs often also exhibit similarities with respect to the kinds of patterns they can occur in. However, such tendencies are not strong enough to allow for a prediction of verb-pattern associations, which disproves a deterministic point of view of verb complementation. From a diachronic point of view, it is sensible to assume that each use of a verb in a new pattern has some kind of a semantic basis. In present-day English, however, such a link is not necessarily automatically apparent and the choice of one complement type over another cannot be deduced from the meaning of either complement or verb. Welke (2005: 117) argues that even if one can assume that certain formal relations used to be functionally distinct, these functions have been lost over time, e.g. German helfen (‘help’) + dative and unterstützen (‘support’) + accusative. Habermann (2007: 86) also concludes for German that “[b]oth, formal structures and the assignment of case roles, are not necessarily the same throughout the different historical language periods.” Thus, a deterministic view simply underestimates the complexity of this relationship. 67 One can, however, very reasonably expect that the link can be abduced instead, i.e. that the combination is not random but transparent enough to facilitate the storage of such combinations (Hopper and Closs Traugott 2003: 41–43). Thus a pattern-verb-association receives an additional se- mantic backing in the sense that an ex post explanation often appears rea- sonable. From a psycholinguistic point of view, this might play a crucial role for memorizing the pattern inventory of a verb and the limited size of

67. As reasons for valency shifts, Habermann (2007: 89–90) also lists phonetic shifts which lead to a syncretism of form which again leads to a reanalysis of one case to a different one and the “decline of morphological case for the benefit of analytically formed prepositional phrases” or “the decrease in the variety of possible constructions linked with a gradual development of proto- typical clause patterns”. Implications 323 constructemes might be crucial with respect to our processing capacities. Similarly, also the form of complement types can provide such a semantic backing. Complements which are morphologically marked are limited as regards the kinds of participant they can express by the possible meanings of the morphological element they carry. Nonetheless, even morpho- logically marked complement types can have a considerable number of different meanings, and consequently they can serve as realizations of a considerable number of participants. As indicated above, we are clearly not dealing with a one-to-one correspondence, even if metaphorical extensions are taken into consideration. Patterns which are configurationally and not morphologically marked, i.e. by a specific sequence of noun phrase complements, might be semanti- cally less flexible than morphologically marked patterns. The formal se- quence, as construction grammar suggests, seems to have developed a meaning of its own in such a way that it can be extended to verbs which do not necessarily have all the relevant participants in their participant inven- tory. Fried and Östman (2004: 52–53) point out that such constructions “pose a serious challenge to syntactic theories that assume the clause struc- ture to be always a direct projection of verbal argument structure” since participants can be added which “cannot be projected by the lexical predi- cate (i.e., they are not even in the inventory of frame elements as potential additions) and ... the addition does not affect the meaning or interpretation of the lexical predicate”. Additionally, constructions are likely to play a role, especially in ac- counting for how certain patterns are extended to new verbs. However, there are patterns which are identical in form but not in meaning, and con- structions with the same form and meaning which still exhibit differences, for instance, with respect to passivization. Thus while constructions as an additional level provide valuable insights into a certain proportion – possi- bly even a considerable proportion – of pattern choices, they are not enough to account for complementation. Thus while frequent complement-partici- pant associations or pattern-participant associations can help to limit choices, as well as the participant inventory of the verb, the exact mappings between meaning and form are not predictable. Figure 39 illustrates all these different interrelated aspects. Information which is verb-specific and thus directly related to the lexical unit (upper level, verb-specific knowledge) interacts with information which goes beyond specific verbs (lower level, general knowledge). Those spheres are clearly not completely separate: general knowledge can only be seen as the result of generalization processes over verb-specific knowledge. Such gene- 324 Conclusion ralized knowledge could include knowledge about which constructions are typically used to describe specific scenes (e.g. transfer for [SCU]_AGENT + verb +[NP]_BENREC +[NP]_AFFECTED or [SCU]_AGENT+ verb +[NP]_ AFFECTED +[to_NP]_BENREC).

STORAGE

verb participant inventory meaning

participant patterns verb specific valency knowledge constructions valency patterns

complement meaning complement inventory

• constructions which can be used to general express a specific scene (constructeme) knowledge • complements which are frequently used to express a concept/participant

Figure 39. Elements of verb complementation

Similarly, on a lower level, frequent participant-complement associations could also be part of a speaker’s general knowledge, i.e. that [for_NP] can express a GOAL or a BENEFICIARY and [about_NP] a TOPIC. In part, this kind of knowledge clearly overlaps with knowledge of constructemes. A distinction is implied here to emphasize that such knowledge might also exist in the form of smaller constructions, i.e. outside the scope of argument structure constructions. Verb-specific knowledge (upper level) includes those aspects listed in Figure 37 above, i.e. the selection restrictions of a verb, its aspectual profile, and obviously also knowledge of its participants. The latter is indicated specifically as the participant inventory in Figure 39, firstly because it was clearly the factor which had the biggest effect on complementation but also because this kind of knowledge must directly interact with knowledge about how participants can be combined in con- nection with a specific verb (participant patterns) and ultimately also with how such participant patterns can be expressed in the form of valency con- Implications 325 structions (see arrows in Figure 39). Since this clearly differs from verb to verb, even if the participant inventories are identical, this kind of know- ledge must be considered to be stored. At the same time, a speaker knows what kinds of complements are associated with a verb (complement inven- tory) and how they can be combined (valency pattern inventory) in order to express certain participants (valency constructions). The analyses in chapter 4 have clearly shown that there is a considerable amount of idiosyncrasy that cannot be accounted for, if only a complement inventory is assumed, i.e. the speaker must also store pattern information.68 To what extent speakers only store valency constructions (i.e. as in the construction grammar approaches) or also purely formal valency patterns and purely semantic participant patterns cannot be decided on the basis of this work, while it might well be that such combined forms as constructions play a crucial role. At the same time, redundancies do not necessarily have to be a disadvantage from a psycholinguistic point of view, as has been discussed above. Moreover, not all speakers of a language necessarily need to make use of all possible sets of information for each lexical unit in their mental lexicon, especially as they are probably best taken as the result of individual generalization processes. General knowledge, as described above, results from verb-specific knowledge, yet it is conceivable that it also facilitates the storage of con- structions in the sense that plausible constructions are backed by know- ledge of further, similar associations. Such factors clearly deserve further research especially as regards the psychological validity of such notions. While no claim is to be made in this present context about what might con- stitute a psychological model of language, the analyses presented here at least clearly point in one direction:

[1] While verb meaning (participant inventory, aspectual quality, selec- tion restrictions), complement type meaning, as well as the meaning associated with certain patterns (i.e. constructions) can account for a considerable number of a verb’s complementation patterns, the num- ber of restrictions found is, by far, too high for these to be regarded as a peripheral phenomenon.

68. The importance attributed to storage in this model is also is in line with Fried and Östman (2004: 42–43) who assume a lexical representation with two lay- ers, where one represents idiosyncratic information and the other “the corres- ponding event pattern in a highly schematized form … [which] is not neces- sarily fully predictable…”. 326 Conclusion

[2] Consequently, a model without a strong role for storage cannot cope with the irregularities identified. [3] Meaning must be considered a crucial factor in facilitating storage by backing pattern-verb associations (cf. abduction).

Abduction can explain why we remember combinations that are con- ventionalized: the combinations of patterns and verbs speakers of the lan- guage encounter appear to be good semantic matches. At the same time, the process of abduction cannot explain why there are still patterns which are plausible from a semantic point of view but which are nonetheless not used by native speakers, e.g. why some patterns are preempted by the existence of other patterns and others not. Even if patterns are sometimes extended in analogy to a semantically similar verb, such unconventional pattern choices are usually considered marked, unusual or outright ungrammatical by many native speakers. In the light of the fact that semantic specializations of complement types and patterns are often rather unsystematic, and the fact that semantic groups are not a sufficient indicator for the patterns which are actually associated with a verb either, it appears highly problematic to think of verb comple- mentation as a process governed by semantic rules. Such rules could only take the form of verb-specific rules, which is tantamount to saying that it is item-specific knowledge that has to be stored and cannot be the result of an online rule application. Whichever way one seems to turn, there are reasonable mechanisms, sometimes even rules and with them a number of examples which seem to prove them as necessary “tools” for accounting for the linking, yet there are always exceptions where this does not apply. The important point to bear in mind is that these exceptions are too common to simply be ignored. Thus one can only agree with Cruse (2000: 282) who said in reference to Fill- more’s six cases (1968) that “history shows elegant simplicity to be a fra- gile thing in linguistics.” Appendix 1

