Serial Verb Constructions: Argument Structural Uniformity and Event Structural Diversity

Home , Adjunct (grammar), Antisymmetry, Argument (linguistics), Branching (linguistics), Feature (linguistics), Inflection

SERIAL VERB CONSTRUCTIONS: ARGUMENT STRUCTURAL UNIFORMITY AND EVENT STRUCTURAL DIVERSITY

A DISSERTATION SUBMITTED TO THE DEPARTMENT OF LINGUISTICS AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

Melanie Owens November 2011

This work is licensed under a Creative Commons Attribution- Noncommercial 3.0 United States License. http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/db406jt2949

ii I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Beth Levin, Primary Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Joan Bresnan

I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Vera Gribanov

Approved for the Stanford University Committee on Graduate Studies. Patricia J. Gumport, Vice Provost Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file in University Archives.

iii Abstract

Serial Verb Constructions (SVCs) are constructions which contain two or more verbs yet behave in every grammatical respect as if they contain only one. This observation guides my approach to SVCs in this dissertation, in accord with which SVCs are defined through displaying (i) no markers of subordination or coordination, (ii) uniform tense, aspect and/or mood values, (iii) the phonological properties of a clause headed by a single verb, and (iv) any other properties that signify monoclausal status in the language concerned. But while (i)-(iv) are standard defining properties for SVCs, I propose a further new defining property: the Theta Structure Property for SVCs, which requires that a SVC display monoclausal argument structure. The Theta Structure Property – so called because it holds at the argument structure level of Theta Structure in Lexical Decompositional Grammar (LDG) – requires that a SVC display the same number of structural arguments and the same list of configurationally-defined non-structural (oblique) arguments as some clause headed by a single verb (in some language). But while some SVCs, called ‘Possible Verb’ SVCs, satisfy this property straightforwardly by virtue of the fact that they have exactly the semantic content of a clause headed by a single verb, there is an additional, non-core class of ‘Impossible Verb’ SVCs for whom satisfaction of the Theta Structure Property is more complex. The basis for the Possible Verb-Impossible Verb distinction are LDG’s Possible Verbs constraints, which determine the semantic content of what can be expressed in a clause headed by a single verb. Possible Verb SVCs satisfy these constraints, but Impossible Verb SVCs, in not satisfying these constraints, have more complex semantic contents and argument structures. Special predicate combining operations

iv (mainly consisting of argument coidentification operations) must apply in order for an Impossible Verb SVC to satisfy the Theta Structure Property for SVCs; these operations deliver (monoclausal) argument structural uniformity across SVCs, although SVCs still display considerable event structural diversity. Event structural analysis features prominently in this dissertation, primarily for the purpose of demarcating the boundaries of Possible Verb and Impossible Verb SVCs. In some instances these two types of SVC can be difficult to distinguish, as the case studies of instrumental and motion SVCs of Chapters 4 and 5 demonstrate. But if certain instrumental and motion SVCs present some rather subtle cases of Impossible Verb serialization, there are Impossible Verb SVCs outside these domains, such as (A), which more blatantly lack the semantic content or event structure of a clause headed by a single verb: ` (A) Ozó kpèé èmà dé. . Ozo beat drum buy ‘Ozo beat the drum (and then) bought it.’ (Edo; Stewart 2001, 49)

Why are SVCs like (A), whose content could never be expressed in a clause headed by a single verb, behaving as if they contain only one verb? One of the most interesting answers to this question is that multi-event concepts are expressed in SVC form in those instances where they describe ‘recognizable event types’ – a combination of actions that cooccur so often as to be conceptualized as a single event. However, Stewart (2001) gives no indication that (A) expresses a recognizable event type. Are SVCs always constrained by some conceptualization of eventhood (recognizable event type or otherwise)? Are SVCs in some languages completely event structurally unconstrained? And does (A) really behave in every grammatical respect as if it contains only one verb? These questions are addressed in the concluding chapter, drawing on the cross-linguistically expanded survey of SVCs built up in the preceding chapters.

v Acknowledgements

Although I don’t think I’ve broken any records, my dissertation completion timeline has definitely been of the extended variety, and so I’d like to thank my advisor Beth Levin for never giving up hope that I’d finish, and for her helpful advice and input right up until the very end. Beth is extremely dedicated to all her students, and in my case one manifestation of this was her effort to meet up with me several times in England, after I’d moved away from Stanford in 2005. I also thank Joan Bresnan for unquestioningly turning up at my dissertation defense in June, as if five years had never elapsed, and for her positive, encouraging and often funny outlook on linguistics and life in general. The other key faculty member who made the completion of my PhD possible this year was Vera Gribanova, who I’m so grateful to for joining my committee for the defense this summer, and for moreover engaging with what I’d written and asking helpful and thought-provoking questions. Many thanks also to Noah Goodman as the committee chair, to Tom Wasow for stepping in as the fourth reader, and to Arnold Zwicky for his help and insight in the early stages of my dissertation. Unfortunately it feels like years since I’ve been anywhere near Indonesia or things Indonesian, but I certainly owe a lot to my Bima friends who helped with the Bima components of my dissertation. First and foremost there is Yanti in New Zealand, who ignited my interest in her native language, and I also thank Mike Dukes for finding Yanti in the first place, and running a great field methods class around her at University of Canterbury. Many thanks also to all my Bima friends in Bima and Mataram who looked after me so well when I was there. When I think of my time at Stanford, what made it so enjoyable was of course

vi the company of my fellow students, and for that I especially thank John, Itamar and Na’ama, Ela, Ivan, Andrew and Melissa, Judith, Lev, Philip, Lis, Jeanette, and Susanne. A huge thanks also to Dave SC, for being a kind of honorary linguistics student, but more importantly a great friend, and someone who put me up every time I came back to the Bay area. My ﬁnal and most heartfelt thanks are naturally reserved for my family; for Mum, Dad, Sam and Jackie in New Zealand, for Philip and Ruth, who were instrumental in helping me complete the dissertation by way of providing childcare and support, and for Paddy, Ele, Alison, and Dorothy. Special thanks also to Lawrence, for being such a good little boy, and to Meredith for being a very good baby since her arrival seven weeks ago. But more than anyone else I want to thank my wonderful husband David, who has supported me, and my dissertation, in every way possible over the last few years. Over the last few weeks he has taken charge of my disastrously incomplete bibliography, which, although a major undertaking in itself, only forms a small part of everything he has done for me.

vii Contents

Abstract iv

Acknowledgements vi

1 Introduction 1 1.1 Introduction...... 1 1.2 PreviousworkonSVCs...... 8 1.2.1 DeﬁnitionsofSVCs...... 11 1.2.2 ArgumentstructureofSVCs...... 16 1.2.3 SyntaxofSVCs ...... 22 1.2.4 EventstructureofSVCs ...... 26 1.2.5 Conclusion...... 38 1.3 Prospectus...... 39 1.3.1 The Theta Structure Property for SVCs ...... 39 1.3.2 The Possible Verb-Impossible Verb distinction in SVCs.... 43 1.3.3 In search of Impossible Verb SVCs, and their limits ...... 46

2 The Theta Structure Property for SVCs 51 2.1 Introduction...... 51 2.2 LexicalDecompositionGrammar ...... 53 2.3 TheThetaStructurePropertyforSVCs ...... 66 2.3.1 TheProperty ...... 66 2.3.2 Structural arguments and the Theta Structure Property for SVCs 68

viii 2.3.3 Non-structural arguments and the Theta Structure Pro-perty forSVCs...... 71 2.4 Projecting the arguments of contiguous and non-contiguous SVCs . . 77 2.4.1 Non-contiguousSVCs...... 80 2.4.2 ContiguousSVCs ...... 93 2.4.3 Conclusion...... 100 2.5 Conclusion...... 103

3 Possible Verb and Impossible Verb SVCs 109 3.1 Introduction...... 109 3.2 PossibleVerbSVCs...... 111 3.2.1 Satisfying the Possible Verbs constraints ...... 111 3.2.2 Serialization lexical rules and Possible Verb SVCs ...... 115 3.2.3 Event structures for Possible Verb SVCs ...... 124 3.2.4 Conclusion...... 130 3.3 ImpossibleVerbSVCs ...... 131 3.3.1 Event structures for Impossible Verb SVCs ...... 132 3.3.2 Predicate Combining Operations for Impossible Verb SVCs . . 136 3.4 Arguments for including Impossible Verb SVCs in the classofSVCs . 147 3.5 Consequences of including Impossible Verb SVCs in the class of SVCs 151 3.5.1 SVCs are not just doing the things that would be done with clauses headed by single verbs in other languages ...... 151 3.5.2 SVCs do not just describe single (non-sequential) events . . . 154 3.5.3 Only by including Impossible Verb SVCs can we see how SVCs can be constrained by the notion of ‘recognizable event type’ . 157 3.5.4 The Possible Verb-Impossible Verb distinction is superior to other two-way distinctions in SVCs ...... 159 3.6 Conclusion...... 163

4 Instrumental SVCs 166 4.1 Introduction...... 166 4.2 Instrumental events and instrumental SVCs ...... 170

ix 4.2.1 Instrumentalevents...... 171 4.2.2 InstrumentalSVCs ...... 181 4.3 Possible Verb instrumental SVCs ...... 186 4.3.1 Introduction...... 186 4.3.2 Instrumental‘use’SVCs ...... 187 4.3.3 Possible Verb instrumental ‘take’ SVCs ...... 203 4.3.4 Conclusion...... 208 4.4 Impossible Verb instrumental SVCs ...... 209 4.5 Conclusion...... 222

5 Motion SVCs 226 5.1 Introduction...... 226 5.2 The semantic composition of Possible Verb and Impossible Verb motion events ...... 229 5.2.1 Threetypesofmotioninformation ...... 230 5.2.2 Constraints on Possible Verb motion events ...... 236 5.2.3 Event structural proﬁles of Possible Verb and Impossible Verb motionSVCs ...... 261 5.2.4 Conclusion, and Nigerian Pidgin English example case .... 264 5.3 IjomotionSVCs ...... 265 5.4 ThaimotionSVCs ...... 273 5.5 BimamotionSVCs ...... 282 5.5.1 Introduction...... 282 5.5.2 The Bima motion SVC template, and SVCs that conform to theBimamotionSVCtemplate ...... 283 5.5.3 Multi-templatic Bima motion SVCs ...... 287 5.5.4 Bima motion SVCs which do not ﬁt the template ...... 291 5.6 Conclusion...... 293

6 Conclusion 296 6.1 The Theta Structure Property for SVCs: taking SVC deﬁnition below thesurface...... 296

x 6.2 SVCs: substitutes for clauses headed by single verbs, and what they tell us when they’re not substituting clauses headed by single verbs . 303

Bibliography 313

xi Chapter 1

Introduction

1.1 Introduction

Serialization is a topic which has been the subject of a rapidly expanding literature over the last 40-50 years, both in terms of its cross-linguistic coverage as well as its sheer volume. But in spite of this (or perhaps because of it), the most prominent characteristic of this literature is its failure to get to grips with what exactly a Serial Verb Construction (SVC) is. In a kind of acknowledgement of this failure, a statement frequently endorsed in the SVC literature is Lord’s contention that ‘rather than [being] a separate universal category, serialization is more accurately characterized as a syndrome of features and phenomena’ (1993, 2). In this dissertation I claim to know what the disease is. Believing and acting as if they only contain one verb is the disease of SVCs. This amounts to something like self-delusion, since in fact they

contain at least two verbs (boldfaced in every SVC in this dissertation):

(1.1) Kkú s às n wá àx`ı m . Koku take crab come market in ‘Koku brought the crab to the market.’ (Fongbe; Lefebvre & Brousseau 2002, 410)

(1.2) Amu lao ne’e maru.na. Amu go ascend sleep.3 ‘Amu went up (to her room/into the house) and slept.’ (Bima)

1 CHAPTER 1. INTRODUCTION 2

(1.3) Nws muab riam txiav nqiaj qaib. 3sg take knife cut meat chicken

‘She cut some chicken with a knife.’ (White Hmong; Jarkey 1991, 83)

(1.4) Chán won j n klàp khâw h . I circle reverse return enter room ‘I returned circling back into the room.’ (Thai; Zlatev & Yangklang 2004, 163)

These sample SVCs conform to the properties that can reliably be used to define SVCs (to be listed in Section 1.2.1): they contain no coordinators or subordinators, the verbs do not display divergent tense, aspect, mood, person or polarity categories, and were the phonological characteristics of (1.1)-(1.4) to be made explicit, they would mirror the phonological characteristics of a clause from the same language which contains a single verb (in as much as the inclusion of an additional verb will allow). (1.1)-(1.4) appear monoclausal. Monoclausality is at the heart of SVC definition, and monoclausality is also the grammatical stamp of believing and acting as if there is only one verb in the clause. But there is one facet of monoclausal behavior which most SVC definitions miss: SVCs must be monoclausal with respect to their argument structure as well. The core proposal for this dissertation is that SVCs should be defined and identified through the Theta Structure Property for SVCs, a property which enforces monoclausal argument structure alongside all the other monoclausal qualities of SVCs. But just as important as the Theta Structure Property itself is the claim that this property enforces monoclausal argument structure in such a way that two main types of SVC can be identified: there are the core, ‘Possible Verb’ SVCs, and then there is an additional class of ‘Impossible Verb’ SVCs. While Possible Verb SVCs have the semantic content and argument structure of a clause headed by a single verb, Impos- sible Verb SVCs do not. And while Possible Verb SVCs fulfill the Theta Structure Property for SVCs straightforwardly, as will be demonstrated in Chapter 2, Impossi- ble Verb SVCs can only fulfill the Theta Structure Property through the application of special predicate combining operations, which will be set out in Chapter 3. But before this distinction is considered in more detail, some further comments are in CHAPTER 1. INTRODUCTION 3

order concerning the emphasis placed upon the notion of monoclausality in defining SVCs in this dissertation. To be sure, anyone could assemble any collection of grammatical properties (within reason) and claim that they jointly serve to define the class of SVCs. I am claiming that the properties which define SVCs should all revolve around the notion of monoclausality, yet there is no sense in which monoclausality can be claimed to be pre-determined or pre-ordained as the core defining property for SVCs. What I would claim is that the notion of monoclausality captures what is most interesting about SVCs. What is most interesting about SVCs is that they are constructions which contain (at least) two verbs, yet they behave in every respect as if they contain only one verb. In other words, SVCs show every sign of being monoclausal, in spite of containing more than one verb. Why are all these sentences behaving as if they contain only one verb when they actually contain at least two? This is the central question which granting prominence to the notion of monoclausality allows us to address. While monoclausality cannot be claimed to be pre-determined or pre-ordained as the core defining property for SVCs, it is nonetheless the property on which most SVC definitions have concentrated implicitly – which is to say there is nothing very radical in my approach. What is novel about my approach, however, is the requirement that SVCs demonstrate monoclausal argument structure, through the fulfillment of the Theta Structure Property for SVCs. The Theta Structure Property for SVCs will repeatedly be described as enforcing ‘monoclausal argument structure’ in this dissertation, but what is meant by ‘monoclausal argument structure’? According to Durie (1997, 291), “when serialization results in a complex of more than two arguments, the configuration of the arguments corresponds closely to the kinds of configurations of arguments+adjuncts found for single clauses in non-serializing languages”. This could describe monoclausal argument structure, although Durie does not refer to it as such, and in fact ‘monoclausal argument structure’ is a term scarcely encountered in linguistic works. This may be because it sounds like it should be torn between two different levels: ‘monoclausality’ is a property which obtains at the level of (surface) syntax, whereas ‘argument structure’ is a semantic level (albeit a ‘syntacticized’ semantic level in many cases). CHAPTER 1. INTRODUCTION 4