Distribution of pattern overlap and restrictions for verb groups as discussed in chapter 3.2

complemen- “unsyste- semantic ‘group different group members tation 999 ? matic” restrictions meaning’ P S meaning options restrictions (total) 1 ‘answer’ answer, reply, respond 48 34 0 10 4 4 0 0 2 ‘assume’ assume, suppose, presume, guess, (gather) 60 41 0 4 14 14 0 1 3 ‘choose’ choose, select, pick, elect, appoint, name, 144 89 11 20 21 9 12 3 nominate, designate 4 ‘declare’ declare, pronounce, announce, proclaim, 110 81 2 18 5 5 0 4 state 5 ‘get’ obtain, receive, get, acquire, buy, purchase 24 22 0 1 1 1 0 0 6 ‘hire’ hire, rent, lease 30 29 1 0 0 0 0 0 7 ‘intend’ mean, intend, plan, aim 56 37 2 7 8 8 0 2 8 ‘like’ love, like, adore 27 24 0 3 0 0 0 0 9 ‘not lose’ keep, maintain, preserve, 24 20 0 1 3 3 0 0 sustain 10 ‘permit’ allow, permit, authorize, 56 34 0 10 10 10 0 2 entitle 11 ‘practice’ train, rehearse, practice 33 25 1 3 2 2 0 2 12 ‘propose’ propose, suggest, 45 41 1 3 0 0 0 0 recommend 328 Appendix different meaning 0 2 0 1 0 0 0 4 0 0 13 0 0 0 12 34 34 12 1 3 0 3 0 0 0 0 10 10 18 11 13 PS 111 9 semantic semantic restrictions (total) “unsyste- matic” restrictions  ? 9 9 99 44 28 2 2 28 44 10 3 16 18 52 90 3 14 43 60 0 0 2 40 45 5 23 3 11 45 88 0 0 24 24 0 0 3 15 18 0 0 3 0 8 12 15 0 34 42 0 100 100 3 68 16 10 150 74 8 42 74 150 13 1233 844 39 39 1233 844 193 123 complemen- tation options 10.1515/9783110240788_appendix3 10.1515/9783110240788_appendix3 10.1515/9783110240788_appendix4 10.1515/9783110240788_appendix5 group members quarrel, argue, dispute, bicker recall, recollect, remember, reminisce show, indicate, demonstrate commence begin, start, train, teach, instruct, educate call ring, phone, consider, think, reflect, contemplate, ponder, judge respect, admire, esteem tolerate endure, bear, push urge, press, ‘group ‘group meaning’ ‘quarrel’ ‘quarrel’ ‘remember’ ‘show’ ‘start’ ‘teach’ ‘telephone’ ‘think’ ‘think highly’ ‘tolerate’ ‘urge’ 13 13 14 15 16 17 18 19 20 21 22 percent TOTAL in Appendices 2–5 are available at the following URLs: following at the 2–5 are available Appendices 2: Appendix 3: Appendix htttp://dx.doi.org/10.1515/9783110240788_appendix2 4: Appendix htttp://dx.doi.org/ 5: Appendix htttp://dx.doi.org/ htttp://dx.doi.org/ References

Aarts, Flor and Jan Aarts 1982 English Syntactic Structures. Functions and categories on sentence analysis. Oxford: Bohn, Scheltma & Holkema. Abbot-Smith, Kirsten, Elena Lieven, and Michael Tomasello 2004 Training 2;6-year-olds to produce the transitive construction: the role of frequency, semantic similarity and shared syntactic distribution. Developmental Science 7:1, pp. 48–55. Ackerman, Farrell and John Moore 2001 Proto-Properties and Grammatical Encoding. A Correspondence Theory of Argument Selection. Stanford: CSLI Publications. Ágel, Vilmos 2000 Valenztheorie. Tübingen: Narr. Algeo, John 2006 British or American English. A handbook of word and grammar patterns. Cambridge: Cambridge University Press. Allerton, David 1975 Deletion and proform reduction. Journal of Linguistics 11, p. 213– 238. Allerton, David 1982 Valency and the English Verb. London / New York: Academic Press. Baker, Mark C. 1988 Incorporation: A Theory of Grammatical Function Changing. Chicago / London: Chicago University Press. Baker, Collin F. 1999 Seeing Clearly: Frame Semantic, Psycholinguistic, and Cross-Lingu- istics Approaches to the Semantics of the English verb See. Doctoral Dissertation, University of California at Berkeley; available at: http://www.icsi.berkely.edu/~collinb/indexedDiss.pdf Barsalou, Lawrence W. 1992 Cognitive Psychology: An Overview for Cognitive Scientists. Hillsdale, NJ: Lawrence Erlbaum Associates. Behrend, Douglas 1998 Language under Constructions. Journal of Child Language 25:2, pp. 447–450. 330 References

Behrens, Heike 2007 The acquisition of argument structure. In Thomas Herbst and Katrin Götz-Votteler (eds.), Valency – Theoretical, Descriptive and Cogni- tive Issues. Berlin / New York: Mouton de Gruyter, pp. 193–214. Biber, Douglas, Susan Conrad, and Randi Reppen 1998 Corpus Linguistics: Investigating Language Structure and Use. Cambridge: Cambridge University Press. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad, and Edward Finegan 1999 The Longman Grammar of Written and Spoken English. Harlow: Longman. Boas, Hans Christian 2003 A constructional approach to resultatives. Stanford: CSLI Publica- tions. Bock, J. Kathryn 1986 Syntactic Persistence in language production. Cognitive Psychology 18, pp. 355–387. Bouchard, Denis 1995 The Semantics of Syntax – A minimalist approach to grammar. Chicago / London: The University of Chicago Press. Bolinger, Dwight 1968 Entailment and the meaning of structures. Glossa 2, pp.119–127. Bondzio, Wilhelm 1971 Valenz, Bedeutung und Satzmodelle. In Gerhard Helbig (Hrsg.), Beiträge. The Hague / Paris: Mouton, pp. 85–103. Bresnan, Joan 1994 Locative Inversion and the Architecture of Universal Grammar. Language 70, pp. 72–131. Bresnan, Joan 2001 Lexical Functional Syntax. Malden, MA / Oxford: Blackwell. Brinton, Laurel J. 1988 The Development of the English Aspectual Systems. Aspectualizers and post-verbal particles. Cambridge: Cambridge University Press. Brown, Gillian and George Yule 1983 Discourse Analysis. Cambridge: Cambridge University Press. Bühler, Karl 1982 Sprachtheorie. Die Darstellungsfunktion der Sprache. Stuttgart / New York: Gustav Fischer. Reprint. First edition 1934. Bußmann, Hadumod 1983 Lexikon der Sprachwissenschaft. Stuttgart: Alfred Kröner. References 331

Busse, Winfried and Jean-Pierre Dubost 1983 Französisches Verblexikon. Stuttgart: Klett. Second revised edition. First edition 1977. Bybee, Joan 1985 Morphology: A Study of the Relation between Meaning and Form. Amsterdam / Philadelphia: Benjamins. Bybee, Joan 1995 Regular morphology and the lexicon. Language and Cognitive Pro- cesses 10, pp. 425–455. Bybee, Joan 2001 Phonology in Language Use. Cambridge: Cambridge University Press. Bybee, Joan 2003 Mechanisms of Change in Grammaticalization: The Role of Fre- quency. In Brian D. Joseph and Richard D. Janda (eds.), The Hand- book of Historical Linguistics. Malden, MA: Blackwell, pp. 602– 623. Bybee, Joan 2006 From usage to grammar: the mind’s response to repetition. Language 82:4, pp. 711–733. Bybee, Joan 2007 Frequency of use and the organization of language. Oxford: Oxford University Press. Campbell, Aimee L. and Michael Tomasello 2001 The acquisition of English dative constructions. Applied Psycho- linguistics 22, pp. 253–267. Chomsky, Noam 1957 Syntactic Structures. The Hague: Mouton. Chomsky, Noam 1965 Aspects of the theory of syntax. Cambridge, Mass: MIT Press. Chomsky, Noam 1975 The Logical Structure of Linguistic Theory. New York / London: Plenum Press. Chomsky, Noam 1980 Rules and Representations. Behavioral and Brain Sciences 3, pp. 1– 61. Chomsky, Noam 1984 Lectures on government and binding. Dordrecht: Foris. Third revised edition. First edition 1981. Chomsky, Noam 1982 Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, MA: The MIT Press. 332 References

Chomsky, Noam 1986 Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: The MIT Press. Comrie, Bernard 1976 Aspect. An Introduction to the Study of Verbal Aspect and Related Problems. Cambridge: Cambridge University Press. Coseriu, Eugenio 1979 System, Norm und ‘Rede’. In Uwe Petersen (ed.), Eugenio Coseriu, Sprache – Strukturen und Funktionen. XII Aufsätze zur Allgemeinen Romanischen Sprachwissenschaft. Tübingen: Narr, pp. 45–59, trans- lated into German by Hansbert Bertsch. Third revised edition. First edition 1970. Croft, William 1991 Syntactic Categories and Grammatical Relations. The Cognitive Organization of Information. Chicago / London: The University of Chicago Press. Croft, William 1998 Event Structure in Argument Linking. In Miriam Butt and Wilhelm Geuder (eds.), The Projection of Arguments. Lexical and Composi- tional Factors. Stanford: CSLI Publications, pp. 21–63. Croft, William 2001 Radical Construction Grammar. Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. Croft, William and D. Alan Cruse 2004 Cognitive Linguistics. Cambridge: Cambridge University Press. Cruse, D. Alan 1986 Lexical Semantics. Cambridge: Cambridge University Press. Cruse, D. Alan 2000 Meaning in Language. An Introduction to Semantics and Pragmat- ics. Oxford / New York: Oxford University Press. Cuyckens, Hubert 1997 Prepositions in Cognitive Lexical Semantics. In Dagmar Haumann and Stefan J. Schierholz (eds.), Lexikalische und Grammatische Ei- genschaften präpositionaler Elemente. Tübingen: Niemeyer, pp. 63– 82. Dahl, Östen 1985 Tense and Aspect Systems. Oxford / New York: Basil Blackwell. Dik, Simon C. 1979 Functional Grammar. Amsterdam: North Holland Publishing Com- pany. Second Print. First edition 1978. References 333

Dirven, René 1989 A Cognitive Perspective on Complementation. In: Dany Jaspers, Wim Kloosters, Yvan Putseys, and Peter Seuren (eds.), Sentential Complementation and the Lexicon. Dordrecht: Foris, pp. 113–140. Dixon, Robert M.W. 1991 A New Approach to English Grammar, on Semantic Principles. Oxford / New York: Oxford University Press. Dowty, David 1982 Grammatical Relations and Montague Grammar. In: Pauline Jacob- son and Geoffrey K. Pullum (eds.), The Nature of Syntactic Repre- sentation. Dordrecht: D. Reidel, pp. 79–130. Dowty, David 1991 Thematic Proto-Roles and Argument Selection. Language: Journal of the Linguistic Society of America, 67:3, pp. 547–619. Dürscheid, Christa 1999 Die verbalen Kasus des Deutschen – Untersuchungen zur Syntax, Semantik und Perspektive. Berlin / New York: de Gruyter. Egan, Thomas 2008 Non-finite Complementation – A usage-based study of infinitive and –ing clauses in English. Amsterdam / New York: Rodopi. Eichinger, Ludwig M. 1995 Von der Valenz des Verbs und den Abhängigkeiten in der Nominal- gruppe. In Ludwig M. Eichinger and Hans-Werner Eroms (eds.), De- pendenz und Valenz. Hamburg: Helmut Buske Verlag, pp.37–52. Emons, Rudolf 1974 Valenzen Englischer Prädikatsverben. Tübingen: Niemeyer. Emons, Rudolf 1978 Valenzgrammatik für das Englische. Eine Einführung. Tübingen: Niemeyer. Engel, Ulrich 1977 Syntax der deutschen Gegenwartssprache. Berlin: E. Schmidt. Engel, Ulrich 1992 Der Satz und seine Bausteine. In: Vilmos Àgel and Regina Hessky (eds.), Offenen Fragen – offene Anworten in der Sprachgermanistik. Tübingen: Niemeyer, pp. 53–76. Engel, Ulrich and Helmut Schumacher 1976 Kleines Valenzlexikon deutscher Verben. Tübingen: Narr. Fabricius-Hansen, Cathrine 2004 Sprachverarbeitung und Valenz aus kontrastiver Sicht. In Ulrich Engel and Meike Meliss (eds.), Dependenz, Valenz und Wortstel- lung. München: Iudicium, pp. 187–198. 334 References

Fanego, Teresa 1996 The development of gerunds as objects of subject-control verbs in English (1400–1760). Diachronica XIII: 1, pp. 29–62. Fillmore, Charles J. 1968 The case for case. In Emmon Bach and Robert T. Harms (eds.), Uni- versals in Linguistic Theory. New York: Holt, Rinehart and Winston, pp. 1–88. Fillmore, Charles J. 1970 Subjects, speakers, and roles. Synthese 21, pp. 251–274. Fillmore, Charles J. 1972 On Generativity. In Stanley Peters (ed.), Goals of Linguistic Theory. Englewood Cliffs, NJ: Prentice Hall, pp. 1–19. Fillmore, Charles J. 1975 An Alternative to Checklist Theories of Meaning. In C. Cogen, H. Thompson, G. Thurgood, K. Whistler, and I. Wright (eds.), Proceed- ings of the First Annual Meeting of the Berkeley Linguistics Society, February 15–17, Berkeley Society, CA: Berkeley Linguistics, pp. 123–131. Fillmore, Charles J. 1977 Topics in Lexical Semantics. In Roger W. Cole (ed.), Current Issues in Linguistic Theory. Bloomington / London: Indiana University Press, pp. 76–138. Fillmore, Charles J. 1988 The Mechanisms of Construction Grammar. In Shelley Axmaker, Annie Jassier, and Helen Singmaster (eds.), Proceedings of the Four- teenth Annual Meeting of the Berkely Linguistic Society, pp. 35–55. Fillmore, Charles J. 2007 Valency Issues in FrameNet. In Thomas Herbst and Katrin Götz- Votteler (eds.), Valency – Theoretical, Descriptive and Cognitive Is- sues. Berlin / New York: Mouton de Gruyter, pp. 129–160. Fillmore, Charles J. 2008 Border Conflicts: FrameNet Meets Construction Grammar (unpub- lished manuscript of a paper given at the Euralex conference in Bar- celona, 2008). Fillmore, Charles J., Paul Kay, and Mary Catherine O’Connor 1988 Regularity and Idiomaticity in Grammatical Constructions: The case of let alone. Language 64:3, pp. 501–538. Fillmore, Charles J. and Paul Kay 1993 Construction Grammar Coursebook Chapters 1 thru 11 (Reading Materials for Ling X20): University of California, Berkely. [original source text not available; unpublished reader cited in Croft (1998)]. References 335

Fischer, Klaus 1997 German-English verb valency. A contrastive analysis. Tübingen: Narr. Francis, Gill 1993 A Corpus–Driven Approach to Grammar: Principles, Methods and Examples. In Mona Baker, Gill Francis, and Elena Tognini-Bonelli (eds.), Text and technology: in honour of John Sinclair. Philadelphia / Amsterdam: John Benjamins, pp. 137–156. Francis, Gill, Susan Hunston, and Elizabeth Manning 1996 Collins Cobuild Grammar Patterns 1: Verbs. London: Harper Collins. Frawley, William 1992 Linguistic Semantics. Hillsdale, NJ: Lawrence Erlbaum Associates. Fried, Mirjam and Jan-Ola Östman 2004 Construction Grammar – A thumbnail sketch. In Mirjam Fried and Jan-Ola Östman (eds.), Construction Grammar in a Cross-Language Prespective. Amsterdam / Philadelphia: John Benjamins, pp. 11–86. Gärtner, Eberhard 1998 Grammatik der portugiesischen Sprache. Tübingen: Niemeyer. Geuder, Wilhelm and Miriam Butt 1998 Introduction. In: Miriam Butt and Wilhelm Geuder (eds.), The Pro- jection of Arguments. Lexical and Compositional Factors. Stanford: CSLI Publications, pp. 1–20. Gillette, Jane, Henry Gleitman, Lila Gleitman, and Anne Lederer 1999 Human simulations of vocabulary learning. Cognition 73, pp. 135– 176. Gisborne, Nikolas 2001 Introduction. In Nikolas Gisborne (ed.), Event Structure, Argument Structure and Argument Linking. Special Issue. In: Nigel Love (ed.), Language Sciences 23:3, pp. 341–637, pp. 341–351. Givón, Talmy 1984 Syntax: A functional-typological introduction, I. Amsterdam / Philadelphia: Benjamins. Goldberg, Adele E. 1995 Constructions: A Construction Grammar Approach to Argument Structure. Chicago / London: The University of Chicago Press. Goldberg, Adele E. 1998 Patterns of Experience in Patterns of Language. In Michael Tomasello (ed.), The new psychology of language. Mahwah, NJ: Erlbaum, pp. 203–219. 336 References

Goldberg, Adele E. 2006a Constructions at Work. The Nature of Generalization in Language. New York: Oxford University Press. Goldberg, Adele E. 2006b The inherent semantics of argument structure: The case of the Eng- lish ditransitive construction. In Dirk Geeraerts (ed.), Cognitive Lin- guistics: Basic Readings. Berlin / New York: Mouton de Gruyter, pp. 401–437. Götz-Votteler, Katrin 2007 Describing semantic valency. In Thomas Herbst and Katrin Götz- Votteler (eds.), Valency – Theoretical, Descriptive and Cognitive Is- sues. Berlin / New York: Mouton de Gruyter, pp. 37–49. Greenbaum, Sidney 1977 Introduction. In Sidney Greenbaum (ed.), Acceptability in Language. TheHague: Mouton Publishers. Greenbaum, Sidney and Randolph Quirk 1970 Elicitation Experiments in English. Linguistic Studies in Use and Attitude. London: Longman. Gries, Stefan Th. 2003 Towards a corpus-based identification of prototypical instances of constructions. Annual Review of Cognitive Linguistics 1 (2003), pp. 1–27. Gries, Stefan Th. 2005 Syntactic Priming: A Corpus-based Approach. Journal of Psycholin- guistic Research 34:4, pp. 365–399. Gries, Stefan Th. and Anatol Stefanowitsch 2004 Extending collostructional analysis. A corpus-based perspective on ‘alternations’. International Journal of Corpus Linguistics 9:1, pp. 97–129. Gries, Stefan Th. and Anatol Stefanowitsch 2006 Cluster Analysis and the identification of collexeme classes. In Sally Rice and John Newman (eds.), Empirical and Experimental Methods in Cognitive/Functional Research. Stanford: CSLI Publications, pp. 1–18. Gries, Stefan Th., Beate Hampe, and Doris Schönefeld 2005 Converging evidence: Bringing together experimental and corpus data on the association of verbs and constructions. Cognitive Lin- guistics 16:4, pp. 635–676. Groefsema, Marjolein 2001 The real-world colour of the dative alternation. In Nikolas Gisborne (ed.), Event Structure, Argument Structure and Argument Linking. Special Issue in: Nigel Love (ed.), Language Sciences, pp. 525–550. References 337