If monoclausal argument structure is to be identified at the surface level of syntax, few theoretical problems are encountered, since coding patterns for sets of grammatical relations in simplex clauses are rather well documented in the linguistics literature, and it should be easy enough to ascertain whether or not an SVC fits one of these patterns. But the following ‘instrumental’ SVC (=(1.3)) demonstrates why this theoretically easy option cannot be exploited – why ‘monoclausal argument structure’ cannot be identified at the surface syntactic level:

(1.5) Nws muab riam txiav nqiaj qaib. 3sg take knife cut meat chicken ‘She cut some chicken with a knife.’ (White Hmong; Jarkey 1991, 83)

(1.5) will not replicate the pattern of grammatical relations in any simplex clause, since a simplex sentence containing a single verb can obviously not have two objects which are each the complements of different verbs. The arrangement of arguments in (1.5) is SVC-specific at the surface syntactic level. However, (1.5) can still be claimed to have monoclausal argument structure if this property is defined at a ‘deeper’ semantic level, where riam ‘knife’ is recognized as an instrument, and not as an object. (1.5) has monoclausal argument structure at this level because it consists of an actor (nws ‘she’), an undergoer (nqiaj qaib ‘chicken’) and an instrument (riam ‘knife’), and there do exist simplex clauses headed by single verbs whose ‘argument structure lists’ similarly consist of an actor, an undergoer and an instrument. What is meant by ‘monoclausal argument structure’, then, is argument structure which would ordinarily project into a surface syntactic monoclausal pattern of grammatical relations in a clause containing a single verb, but for reasons specific to SVCs (to do with the fact that there is more than one verb), this argument structure does not result in monoclausal surface syntactic structure for this type of construction. The property used to enforce monoclausal argument structure in this dissertation is the Theta Structure Property for SVCs, and this property is formulated in the framework of Lexical Decomposition Grammar (LDG) (Wunderlich 1997a, Wunder- lich 1997b, Wunderlich 2000). The level at which monoclausal argument structure obtains is Theta Structure in this framework. Theta Structure is in some ways equivalent to syntacticized levels of argument structure in other frameworks (e.g. Grimshaw CHAPTER 1. INTRODUCTION 5

1990, Alsina 1996, Manning 1996). It is the level at which argument structure operations, such as passive or middle, apply, and it is the locus of binding. With respect to the discussion immediately above, a crucial point is that it doesn’t consist of lists of thematic roles such as ‘agent’ or ‘patient’ (or ‘generalized’ thematic roles such as ‘actor’ or ‘undergoer’); instead it consists of an ordered string of λ-abstracted arguments. Certain principles (to be detailed in Chapter 2) determine that some of these λ-abstracted arguments are structural arguments, which are projected into structural positions in the syntax (such as ‘subject’ and ‘object’), according to their position in the string. All the other λ-abstracted arguments are non-structural, but may be realized as oblique arguments in the syntax. Structural arguments are loosely equivalent to ‘core’ arguments and non-structural arguments to ‘peripheral’ arguments. Theo- ries of the syntax-semantics interface have tended to concentrate on the realization of core arguments, but the challenge that SVCs such as (1.5) throw up is to account for the realization of peripheral arguments (such as instruments), which nevertheless look to be structural in the surface syntax. Non-standard realization of peripheral arguments is one of the challenges addressed in Chapter 2, where the Theta Structure Property for SVCs and its accompanying complications are stated and illustrated. Chapter 3 then moves into the second main claim of this dissertation, that there are two diﬀerent types of SVCs: ‘Possible Verb’ and ‘Impossible Verb’ SVCs. Possible Verb SVCs, such as (1.6) below, satisfy LDG’s ‘Possible Verbs’ constraints. These constraints will be discussed in Section 2.2, but suﬃce it to say for now that what they are constraining is the semantic content of what can be expressed in a clause headed by a single verb. It is as a direct consequence of the fact that (1.6) does satisfy the Possible Verbs constraints that this SVC – like all other Possible Verb SVCs – has an (accurate) translation equivalent consisting of a clause containing a single verb:

(1.6) Ana kea rama lu’u.na d’ei kota. baby crawl enter.3 in box ‘The baby crawled into the box.’ (Bima)

Impossible Verb SVCs such as (1.7), in contrast, do not satisfy LDG’s ‘Possible Verbs’ constraints, and will have no translation equivalents consisting of clauses containing CHAPTER 1. INTRODUCTION 6

single verbs:

(1.7) Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozocookedandatesomefood.’ (Edo;Stewart2001,10)

The main technical reflex of the Possible Verb-Impossible Verb distinction is that these two types of SVC satisfy the Theta Structure Property for SVCs in very different ways. Specifically, Impossible Verb SVCs are more complicated in requiring certain predicate combining operations to apply in order to fulfill the Theta Structure Property, to the extent that it is reasonable to ask whether these SVCs (or their inclusion in the class of SVCs) really warrants such complication. In the end, however, I argue that Impossible Verb SVCs should be included in the class for the reason that they are just as monoclausal in their grammatical behavior as any Possible Verb SVC. But their inclusion does have a number of consequences, many of which challenge some of the assumptions linguists typically hold about SVCs. One assumption linguists tend to hold about SVCs is that these constructions are always doing the things that would be done with single-verb-headed clauses in other languages. An extension of this assumption is that they are ‘filling in’ for functional gaps in the absence of certain grammatical categories like prepositions, using an additional verb to execute the function of the preposition. Indeed, serialization is often equated with grammatical impoverishment. Although there is some truth to these assumptions, they are far from being the whole story. Impossible Verb SVCs such as (1.7) bear out in the most marked sense how SVCs are not always doing what would be done with a clause headed by a single verb. Another commonly held assumption about SVCs, to be discussed in Section 1.2.4.1 in the literature review of this chapter, is that these constructions always describe ‘single events’. An Impossible Verb SVC like (1.7), however, is unlikely to be attributed single event status, for the reason that it describes temporally distinct actions with no necessary causal connection. On the other hand, might (1.7) describe a ‘recognizable event type’ – a combination of actions that cooccur so often as to be conceptualized as a single event? Section 1.2.4.2 below observes how, in at least some languages, CHAPTER 1. INTRODUCTION 7

SVC formation is constrained by what constitutes a ‘recognizable event type’ in the society in which the language is spoken. Event structural concepts play a prominent part in this dissertation; in particular, the event structural analyses set out in Chapters 3, 4 and 5 provide a firm grasp of what the bounds on Possible Verb, or Impossible Verb, events are, which in turn provides a clear view of the ultimate insight of the Possible Verb-Impossible Verb distinction: that some SVCs (Possible Verb SVCs) are doing what could otherwise be done with clauses headed by single verbs, while others (Impossible Verb SVCs) cannot be. Because event structural notions feature prominently in this dissertation, the literature review of Section 1.2 devotes considerable space to event-related issues. However, the first main thrust of this dissertation is not event structural but argument structural; first I argue that SVCs need to be defined through argument structural (in addition to other, core) properties, and then I propose the Theta Structure Prop- erty for SVCs as a means of enforcing uniform monoclausal argument structure in SVCs. Only once this property is put into effect can SVCs be seen to diverge event structurally. But as far as the argument structure of SVCs is concerned, perhaps the most significant revelation in Section 1.2 is that there has already been one attempt to enforce monoclausal argument structure for SVCs – even if the proponents of this approach do not see it this way. Internal Argument Sharing is the core defining property for SVCs in the formal literature, and in Section 1.2.2 I will demonstrate that, although it goes a long way towards enforcing monoclausal argument structure, it has some serious faults which effectively disqualify it as a suitable defining property for SVCs. So, the primary upshot of Section 1.2.2 is that it paves the way for the Theta Structure Property for SVCs to replace Internal Argument Sharing. But Section 1.2.2 also makes a secondary observation, that Internal Argument Sharing is apparently shaped by African SVC data in isolation, and does not respond to the full cross- linguistic range of SVC data. In contrast, this dissertation approaches serialization from as broad a cross-linguistic base as possible. I also include a large amount of data collected myself from the Austronesian language Bima (such as (1.2) and (1.6) above). Although hardly presenting a comprehensive cross-linguistic survey of serialization, CHAPTER 1. INTRODUCTION 8

this dissertation is at least informed by a body of SVC data that is more cross- linguistically representative than the traditional base of African and Asian SVCs.

1.2 Previous work on SVCs

Serialization was a phenomenon first brought to light in Christaller (1875), who, in his study of Twi, described verbal combinations where it is possible for “two or more verbs, not connected by conjunctions [to] have the same subject” (1875, 144). Westermann (1930) next described serial constructions in Ewe. These two works were purely descriptive; it was not until the publication of Chomsky (1957) that SVCs came to be seen as posing any kind of theoretical problem, seeming to refute the claim that a single clause can have only one finite verb. Once the power of transformations and deletions was unleashed in the 1960s, several analyses of serial verb phenomena appeared, starting with Stewart’s (1963) transformational analysis of Twi, later followed by Bamgbos.e’s (1973, 1974) and Awobuluyi’s (1973) treatments of Yoruba. The 1970s saw a cross-linguistic expansion of SVC data; after a long-standing concentration on West African languages, serializing creoles came under consideration, while Li and Thompson (1974) discussed Chinese serial verbs. Discussion of serialization in Southeast Asian, Austronesian, Papuan and South American languages followed in the 1980s. The history of the SVC literature is a history of having to come to grips with a rapidly expanding database, then, but it is also a history of culling the phenomena to be accounted for, of delimiting and defining SVCs (in some instances perhaps unjustifiably). This is a very brief history of SVCs which in this section will be expanded upon by branching out into four core areas of the literature: definitions of SVCs in Section 1.2.1, the argument structure of SVCs in Section 1.2.2, the syntax of SVCs in Section 1.2.3, and the event structure of SVCs in Section 1.2.4. Of these four areas, it is the first that is naturally the most fundamental, and which has met with some of the most determined efforts, to delimit and cull the phenomena to be accounted for, as just mentioned. In surveying the ways in which SVCs have been defined in the CHAPTER 1. INTRODUCTION 9

literature, Section 1.2.1 arrives at what I take to be the best middle ground. But while issues of SVC definition are necessary and foundational to any SVC study, it is the discussion of the argument structure of SVCs in Section 1.2.2 that is of crucial importance to this study. Internal Argument Sharing, the property mentioned in conclusion to Section 1.1, will be described in Section 1.2.2 in the context of how the argument structure of SVCs has been approached in the literature. This discussion motivates the use of argument structural properties in defining SVCs, but Section 1.2.2 also lays bare some of the problems with Internal Argument Sharing. These problems crucially motivate the Theta Structure Property for SVCs as the preferred argument structure-based defining property. The two remaining sections in this review of the SVC literature investigate the syntax and semantics of SVCs respectively. Because this dissertation is not heavily syntax-oriented, Section 1.2.3 provides only superficial coverage of the numerous syntactic analyses proposed for SVCs, although it does focus on one feature that they all share: the exclusion of SVCs with contiguous verbs from the class of SVCs. The review of the semantics of SVCs in 1.2.4 is more extensive than the review of the syntax of SVCs (even though there has been more work in syntax). Seman- tic analysis of SVCs could mean classifying these constructions according to their semantic functions, as in the following (incomplete) list:

(1.8) Instrumental SVCs:

No teki baskita tyari watra. neg take basket carry water ‘Don’t carry water with a basket.’ (Sranan; Sebba 1987, 59)

(1.9) Benefactive SVCs:

S´ei wan ijsie d´a mi! sell det ice-cream give 1sg ‘Sell an ice-cream for me!’ (Saramaccan; Veenstra 1996, 95) CHAPTER 1. INTRODUCTION 10

(1.10) Locative SVCs:

Sàlám`ı lá èbi ta èsákó o. Salami took knife be table loc ‘Salami put theknifeonthetable.’ (Nupe;Lord1993,126)

(1.11) Comparative SVCs:

Nimo enyau nmah m-telwo-i horo-m. house poss:1sg big sg:es-go.past-const poss-2sg ‘My house is bigger than yours.’ (Erromangan; Crowley 2002, 185)

(1.12) Exceptive SVCs:

B`e di b´e kp mi. they+cons eat they+cons cut me ‘They eat without me.’ (Baule; N’Guessan 2000, 85)

(1.13) Aspectual SVCs:

Mi jab´ı d´ı dóo káá. 1sg open det door finish ‘I already opened the door.’ (Saramaccan; Veenstra 1996, 98)

(1.14) Manner/Adverbial SVCs:

Ra-vavine na loli na hinau mo mahere. pl-woman 3pl do art thing 3sg.real straight ‘Women do things the right way.’ (Tamambo; Jauncey 1997, 388)

However, the facet of semantic analysis discussed in 1.2.4 is the event structure of SVCs. Event structural notions are frequently invoked in the SVC literature and they also feature prominently in this dissertation, but with more deﬁnition and rigor than in any other SVC research, I would argue. The Possible Verb–Impossible Verb distinction and its consequences will be cast in event structure terms in Chapter 3, and event structural analysis will be the main mode of analysis in the case studies of instrumental and motion SVCs in Chapters 4 and 5 as well. CHAPTER 1. INTRODUCTION 11

1.2.1 Deﬁnitions of SVCs

Three defining properties of SVCs appear to be uncontroversial: (a) SVCs must be monoclausal,1 (b) all verbs in an SVC must share the same tense, aspect, mood and polarity categories, and (c) there can be no marking of coordination or subordination between the verbs in an SVC. These requirements presuppose another defining property: that the elements in the sequence actually be verbs, and not, say, prepositions (or ‘coverbs’, or ‘verbids’). Sebba (1987, 39) further proposes that all verbs in an SVC must be lexical verbs which are capable of independently heading a simple sentence, and few would object to this.2 In theory, all that needs to be demonstrated in order to establish SVC status is that each participating verb is actually a verb, although the difficulty that can arise in practice is that many serializing languages have little inflection, and little evidence for grammatical categories in general. Nonetheless, assuming it can be shown that the elements in the construction are verbs, an additional requirement sometimes imposed upon SVCs is that each verb have exactly the same semantics in an SVC as it would heading a simple clause in isolation. This requirement will be discarded as too strict; this dissertation will discuss numerous examples of SVCs for which this property does not hold, but where the meaning of the verb outside the SVC can be systematically related to its altered (or ‘specialized’ (Bril 2004)) meaning within an SVC. The theoretical tool used here to account for these altered meanings are serialization lexical rules, and these will be illustrated in Chapter 2, and in Section 3.2.2. To the requirement (b) that the verbs in an SVC must share the same tense, aspect, mood and polarity categories, could be appended the requirement that they also share the same person-number categories. Comrie (1995) specifies this additional requirement, and identifies three patterns of verb marking in SVCs: no marking on either verb, as in (1.15), marking on one (‘main’) verb, as in (1.16), and identical marking on both verbs, as in (1.17):

1But see Foley (2006) for a rare rejection of this claim. 2But see Crowley (2002, 82-107) for a description of Paamese nuclear-layer serialization where a number of verbs in serial combinations do not have an independent existence outside SVCs.

CHAPTER 1. INTRODUCTION 12

(1.15) Kkú s às n wá àx`ı m . Koku take crab come market in ‘Koku brought the crab to the market.’ (direction towards speaker) (Fongbe; Lefebvre & Brousseau 2002, 410)

(1.16) An abso rag h-n--a. we vegetables carry come-fut-1pl-decl ‘We will bring vegetables.’ (Haruai; Comrie 1995, 28)

` (1.17) Acw` `alˆ rw`ot. 1sg.fat.hab 1sg.exceed.hab king ‘Iamfatterthantheking.’ (Lango;Noonan1992,211)

Marking of tense, aspect, mood, polarity or person-number categories on each verb in an SVC is relatively rare, and multiple marking of the person-number category in particular can throw up some incongruity besides, in those circumstances where identical person-number marking is at odds with what is figured to be the person- number status of the underlying subject of one (or more) of the verbs. The first person singular marking on migu ‘flow’ in (1.18), for example, although identical to the marking on mede ‘take’, is at odds with the third person status of msum ‘water’ (the underlying subject of migu):

(1.18) Mede aburow migu msum. 1sg.take corn 1sg.ﬂow water.in ‘I pour corn into water.’ (Akan; Schachter 1974, 258)

Sometimes a particular construction, such as the Paamese ‘core-layer’ SVC exempli- ﬁed in (1.19), will look to fulﬁll the identical verb category marking requirement, until the person-number values of the arguments associated with each verb are varied and identical marking is not maintained, as in (1.20):

(1.19) Meatin kail avalus aumai. person pl 3pl:real-row 3pl:real-come ‘The people rowed hither.’ (Paamese; Crowley 2002, 55) CHAPTER 1. INTRODUCTION 13

(1.20) Inau nuas vuas hemat. 1sg 1sg:dist.fut-hit pig 3sg:dist.fut-die ‘I will hit the pig to death.’ (Paamese; Crowley 2002, 55)

Paamese ‘core-layer’ SVCs will be excluded from the class of SVCs in this dissertation through not fulﬁlling the strict requirement that each verb have identical person- number marking. SVCs are also deﬁned phonologically. The phonological properties of a clause containing a serial verb as a whole are expected to be the same as those of a clause containing a single verb. (1.21), for example, cannot be an SVC since (among other reasons) a pause is indicated between the two verbal constituents in this sentence:

(1.21) Kofi opo Amba, tyari en gwe. Kofi lift Amba carry her away ‘Kofi lifted Amba and carried her off.’ (Sranan; Sebba 1987, 111)

Neither simple Sranan sentences containing single verbs nor true SVCs allow such pauses. While clauses containing SVCs are expected to show the same phonological properties as clauses containing single verbs, the serial verb complex itself is not expected to resemble a single predicate phonologically under most serial verb definitions. A common defining requirement of serial verbs is that they consist of at least two separate phonological words; if phonological processes sensitive to word boundaries instead identify the entire verb sequence as a single word, then the combination is a compound verb rather than an SVC. But compound verbs can be extremely difficult to tease apart from SVCs, and in fact there may even be grounds for not excluding them from the class of SVCs. The predominant tendency to exclude sequences of verbs which form a single phonological word from the class of SVCs could be attributed to the concentration on West African languages in the first stages of serial verb research. The languages examined had isolating morphological structure, with SVCs consisting of separate phonological words. Another reason for drawing a sharp distinction between SVCs and compound verbs could simply be that there is the terminology in place to do CHAPTER 1. INTRODUCTION 14

so. But compound verbs are sometimes classiﬁed as a type of SVC, on the grounds that these constructions display a wide array of similarities, and that to keep them apart would be to ignore the fact that they often present instantiations of the same phenomenon. Margetts (1999, 101), for example, claims that the term ‘compound’ does not contradict an analysis as serialization, while Nishiyama (1998, 176) casts Japanese compound verbs as a type of serialization. The point that compound verbs appear to be doing exactly the same things as SVCs functionally will also be borne

out by some }Hoan ‘compound verb’ data in Section 2.4.2. Another SVC researcher who does not limit SVCs to constructions containing multiple phonological words is Aikhenvald (2006). She allows single phonological word constructions into the class because of the “functional and formal overlap between ‘one-word serialization’ and ‘multi-word serialization’ ” (2006, 39).3 ‘One-word serialization’, or compounding, can be illustrated through the following Yimas example, where it is not just the serial combination, but the entire verbal complex which forms a single phonological word:

(1.22) Narm pu-tpul-kamprak-r-akn. skin.viisg 3.pl.s-hit-break-perf-3.sg.d ‘They hit and broke his skin.’ (Yimas; Foley 2006, 7)

The compound status of the verbs in this Yimas SVC can be attributed to the incorporating nature of the language. When two verbs form a single phonological word in a non-incorporating language there may be stronger grounds for excluding this compound verb from the class of SVCs, although it still seems prudent to keep compound verbs within sight of any investigation of SVCs, and this is the approach I will take here. In summary, the following properties, which recur in the SVC literature, can be taken to deﬁne SVCs: 3Aikhenvald also describes some of the diﬃculties of distinguishing between phonological and grammatical words in some serializing languages. CHAPTER 1. INTRODUCTION 15

(a) SVCs are monoclausal.

(b) The verbs in an SVC must have the same tense, aspect, mood, polarity, and person-number categories.

(d) The intonational properties of a clause containing a serial verb combination are the same as those of a clause containing a simple predicate in the language concerned.

What goes unmentioned in the SVC literature is that properties (b), (c) and (d) are all to some degree derivatives of property (a), that SVCs be monoclausal. Identical tense, aspect, mood, polarity and person-number marking on verbs, the absence of markers of coordination or subordination, and intonational properties of single clauses are all properties that we would expect a monoclause to display. Property (a) cannot, however, be decomposed and removed for the reason that, in practice, proving that a particular construction is monoclausal can be a language-dependent matter (Butt 2003). Tests or properties which establish monoclausal status can be completely language-specific, and so (a), besides being the core SVC-defining property, also acts as something of a cover-all. The emphasis placed on property (a) serves to reinforce the ideas expressed in the introduction to this chapter, that monoclausality is of paramount importance to SVCs and what they are about, that what is interesting about SVCs is that they are acting in every grammatical respect as if they contain only one verb when in fact there are (at least) two. But there is one major element missing from the list of SVC- defining properties above which concerns the fact that, although it is uncontroversial that there can be no marking of coordination or subordination between the verbs in SVCs, the absence of such marking by no means confers serial verb status upon a potential SVC. The list of properties above cannot exclude covert coordinations from the class of SVCs. In order to do this, SVC definition needs to move into the realms of argument structure. CHAPTER 1. INTRODUCTION 16

1.2.2 Argument structure of SVCs

Whereas most studies of SVCs initially concentrated on the question of how it is that a single clause can contain more than one finite verb and what the limitations on these combinations of verbs are, Stewart (1963) was the first to pose the question of how to account for the fact that each verb in an SVC does not have all of its arguments overtly and independently realized. This issue is approached in different ways in the formal and functional literature on SVCs. The response in the formal literature has been to require SVCs to display Internal Argument Sharing, as will be explained further shortly. The typologically richer functional work on SVCs, typified by Aikhenvald (2006) or Bril (2004), does not seek to delimit the class of SVCs through argument sharing properties, but contents itself in listing off attested argument sharing patterns. This ‘listing off’ can be quite complex; Bril’s (2004) description of argument sharing (and argument fusion) is intermeshed with Aikhenvald’s distinction between symmetrical and asymmetrical SVCs (see Section 1.2.4.3), while Aikhenvald’s (2006) enumeration of argument sharing schemes goes on for nine pages. For simplicity’s sake I will here exemplify the functional approach to argument sharing by setting out Crowley’s (2002) list of five argument sharing schemes. Crowley’s (2002) five argument sharing schemes, with accompanying examples, are as follows: (1.23) demonstrates same-subject sharing, where the subject of the first verb is also the subject of the second verb; in switch-subject sharing (1.24), the object of the first verb is the subject of the second verb; for an inclusory serialization like (1.25) it is argued that the subject of the second verb includes both the subject and the object of the first verb; in multiple object serialization (1.26), each verb is transitive and has its own distinct object; and finally, there is what Crowley terms ambient serialization, which actually does not involve any argument sharing, since one verb functions semantically as an adverb, modifying the other verb, and is not associated with a particular participant, as in (1.27):

(1.23) Amu lao ne’e maru.na. Amu go ascend sleep.3 ‘Amu went up (to her room/into the house) and slept.’ (Bima) CHAPTER 1. INTRODUCTION 17

` (1.24) E. sósà gbé émá!tó. n pè. rhé. . Esosa hit metal be-flat ‘Esosa hit the metal flat.’ (Edo; Stewart 2001, 58)

(1.25) Kofi teki Amba go na a foto. Kofi take Amba go loc def city ‘Kofi took Amba to the city.’ (Sranan; Sebba 1987, 59)

(1.26) Olè fi ò. be. gún o. ba. thief use knife stab chief ‘The thief stabbed the chief with the knife.’ (Yoruba; George 1985)

(1.27) Wela e qiliano-na taqaa baqu. child 3sg:nfut pile.soil.around-3:obj be.bad banana ‘The child piled the soil around the banana tree badly.’ (Toqabaqita; Aikhenvald 2006, 29)

Crowley’s list of argument sharing patterns is not exhaustive (for a more exhaustive account, read Aikhenvald’s (2006) discussion), but it is indicative of the more typologically-oriented, non-formal approaches to argument sharing and argument structure in SVCs. In the formal SVC literature, Stewart’s (1963) question of how to account for the fact that each verb in an SVC does not have all of its arguments overtly and independently realized was met with a ground-breaking answer in Baker (1989) in the form of Internal Argument Sharing, which Collins (1997) describes in the following simple terms:

(1.28) Internal Argument Sharing in SVCs (Collins 1997, 463)

In a serial verb construction, V1 and V2 must share an internal argument.

Internal Argument Sharing was not necessarily formulated in direct response to Stew- art’s question; instead Baker (1989) was more concerned with excluding covert coordinations, such as (1.29), from the class of SVCs:

` (1.29) Ozó gbò. ó. ´ıv`ın bòló ó. là. Ozo planted coconut peel corn ‘Ozo planted coconut and peeled corn.’ (Edo; Stewart 2001, 51) CHAPTER 1. INTRODUCTION 18

Each verb in (1.29) realizes a distinct internal argument. The internal argument of gbò. ó. ‘plant’ is ´ıv`ın ‘coconut’ and the internal argument of bòló ‘peel’ is ó. là ‘corn’, and since these verbs do not share a single internal argument, (1.29) does not display Internal Argument Sharing, and so is not an SVC, but rather a covert coordination. (1.30), in contrast, is an SVC, since it does display Internal Argument Sharing, since sè ‘cook’ and tà ‘sell’ share the single internal argument e. ran ‘meat’:

(1.30) Bó. lá sè e. ran tà. Bola cook meat sell ‘Bola cooked some meat and sold it.’ (Yoruba; Baker 1989, 529)

Internal Argument Sharing has come to be recognized as the major defining requirement for SVCs in the current formal literature (Baker 1989, Collins 1997, Stewart 2001, etc.), even if there is some variation in how it is implemented (whereas Baker’s (1989) original account is said to involve ‘true’ Internal Argument Sharing, subsequent analyses have diverged by instead mediating the sharing through an empty category (Collins 1997, Carstens 1988, Law and Veenstra 1992, Campbell 1996)). However, in this section I will discuss some of the problems with Internal Argument Sharing, which ultimately render it unsuitable as a defining property for SVCs, and which lay the groundwork for adopting the Theta Structure Property for SVCs in its place. Durie (1997) criticizes Baker’s (1989) analysis of SVCs from several angles, and while some of these criticisms are specific to Baker’s particular implementation of Internal Argument Sharing, there is one notable criticism which is not, and to which all Internal Argument Sharing approaches are subject. This is that Internal Argument Sharing excludes intransitive+transitive verb combinations such as the following from the class of SVCs, even though these are otherwise uncontroversially considered to be SVCs:

(1.31) Taho mpara lao te’e.ku tamba. ﬁne then go install.1 trap ‘Okay then, I’ll go and install a trap.’ (Bima) CHAPTER 1. INTRODUCTION 19

(1.32) Va-n-a su-n. -ap. come-nonpast-1pl.exc see-nonpast-1pl.exc ‘We will come and see.’ (Kon. d. a; Steever 1988, 71-3)

(1.33) I u no-da axe mweyaayec xe pwa-giik. 3sg pft look-up see house top class-one ‘He looks up (and) sees a house.’ (Nˆelˆemwa; Bril 2004, 190)

(1.34) Eam toro peha kaveru o karuu te Kitata. 2pl must climb steal art karuu nut prep Kitata ‘You must climb and steal Kitata’s karuu nuts.’ (Teop; Reinig 2004, 94)

(1.35) Na yawy-r si˜na-tat-me-an-r. 1sg dog-3sm rise-hit-r.pst-1s-3sm ‘I rose (and) hit a dog.’ (Alamblak; Bruce 1986, 246)

(1.36) A e opo weri en krosi esi. he t/a get.up wear his clothes quick ‘He gets up quickly and puts on his clothes.’ (Sranan; Sebba 1987, 57)

These examples mostly involve an intransitive motion (or posture) verb whose moving argument is also the agent of the transitive action verb that follows. None of these SVCs involves internal argument sharing since, in each case, even if the sole argument of the ﬁrst verb is analyzed as an internal, unaccusative argument, it is certainly not the case that it is shared as the internal argument of the second verb. The sole argument of opo ‘get up’ in (1.36), for example, is the third person subject, yet it is en krosi ‘his clothes’ which is the internal argument of the second verb, weri ‘wear’, and not the third person human entity. Baker (1989, 532-3) acknowledges that the following Sranan sentence does not display internal argument sharing, but argues that it should not be analyzed as an SVC on the basis of Sebba’s suggestion that go subcategorizes for an inﬁnitival S′ complement headed by luku:

(1.37) Wan man go luku wan dansi. a man go look a dance ‘A man went to watch a dance.’ (Sranan; Sebba 1987, 53) CHAPTER 1. INTRODUCTION 20

However, even supposing it were correct, there is no justification for extending this analysis to (1.31)-(1.36), for which there is no evidence of infinitival structure. (1.31)- (1.36) behave in every respect like other SVCs in the languages concerned. Durie (1997) furthermore points out that the following Kalam SVC cannot involve S′ complementation, since Kalam is verb-final, which would mean that an infinitival S′ complement of am ‘go’ would be expected to precede, rather than follow it:

(1.38) Kty am kmn pak dad ap-l, ... they go game.mammal hit carrying come-ss:prior ‘Sometimes, having gone and hit and carried back such a game mammal, ...’ (Kalam; Lane 1991, 71)

Durie also makes the following comment: “Every serializing language I have encountered includes a category of motion serialization, where a verb of motion is combined with some other verb in such a way that the motion verb comes ﬁrst and the moving argument is the Agent of the second verb”(1997, 310). Without any further information about the composition of Durie’s data survey it cannot be concluded whether the majority (or indeed the totality) of serializing languages do exhibit intransitive+transitive verb combinations of this type. Many African serializing languages apparently do not permit such combinations (unless these are systematically excluded from the samples of serialization provided, for some reason). It is only in relation to this cross-linguistically limited sample of African languages that In- ternal Argument Sharing could be seen as a reasonable deﬁning property for SVCs; if the survey of SVCs is widened to include languages with intransitive+transitive non-internal argument sharing SVCs, then Internal Argument Sharing is doomed. SVCs such as the following present additional problems for Internal Argument