Gropen, Jess, Steven Pinker, Michelle Hollander, Richard Goldberg, and Ronald Wilson 1989 The Learnability and Acquisition of the Dative Alternation in Eng- lish. Language 65:2, pp. 203–257. Gruber, Jeffrey S. 1965 Studies in Lexical Relations, Doctoral Dissertation, MIT, Cambridge, MA. Habermann, Mechthild 2007 Aspects of diachronic valency syntax of German. Iin Thomas Herbst and Katrin Götz-Votteler (eds.), Valency – Theoretical, Descriptive and Cognitive Issues. Berlin / New York: Mouton de Gruyter, pp. 85–100. Habermann, Mechthild and Gaston van der Elst 1997 Syntaktische Analyse. Erlangen / Jena: Palm & Enke. Sixth revised edition. First edition 1985. Haegeman, Liliane 1991 Introduction to Government & Binding Theory. Oxford / Cambridge, MA: Blackwell. Halliday, Michael A. K. 1967a Notes on and theme in English: Part 1. Journal of Lin- guistics 3, pp. 37–81. Halliday, Michael A. K. 1967b Notes on transitivity and theme in English: Part 2. Journal of Lin- guistics 3, pp. 199–244. Halliday, Michael A. K. 1970 Language Structure and Language Function. In: John Lyons (ed.), New Horizons in Linguistics. Harmondsworth: Penguin Books Ltd, Second edition, pp. 140–165. Hamawand, Zeki 2002 Atemporal Complement Clauses in English: A Cognitive Grammar Analysis. München: LINCOM Europa. Hawkins, John A. 1994 A Perfomance Theory of Order and Constituency. Cambridge: Cambridge University Press. Helbig, Gerhard 1971 Theoretische und Praktische Aspekte eines Valenzmodells. In Gerhard Helbig (ed.), Beiträge. The Hague / Paris: Mouton, pp. 31–49. Helbig, Gerhard 1979 Zum Status der Valenz und der semantischen Kasus. Deutsch als Fremdsprache 2, pp. 65–78. Helbig, Gerhard 1992 Probleme der Valenz- und Kasustheorie. Tübingen: Niemeyer. 338 References

Helbig, Gerhard, and Wolfgang Schenkel 1973 Wörterbuch zur Valenz und Distribution deutscher Verben. Leipzig: VEB Bibliographisches Institut. Second revised and extended edi- tion. First edition 1969. Herbst, Thomas 1983 Untersuchungen zur Valenz englischer Adjektive und ihrer Nomina- lisierungen. Tübingen: Narr. Herbst, Thomas 1988 A valency model for nouns in English. Journal of Linguistics 24, pp. 265–301. Herbst, Thomas 1996 How obligatory are obligatory complements? – An alternative ap- proach to the categorization of subjects and other complements in valency grammar. English Studies 2, pp. 179–199. Herbst, Thomas 1999a Valency between syntax and lexicology. In Wolfgang Falkner and Hans-Jörg Schmid (eds.), Words, lexemes, concepts – approaches to the lexicon. Studies in honour of Leonhard Lipka. Tübingen: Narr, pp.167–173. Herbst, Thomas 1999b Designing an English Valency Dictionary: combining linguistic theory and user-friendliness. In Thomas Herbst and Kerstin Popp (eds.), The Perfect Learners' Dictionary? Tübingen: Niemeyer, pp. 229–253. Herbst, Thomas 2003 Was soll zum Beispiel eine obligatorische Ergänzung sein? Oder ein zweivalentes Verb? – Zur Interdependenz valenzpolitischer Festle- gungen. In Alan Cornell, Klaus Fischer, and Ian Roe (eds.), Valency in Practice – Valenz in der Praxis. Oxford: Peter Lang, pp. 65–88. Herbst, Thomas 2007 Valency complements or valency patterns. In Thomas Herbst and Katrin Götz-Votteler (eds.), Valency – Theoretical, Descriptive and Cognitive Issues. Berlin / New York: Mouton de Gruyter, pp. 3–35. Herbst, Thomas 2009 Valency – item-specificity and idiom principle. In Ute Römer and Rainer Schulze (eds.), Exploring the lexis-grammar interface. Amsterdam / Philadelphia: Benjamins, pp. 49–68. Herbst, Thomas 2010 Valency constructions and clause constructions or how, if at all, valency grammarians might sneeze the foam off the cappuccino. In Hans-Hörg Schmid and Susanne Handl (eds.), Cognitive Founda- References 339

tions of Linguistic Usage Patterns. Berlin / New York: Mouton de Gruyter, pp. 225–256. Herbst, Thomas and Susen Faulhaber forthc. Optionen der Valenzbeschreibung. Ein Valenzmodell für das Engli- sche. In: Marek Konopka, Jacqueline Kubczak, Christian Mair, František Štícha, and Ulrich H. Waßner (eds.), Grammar & Corpora, Grammatik und Korpora 2009. Third International Conference, Dritte Internationale Konferenz, Mannheim, September 22–24 2009. Tübingen: Narr (working title). Herbst, Thomas, David Heath, Ian Roe, and Dieter Götz 2004 A Valency Dictionary of English. Berlin / New York: Mouton de Gruyter. [= VDE] Herbst, Thomas and Michael Klotz 2003 Lexikografie. Paderborn: Schöningh. Herbst, Thomas and Michael Klotz 2002 Meeting and kissing as valency problems - some remarks on the treatment of reciprocity and reflexivity in a valency description of English. In Ekkehard König (ed.), Reflexives and Intensifiers – the use of self-forms in English. Themenheft der Zeitschrift für Anglistik und Amerikanistik, ZAA 50:3, pp. 239–249. Herbst, Thomas and Michael Klotz forthc. Dependency Grammar. In: Artemis Alexiadou, Tibor Kiss, and Miriam Butt (eds.), Syntax II. In: Hugo Steger and Herbert Ernst Wiegand (eds.), Handbooks of Linguistics and Communication Sci- ence (HSK). Herbst, Thomas and Ian Roe 1996 How Obligatory Are Obligatory Complements? – An Alternative Approach to the Categorization of Subjects and Other Complements in Valency Grammar. English Studies 2, pp. 179–199. Herbst, Thomas and Susen Schüller [now Faulhaber] 2008 Introduction to Syntax. A Valency Approach. Tübingen: Narr. Heringer, Hans Jürgen 1967 Wertigkeiten und nullwertige Verben im Deutschen. Zeitschrift für Deutsche Sprache 23, pp. 13–34. Heringer, Hans Jürgen 1972 Deutsche Syntax. Berlin / New York: Walter de Gruyter. Second edition. First edition 1970. Heringer, Hans Jürgen 1984 Neues von der Verbszene. In Gerhard Stickel (ed.), Pragmatik in der Grammatik. Düsseldorf: Pädagogischer Verlag Schwann-Bagel, pp. 34–64. 340 References

Heringer, Hans Jürgen 1993 Dependency Syntax - Basic ideas and the classical model. In Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo Ven- nemann (eds.), Syntax. Ein internationales Handbuch zeit- genössischer Forschung = an internatinal handbook of contempo- rary research. Berlin / New York: Walter de Gruyter, pp. 298–316. Heringer, Hans Jürgen 1996 Deutsche Syntax Dependentiell. Tübingen: Stauffenburg. Hoey, Michael 2005 Lexical Priming. A new theory of words and language. London / New York: Routledge. Hopper, Paul J. and Elizabeth Closs Traugott 2003 Grammaticalization. Cambridge / New York: Cambridge University Press. Second edition. First edition 1993. Huddleston, Rodney and Geoffrey K. Pullum 2002 The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. [CamG] Hudson, Richard 1984 Word Grammar. Oxford: Blackwell. Hudson, Richard 1990 English Word Grammar. Oxford: Blackwell. Hudson, Richard, Andrew Rosta, Jasper Holmes, and Nikolas Gisborne 1996 Synonyms and syntax. Journal of Linguistics 32, pp. 439–446. Hunston, Susan and Gill Francis 2000 Pattern Grammar. A corpus-driven approach to the lexical grammar of English. Amsterdam / Philadelphia: Benjamins. Ickler, Theodor 1985 Valenz und Bedeutung. Beobachtungen zur Lexikographie des Deut- schen als Fremdsprache. In Henning Bergenholtz and Joachim Mugdan (eds.), Lexikographie und Grammatik. Akten des Essener Kolloquiums zur Grammatik im Wörterbuch 28.-30.6.1984. Tübingen: Niemeyer, pp. 358–377. Jackendoff, Ray S. 1972 Semantic Interpretation in Generative Grammar. Cambridge, MA / London: The MIT Press. Jackendoff, Ray S. 1985 Semantics and Cognition. Cambridge, MA / London: The MIT Press. Second Printing. First edition 1983. Jackendoff, Ray S. 1987 The status of thematic relations in linguistic theory. Linguistic In- quiry 18, pp. 369–411. References 341