Sharing: (1.39) Kkú kló hú As´ıbá.` Koku be.big surpass Asiba ‘Koku is bigger than Asiba.’ (Fongbe; Lefebvre & Brousseau 2002, 401)

(1.40) Ne-len mi-yip hal-yak na.kat. art-wind perf-blow ﬂy-away art-cards ‘The winds blew the cards away.’ (Mwotlap; Francois 2004, 119) CHAPTER 1. INTRODUCTION 21

(1.41) Wela e qiliano-na taqaa baqu. child 3sg:nfut pile.soil.around-3:obj be.bad banana ‘The child piled the soil around the banana tree badly.’ (Toqabaqita; Aikhenvald 2006, 29)

Lefebvre and Brousseau (2002) comment that, while Asiba could be the internal argument of hú ‘surpass’ in (1.39), it cannot be analyzed as the internal argument of kló ‘be big’. The verbs in (1.40) are also claimed (by Aikhenvald (2006)) not to share any arguments, while the verb taqaa ‘be bad’ in (1.41), if it could be said to take an argument, selects the event argument which it modifies. In failing to display Internal Argument Sharing, none of (1.39)-(1.41) would be classified as SVCs. This presents another respect in which Internal Argument Sharing is too restrictive, since (1.39)-(1.41) have just as strong a claim to SVC status as any other SVC considered here; (1.39)-(1.41) look like SVCs, they display every facet of monoclausal behavior expected of SVCs,4 and SVCs of this type are included without comment in numerous SVC surveys (such as Bril and Ozanne-Rivierre (2004), Aikhenvald (2006), and van Staden and Reesink (2008)). Some versions of Internal Argument Sharing even exclude instrumental SVCs such as (1.42) from the class of SVCs, although Baker (1989) did originally claim that the instrumental argument (egbà ‘axe’ in (1.42)) is shared between the verbs (lá ‘take’ and ba ‘cut’ in (1.42)) and so such SVCs do satisfy Internal Argument Sharing according to him:

(1.42) Tsoda lá egbà ba cigbà. Tsoda take axe cut tree ‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63)

SVCs such as (1.39)-(1.41), along with the widespread existence of intransitive+ transitive verb combinations in SVCs such as (1.31)-(1.36) lay bare the faults of Internal Argument Sharing. Baker’s innovation of Internal Argument Sharing was ground-breaking but flawed, then, and all subsequent formal analyses suffer from the same flaws. But this is not to claim that the idea of incorporating an argument sharing requirement into the definition of SVCs is a bad one; indeed, some of the more

4That is, they display uniform tense, aspect, mood and person-number marking, no marking of coordination or subordination, and monoclausal phonological behavior. CHAPTER 1. INTRODUCTION 22

typologically-focused, functional works on SVCs which simply list off possible argument sharing schemes, instead of actually requiring some form of argument sharing, may be subject to the criticism that they have no way of excluding covert coordinations from the class of SVCs. The Theta Structure Property for SVCs implements an argument sharing requirement in such a way as to exclude covert coordinations, as will be described in Section 1.3, and more fully in Chapter 2. The final point to mention here is that the way in which it does this has been influenced by Durie’s (1997) sketch of an argument structure ‘fusion’ analysis for SVCs, which accounts for how SVCs show the same range of argument structures as simplex predicates. There is a sense in which the Theta Structure Property finishes off what Durie (1997) started, even if the technical details, and the framework in which the Theta Structure Property is embedded diverge from Durie’s analysis (which is constructed in Jackendoff’s (1990) framework).

1.2.3 Syntax of SVCs

SVCs have received numerous syntactic analyses in the literature, although as Good (2003) points out, these can be categorized into three main groups: those that treat serialization as a type of complementation, schematized in (1.43), those that treat it as a type of coordination, as in (1.44), and those that treat it as a type of adjunction, as in (1.45):

(1.43) COMPLEMENTATION S

NP VP

V (NP) VP

(1.44) COORDINATION S

NP VP

VP VP ... CHAPTER 1. INTRODUCTION 23

(1.45) ADJUNCTION S

NP VP

VP VP

V (NP)

Sebba (1987), Baker (1989), Larson (1991), Lefebvre (1991) and Collins (1993) are some proponents of the ﬁrst approach, although their reasons for settling upon a complementation analysis are quite diverse, as Good (2003) explains. Proponents of the coordination approach likewise vary in the arguments for their analyses, and the same is also true for proponents of adjunction analyses. So although syntactic analyses of SVCs can be organized into three broad groupings, there are really very few points of convergence among them. There is, however, one (and only one) feature which they all have in common, and with which I will take issue: all the syntactic analyses of SVCs Good (2003) mentions are analyses of serial verb phrases, as the diagrams in (1.43)-(1.45) would suggest. Because (1.43)-(1.45) involve the complementation, coordination and adjunction of VPs, any objects in these structures will be realized directly adjacent to the verbs which select them. This is manifest in (1.46), where as.o. ‘clothing’, the object of wo. ‘wear’, occurs directly after this verb, before the VP headed by f´un ‘give’ (regardless of whether complementation, coordination or adjunction is involved):

(1.46) Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5)

The verbs in (1.46) are non-contiguous. (1.47), on the other hand, is a contiguous SVC, where the verbs occur directly adjacent to each other, with no objects intervening between them:

(1.47) Fu burede ije sime abe ufu. he bread def knife take cut ‘He cut the bread with a knife.’ (Barai; Foley & Olson 1985, 44) CHAPTER 1. INTRODUCTION 24

The problem with the three types of serial verb analysis schematized in (1.43)-(1.45) is that none of them can cater to contiguous verbs in SVCs.5 (1.43)-(1.45) only allow for non-contiguous SVCs, which reﬂects the fact that the researchers who have adopted versions of these analyses are only concerned with the traditional core of African and African substrate creole non-contiguous SVCs. For the ﬁrst 100 years of serial verb research, non-contiguous serialization was all there was. The Twi SVCs in Christaller’s pioneering (1875) work were non- contiguous, Westermann’s (1930) Ewe SVCs were non-contiguous, and so were all the other African, African substrate creole and Chinese SVCs described in the literature through til the 1970s. Later discussions of serial verb phenomena in Austronesian and Papuan languages in particular call strongly for the recognition of contiguous SVCs, yet many current Afro-centric approaches to SVCs either ignore these calls or else deny claims that contiguous SVCs can have SVC status. Baker (1989), for example, sidelines contiguous SVCs by claiming they all present cases of compound verbs,

5(1.47) is an example of a contiguous SVC in which all the verbs cluster together, with objects realized in distinct syntactic constituents. Under Kroeger’s (2004) analysis, (1.47) would have the following syntactic structure: S

NP VP

he NP NP V′

bread knife V V

take cut An alternative analysis of (1.47) is the following, where ‘knife’ appears within the same minimal VP constituent as ‘take’, as its direct object, and where this VP constituent necessarily intervenes between ‘cut’ and its object ‘bread’ in the larger VP constituent: S

NP VP

he bread VP cut

knife take I know of no evidence to argue in favor of either analysis in Barai contiguous SVCs. CHAPTER 1. INTRODUCTION 25

and thus should not be classiﬁed as SVCs. The problem with this claim is that it is oblivious to the key issue of whether or not the verbs concerned constitute a single phonological word: if they do then the construction is a compound verb; if they do not then the construction is a contiguous SVC. Of course, distinguishing contiguous SVCs from compound verbs can certainly be tricky – this much was acknowledged in Section 1.2.1 – but at the same time there are no grounds for claiming that all contiguous multi-verb constructions must be compounds. Once it is acknowledged that contiguous SVCs exist, these constructions can apparently be found in every language family – even in African serializing languages. In

Section 2.4.2 it will be suggested that the }Hoan ‘compounds’ Collins (2002) describes may actually be contiguous SVCs, but regardless of whether or not this is true, one point needs to be emphasized: contiguous SVCs have just as much claim to SVC status as non-contiguous SVCs. To exclude the former from the class of SVCs would be to limit the cross-linguistic class of SVCs undesirably and unreasonably. This dissertation pays just as much attention to contiguous SVCs as to non-contiguous SVCs. The Theta Structure Property for SVCs, formulated at a level of argument structure, deals with contiguous SVCs just as well as it does with non-contiguous SVCs. Where the syntax is concerned, the contiguous–non-contiguous distinction requires SVCs to instantiate at least two different types of phrase structure, as will be discussed in Chapter 2, and Chapter 2 also states the argument realization generalizations for non-contiguous SVCs, which account for the SVC-specific clause structure of this type of SVC. The arguments of contiguous SVCs, in contrast, are allowed to project into the syntax in exactly the same way as in a simplex clause, in the absence of any evidence for SVC-specific clause structure. Aside from this discussion in Chapter 2, very little issue will be made of the contiguous–non-contiguous distinction in the rest of this dissertation, the underlying attitude being that contiguous SVCs simply have to be included in the class of SVCs and beyond that there is little else to say. CHAPTER 1. INTRODUCTION 26

1.2.4 Event structure of SVCs

The properties used to define SVCs in Sections 1.2.1 and 1.2.2 focused on the ways in which SVCs resemble simplex clauses headed by single verbs syntactically, mor- phologically, phonologically, and argument structurally. As far as their meaning is concerned, the idea that SVCs resemble simplex clauses headed by single verbs is borne out semantically to an extent, in the form of the claim that SVCs describe a single event. The notion of ‘event’ has featured prominently in some recent semantic analyses of SVCs. Events will also feature strongly in this dissertation, once Possible Verb and Impossible Verb SVCs are event structurally decomposed in Chapter 3. But event structural terminology should not be utilized without proper explanation and care, and in fact the main complaint of this section is that studies of SVCs tend to make an initial, defining statement that SVCs describe a ‘single event’ without acknowledging, or even realizing, how unspecific a statement this actually is. Section 1.2.4.1 below discusses several approaches to SVCs which subscribe to the idea that SVCs describe a single event, attempting, where possible, to pinpoint what is meant by ‘single event’. Notions of temporal structure feature prominently in this; single events tend to be identified with non-sequential semantics. But Section 1.2.4.1 ultimately concludes that SVCs cannot be equated with single, necessarily non-sequential, events across-the-board, and it also surveys the small number of SVC researchers who identify SVCs not with single non-sequential event status, but with the opposite status of multiple sequential events. Section 1.2.4.2 discusses approaches to SVCs which take a different slant on eventhood in claiming that SVCs describe ‘recognizable event types’, or what is conceptualized as a single event. These claims are backed up by some interesting data which demonstrate how certain culture-specific, closely associated actions can be expressed in SVC-form only because they constitute a recognizable event type. Interesting as this kind of data is, it is unfortunately rather limited, which raises questions of what part ‘recognizable event types’ might play in constraining SVCs cross-linguistically. Section 1.2.4.3 surveys a common trend in SVC research: to divide the class of SVCs along event structural lines into two types. Broadly speaking, these two-way distinctions contrast single event non-sequential SVCs with multiple event sequential CHAPTER 1. INTRODUCTION 27

SVCs, and while the Possible Verb-Impossible Verb distinction is not based exclusively in these event structural terms, the event-structure-based distinctions in Sec- tion 1.2.4.3 nonetheless provide a kind of precursor to the Possible Verb-Impossible Verb distinction, which I will eventually lay out in Chapter 3.

1.2.4.1 ‘SVCs describe a single event’

The claim that SVCs describe a single event is commonplace and widely accepted in the SVC literature, but it is also subject to considerable complexity, one indication of which is Aikhenvald’s (2006, 12) comment that “semantically, serial verbs may encode one event, or several subevents closely linked together, or even several events in sequence which may be conceptualized as connected to each other”. Are all of these notions compatible with one over-arching category of ‘single event’? According to Durie (1997, 291), that SVCs describe a single event “is repeatedly reported to be an intuition of native speakers, and can be demonstrated through semantic analysis”, although unfortunately, he is not explicit about what form this semantic analysis takes. The main difficulty in demonstrating that an SVC describes a single event stems from the indeterminacy in pinpointing exactly what a ‘single event’ is. Writers tend to seize upon notions of ‘indivisibility’ towards this end, although even more prominent are notions concerning temporal structure: single events tend to be identified with non-sequential semantics. This section surveys approaches to SVCs which identify single event status and non-sequential structure as the cornerstones of SVC definition. The opposing standpoint, that SVCs must describe multiple, sequential events is encountered far less often and will be mentioned only briefly in conclusion to this section. Although the primary concern here is with SVC studies which make overt reference to event structural notions, some of the earlier SVC studies appear to require that an SVC describe a single, non-sequential event without actually making this requirement explicit. Stahlke (1970), for example, only includes non-sequential Kwa SVCs in his sample, in the absence of any defining requirement that they describe a single event, and the same is true of Jansen, Koopman and Muysken (1978) in their study of CHAPTER 1. INTRODUCTION 28

serialization in Kwa substrate creoles. Some of the SVC studies that do explicitly require SVCs to describe single, non-sequential events may have been inﬂuenced by the absence of sequential, multiple event SVCs in these earlier studies. One example of a study which requires SVCs to describe single, non-sequential events is Byrne (1987). Byrne disallows sequential SVCs by stating that “true serials contain simultaneous actions”; the Saramaccan sentence in (1.48), for example, is interpreted as an SVC only under the non-sequential single event reading indicated in the translation:6

(1.48) A t´ei goni suti di pingo.’ he take gun shoot the pig ‘He shot the pig with a gun.’ (Saramaccan; Byrne 1987, 160)

(1.48) has a second, sequential reading of “consecutive actions, or independent events actually unfolding in real time” (‘he took a gun and (then) shot the pig’), but in this case it is adjudged by Byrne not to be an SVC (1987, 160). In accord with Byrne’s treatment of SVCs, Comrie (1995, 26) speciﬁes that “the verbs in the serial verb construction are interpreted as expressing a single event”, but he goes on to further qualify and explain this by stating that an SVC has two main functions:

(i) to allow the creation of new lexical items (ii) to allow the expression of more arguments of the predicate (Comrie 1995, 26)

Comrie exempliﬁes (i) with the case of the Sranan serial combination of dropu ‘drip’ and fadon ‘fall’ creating an equivalent of ‘drip down’, while (1.48) above is an example of an SVC which allows for the expression of more arguments of the predicate, since the verb t´ei ‘take’ introduces an instrument argument which would not otherwise be realized. (1.49) is an analogous instrumental SVC in Fongbe, but according to Lefeb- vre, (1.50), although presenting an identical sequence of verbs and objects, cannot

6While a sequential interpretation of taking the gun and then shooting the pig could be construed in the sentence ‘he shot the pig with a gun’, this sentence certainly does not entail any initial taking event (in contrast to biclausal ‘he took a gun and (then) shot the pig’). Chapter 4 explains and expands upon the idea that the most that necessarily precedes the ‘main’ action of shooting the pig in this sentence is volitional causation from the actor to instrument in Croft’s (1991) terms. CHAPTER 1. INTRODUCTION 29

constitute an SVC since, “as can be observed from the translation, this series does not refer to a single event, but rather to two successive events” (1991, 63):

(1.49) Kkú s àt´ın xò As´ıbá. Koku take stick hit Asiba

‘Koku hit Asiba with a stick.’ (Fongbe; Lefebvre 1991, 61)