Jackendoff, Ray S. 1991 Semantic Structures. Cambridge, MA: The MIT Press. Second print- ing. First edition 1990. Jacob, Daniel 2004 Generalisierte Semantische Rollen und freie Linkingprinzipien. In Rolf Kailuweit und Martin Hummel (eds.), Semantische Rollen. Tübingen: Narr, pp. 104–117. Jolly, Julia 1991 Prepositional Analysis within the Framework of Role and Reference Grammar. New York: Peter Lang. Jolly, Julia 1993 Preposition Assignment in English. In Robert D. Van Valin (ed.), Advances in Role and Reference Grammar. Amsterdam / Philadelphia: John Benjamins, pp. 275–310. Kabakþiev, Krasimir 2000 Aspect in English. Dordrecht: Kluwer Academic Publishers. Kay, Paul and Charles J. Fillmore 1999 Grammatical Constructions and Linguistic Generalizations: The What’s X doing Y? Construction. Language 75, pp. 1–33. Kidd, Evan, Elena Lieven, and Michael Tomasello 2006 Examining the role of lexical frequency in the acquisition and pro- cessing of sentential complements. Cognitive Development 21, pp. 93–107. Kiparsky, Paul and Carol Kiparsky 1970 Fact. In Manfred Bierwisch and Karl Erich Heidolph (eds.), Progress in Linguistics. A Collection of Papers. The Hague / Paris: Mouton, pp. 143–173. Kleinke, Sonja 2002 Englische Komplementstrukturen. Schematische und prototypische Bedeutungen. Frankfurt am Main: Peter Lang. Klotz, Michael 2000 Grammatik und Lexik. Studien zur Syntagmatik englischer Verben. Tübingen: Stauffenburg. Klotz, Michael 2007 Valency Rules? The case of verbs with propositional complements. In Thomas Herbst and Katrin Götz-Votteler (eds.), Valency. Theo- retical, Descriptive and Cognitive Issues, Berlin / New York: Mouton de Gruyter, pp. 117–128. Labov, William 1972 Sociolinguistic Patterns. Oxford: Basil Blackwell. 342 References

Lakoff, George 1987 Women, Fire and Dangerous Things. Chicago / London: Chicago University Press. Lamb, Sidney 2001 Learning syntax – a neurocognitive approach. In Martin Pütz, Susanne Niemeyer, and René Dirven (eds.), Applied Cognitive Linguistics I: Theory and Language Acquisition. Berlin / New York: Mouton de Gruyter, pp. 167–191. Langacker, Ronald W. 1987 Foundations of Cognitive Grammar. Vol 1: Theoretical Prerequi- sites. Stanford: Stanford University Press. Langacker, Ronald W. 1991 Foundations of Cognitive Grammar. Vol 2. Stanford: Stanford Uni- versity Press. Langacker, Ronald W. 1999 Grammar and Conceptualization. Berlin / New York: Mouton de Gruyter. Leech, Geoffrey N. 1971 Meaning and the English Verb. London: Longman. Leech, Geoffrey N. 1983 Principles of Pragmatics. London / New York: Longman. Leisi, Ernst 1971 Der Wortinhalt. Seine Struktur im Deutschen und Englischen. Heidelberg: Quelle & Meyer. Levin, Beth 1985 Lexical Semantics in Review: an Introduction. In Beth Levin (ed.), Lexical Semantics in Review. Lexicon Project Working Papers Num- ber 1, Cambridge, MA: Center for Cognitive Science Massachusetts Institute of Technology, pp. 1–62. Levin, Beth 1993 English Verb Classes and Alternations – A Preliminary Investiga- tion. Chicago / London: The University of Chicago Press. Levin, Beth and Malka Rappaport Hovav 2005 Argument Realization. Cambridge: Cambridge University Press. Löbner, Sebastian 2002 Understanding Semantics. London: Arnold. Lyons, John 1977 Semantics 1. Cambridge: Cambridge University Press. Lyons, John 1981 Language, Meaning & Context. Bungay: The Chaucer Press. References 343

MacWhinney, Brian 1989 Competition and Lexical Categorization. In R. Corrigan, F. Eckman, and M. Noonan (eds.), Linguistic categorization. New York: John Benjamins, pp. 195–241. MacWhinney, Brian 1998 Models of the Emergence of Language. Annual Review of Psycho- logy 49, pp. 119–227. MacWhinney, Brian 2001 Emergentist Aproaches to language. In Joan Bybee and Paul Hopper (eds.), Frequency and the Emergence of Structure. Amsterdam / Philadelphia: John Benjamins, pp. 449–470. Mair, Christian 1995 Changing patterns of complementation, and concomitant grammati- calization, of the verb help in present-day British English. In Bas Aarts and Charles F. Meyer (eds.), The verb in contemporary Eng- lish – Theory and description. Cambridge: Cambridge University Press, pp. 258–272. Mair, Christian 2003 Gerundial complements after begin and start: Grammatical and so- ciolinguistic factors, and how they work against each other. In Günter Rohdenburg and Britta Mondorf (eds.), Determinants of Grammatical Variation in English. Berlin / New York: Mouton de Gruyter, pp. 329–345. Marantz, Alec 1984 On the Nature of Grammatical Relations. Cambridge, MA / London: The MIT Press. Martinet, André 1981 Sprachökonomie und Lautwandel. Eine Abhandlung über die diach- ronische Phonologie. Aus dem Französischen von Claudia Fuchs. Stuttgart: Klett-Cotta. First edition 1955, original title: Économie des changements phonétiques. Matthews, Peter 1981 Syntax. Cambridge: Cambridge University Press. Mindt, Dieter 2000 An Empirical Grammar of the English Verb System. Berlin: Cornelsen. Mukherjee, Joybrato 2005 English Ditransitive Verbs – Aspects of Theory, Description and a Usage-based Model. Amsterdam / New York: Rodopi. Naigles, Letitia 1990 Children use syntax to learn verb meaning. Journal of Child Lan- guage 17:2, pp. 357–374. 344 References

Noël, Dirk 2003 Is there semantics in all syntax? The case of accusative and infinitive constructions vs. that-clause. In Günter Rohdenburg and Britta Mondorf (eds.), Determinants of Grammatical Variation in English. Berlin / New York: Mouton de Gruyter, pp. 347–377. Palmer, Frank 1981 Semantics. Cambridge: Cambridge University Press. Second edition. First edition 1979. Palmer, Frank 1986 Mood and Modality. Cambridge: Cambridge University Press. Palmer, Frank 1990 The Semantics of Grammar. (Review of Wierzbicka 1988). Journal of Linguistics 26, pp. 223–233. Pinker, Steven 1989 Learnability and Cognition: the Acquisition of Argument Structure. Cambridge, MA: The MIT Press. Pollard, Carl and Ivan A. Sag 1994 Head-driven . Chicago / London: Univer- sity of Chicago Press. Quirk, Randolph 1974 The Linguist and the English Language. London: Arnold. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik 1972 A Grammar of Contemporary English. London: Longman. [GCE] Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik 1985 A Comprehensive Grammar of the English Language. London / New York: Longman. [CGEL] Ramchand, Gillian Catriona 1998 Deconstructing the Lexicon. In Miriam Butt and Wilhelm Geuder (eds.), The Projection of Arguments. Lexical and Compositional Factors. Stanford: CSLI Publications, pp. 65–96. Ramchand, Gillian Catriona 2008 Verb Meaning and the Lexicon. A First-Phase Syntax. Cambridge: Cambridge University Press. Randow, Elise von 1986 Valente Substantive des Englischen. Tübingen: Narr. Rappaport, Malka and Beth Levin 1988 What to Do with Theta-Roles. In Wendy Wilkins (ed.), Syntax and Semantics, Volume 21: Thematic Relations. San Diego: Academic Press Inc., pp. 7–36. References 345

Rappaport Hovav, Malka and Beth Levin 1998 Building Verb Meanings. In Miriam Butt and Wilhelm Geuder (eds.), The Projection of Arguments. Lexical and Compositional Factors. Stanford: CSLI Publications, pp. 97–134. Rericha, Václav 1987 Notes on infinitival and –ing participle complements of the verbs begin and start. Philologica Pragensia 30:3, pp.129–133. Riddle, Elizabeth 1975 Some Pragmatic Conditions on Complementizer Choice. Papers from the Eleventh Regional Meeting of the Chicago Linguistic Soci- ety. Chicago: Chicago Linguistic Society, pp. 467–474. Ritter, Elizabeth and Sara T. Rosen 1998 Delimiting Events in Syntax. In William Geuder and Miriam Butt (eds.), The Projection of Arguments. Lexical and Syntactic Con- straints. Stanford: CSLI Publications, pp. 135–164. Rosch, Eleanor and Carolyn B. Mervis 1975 Family Resemblances: Studies in the internal structure of categories. Cognitive Psychology 8, pp. 382–439. Rohdenburg, Günter 2003 Cognitive complexity and horror aequi as factors determining the use of interrogative clause links in English. In Günter Rohdenburg and Britta Mondorf (eds.), Determinants of Grammatical Variation in English, Berlin / New York: Mouton de Gruyter, pp. 205–249. Rudanko, Juhani 1989 Complementation and Case Grammar. A Syntactic and Semantic Study of Selected Patterns of Complementation in Present-Day Eng- lish. Albany: State University of New York Press. Schlesinger, I. M. 1995 On the semantics of the object. In Bas Aarts and Chares F. Meyer (eds.), The verb in contemporary English. Theory and description. Cambridge: Cambridge University Press, pp. 54–74. Schmid, Hans-Jörg 1993 Cottage und Co., idea, start vs. begin: Die Kategorisierung als Grundprinzip einer differenzierten Bedeutungsbeschreibung. Tübingen: Niemeyer. Schumacher, Helmut 1986 Verben in Feldern. Valenzwörterbuch zur Syntax und Semantik deutscher Verben. Berlin / New York: de Gruyter. Schumacher, Helmut, Jacqueline Kubczak, Renate Schmidt, and Vera de Ruiter 2004 VALBU – Valenzwörterbuch deutscher Verben. Tübingen: Narr. 346 References