(1.50) K k´u s c´ıc`ı kp x . Koku take glasses look house ‘Koku takes glasses and looks at the house.’ (Fongbe; Lefebvre 1991, 61)

Lefebvre (1991) appears to subscribe to the view that an SVC should describe a single, non-sequential event, then, and this also seems to be the belief held by Hale (1991) in relation to the contrast between clause chains and SVCs in Misumalpan languages. According to Hale, “the serial construction, unlike clause chaining, depicts a single cohesive event” (1991, 26). He exemplifies SVCs with the case of direction verbs combining with manner of motion verbs to describe ‘flying thither’ and ‘flying hither’; “neither verb in the series corresponds to a distinct event, and the verbs of direction serve merely to express just that, direction” (1991, 8). Some Misumalpan sentences can correspond to either SVCs or clause chains. (1.51), for example, is “ambiguous in regard to event cohesion” (Hale 1991, 26), since it could either have the single event (a) reading, or else the multiple event (b) reading:

(1.51) Witin ai pruk-an kauhw-ri. (Miskitu) Alas yaa-baut-ak wauhd-ikda. (Ulwa) he me strike-obv:3 fall-past:1 (a) ‘He knocked me down.’, (b) ‘He hit me and I fell down.’ (Hale 1991, 26)

Since (1.51) is regarded as an SVC under the single event reading but a clause chain under the multiple event reading, event structure is the sole determinant of SVC status under Hale’s approach.7

7Under the approach of this dissertation, however, (1.51) is disqualiﬁed from SVC status (even under the (a) reading) for the reason that this construction displays non-uniform person marking, which I regard as a clear enough grammatical sign of non-monoclausal behavior. Languages whose verbs display identical person marking on both verbs under all circumstances are displaying more convincing monoclausal behaviour. CHAPTER 1. INTRODUCTION 30

Now, it is by no means impossible that in some serializing languages, SVCs may be limited to describe only single non-sequential events. But it is certainly not the case that all SVCs in all serializing languages are limited in this way. In Section 1.2.4.2 it will be suggested that diﬀerent languages can show diﬀerent event structural limits on their SVCs. In fact, one serializing language whose event structural threshold turns out to be ‘higher’ (or ‘looser’) than single event status is actually Fongbe, since it transpires in Lefebvre and Brousseau (2002) that this language does have SVCs with

multiple event, sequential semantics, such as the following:8

(1.52) Kkú s àvún s -d`ın. Koku take dog def take-bury ‘Koku took the dog and buried it.’ (Fongbe; Lefebvre & Brousseau 2002, 400)

Fongbe SVCs are not limited to describing non-sequential, single events then, and as far as universal statements to the eﬀect that single event status should be a cross- linguistic deﬁning property of SVCs are concerned, even Comrie (1995) eventually shies away from this claim. Comrie appears to admit the following sequential, multi- event Haruai (Papua New Guinea) structure as an SVC, but remarks “it is not clear in what sense this sentence can be described as encoding a single event” (1995, 36):

w (1.53) Nob’-ak homlo rmal rag wl nm-o-a. man-that banana cut carry little eat-past:3sg-decl ‘That man cut some bananas, brought them, and ate a little.” (Haruai; Comrie 1995, 36)

Having surveyed a range of SVC researchers who equate SVC status with single, non-sequential unitary event structure, we can conclude that various expressions of this viewpoint need to be qualiﬁed, updated or corrected. At best, single non- sequential event status could be a language-speciﬁc characteristic of SVCs in a particular language, otherwise, SVCs with sequential, multiple event semantics need to

8This SVC, and others like it, fulfills the Fongbe-specific SVC tests described in Lefebvre and Brousseau (2002), which include: having one subject, only one temporal reference, only one occurrence of the definite future marker that has scope over the whole clause, and only one occurrence of the negation marker that has scope over the whole clause. CHAPTER 1. INTRODUCTION 31

be included in the class as well. This is a position which some of the sequential ‘recognizable event type’ SVCs included in Section 1.2.4.2 argue for, and it is also the stance taken by numerous SVC researchers who draw two-way distinctions in SVCs in such a way that multiple event sequential SVCs are included, as will be discussed in Section 1.2.4.3. To conclude this section, I will briefly mention the small number of SVC researchers who take the opposing standpoint from those who equate SVCs with single event status, by instead requiring SVCs to have multiple event status. Li and Thompson (1974, 265) identify SVCs with multiple event status in their statement that “true serial verb sentences ... are always of the form [Subject V (NP) V (NP)] where each of the verbs represents a separate action” (emphasis mine). Choi (2005, 35) also argues for multiple event status in SVCs, claiming that the verbs “form separate syntactic units and express separate events”, while Butt’s (1995) argument is more subtle, in claiming that “the several events described in a serial verb complex are stacked together to form a single, complex, conceptual event”9 (225). The requirement that an SVC have multiple event structure is less commonly encountered in the SVC literature than the requirement that an SVC have single event structure. But the most common stance, of which this dissertation is in favour, is for the class of SVCs to encompass both single event and multiple event SVCs. Section 1.2.4.3 surveys the trend to include both single event and multiple event SVCs in the class of SVCs, but to draw a distinction within SVCs based in these event structure terms (or in a precursor to event structure terms in the case of Christaller (1875)). But first, Section 1.2.4.2 will examine a constraint on serial verb formation with a deeper cognitive basis than some of the vague claims described in this section which equate SVCs with single events. 9The following SVC, with coidentified subevents, is held to express several events stacked together, indicating that this stacking can be very tight indeed:

Kofi hari a ston puru na ini a olo. Kofi pull the stone remove loc in the hole ‘Kofi pulled the stone out of the hole.’ (Sranan; Butt 1995, 222)

Butt (1995) is in fact primarily concerned with distinguishing SVCs from ‘complex predicates’; according to her, the former express several events, whereas the latter can only describe single, indivisible events. CHAPTER 1. INTRODUCTION 32

1.2.4.2 ‘SVCs describe recognizable event types’

The claim that serial verb expressions are conceptualized as events is supported by the observation that, in order for two verbs to combine as a serial verb, they must jointly describe a ‘recognizable event type’ (Durie 1997, Enfield 1998). This may be shaped by cultural factors; Durie hypothesizes that the reason why the serial verb combination of Yoruba verbs meaning ‘eat’ and ‘sleep’ is acceptable is because it is normal cultural practice for Yoruba speakers to rest after a meal, while a serial verb combination of Yoruba verbs meaning ‘eat’ and ‘return’ is not acceptable because eating outside the home is not the norm. Similarly, White Hmong speakers can express the notion of dancing and playing bamboo pipes in SVC form just because this constitutes a recognizable event type in White Hmong society (Jarkey 1991), while Alamblak speakers can express the notion of climbing a tree and looking for insects in SVC form because this constitutes a recognizable event type in Alamblak society (Bruce 1986). Aikhenvald (2006, 12) compares these constraints on SVCs to English verbal lexical compounds such as mountain-climbing or berry-picking, whose ‘name-worthiness’ seems to depend on how recognizable or typical the activities they describe are. Ladder-climbing, in contrast, is acceptable only in limited contexts where this is a typical, recurring activity. In a similar fashion, the use of denominal verbs such as blackberrying or shrimping (Clark and Clark 1979, Levin 1993) may also capture the flavor of these ‘recognizable event type’ SVCs (to the extent that Clark and Clark’s claim that these are “regulated on convention” and “on the basis of mutual knowledge” tallies with the notion of event ‘recognizability’). Adjectival compounds like bitter-sweet could also be compared to ‘recognizable event type’-constrained SVCs, although these are rather rare in English.10 In contrast to English, many languages display extensive co-compounding. According to Wälchli (2005), co-compounds contain two or more parts that are in a relationship of natural coordination, where natural coordination “implies, among other things, that the parts express semantically closely

10The only other adjectival compounds Marchand (1969) supplies (aside from technical combinations such as allegoric-didactic or phonetic-semantic) are from non-current literature, such as Shakespeare’s fortunate-unhappy, Thomas Wolfe’s haunting-strange, or Galsworthy’s noble-good. CHAPTER 1. INTRODUCTION 33

associated concepts, such as ‘brother and sister’, ‘hands and feet’, ‘eat and drink’, etc.” (2005, 1). For the purpose of drawing the closest possible comparison to ‘recognizable event type’ SVCs, verbal co-compounds such as (1.54) would presumably present the best ﬁt, although W¨alchli (2005, 124-6) does supply several arguments in support of keeping verbal co-compounds apart from what he regards as true cases of serialization.

(1.54) ... son pid’e.´s-pan.´s t’enst. she cook.pst3sg-bake.pst3sg dat3pl ‘... she prepared food for them.’ (Kemajkina; W¨alchli 2005,2)

One such argument is that SVCs typically have the semantics of temporal sequencing which co-compounds do not or cannot. Observations about how SVCs can be constrained by the notion of ‘recognizable event type’ point towards the tantalizing prospect that SVCs may provide a window into what an event – as conceived both linguistically and conceptually – actually is, yet at the same time it is uncertain whether the concept of ‘recognizable event type’ constrains the SVCs in all serializing languages. Consider the contrasting grammati- cality judgments for the Edo and Bima SVCs in (1.55) and (1.56) respectively; how could we account for the fact that the action of cooking and (then) eating can be expressed in an Edo SVC but not a Bima SVC, even though cooking and eating presents no more recognizable an event type in Edo society than in Bima society?:

(1.55) Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozocookedandatesomefood.’ (Edo;Stewart2001,10)

(1.56) Fero lowi *(ro) ngaha.na janga ede. Fero cook conj eat.3 chicken dem ‘Fero cooked and ate that chicken.’ (Bima)

Edo also allows many other serial verb combinations than Bima (and most other serializing languages) do not, such as ‘beat and buy (a drum)’ (Stewart 2001, 49) or ‘buy and read (LGB)’ (Stewart 2001, 60). Could SVCs in diﬀerent languages be CHAPTER 1. INTRODUCTION 34

constrained by different notions of ‘recognizable event type’? This seems possible, if hard to prove, although what is clear at this stage is that different serializing languages show different event structural limits on their SVCs. The SVCs in some languages, for example, are limited to describing single, non-sequential events, while the SVCs in other languages are event structurally ‘looser’, allowing the description of multiple events in sequence. At present it seems best to simply try and characterize and/or formulate the upper limits on SVCs in different languages in purely event structural terms, eschewing the notion of ‘recognizable event type’ unless the data explicitly argue for it. In Chapter 6 I attempt to characterize the upper bound on Bima SVCs in event structural terms, while also evaluating whether there may in fact be any such bound on Edo SVCs. But while these characterizations may make no reference to ‘recognizable event types’, the Yoruba (eating then sleeping), White Hmong (dancing and playing bamboo pipes) and Alamblak (climbing a tree and looking for insects) cases described above are valuable for the way in which they do clearly call the notion of ‘recognizable event type’ into account. Future research will no doubt uncover further cases of SVCs being constrained by ‘recognizable event type’. But for now one firm conclusion that can be drawn from the SVCs constrained in this way that we do have, such as (1.57), is that they lend weight to the claim that SVCs that describe multiple, sequential events (which we can see impressionistically do not have the semantic content of a clause headed by a single verb) need to be included in the class of SVCs:

(1.57) Myt ritm muh-hambray-an-m. tree insects climb-search.for-1s-3pl ‘I climbed the tree, looking for insects.’/‘I climbed the tree and looked for insects.’ (Alamblak; Bruce 1986, 26)

The next section investigates how certain two-way distinctions in the SVC literature ensure the inclusion of such combinations describing multiple, sequential events in the class of SVCs. CHAPTER 1. INTRODUCTION 35

1.2.4.3 Event-structure-based distinctions in SVCs

Almost every wide-ranging study of SVCs divides these constructions into two main categories, and this section surveys a small number of studies which do so along event structural lines. This section thus presents a kind of precursor to the Possible Verb- Impossible Verb SVC distinction – the distinction that I claim can tell us the most about SVCs, which will be event structurally characterized in Chapter 3, and which will take center stage in the case studies of Chapters 4 and 5. The first SVC researcher to draw an event structure-based distinction in SVCs was in fact the first person to discuss the phenomenon of serialization: Christaller (1875). Christaller (1875) draws a distinction between ‘accidental’ and ‘essential’ SVCs. According to him, accidental combinations involve “two or more predicates (verbs with, or without, complements or adjuncts), expressing different successive actions, or a state simultaneous with another state or action” which semantically resemble “sentences ... contracted into one”, and with “the verbs ... coordinate in meaning” (1875, 144). In an essential combination, on the other hand, “one verb is principal, and another is an auxiliary verb, supplying ... an adverb of time or manner, ... or forming or introducing a complement ... or adjunct; or the second verb is supplemental, forming part of a verbal phrase” (1875, 144). Whereas verbs in accidental combinations can express successive actions, Christaller specifies that the actions expressed by verbs in essential combinations must be simultaneous, they are in “an internal or inseparable relation or connection” (1875, 144), and moreover, “the auxiliary or supplemental verb is coordinate only in form, but subordinate in sense” (1875, 144). Since he was writing in the nineteenth century, Christaller (1875) does not refer to event structure notions, but what he is describing would seem to be consistent with essential SVCs describing single, non-sequential unitary events, whereas accidental SVCs describe multiple, possibly sequential, events. As a testament to Christaller’s insight, at least four modern studies of serial verbs cite this distinction and attempt to refine it. Under Sebba’s (1987) formal treatment of SVCs, accidental and essential combinations correspond to his syntactic distinction between ‘coordinating’ and ‘subordinating’ SVCs. Coordinating SVCs have the CHAPTER 1. INTRODUCTION 36

syntactic make-up of coordinated VPs and semantically “their distinguishing characteristic is that they refer to several actions” (1987, 110). In a subordinating SVC, on the other hand, “although two or more verbs are present, the sentence is interpreted as referring to a single action rather than a series of related actions” (1987, 112). While coordinating SVCs have a multiple event interpretation, subordinating SVCs have a single event interpretation. Stewart (2001), meanwhile, draws a distinction between ‘consequential’ and ‘resultative’ SVCs. ‘Consequential’ SVCs are Stewart’s version of accidental SVCs; they are the SVCs “in which the verbs express a natural sequence of events, and [their temporal ordering follows a] precedence-consequence iconic relation” (2001, 14). In a ‘resultative’ SVC, in contrast, “the action of the first verb brings about the result that is denoted by the second verb. This implies that there is a strict cause-effect relationship between the verbs and there is no time lapse between the eventualities that they express” (2001, 13). Whereas consequential SVCs are sequential and are made up of two events, resultative SVCs are non-sequential and are made up of one event (2001, 87). Two further two-way distinctions in SVCs which invoke event structure terms and align themselves with Christaller’s accidental-essential distinction are provided by Hellan et al. (2003) and Osam (1994). I will not describe these, but will instead mention one final two-way distinction in SVCs, for which Christaller’s distinction is not a precedent, and the primary basis for which is not an event structure distinction. The primary basis for Aikhenvald’s (2006) ‘asymmetrical’-‘symmetrical’ distinction in SVCs is the issue of whether both verbs in the SVC are drawn from a semantically and grammatically unrestricted open class, as in ‘symmetrical’ SVCs, or whether one of the verbs is drawn from a grammatically or semantically restricted closed class, as in ‘asymmetrical’ SVCs. (1.58) below would be classified as a symmetrical SVC, since both rumpa ‘find’ and weha ‘take’ are drawn from an open class, whereas (1.59) would be classified as an asymmetrical SVC, since lu’u ‘enter’ is drawn from the closed class of direction verbs:

(1.58) Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima) CHAPTER 1. INTRODUCTION 37

(1.59) Ana kea rama lu’u.na d’ei kota. baby crawl enter.3 in box ‘The baby crawled into the box.’ (Bima)

Although the open or closed status of the verbs involved is the primary basis for the symmetrical-asymmetrical distinction, there are strong event structure correspon- dences for each type of SVC. Symmetrical SVCs for the most part have multiple, sequential event structure, in which “the order of components tends to be iconic, reflecting the temporal sequence of subevents” (Aikhenvald 2006, 22). Asymmetrical SVCs, on the other hand, describe a single, unitary event. According to Aikhenvald, the verb from the open class denotes this single event, while the verb from a closed class provides a modificational specification, which could pertain to motion, posture or tense-aspect (2006, 21). Just taking into account the number of publications that refer to it, the asymmetrical-symmetrical distinction is probably the most frequently invoked two-way distinction of all (although it must be said that it is mostly recent studies of Oceanic SVCs that build this distinction into their descriptions). But there are some problems with the asymmetrical-symmetrical distinction, centering upon the difficulty of establishing whether certain items are open or closed class. These problems will be discussed further in Chapter 3, along with the problems of some of the other two-way distinctions in SVCs.

1.2.4.4 Summary

For the sake of providing the briefest of summaries to this section on the event structure of SVCs I will simply state the main points to be gleaned from each subsection. With respect to Section 1.2.4.1: various researchers have attempted to define and restrict SVCs in event structure terms – usually through the rather non-specific claim that SVCs must describe a ‘single event’. I argued against defining SVCs in this way. With respect to Section 1.2.4.2: data showing how SVCs are constrained by the notion of ‘recognizable event type’ is serial verb ‘gold’ – extremely interesting and elucidating, but seemingly so rare as to not be applicable in most cases. I suggested that SVCs may be constrained in different ways, showing different event structural CHAPTER 1. INTRODUCTION 38

limits. With respect to Section 1.2.4.3: time after time, SVC researchers have apparently felt compelled to divide the class of SVCs into two types, usually along event structural lines. I want to draw a two-way distinction among SVCs too, although not (exclusively) along event structural lines. The Possible Verb-Impossible Verb distinction will be discussed in Chapter 3.

1.2.5 Conclusion

This section has not just provided a literature review of SVCs for its own sake. Instead this review has been specifically tailored to the purposes, themes and methodologies of this dissertation. The purpose of Section 1.2.1 was to sift out, from all the claims about how to define and identify SVCs, a definition for SVCs to use in this dissertation. The result was the following list of core SVC-defining properties:

(a) SVCs are monoclausal.

(b) The verbs in an SVC must have the same tense, aspect, mood, polarity, and person-number categories.

(d) The intonational properties of a clause containing a serial verb combination are the same as those of a clause containing a simple predicate in the language concerned.

But there is one property still absent from this list, which relates directly to the core message of this dissertation that SVCs need to be defined through argument structural properties. Section 1.2.2 is the most crucial part of this literature review for the reason that it articulates and supports this message. This section argues that simply listing off all the possible argument sharing schemes SVCs can display (as is typical in more typologically-oriented SVC research) is not quite good enough for the purposes of SVC definition, while the property of Internal Argument Sharing, although it goes a long way towards enforcing monoclausal argument structure in SVCs, has some CHAPTER 1. INTRODUCTION 39

fatal flaws that render it unsuitable. The next section describes the Theta Structure Property for SVCs, which avoids these flaws. Section 1.2.2 thus contains the germ of the Theta Structure Property, the core proposal for this dissertation. While Section 1.2.3 provided only a brief survey of syntactic approaches to SVCs and served mainly to emphasize that contiguous, as well as non-contiguous, structures should be included in the class of SVCs, Section 1.2.4, on the event structure of SVCs, contained much more depth. This is fitting in the light of the event structural focus of this dissertation. Event structural analysis will be the main mode of analysis in the instrumental and motion SVC case studies of Chapters 4 and 5, mainly for the reason that the Possible Verb-Impossible Verb distinction will be analyzed in event structural terms. Section 1.3 now describes this core distinction – and the core Theta Structure Property – in more detail.

1.3 Prospectus

Sections 1.3.1 and 1.3.2 of this concluding section discuss the two main claims of this dissertation: the claim that the Theta Structure Property for SVCs should be used in the deﬁnition of SVCs, as will be expanded upon in Chapter 2, and the claim that two types of SVC should be distinguished – Possible Verb and Impossible Verb SVCs – as will be expanded upon in Chapter 3. Section 1.3.3 then sets the course for the three chapters to follow, by discussing how the Theta Structure Property and the Possible Verb-Impossible Verb distinction bear upon instrumental SVCs (the topic of Chapter 4) and motion SVCs (the topic of Chapter 5), and what the wider implications of including Impossible Verb SVCs in the class of SVCs are (Chapter 6).

1.3.1 The Theta Structure Property for SVCs

SVCs are multi-verb sentences which are monoclausal in every surface grammatical respect and one respect which should not be overlooked is the projection of monoclausal argument structure. Internal Argument Sharing is the theoretical device from CHAPTER 1. INTRODUCTION 40

the formal literature which was observed in Section 1.2.2 to enforce monoclausal argument structure up to an extent, but where it fails is in the exclusion of SVCs such as the following, which are monoclausal in every respect, including their argument structure:

(1.60) Taho mpara lao te’e.ku tamba. ﬁne then go install.1 trap ‘Okay then, I’ll go and install a trap.’ (Bima)

(1.61) Dadao ri-a maa moon, eori ma ani. call:red 3pl-art pl woman 3pl come eat

‘Call the women over so that they come and eat.’ (Teop; Reinig 2004, 102) (1.62) Kkú kló hú As´ıbá.` Koku be.big surpass Asiba ‘Koku is bigger than Asiba.’ (Fongbe; Lefebvre & Brousseau 2002, 401)

(1.63) Ne-len mi-yip hal-yak na.kat. art-wind perf-blow ﬂy-away art-cards ‘The winds blew the cards away.’ (Mwotlap; Francois 2004, 119)

(1.64) Wela e qiliano-na taqaa baqu. child 3sg:nfut pile.soil.around-3:obj be.bad banana ‘The child piled the soil around the banana tree badly.’ (Toqabaqita; Aikhenvald 2006, 29)

This dissertation puts forward the Theta Structure Property for SVCs as the property which enforces monoclausal structure without exception. Like Internal Argument Sharing, it excludes covert coordinations from the class of SVCs, but it also succeeds where Internal Argument Sharing fails by including (1.60)-(1.64). The Theta Struc- ture Property will be stated and illustrated extensively in Chapter 2. Here I will present only the bare bones of the property, without going into any of its minor details. The Theta Structure Property is formulated in the framework of Lexical Decom- position Grammar (LDG) and the feature of LDG which is the most crucial to the Theta Structure Property is that it posits a level of Theta Structure, which is where CHAPTER 1. INTRODUCTION 41

a monoclausal projection of arguments is to be located. That part of (1.65b) enclosed within a box is the Theta Structure (TS); the remaining part of (1.65b) is the Se- mantic Form (SF) representation, a partial semantic representation that presents all the grammatically relevant information for a lexical item:

(1.65) a. Ana to’i b’ale ka.lu’u.na bola d’ei uma. child small throw caus.enter.3 ball in house ‘The child threw the ball into the house.’ (Bima) b. b’ale λz λy λx λs {THROW(x,y) & BECOME(LOC(y,IN(z)))}(s) ka.lu’u: −sc +sc +sc

Theta Structure consists of a list of λ-abstracted arguments, and these are matched with [±sc] (structural case) specifications in (1.65b), indicating whether the arguments are structural arguments or not. A structural argument is one that is projected into one of the core grammatical relations of the clause, such as subject or object; the grounds for determining the [±sc] status for an argument will be explained in Section 2.2. While structural ([+sc]) arguments are projected into syntactic positions such as subject or object, non-structural ([-sc]) arguments, if not suppressed (i.e. linked to ∅ and not realized syntactically), are realized in oblique positions in the syntax. (1.65) has two structural arguments (realized as ana to’i ‘the child’ and bola ‘the ball’) and one non-structural argument (realized as uma ‘the house’). What the Theta Structure Property for SVCs requires is for an SVC to display the same number of structural arguments and the same list of non-structural, oblique arguments as some clause headed by a single verb. Structural arguments are blind to thematic role information, but non-structural arguments are identified with thematic roles through the particular configurations that define them at Semantic Form. (1.65) satisfies the Theta Structure Property for SVCs because its Theta Structure consists of two structural arguments and one non-structural argument which will be configurationally defined as a ‘containing goal’ in Chapter 2, and there do exist clauses headed by single verbs (such as the English translation for (1.65a)) which likewise consist of two structural arguments and a ‘containing goal’ non-structural argument. (1.66) below satisfies the Theta Structure Property for SVCs for all the same reasons, but CHAPTER 1. INTRODUCTION 42

there are some major differences from (1.65). Firstly, (1.66) is a non-contiguous SVC, which refers to the fact that the verbs are non-contiguous in allowing objects (in this case è. khù ‘door’) to intervene between them. (1.65), in contrast, is a contiguous SVC, since bola ‘ball’ cannot be realized between the two verbs (b’ale ‘throw’ and ka.lu’u ‘make enter’). Secondly, the ‘containing goal’ non-structural argument òwá ‘house’ in (1.66) is realized not in an oblique phrase headed by a preposition (as in d’ei uma ‘in the house’ in (1.65)), but as the object of the second verb, làá ‘enter’:

(1.66) a. Ozó` gbé è. khù làá òwá. Ozo hit door enter house ‘Ozo hit the door into the house.’ (E`. dó; Stewart 2001, 145) b. gbélàá: λz λy λx λs {HIT(x,y) & BECOME(LOC(y,IN(z)))}(s) −sc +sc +sc

At the heart of the difference between the contiguous SVC in (1.65) and the non- contiguous SVC in (1.66) is the fact that (1.66), unlike (1.65), displays SVC-specific clause structure. Whereas the arguments in (1.65) are projected into the syntax in exactly the same way they would be in a simplex clause in some language, this is hardly true of (1.66), and this is why, in addition to stating and illustrating the Theta Structure Property for SVCs, Chapter 2 also states and illustrates the argument realization generalizations for non-contiguous SVCs, to which there are three major strands: (i) they account for which argument is realized as the complement of which verb (ensuring, for example, that è.khù ‘door’ in (1.66) is the complement of gbé ‘hit’ and not làá ‘enter’); (ii) they account for how some non-structural arguments, such as òwá ‘house’ in (1.66), come to look as if they are structural, in being realized as bare NP complements of a verb; and (iii) in an extended sense they account for the very existence of the contiguous versus non-contiguous distinction in SVCs, by generalizing across the ways in which non-contiguous SVCs diverge in displaying SVC- specific clause structures. It is this distinction which, as mentioned in Section 1.2.3, most other approaches to SVCs have failed to acknowledge. It is important that the contiguous versus non-contiguous distinction in SVCs be acknowledged, and it is important the Theta Structure Property be able to enforce monoclausal argument structure in both contiguous and non-contiguous SVCs. CHAPTER 1. INTRODUCTION 43

Chapter 2 demonstrates how it does so, as well as demonstrating how this property includes in the class of SVCs the structures in (1.60)-(1.64) above that Internal Argument Sharing undesirably excludes.

1.3.2 The Possible Verb-Impossible Verb distinction in SVCs

While Chapter 2 contains the first, core claim for this dissertation, that the Theta Structure Property for SVCs should be a major component of how SVCs are defined, Chapter 3 contains a claim that follows on from the Theta Structure Property and concerns what the class of SVCs actually looks like. The core, basic class of SVCs which satisfy the Theta Structure Property for SVCs straightforwardly are here termed ‘Possible Verb’ SVCs for reasons to be explained shortly; what Chapter 3 argues for is the recognition of an additional class of ‘Impossible Verb’ SVCs, whose satisfaction of the Theta Structure Property is slightly more complex. Possible Verb SVCs are so called because they satisfy LDG’s ‘Possible Verbs’ constraints. Although the Theta Structure Property for SVCs need not be tied to any particular framework, if there is one reason for implementing this property in the LDG framework that stands out above all others, it is that LDG’s ‘Possible Verbs’ constraints can be utilized. These constraints have already been described as encapsulating the semantic content of what can be expressed in a clause headed by a single verb, and they will be stated in full in Section 2.2 and demonstrated more extensively in Chapter 3. Chapter 3 also demonstrates how Impossible Verb SVCs, in contrast to Possible Verb SVCs, do not fulfill the Possible Verbs constraints. The basis for the Possible Verb-Impossible Verb distinction is fulfillment (or non-fulfillment) of these Possible Verbs constraint then, and this distinction has two main reflexes. The first main reflex of the Possible Verb-Impossible Verb distinction is that Pos- sible Verb SVCs will have translation equivalents which are clauses headed by single verbs, whereas Impossible Verb SVCs cannot have (accurate) translation equivalents which are clauses headed by single verbs. This is because the Possible Verbs constraints determine which Semantic Form representations can be realized as single simplex clauses headed by single verbs, as just mentioned, and the broader significance CHAPTER 1. INTRODUCTION 44

of this fact is the key insight of the Possible Verb-Impossible Verb distinction that some SVCs (Possible Verb SVCs) are doing what would be done by clauses headed by single verbs in other languages whereas other SVCs (Impossible Verb SVCs) are not. Semantic Form representations which do fulﬁll the Possible Verbs constraints, and therefore can correspond to single-verb-headed clauses, appear within {} braces in this dissertation. The {} braces themselves can thus be understood to be representative of Possible Verb structure. (1.67a) (=(1.6)), for example, is a Possible Verb SVC because it fulﬁlls the Possible Verbs constraints, as indicated by the single set of {} braces in (1.67b), and it moreover has a single-verb-headed translation in (1.67a):

(1.67) a. Ana kea rama lu’u.na d’ei kota. baby crawl enter.3 in box ‘The baby crawled into the box.’ (Bima) b. rama lu’u: λy λx λs {CRAWL(x) & BECOME(LOC(x, IN(y)))}(s) −sc +sc

(1.68a) (=(1.7)), in contrast, is an Impossible Verb SVC because it does not fulﬁll the Possible Verbs constraints, as the two sets of {} braces in (1.68b) are intended to indicate, and it cannot have a single-verb-headed translation equivalent:

(1.68) a. Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10)

b. léré: λy λx λs1 {COOK(x,y)}(s1) & λv λu λs2 {EAT(u,v)}(s2) +sc +sc +sc +sc k k k k Predicate Combining Operations for Impossible Verb SVCs ⇓ ⇓ ⇓ ⇓ c. léré: λy λx λs [{COOK(x,y)} & {EAT(x,y)}](s) +sc +sc