Sinclair, John 1991 Corpus, Concordance, Collocation. Oxford: Oxford University Press. Sinclair, John 2004 Trust the Text. Language, corpus and discourse. Abingdon, Oxon / New York: Routledge. Sinclair, John 2008 The phrase, the whole phrase, and nothing but the phrase. In Sylviane Granger and Fanny Meunier (eds.), Phraseology. An interdiscipli- nary perspective. Amsterdam / Philadelphia: John Benjamins, pp. 407–410. Slobin, Dan I. 1997 The Origins of Grammaticizable Notions: Beyond the Individual Mind. In Dan I. Slobin (ed.), The Crosslinguistic Study of Language Acquisition, vol 5, Expanding the Concepts. Mahwah, NJ: Lawrence Erlbaum Associates, pp. 265–323. Smith, Carlota S. 1991 The Parameter of Aspect. Dordrecht: Kluwer Academic Publishers. Smith, Neil 1999 Chomsky. Ideas and Ideals. Cambridge: Cambridge University Press. Somers, Harold L. 1984 On the validity of the complement-adjunct distinction in valency grammar. Linguistics 22, pp. 507–530. Sommerfeldt, Karl-Ernst, and Herbert Schreiber 1996 Wörterbuch der Valenz etymologisch verwandter Wörter. Tübingen: Niemeyer. Standop, Ewald 2000 Englische Verbkomplementation. Anglia 118, pp. 217–257. Starosta, Stanley 1988 The case for lexicase. An outline of lexicase grammatical theory. London / New York: Pinter (Open linguistics series). Stefanowitsch, Anatol 2006 Negative evidence and the raw frequency fallacy. Corpus Linguistics and Linguistic Theory 2:1, pp. 61–77. Stefanowitsch, Anatol 2008 Negative entrenchement: A usage-based approach to negative evi- dence. Cognitive Linguistics 19:3, pp. 513–531. Stefanowitsch, Anatol and Stefan Th. Gries 2003 Collostructions: Investigating the interaction of words and construc- tions. International Journal of Corpus Linguistics 8:2, pp. 209–243. References 347

Stubbs, Michael 2001 Words and Phrases: Corpus Studies of Lexical Semantics. Oxford / Malden, MA: Blackwell. Talmy, Leonard 2007 Lexicalization patterns: Semantic structure in lexical forms. In Timothy Shopen (ed.), Language typology and language description. Vol. 3, Grammatical categories and the lexicon. Cambridge: Cambridge University, pp. 66–168. Second edition. First edition 1985. Tarvainen, Kalevi 1985 Semantic Case in the Framework of Dependency Theory. Series A, Paper no. 145. Trier: Linguistic Agency University Trier. Tenny, Carol L. 1994 Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer Academic Publishers. Tesnière, Lucien 1959 Éléments de syntaxe structurale. Paris: Klincksieck. Tomasello, Michael 1998a The Return of Constructions. Journal of Child Language 25, pp. 431–442. Tomasello, Michael 1998b Introduction: A Cognitive-Functional Perspective on Language Structure. In Michael Tomasello (ed.), The New Psychology of Lan- guage. Cognitive and Functional Approaches to Language Structure. Mahwah, NJ / London: Erlbaum Associates, pp. vii–xxiii. Tomasello, Michael 2003 Constructing a Language. A usage-based theory of language acqui- sition. Cambridge, MA / London: Harvard University Press. Volume 2: Cognition, Perception, and Language. Tomasello, Michael 2006 Acquiring Linguistic Constructions. In Deanna Kuhn and Robert Siegler (eds.), Cognition, Perception, and Language. In: William Damon and Richard M. Lerner (eds.), Handbook of Child Psycho- logy. Sixth edition. Volume Two. New York: John Wiley and Sons, pp. 255–298. Van Valin, Robert D. 2004 Semantic Macro-Roles in Role and Reference Grammar. In Rolf Kailuweit and Martin Hummel (eds.), Semantische Rollen. Tübingen: Narr, pp. 62–82. Van Valin, Robert D. and Randy J. Lapolla 1997 Syntax. Structure, Meaning and Function. Cambridge: Cambridge University Press. 348 References

Vater, Heinz 1978 Probleme der Verbvalenz. Kölner Linguistische Arbeiten zur Germa- nistik 1. Trier: L.A.U.T. Vendler, Zeno 1967 Linguistics in Philosophy. Ithaca: Cornell University Press. Vennemann, Theo 1977 Konstituenz and Dependenz in einigen neueren Grammatiktheorien. Sprachwissenschaft 2, pp. 259–301. Verspoor, Marjolijn H. 1990 Semantic Criteria in English Complement Selection. Leiden: Rijks- universiteit te Leiden. Vosberg, Uwe 2003 The Role of Extractions and Horror Aequi in the Evolution of -ing Complements in Modern English. In Günter Rohdenburg and Britta Mondorf (eds.), Determinants of Grammatical Variation in English. Berlin / New York: Mouton de Gruyter, pp. 305–327. Wasow, Thomas 2002 Postverbal Behavior. Stanford, CA: CSLI Publications. Wasow, Thomas and Jennifer Arnold 2003 Post-verbal constituent ordering in English. In Günter Rohdenburg and Britta Mondorf (eds.), Determinants of Grammatical Variation in English, Berlin / New York: Mouton de Gruyter, pp. 119–154. Welke, Klaus 1988 Einführung in die Valenz- und Kasustheorie. Leipzig: VEB Biblio- graphisches Institut Leipzig. Welke, Klaus 2005 Deutsche Syntax funktional. Perspektiviertheit syntaktischer Struk- turen. Tübingen: Stauffenburg. Second revised edition. First edition 2002. Wierzbicka, Anna 1988 The Semantics of Grammar. Amsterdam / Philadelphia: John Benja- mins. Wierzbicka, Anna 1996 Semantics. Primes and Universals. Oxford / New York: Oxford University Press. Wulff, Stefanie 2008 Das Prinzip der nicht-Synonymität: V1-and-V2 und V1-V2 im Eng- lischen. In Anatol Stefanowitsch and Kerstin Fischer (eds.), Kon- struktionsgrammatik II. Von der Konstruktion zur Grammatik. Tübingen: Stauffenburg, pp. 189–201. References 349

Wulff, Stefanie, Anatol Stefanowitsch, and Stefan Th. Gries 2007 Brutal Brits and persuasive Americans. Variety-specific meaning construction in the into-causative. In Günter Radden, Klaus-Michael Köpke, Thomas Berg, and Peter Siemund (eds.), Aspects of Meaning Construction. Amsterdam / Philadelphia: John Benjamins, pp. 265– 281. Zipf, George K. 1949 Human Behavior and the Principle of Least Effort. Cambridge, MA: Addison-Wesley Press.

Dictionaries Collins COBUILD English Guides, 1: Prepositions 1991 edited by John Sinclair et al. London: Harper Collins. Collins English Dictionary 2006 edited by Sandra Anderson et al., Glasgow: Harper Collins Publish- ers. Eighth edition. Longman Dictionary of Contemporary English 2005 edited by Della Summers. Harlow: Longman. Fourth edition. [LDOCE4] Longman Language Activator 1993 edited by Della Summers. Harlow: Longman. Oxford Learner’s Thesaurus. A dictionary of synonyms. 2008 edited by Diana Lea, Jennifer Bradberry, Richard Poole, and Helen Warren. Oxford / New York: Oxford University Press. Oxford Advanced Learner’s Dictionary 2005 edited by Sally Wehmeier, Colin McIntosh, Joanna Turnbull, and Michael Ashby. Oxford / New York: Oxford University Press. Sev- enth edition. [OALD7] The Oxford Dictionary of Synonymy and Antonyms 2007 Oxford / New York: Oxford University Press. Second edition. The Oxford English Dictionary 1989 edited by James A. H. Murray, Henry Bradley, W. A. Craigie, and C. T. Onions. Second edition prepared by John A. Simpson and Ed- mund S. C. Weiner. 20 vols. Oxford: Clarendon Press. [OED]

Internet Sources http://framenet.icsi.berkeley.edu/ http://www.patternbank.uni-erlangen.de http://wordnetweb.princeton.edu/perl/webwn 350 References

Corpora British National Corpus: Examples of usage taken from the British National Cor- pus (BNC) were obtained under the terms of the BNC End User Licence. Copy- right in the individual texts cited resides with the original IPR holders. For infor- mation and licensing conditions relating to the BNC, please see the web site at http://www.natcorp.ox.ac.uk Index

Aarts, Flor 132 Behrend, Douglas 308 Aarts, Jan 132 Behrens, Heike 123 Abbot-Smith, Kirsten 246 Biber, Douglas 199 abduction 278, 326 blocking 45, 53, 64–65, 67, 126, Ackerman, Farrell 13 138–140, 190, 303 adjunct 4, 6, 8, 24, 31, 46–47, 56, Boas, Hans Christian 212, 233, 257–259 243–245 Ágel 3, 223 Bock, J. Kathryn 192 Aktionsart 17, 223 (see also Bolinger, Dwight 172 situation type) Bondzio, Wilhelm 237 Algeo, John 133, 159, 163, 186 Bouchard, Denis 237–238 Allerton, David 3, 5, 8–9 Bresnan, Joan 8, 149, 154, 236, 238, alternation 5, 65–66, 90, 93, 153, 293 191–192, 203–204, 210, 213, Brinton, Laurel J. 225, 227, 231, 224, 228, 237, 252, 279–282, 234 304, 310 Brown, Gillian 255 alternative realization 20–21, 23, Bühler, Karl 30 29–30, 32–67, 73–74, 77–78, Busse, Winfried 3 80, 151–153, 266, 281, 309– Bußmann, Hadumod 87 311, 313 Butt, Miriam 223, 233 analogy 118–119, 143, 206, 246– Bybee, Joan 123, 192, 247–248, 247, 284, 287, 326 315–316 argument 11–13, 70, 73–75, 154, 209, 236 Cambridge Grammar of the English direct/indirect 70, 153 Language (also CamG) 24, 188 external/internal 224 Campbell, Aimee L. 240 realization 233, 235, 237–238, case 280 morphological 5, 12, 153, 305, selection 14, 224, 234 322 argument structure construction (see semantic 11–12, 29, 74, 81, 85, construction) 154, 233, 322 Arnold, Jennifer 149, 158, 192, 311 categorial selection (c-selection) 73– aspect 14, 17, 114, 121, 223–225, 74 227–229, 233–235, 238–239, Chomsky, Noam 12, 67, 73–75, 79, 299–300, 302, 317, 325 (see also 91, 149, 213, 224, 236–237, situation type) 292, 296, 310, 346 clausal role 128, 130–131, 145, 151, Baker, Collin F. 242 156, 251–252, 255, 311 Baker, Mark C. 264 cognitive grammar 3, 67, 76 Barsalou, Lawrence W. 242 cognitive linguistics 7, 76, 173 352 Index colligation 43, 141, 245 conventionalization 89, 91, 117, collocation 9, 43, 66, 78, 84, 87, 152, 159, 243–246, 295, 307, 141, 245, 313, 315 312–315, 322, 326 collostruction 78, 91, 160, 192 Coseriu, Eugenio 2 commutation 5, 63 Croft, William 3, 11, 13, 15–16, 37, complement 4–6, 8, 11, 17, 31, 47, 71, 85, 151–152, 210–211, 238, 257–259 241–242, 293, 296, 307–308, clausal 29, 48, 68, 71, 138, 334 154–155, 170–171, 174, Cruse, D. Alan 3–4, 11, 15–16, 326 180, 188, 199, 242, 295, Cuyckens, Hubert 24 297, 300, 304 inventory 7, 20, 22, 127, 254, Dahl, Östen 223 263, 319, 325 dependency 3–4 polysemous 37, 49, 58, 62, 68, dialect difference 21, 66, 159, 163, 80, 250 186, 203, 205, 245 synonymous 37, 61, 63, 68, 78 Dik, Simon C. 227, 234 Comprehensive Grammar of the Dirven, René 176 English Language (also CGEL) ditransitive construction (see 7–8, 18, 25, 27–28, 71, 132, construction) 163, 165, 167, 171, 177–178, Dixon, Robert M. W. 95, 115, 172, 182, 184, 187–188, 214, 220, 174, 176, 183, 185, 255, 294– 225, 227, 268, 272 295 Comrie, Bernard 225, 227 Dowty, David 14, 70, 80, 149, 238, Conrad, Susan 199 252, 255 constructeme 128, 320, 323–324 Dubost, Jean-Pierre 3 construction 6, 15, 68, 76, 78, 119, Dürscheid, Christa 85, 305 212, 239–241, 263, 265, 276, 278, 294, 296, 308, 323–325 Egan, Thomas 177 argument structure 15, 83, 122, Eichinger, Ludwig M. 4, 333 207, 211, 240, 264, 274, Emons, Rudolf 3, 5, 70, 149, 155 278, 308, 324 Engel, Ulrich 3, 5–6, 333 caused motion 151, 240, 274– entrenchment 91, 117, 150, 258, 275, 308, 310 315 ditransitive 90, 192, 207, 240, Erlangen Valency Patternbank 6, 8, 247, 250, 267–268, 270– 22, 122, 128, 138, 267–268, 271, 275, 307–308, 310 272, 279, 298, 314, 316, 320 grammar 10, 15, 75, 79, 120, 138, 151, 207, 211, 239– Fabricius-Hansen, Cathrine 297 241, 264–267, 272, 276, Fanego, Teresa 173, 205 293, 298, 308, 310, 323, Faulhaber, Susen 132, 268 325 Fillmore, Charles J. 3, 8, 12–13, 15– predicative 160, 270–271 16, 19, 21, 81, 154, 209, 211, 213, 245, 252, 254, 278, 293, 301, 310, 326 Index 353

Fischer, Klaus 5 Habermann, Mechthild 3, 246, 322 frame 8, 13–15, 67, 85, 141, 154, Haegeman, Liliane 83, 188 209, 211–212, 323, 238 Halliday, Michael A. K. 11–12, 150, FrameNet 8, 13, 56, 59, 63, 209, 241, 255 270 Hamawand, Zeki 68, 76, 173–174 Francis, Gill 7, 22, 83, 90–91, 128, Hampe, Beate 78, 160 152, 198, 235–236, 241, 247, Hawkins, John 192 279, 282–284, 287–289, 291, Heath, David 3 293, 296, 299, 308, 335 Helbig, Gerhard 3–5, 8, 11–12, 29, Frawley, William 19, 254 141, 152, 155, 213, 256–257, frequency 7, 17, 35, 40, 55, 73, 75, 298, 305, 337 171–172, 177, 196–200, 205– Herbst, Thomas 3–6, 8–10, 15–17, 206, 247, 258, 311–313, 316 19–20, 22, 24–25, 29, 35, 70, token 206, 316 82, 84, 103, 123, 128, 131–132, type 206, 267, 272, 316 138, 145, 151, 153, 156, 163, Fried, Mirjam 209, 211, 323, 325 170, 172, 203, 251–252, 254– 255, 260, 263–264, 268, 279, Gärtner, Eberhard 3 297–298, 301, 311, 320 Geuder, Wilhelm 223, 233 Heringer, Hans Jürgen 3–4, 25, 31, Gillette, Jane 294–295, 308 70, 141, 153, 155, 223, 254 Gisborne, Nikolas 90, 223, 242, Hoey, Michael 140–141, 197, 245– 335–336 246 Givón, Talmy 19 Holmes, Jasper 90, 242 Goldberg, Adele 13, 15, 34, 37, 68, Hopper, Paul J. 21, 78, 152, 265, 76, 83, 119–120, 122, 138, 141– 278, 306, 322 142, 151, 157, 207, 211–213, horror aequi 34, 192, 311 240–241, 247–248, 250–252, HPSG 293 263–267, 273–276, 278, 289, Huddleston, Rodney 4, 24 293–294, 296, 298, 301, 303, Hudson, Richard 3, 90, 242–243, 306, 308–310, 337 293 Götz, Dieter 3 Hunston, Susan 7, 22, 83, 90–91, Götz-Votteler, Katrin 17 128, 152, 198, 235–236, 241, gradience 25, 36, 56, 173, 248, 259 247, 279, 282–284, 287–289, grammaticalization 24, 77, 151, 291, 296, 299, 308 153, 265, 307, 316 Greenbaum, Sidney 7, 20, 281 Ickler, Theodor 223 Gries, Stefan Th. 34, 78, 90, 151, idiom 66, 313 158, 160, 192, 241, 245, 310– idiom principle 7, 17, 314 311 idiomatic 35, 116, 135, 143, 217, Groefsema, Marjolein 150, 156, 245, 271–272, 276–277, 287 158, 213 idiomaticity 293, 313 Gropen, Jess 157, 207, 279, 296 idiosyncrasy 91, 152, 161, 251, 272, Gruber, Jeffrey S. 12 280, 282, 295–299, 313, 325 354 Index idiosyncratic 2, 60, 80, 88, 100, 108, Lexical Functional Grammar 8, 154, 120, 123, 139, 153–155, 197, 236, 293 209, 223, 239, 242, 256, 282, lexicon 73, 75, 80, 128, 204, 225, 292, 295, 298–299, 325 236, 242, 247, 279, 281, 292, information structure 151–152, 192, 315, 325 241, 255, 309, 319 Lieven, Elena 246–247 linking 121, 129, 154, 196, 209, Jackendoff, Ray S. 154, 234, 251, 211, 223–224, 238, 242, 246, 254, 256, 258 294, 296–297, 300, 326 Jacob, Daniel 246 linking rules 75, 83, 149–150, 154, Johansson, Stig 199 224, 279, 296 Jolly, Julia 153–154 Löbner, Sebastian 212–213 Lyons, John 87, 256 Kabakþiev, Krasimir 38, 234 Kay, Paul 15–16, 211, 245, 293 MacWhinney, Brian 246–248 Kidd, Evan 246–247 Mair, Christian 203–205 Kiparsky, Paul and Carol 171, 177 Manning, Elizabeth 7, 22, 83, 279, Kleinke, Sonja 171–172, 176–177 282–283, 287–289, 291 Klotz, Michael 3–4, 9, 74, 77, 84, Marantz, Alec 29, 70, 153 123, 153, 203–204, 254, 312, Martinet, André 309 315 Matthews, Peter 188 merger 31, 188, 254, 256, 263–264 Labov, William 20 Mervis, Carolyn B. 14 Lakoff, George 24, 273, 293 metaphor 160, 213, 250, 275, 303, Lamb, Sidney 17, 248 306, 315, 323 Langacker, Ronald W. 67, 141, 150, Mindt, Dieter 171, 199, 203 173, 241–242, 248, 273, 292, Moore, John 13 294–295, 301 motivation 90, 245, 247, 264, 266, language acquisition 7, 213, 246– 273, 275–276, 278, 283, 305– 248, 313 306 language change 21, 151, 173, 193, Mukherjee, Joybrato 2, 16, 207, 205, 246–247, 265, 307 224, 293 Lapolla, Randy J. 225, 227–228, 347 Naigles, Letitia 308 Leech, Geoffrey N. 7, 199, 235, 261 native speaker assessment 8, 11, Leisi, Ernst 256 20–22, 38, 55, 59–62, 66–67, Levin, Beth 12, 22, 75, 83, 90, 93, 79, 88–89, 91–92, 95, 102, 115, 153, 210, 234–239, 241, 252, 117–120, 130–132, 158–159, 279–283, 295–296, 299–300 163, 165, 167, 169, 178, 184, lexical co-text 20, 36, 43, 77, 130, 186–187, 189–190, 192, 195, 140, 312–313, 315 220, 222, 245, 247, 249, 263, lexical filling 21, 55, 122, 132, 199– 268, 270, 281, 300–302, 312– 200, 301, 313 313, 326 Noël, Dirk 1–2, 174–176, 296 Index 355

O’Connor, Mary Catherine 15–16, 234, 238, 254, 273, 278, 245, 293 307, 318 object 5–6, 68, 70, 132, 156, 160, particle 8, 24–26, 29, 65, 68, 70, 171, 173, 191–192, 209–210, 72–74, 77, 152–154, 170, 183, 224, 235, 238, 255, 268, 295, 234, 258, 277, 307, 311 297–298, 311 complement 8, 28–29, 36, 45, objoid 5 68, 70–72, 153–155, 183, obligatoriness 4, 141, 145, 197 242, 272, 275, 297, 304, obligatory 310–312 complement 5, 9–10, 30, 141– phrase 18, 24–25, 68, 70, 72, 142, 194, 254, 258, 303 93, 153, 234, 257–258, participant 132, 141–142, 211, 260, 272, 274, 297, 304, 240, 300 307, 311 valency slot 301, 303 passive 5, 7–11, 47, 133, 175, 209, oblique 122, 154, 210, 238, 274– 250, 267–270, 278, 301, 323 275, 297 passivizability 8, 268, 277 optional pattern inventory 20, 22, 89, 91, complement 5, 8–9, 30, 142, 168, 205, 209, 246–247, 256, 254, 257 319–320, 322 participant 142–144, 301 Pattern Grammar 7, 90, 152, 279, valency slot 143 282–284, 288, 291 optionality 8–9, 100, 108, 141–145, perject 5 181 Pinker, Steven 67, 83, 150, 154, Östman, Jan-Ola 209, 211, 323, 325 213, 224, 237, 247, 296, 300 Oxford English Dictionary (OED), Pollard, Carl 293 26–27, 58, 307 polysemy 25, 37, 49–58, 62, 68, 71, 80, 141, 152, 242, 250, 265, Palmer, Frank 174, 205, 252 272–273, 275, 306–307, 317, participant 11, 13, 17, 20, 30, 70, 84 predicate complement unit (also inventory 25, 84–86, 92, 99, PCU) 10–11 114–115, 118, 120–121, predicator (see also valency carrier) 129, 141, 157, 191, 209– 9 210, 218, 229, 232, 234, preemption 173, 186, 303, 315–316, 256, 263, 276, 299–302, 326 317, 319, 323–325 preposition (see also particle) 4–5, pattern 15–16, 72, 79, 117, 23–25, 29, 153–154 124–133, 141, 150–151, prepositional 170, 210, 222, 235, 251, phrase 70, 235, 322 256, 260–261, 264–266, verb 8, 24, 163, 282 268, 271–274, 299–300, priming 34, 79, 140, 192, 197, 245– 309, 318, 320, 324–325 246, 311 role 8, 13, 16–17, 30, 59, 131, principle of maximized motivation 150, 156, 211–214, 223, 264–266, 306 356 Index principle of no synonymy 37, 151, Schüller, Susen (see also Faulhaber, 264–265, 274–275 Susen), 3–4, 6, 8–10, 15–17, 19– processing 7, 51, 247, 260, 265, 20, 24, 29, 35, 84, 131, 145, 293, 297, 311–313, 315, 318, 151, 156, 170, 251–252, 255, 323 260, 264, 268, 297–298, 311 profiled roles 141, 211, 240 Schumacher, Helmut 3, 5–6, 83, profiling 141, 211, 238, 240, 301 298 projection 79–81, 211, 224, 236, selection restrictions 12, 114, 121, 240, 323 212–223, 229, 239, 243, 296, projection principle 73, 75, 236, 310 299–300, 317, 324–325 projectionist approaches 75, 79, semantic determinism 125, 173, 235–237, 241, 303, 307 223, 279, 281, 283, 287, 296, proto-role 14, 238 299, 322 Pullum, Geoffrey K. 4, 24 semantic restrictions 114, 121, 212, 221, 300 Quirk, Randolph 7, 67, 171, 203, semantic role 12–15, 17, 19–20, 30– 281 31, 76, 145, 209–211, 223, 236– 237, 240, 258, 298 Ramchand, Catriona 79, 149, 200, semantic role hierarchy 19, 76, 211, 225, 237–239 238, 296 Randow, Elise von 3 semantic specialization 21, 23, 29, Rappaport Hovav, Malka 12, 75, 32, 36–37, 47–48, 55, 61, 77, 234–238, 295, 300 80, 304–305, 309, 312–313, 326 Reppen, Randi 199 sentence type 9–11, 16 Rericha, Václav 203 Sinclair, John 4, 7, 17, 235, 293, Riddle, Elizabeth 176 308, 314, 349 Ritter, Elizabeth 233 situation type 65, 223–234, 299–300 Roe, Ian 3, 10 Slobin, Dan I. 242 Rohdenburg, Günter 24, 34, 153, Smith, Carlota 225 192, 311 Smith, Neil 13 Rosch, Eleanor 14 Somers, Harold 257 Rosen, Sara T. 233 Sommerfeldt, Karl-Ernst 3 Rosta, Andrew 90, 242 Standop, Ewald 188 Rudanko, Juhani 88, 173, 176, 178, Starosta, Stanley 8 181, 214 Stefanowitsch, Anatol 78, 90–91, 117, 151, 158, 160, 241, 245 Sag, Ivan A. 293 storage 35, 53, 60, 75, 78, 84, 123, Schenkel, Wolfgang 3, 5, 8, 141, 150, 153, 192, 204, 206, 210, 155, 298 242, 248, 278, 282, 296, 313, Schlesinger, I. M. 306, 310–312 315–316, 318–319, 322, 325– Schmid, Hans-Jörg 203 326 Schönefeld, Doris 78, 160 Stubbs, Michael 315 Schreiber, Herbert 3 subcategorization 67, 154, 198, 213, 236 Index 357 subject 5, 7, 9–11, 19, 57, 59, 68, 271, 274–275, 277–278, 156, 173–175, 181, 209–210, 318, 320–321, 325 214, 224, 238, 252, 255, 268– qualitative 5, 12, 30, 204 270, 275, 295, 297 quantitative 5, 12, 30 subject complement unit (also SCU) semantic 11–12, 29, 152, 258, 10–11, 220 298 subtype 14, 30, 50–51, 70–71, 78, slot 11, 21, 23, 29–32, 35, 49, 80, 128–129, 151, 170, 179, 51, 64, 72, 78, 80, 318–319 250, 306, 310–311, 318–319 structure 23, 29–31, 34, 38, 45– Svartvik, Jan 7 46, 48, 51, 59–60, 62–63, syntactic gap 2, 21, 117, 121, 173, 79, 82, 250, 253, 318–319 242, 302, 309 syntactic 11, 29, 140, 152, 298 syntax-semantics interface 2, 292– theory 1, 3–4, 7–8, 70, 150, 298 153–154, 237, 242, 259, 293, 297–298 Talmy, Leonard 213 Valency Dictionary of English (also Tarvainen, Kalevi 14, 85, 311 VDE) 3, 5, 8–9, 17, 20, 23, 25, Tenny, Carol L. 14, 70, 153, 224, 32, 86–87, 154, 267, 298 296, 300 van der Elst, Gaston 3 Tesnière, Lucien 3–4 Van Valin, Robert D. 68, 128, 225, thematic role 13, 85, 153, 238–239 227–228, 306 theta role 12, 29, 224, 236–237, 296 Vater, Heinz 4 theta-criterion 75, 254 Vendler, Zeno 225, 227 Tomasello, Michael 7, 173, 186, Vennemann, Theo 3 240, 246–247, 275, 292 verb-specific role 13–14, 20, 85 Traugott, Elizabeth Closs 21, 78, Verspoor, Marjolijn H. 71, 174– 152, 265, 278, 306, 322 177, 183, 305 Vosberg, Uwe 192–193 Universal Grammar 74 usage-based 123, 177, 196, 248, 282 Wasow, Thomas 149–150, 158, 191–192, 197, 204, 311 valency 3, 74 Welke, Klaus 3, 8, 70, 154, 209, carrier 2, 4–6, 10, 24, 35, 82, 223–225, 227, 237, 257–258, 199, 203, 223, 248, 275, 297–298, 322 297 Wierzbicka, Anna 71, 175–176, construction 15–16, 123, 126, 205, 305, 344 128, 251, 264, 274, 278, Word Grammar 3, 293 309, 318, 320, 324–325 Wulff, Stefanie 151, 245 pattern 6–8, 11, 15–16, 83, 120, 124–128, 131, 237, 248, Yule, George 255 251, 256, 264, 266–267, Zipf, George K. 309