The second main reﬂex of the Possible Verb-Impossible Verb distinction is that, whereas simplex Theta Structure arrays of arguments that satisfy the Theta Struc- ture Property for SVCs are arrived at quite straightforwardly in Possible Verb SVCs CHAPTER 1. INTRODUCTION 45

like (1.67), Impossible Verb SVCs like (1.68) are argument structurally much more complex. Corresponding to every Possible Verbs ({}) structure at Semantic Form is a distinct Theta Structure (the latter being λ-abstracted from the former). This means that Impossible Verb SVCs, in containing more than one Possible Verbs structure, will initially contain more than one Theta Structure, as in (1.68b). Some amount of work is then required to account for how Impossible Verb SVCs come to display monoclausal argument structure. This ‘work’ takes the shape of the Predicate Combining Operations for Impossible Verb SVCs alluded to in (1.68), that produce a single, composite Theta Structure (i.e. one set of λ-abstracted elements as in (1.68c)) that does encapsulate a monoclausal projection of arguments, and so does satisfy the Theta Structure Property for SVCs. The Predicate Combining Operations for Impossible Verb SVCs (mostly comprised of particular argument coidentiﬁcation operations) will be stated in Chapter 3, followed by demonstrations of how they produce Theta Structures for Impossible Verb SVCs that satisfy the Theta Structure Property for SVCs. I then argue why these operations should be utilized to include this ‘non-core’ type of SVC in the class of SVCs. Ultimately, the reason for the inclusion of Impossible Verb SVCs is the fact that they appear monoclausal in every sense, including this argument structural sense, even if the Predicate Combining Operations do add extra complication. Once Impossible Verb SVCs are accepted into the class, Section 3.5 of Chapter 3 turns to the consequences of their inclusion, which for the sake of brevity I will summarize through the headings of the four subsections of Section 3.5:

(a) SVCs (i.e. Impossible Verb SVCs) are not just doing the things that would be done with clauses headed by single verbs in other languages

(b) SVCs (i.e. Impossible Verb SVCs) do not just describe single (non-sequential) events

(c) Only by including Impossible Verb SVCs can we see how SVCs can be constrained by the notion of ‘recognizable event type’ CHAPTER 1. INTRODUCTION 46

(d) The Possible Verb-Impossible Verb distinction is superior to other two-way distinctions in SVC studies

Of these four consequences of including Impossible Verb SVCs in the class of SVCs it is (a) which is the most transparent, and which has already been outlined above, as the first main reflex of the Possible Verb-Impossible Verb distinction. Consequence (d) is transparent in the sense that the inclusion of Impossible Verb SVCs will of course result in yet another two-way distinction being drawn in SVCs, and in Chapter 3 I discuss the advantages and insights the Possible Verb-Impossible Verb distinction affords over other two-way distinctions in SVCs. Consequences (b) and (c) are tied up with event structural concepts. Regarding (b), Section 1.2.4.1 in essence complained that, although numerous SVC studies have claimed that SVCs describe single events, these studies are seldom clear about what a ‘single event’ is. Chapter 3 tries to eliminate any such unclarity by analyzing and classifying SVCs in an event semantic framework. Event structural templates and descriptions are provided for Possible Verb SVCs on the one hand, and for Impossible Verb SVCs on the other, which, in addition to setting up a formal framework in which to address the identification of ‘single events’, serve to corroborate and support the Possible Verb-Impossible Verb distinction itself. Event structural analysis features strongly in the case studies of Chapters 4 and 5, to which Section 1.3.3 now turns, although Section 1.3.3 also addresses issues pertaining to consequence (c) above about how (or whether) SVC formation is limited, and how the notion of ‘recognizable event type’ bears upon this.

1.3.3 In search of Impossible Verb SVCs, and their limits

Some Impossible Verb SVCs stand out as never possibly satisfying the Possible Verbs constraints, as never being able to be alternatively expressed in a clause headed by a single verb. This is true of all the Impossible Verb SVCs included in this chapter, Chapter 2 and Chapter 3; the semantic content of (1.69), for example, could never be accurately or completely expressed in a clause headed by a single verb: CHAPTER 1. INTRODUCTION 47

` (1.69) Ozó kpèé èmà dé. . Ozo beat drum buy ‘Ozo beat the drum (and then) bought it.’ (Edo; Stewart 2001, 49)

SVCs such as (1.69) contain full lexical verbs which are semantically independent of each other and which do not obviously belong to some common semantic domain (such as the domain of motion). These kinds of SVCs seem to capture Christaller’s (1875) sense of what it is for an SVC to be ‘accidental’; the verbs are not working together within a single semantic domain – instead it seems more an accident that they have ever come together in the ﬁrst place. Because SVCs like (1.69) stand out for their impossibility to ever be alternatively expressed in a clause headed by a single verb, the search for Impossible Verb SVCs of this kind is not a diﬃcult one. But when the verbs in an SVC appear not entirely semantically independent of each other, or when they are drawn from a single semantic domain, then the determination of Impossible Verb status can be less clear-cut. The verbs in (1.70), for example, may be semantically dependent to the extent that tei ‘take’ could be said to supply an instrumental component of meaning within the overall shooting event, while the verbs in (1.71) are both drawn from the domain of motion, and may also not be completely semantically independent of each other:

(1.70) A tei goni suti di pingo. he take gun shoot the pig ‘He shot the pig with a gun’, or ‘He took the gun and (then) shot the pig.’ (Saramaccan; Byrne 1985, 281-2)

(1.71) Er´ı, ok´ı mu to. ru. be..in-mi.. 3sg swim go river cross-past ‘He swam across a river.’/‘He went swimming and crossed a river.’ (Ijo; Williamson 1963, 152)

(1.70) is an instrumental SVC and (1.71) is a motion SVC, but something these two SVCs have in common is that they both receive two alternative translations: one (the ﬁrst in each case) suggestive of Possible Verb semantics and another in which the word(s) ‘and (then)’ are suggestive of Impossible Verb semantics. These alternative CHAPTER 1. INTRODUCTION 48

translations underscore the point that Possible Verb and Impossible Verb SVCs can be difficult to distinguish in such instances. So, while the search for Impossible Verb SVCs may not be difficult where the more ‘accidental’ SVCs like (1.69) are concerned, this search becomes more difficult when confined to the more narrow domains of instrumental SVCs and motion SVCs. Instrumental SVCs are explored in Chapter 4 and motion SVCs in Chapter 5. In some respects, the aims common to these case studies are of a very general nature: to describe what happens in an event involving an instrument participant on the one hand, or what happens in a motion event on the other, and to illustrate how different languages express these events in SVC-form, utilizing different structural and lexical options. Other aims common to these case studies, in contrast, are very specific to the proposals and claims of this dissertation. Both case studies aim to demonstrate how the respective types of SVC satisfy the Theta Structure Property for SVCs. But in each case, the manner in which any given SVC will satisfy the Theta Structure Property will vary markedly, depending on whether the SVC has Possible Verb or Im- possible Verb status. Impossible Verb SVCs are subject to the Predicate Combining Operations for Impossible Verb SVCs, as mentioned in the previous section. Determi- nation of Possible Verb or Impossible Verb status is of essential practical importance for the purposes of demonstrating how an SVC satisfies the Theta Structure Prop- erty, then, and the instrumental and motion SVCs in (1.70) and (1.71) above illustrate how there can be a very fine line between Possible Verb and Impossible Verb status in these domains. It is in the interests of supporting the Possible Verb-Impossible Verb distinction itself that this fine line should be examined. Some of the factors that determine Impossible Verb status will be shown to be domain-specific, although one feature both case studies have in common is that event structural analysis is brought to the fore, as a way of supporting and corroborating the Possible Verbs constraints that determine Possible Verb or Impossible Verb status. While the ultimate result of testing out the Possible Verb-Impossible Verb distinction with instrumental and motion SVCs like (1.70) and (1.71) is the validation and fortification of the distinction itself, this does not spell the end for the consideration of the more blatant, ‘accidental’ Impossible Verb SVCs such as (1.69) above and (1.72): CHAPTER 1. INTRODUCTION 49

` (1.72) Ozó dé. LGB tié. Ozo buy LGB read ‘OzoboughtLGBandreadit.’ (Edo;Stewart2001,60)

For, of the Impossible Verb SVCs that scream out that they could never possibly satisfy the Possible Verbs constraints, some, such as (1.69) and (1.72), may scream rather too loudly. (1.69) and (1.72) are highly unlikely to describe ‘recognizable event types’ in the society in which they are spoken. Section 1.2.4.2 described three – and only three – cases of SVC formation being constrained by the notion of ‘recognizable event type’: there was the case of Alamblak, in which the concept of climbing a tree and looking for insects could be expressed in SVC-form because this constitutes a recognizable event type in this society; White Hmong, where the concept of dancing and playing bamboo pipes could be expressed in SVC-form because this constitutes a recognizable event type in this society; and Yoruba, where the concept of eating then sleeping could be expressed in SVC-form for the same reasons. Observations of this kind arguably present the most interesting answer to the most interesting question about SVCs. In Section 1.1 the most interesting question about SVCs was stated as: why are SVCs behaving as if they contain only one verb when they actually contain at least two? But including Impossible Verb SVCs in the class of SVCs actually entails a more speciﬁc and interesting question: why are Impossible Verb SVCs behaving as if they contain only one verb? Why would an SVC whose semantic content could never even be expressed in a clause containing a single verb behave as if it only contains one verb? The answer that an Impossible Verb SVC is behaving monoclausally for the reason that it constitutes a ‘recognizable event type’ is clearly applicable in some cases, although it is less clear whether this applies to all Impossible Verb SVCs in all languages (such as (1.69) or (1.72), for example). This dissertation takes only a few steps in the direction of determining what constrains (Impossible Verb) SVC formation across languages. But I also ask: should all SVCs be constrained by some kind of (language-speciﬁc) event conceptualization? Or can SVCs be event structurally completely unconstrained, able to combine any number of loosely connected events, as would seem to be the case in the following Saramaccan SVC (=(1.73))?: CHAPTER 1. INTRODUCTION 50

(1.73) A k´ısi d´ı fou náki k´ıi limbó bói njan. 3sg catch det bird hit kill clean cook eat ‘He caught the bird, struck it dead, cleaned, cooked and ate it.’ (Saramaccan; Veenstra 1996, 103)

Is (1.73) (or (1.69), or (1.72)) really an SVC? These are questions for future research which can receive only speculative answers in this dissertation, which is why issues of recognizable event types and event conceptualization for the most part will not be discussed again until Chapter 6. Chapter 2

The Theta Structure Property for SVCs

2.1 Introduction

Chapter 1 promised a definitive way to ensure that SVCs display monoclausal argument structure and now it is time to deliver on this promise, by setting out the Theta Structure Property for SVCs in Section 2.3 below. But first, the framework in which this property is formulated must be described, and so Section 2.2 explains some of the core principles of Lexical Decomposition Grammar (LDG). The aspect of LDG which is most crucial to the Theta Structure Property for SVCs is the distinction between structural and non-structural arguments, and so Section 2.3, after stating the Theta Structure Property, demonstrates how this property plays out with respect to these two types of argument separately. Although the main purpose of this chapter is to state and illustrate the Theta Structure Property for SVCs, there is also a secondary focus: Section 2.4 demonstrates how SVCs, defined and identified through the Theta Structure Property at LDG’s levels of Semantic Form (SF) and Theta Structure (TS), can map into different phrase structures, giving rise to the two distinct syntactic patterns of non-contiguous SVCs, such as (2.1), and contiguous SVCs, such as (2.2):

51 CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 52

` (2.1) E. sósà gbé émá!tó. n pè. rhé. . Esosa hit metal be-flat ‘Esosa hit the metal flat.’ (Edo; Stewart 2001, 58)

(2.2) Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima)

The projection of arguments into the syntax is one of the most diﬃcult aspects of SVC analysis. Non-contiguous SVCs, which constitute the traditional domain of investigation for SVCs, pose the most serious challenges. One challenge, as mentioned in the introduction to Chapter 1, is to account for the realization of peripheral arguments in non-contiguous SVCs like (2.3), where the instrument has a structural appearance in the surface syntax similar to that of any core argument:

(2.3) Nws muab riam txiav nqiaj qaib. 3sg take knife cut meat chicken ‘She cut some chicken with a knife.’ (White Hmong; Jarkey 1991, 83)

Section 2.4.1 tackles these issues by generalizing across argument realization patterns for non-contiguous SVCs. These generalizations capture the fact that non-contiguous SVCs can display a clause structure which is SVC-specific in the sense discussed in Sections 1.1 and 1.3, and it is this SVC-specific clause structure, combined with the propensity for certain verbs in non-contiguous SVCs to perform the functions we would usually associate with adpositions, that produces the effect of non-structural arguments appearing structural. Contiguous SVCs, in contrast, appear not to display SVC-specific clause structure (but instead display essentially simplex clause structure), although the main topic under investigation in Section 2.4.2 is the question of whether the clause structure of these SVCs might ever be SVC-specific. To conclude this chapter, Section 2.5 summarizes the workings of the Theta Struc- ture Property for SVCs and demonstrates how it presents an improvement upon In- ternal Argument Sharing as a defining property for SVCs. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 53

2.2 Lexical Decomposition Grammar

The Theta Structure Property for SVCs is formulated in the framework of Lexi- cal Decomposition Grammar (Wunderlich 1997a, Wunderlich 1997b, Kaufmann and Wunderlich 1998, Wunderlich 2000) and SVCs will be semantically analyzed within this framework throughout this dissertation. The Theta Structure Property for SVCs is not exclusively tied to the LDG framework, and could be reformulated in some other framework, yet the distinction drawn between structural and non-structural arguments is one aspect of LDG which allows for ease of formulation of this property. Just as important, the Possible Verbs constraints of this framework allow straightforward illustration of those SVCs which are semantically equivalent to clauses headed by single verbs and those which are not – the property which forms the basis for the distinction between Possible Verb and Impossible Verb SVCs to be discussed in Chap- ter 3. The current discussion of LDG draws primarily upon Wunderlich (2000), which provides one of the most comprehensive and recent discussions of this framework. LDG consists of four levels of representation, each with distinct structural properties, and a set of principles which constrain the mappings between these levels. The four levels are Conceptual Structure (CS), Semantic Form (SF), Theta Structure (TS), and Morphology/Syntax (MS). Wunderlich (2000, 250) provides the following illustration of this four-level architecture for the German verb geben ‘give’:

(2.4) The four levels of LDG

TS SF CS λz λy λx λs {ACT(x) & ⇒ ⇐ x= Agent or Controller +hr +hr -hr BEC POSS(y,z)}(s) y=Recipient -lr +lr +lr z=Patient or Aﬀected ↓ ↓ ↓ Causal event: ACT(x)(s1) agr Result state: acc dat nom POSS(y,z)(s2)

Semantic Form plays the most crucial role among these levels. On the one hand it CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 54

represents a streamlined version of Conceptual Structure, stripping away certain information that could be inferred (such as event structural and thematic information). On the other hand it provides the base from which thematic roles are λ-abstracted at Theta Structure. Some of these λ-abstractors correspond to structural arguments, and these λ-abstractors are annotated at Theta Structure with the features [±hr] (‘there is/isn’t a higher role’) and [±lr] (‘there is/isn’t a lower role’), which determine the mappings to case roles at Morphology/Syntax . Theta Structure is the level at which argument structure operations such as passive apply, and it also has the autonomy to alter the sequences of λ-abstracted arguments inherited from Semantic Form in order to account directly for expletive arguments, dative experiencers and instantiations of quirky case. But concentrating on Semantic Form, it is at this level that partial semantic representations for lexical items are formulated, by means of a binary categorial syntax. Even the conjunction ‘&’ involves (asymmetric) binary structure; so ACT(x) & BEC POSS(y,z) in (2.4) above could be more accurately notated [ACT(x) [& BEC POSS(y,z)]]. (Semantic Form formulations will also be represented in tree form later in this section.) Semantic Form representations otherwise have much in common with many other types of predicate decomposition in the literature; compare, for example, the following representations of the resultative complex predicate drink (the bottle) empty:

(2.5) a. DRINK(x,z) & BECOME(EMPTY(y)) (Wunderlich 1997b, 36) b. CAUSE(DRINK(x,z), BECOME(EMPTY(y))) (Dowty 1979, 93) c. CAUSE(ACT(x,y), BECOME(EMPTY(y))) BY DRINK(x,z) (Jackendoﬀ 1990, 232)

The one major point of departure for the LDG representation for drink empty in (2.5a) is the absence of a CAUSE predicate. A causal relation between the subevents of drinking and emptying the vessel is, however, inferred (as a consequence of the Coherence constraint of LDG to be introduced shortly), and need not be made explicit through a CAUSE predicate. (2.5a) illustrates the basic shape Semantic Form representations take; the remainder of this section will consider the constraints on the CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 55

mapping between Semantic Form and Conceptual Structure on the one hand and between Semantic Form and Theta Structure on the other. The former are the ‘Possible Verbs’ constraints (Kaufmann 1995) and the latter are the ‘Argument Hierarchy’ and ‘Structural Argument’ constraints. While Semantic Form representations are partial semantic representations seen as part of the grammar, Conceptual Structure representations are richer semantic representations seen as part of the extra-linguistic conceptual system. This division into two semantic levels is an innovation attributed to Bierwisch (1983) in the LDG literature. The primary constraint in the mapping between these two levels is the Possible Verbs constraint, ﬁrst proposed by Kaufmann (1995):

(2.6) Possible Verbs. In a decomposed SF representation of a verb, every more deeply embedded predicate must specify the higher predicate or sortal properties activated by the higher predicate. (Wunderlich 2000, 25)

(2.6) requires that in a coordination structure [A [& B]], B must further specify A, or sortal properties activated by A. Another way of describing this is to say that the head predicate A opens up a possible interpretation space, and B specifies certain options within this space, where “the interpretation space of a predicate is determined by both the predicate itself and the sortal properties of its arguments that are necessary in order for the predicate to apply” (Kaufmann and Wunderlich 1998, 5). The positional predicate STAND, for example, opens up an interpretation space whose options (specified by additional predicates at Semantic Form) could pertain to location (as in stand in the corner) or gestalt properties (as in stand upright). As another example, the transitive predicate CUT(x,y) opens up an interpretation space involving “destruction of the integrity of y through a particular type of physical activity of x” (Kaufmann 1995), and the sortal properties it activates pertain to the integrity of y (that it is an object that can undergo cutting) and the physical properties of x (that it is either sharp, if inanimate, or can manipulate a sharp object if animate). Kaufmann and Wunderlich (1998) use the following minimal pair, whose members differ only in which argument is realized as subject, to demonstrate how the Possible CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 56

Verbs constraint works:

(2.7) a. Ein Motorrad/ein Fuerwerkskörper heulte um die Ecke. a motorbike/a firework howled around the corner ‘A motorbike/a firework howled around the corner.’ b. *Ein Junge heulte um die Ecke. a boy howled around the corner ‘A boy howled around the corner.’

(2.7a) and (2.7b) both describe motion events, but the reason why (2.7a) is acceptable, in contrast to (2.7b), is that heulte ‘howl’, the emission of sound, is interpreted as being caused by the motion the motorbike or ﬁrework undergoes. Because the howling is interpreted as a kind of (internally caused) by-product of the movement of these objects, HOWL can in these cases further specify the properties in the domain activated by MOVE in the following structure:

(2.8) {MOVE(x) & HOWL(x) & AROUND-THE-CORNER(x)}(s)

In (2.7b), in contrast, HOWL cannot further specify MOVE because a boy howling – the vibration of his vocal chords – can in no way be directly caused by his movement around the corner. The property of howling cannot in this case belong to the same domain as the property of being in motion. A boy can of course run around a corner howling, but this is not a situation that can be encoded by a single verb, or to be more precise the single-verb-headed complex consisting of heulte ‘howl’ and the adverbial um die Ecke ‘around the corner’. Throughout this dissertation, the Possible Verbs constraint will be applied to numerous predicate complexes associated with SVCs, as a means of determining whether the SVC has an equivalent translation consisting of a clause headed by a single verb in some other language. The importance of the Possible Verbs constraint in (2.6) to this dissertation cannot be emphasized enough, since it forms the basis for the central distinction between Possible Verb and Impossible Verb SVCs. Possible Verb SVCs, which fulﬁll the Possible Verbs constraint, have translation equivalents consisting of a clause headed by a single verb in some other language, whereas Impossible Verb SVCs CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 57

do not fulﬁll the Possible Verbs constraint and so do not have such single-verb-headed clause translations. There are two further constraints upon the mapping between Semantic Form and Conceptual Structure which are described as subconstraints of, or preconditions for, Possible Verbs. Connexion requires that there is some form of argument sharing between the predicates, while Coherence regulates the temporal or causal proﬁle of the complex event described:

(2.9) Connexion. In a decomposed SF structure, each predicate must share at least one argument with another predicate, either explicitly or implicitly. (Wunderlich 2000, 25)

(2.10) Coherence. Subevents encoded by the predicates of a decomposed SF structure must be connected contemporaneously or causally.

There is some amount of vagueness in the requirement that predicates be “connected contemporaneously or causally”. Impossible Verb SVCs, which do not fulfill the Possible Verbs constraint, are contemporaneously and causally connected to some degree. However, they do not conform to Wunderlich’s stricter description of Coher- ence, which requires that the subevents “either share their temporal structure or form a cause-result (accomplishment or achievement) complex” (2000, 25). All temporal and event structural information pertaining to a complex predicate is explicated at Conceptual Structure (CS). The three SVCs in (2.11)-(2.13) each include Conceptual Structure specifications (in addition to the Theta Structure and Semantic Form specifications in (2.11b), (2.12b) and (2.13b)), but the other feature of (2.11)-(2.13) is that, in fulfilling the Possible Verbs constraint, they can all be classified as Possible Verb SVCs. They can moreover be ‘computed’ at Conceptual Structure to have one of three Possible Verb temporal and event structures which will be defined in Chapter 3 as ‘unitary-event’, ‘coidentified’ and ‘non-coidentified resultative’ structures (the temporal indices, tci (time of commencement) and tti

(time of termination), and the temporal precedence and equality relations,

(2.11) UNITARY-EVENT SVC:

a. Rao lowi ka.lai.na oha lab’o uta. Rao cook caus.separate.3 rice with ﬁsh ‘Rao cooked the rice and the ﬁsh separately.’ (Bima)

b. λy λx λs1 {COOK(x,y)(s1) & SEPARATE(s1)}

c. Conceptual Structure: One subevent, s1, of cooking.

(2.12) COIDENTIFIED SVC:

a. Ana kea rama lu’u.na d’ei kota. baby crawl enter.3 in box ‘The baby crawled into the box.’ (Bima) b. λx λs {CRAWL(x) & BECOME(LOC(x, IN(the box)))}(s)

c. Conceptual Structure: Two coidentiﬁed subevents: s1 of crawling and

s2 of entering the box. tc1 =temporaltc2 and tt1 =temporaltt2.

(2.13) NON-COIDENTIFIED RESULTATIVE SVC:

a. Ana dou to’i d’oho hari ka.losa.na Fero watu uma. child person small pl laugh caus.exit.3 Fero from house ‘The children laughed Fero out of the house.’ (Bima) b. λy λx λs {LAUGH(x) & BECOME(LOC(y,OUT-OF(the house)))}(s)

c. Conceptual Structure: Two subevents, s1 of the children laughing and

s2 of Fero exiting from the house, where s1 causes s2. Either

tc1 =temporaltc2 or tc1

Impossible Verb SVCs, unlike these three Possible Verb SVCs, do not fulﬁll the Pos- sible Verbs constraint and do not exhibit any of these Conceptual Structures, but instead require ‘looser’, composite event structures, which will also be set out in Chapter 3. The beneﬁts of analyzing Possible Verb and Impossible Verb SVCs in these event structure terms are (i) that it creates a more easily grasped picture of what constitutes a Possible Verb SVC on the one hand and an Impossible Verb SVC CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 59

on the other, and (ii) that it facilitates comparison with all of the other work in the SVC literature which analyzes serialization in event structure terms (as described in Section 1.2.4). We turn now to the interface between Semantic Form and Theta Structure. I mentioned above that partial semantic representations for lexical items are formulated at Semantic Form through binary categorial syntactic structures; (2.14b), for example, is the binary categorial structure corresponding to the more compact version of the semantic representation for the lexical item enter in (2.14a):1

(2.14) a. enter: λy λx λs {GO(x) & BECOME LOC(x, IN(y))}(s) b. t

<< s,t >< s,t >>

GO x &

BECOME t

LOC e

IN y Adopting a binary branching structure in the representations at Semantic Form ensures a unique ranking of arguments at this level in terms of hierarchical prominence. 1IN in (2.14a) and (2.14b) would actually appear as ‘INT’ in the LDG literature, which signifies that the figure is located within the INTernal region of its ground argument. I use IN only because it is a more commonly encountered semantic primitive, and one that better fits with the other locational semantic primitives I use (see Section 2.3.3). The conventions I adopt for semantic primitives in this dissertation are otherwise mostly identical to those in the LDG literature. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 60

Binary branching is crucial to the following constraint on the mapping between Se- mantic Form and Theta Structure:

(2.15) Argument Hierarchy. The list of λ-abstractors in TS corresponds to the depth of embedding in SF, with the lowest argument to the left (ﬁrst subjected to Functional Application), and the highest argument to the right. Correspondingly, the lowest argument (of a polyadic verb) is designated as [+hr,-lr], and the highest argument as [-hr,+lr], whereas all medial arguments are designated as [+hr,+lr]. (Wunderlich 2000, 252)

What goes unmentioned in (2.15) is that it is only structural arguments which receive [±lr] and [±hr] designations, for the purpose of being linked to the structural positions of subject, object or indirect object at Morphology/Syntax. The (rightmost) situational argument s is never a structural argument. Otherwise, whether an argument is structural or not is in large part determined by the following constraint, which also relies upon binary branching structure in Semantic Form representations:

(2.16) Structural Argument. An argument is structural only if it is either the lowest argument or (each of its occurrences) L(exically)-commands the lowest argument; so every internal (non-highest) argument of a nonﬁnal predicate in SF is nonstructural. L-command is deﬁned for the nodes in Semantic Form, which represent logical types, as follows: α L-commands β if the node γ, which either directly dominates α or dominates α by a chain of nodes type-identical with γ, also dominates β. (Wunderlich 2000, 252)

y in (2.14b) is a structural argument because it is the lowest argument, and x is also a structural argument because each of its occurrences L-commands y. Structural arguments are linked to the grammatical relations of subject, direct object or indirect object (as mediated by [±lr] and [±hr] feature speciﬁcations), as just mentioned. Most languages allow either two or three structural arguments; in rarer cases a language will have only one structural argument (such as Wunderlich (1997b, 47) argues for Tagalog). German has three structural argument positions, since dative is a structural CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 61

case, and English has three structural arguments in double object constructions where the ‘dative’ argument appears morphosyntactically as a direct object. According to Wunderlich (1997b, 47), “languages in which a three-place predicate is expressed by a serial verb construction ... have only two structural arguments”. In fact, a more accurate version of this statement could be to say that languages in which a three- place predicate can only be expressed by a serial verb construction have only two structural argument positions. This is necessary because in some languages, such as Fongbe, three-place predicates can be expressed in SVCs, such as (2.17), but they can also be expressed in simplex clauses containing three structural argument positions,

such as (2.18):

(2.17) Kkú s às n ná As´ıbá.` Koku take crab def give Asiba

‘Koku gave the crab to Asiba.’ (Fongbe; Lefebvre & Brousseau 2002, 448)

(2.18) Kkú ná As´ıbá` às n. Koku give Asiba crab ‘Koku gave Asiba crab.’ (Fongbe; Lefebvre & Brousseau 2002, 445)

As´ıbá` occupies the third structural argument position in (2.18) (as verified through the fact that this argument can be expressed with the pronominal clitic bearing a low tone that signifies the assignment of structural objective case). These examples show that expression of a three-place predicate in SVC-form does not necessarily entail the absence of a third structural argument position in the language concerned. The ditransitive simplex Fongbe clause in (2.18) illustrates how third structural arguments in this language are unmarked and placed adjacent to the second structural argument (interestingly, the order of the objects can also be reversed in (2.18)). (2.18)

has three structural arguments, then, but how many structural arguments does (2.17)

contain? On the face of it, (2.17) looks like it could contain three in total, with Kk´u

in first structural argument position and àsn ‘the crab’ and As´ıbá` both occupying second structural argument positions. Section 2.4.1 will, however, consider whether As´ıbá` may in fact be a non-structural argument. This claim may seem surprising from the perspective that As´ıbá` , as the object of a verb ná ‘give’, looks just as structural as

CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 62

àsn ‘the crab’. But what could otherwise be claimed – and indeed what has often been claimed in the SVC literature – is that this verb, and many others in SVCs, are essentially doing the same ‘job’ as an adposition or an oblique semantic case: it is introducing a non-structural argument into the clause. Support for the claim that ná ‘give’ is performing this adpositional/oblique semantic case function is tied up with Lefebvre and Brousseau’s (2002) arguments that the semantic status of As´ıbá` in (2.17) is more akin to a ‘goal’ than the ‘recipient/possessor’ status of As´ıbá` in (2.18), for reasons that will be explained in Section 2.4.1. But for now I will present two alternative cases for which it is argued that verbs perform the adpositional/oblique semantic case function of introducing non-structural arguments; lá ‘take’ in (2.19) introduces a non-structural instrument argument (egbà ‘axe’), while fún ‘give’ in (2.20) introduces a non-structural benefactive argument (Olú):

(2.19) Tsoda lá egbà ba cigbà. Tsoda take axe cut tree ‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63)

(2.20) Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5)

At Semantic Form these non-structural arguments are lexically marked as oblique. Arguments which are lexically marked as oblique have received scant illustration in the LDG literature, although the oblique directional argument of the German verb gehen ‘go’, lexically marked as oblique with the feature [+dir] in (2.21), features more than once (usually in opposition to the semantic representation for the verb betreten ‘enter’, in which the directional argument is incorporated into the meaning of the verb and has structural status):

(2.21) gehen ‘go’: λP λx {GO(x) & P(x)} [+dir] in das Zimmer ‘into the room’: [+dir] λx BECOME(LOC(x,IN(das Zimmer))) [ in das Zimmer gehen: λx {GO(x) & BECOME(LOC(x,IN(das Zimmer)))} ] (Wunderlich & Lak¨amper 1999,13) CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 63

Lexically marking an argument as oblique is just one way of rendering it non-structural (independent of its position in the Semantic Form tree representation). Some other arguments at Theta Structure which are not subject to structural linking are those which the Structural Argument constraint in (2.16) determines are not structural, and those which are suppressed by argument structure operations, such as passive. These two latter types of non-structural argument, if they are realized at all, are realized in exactly the same way as arguments lexically marked as oblique, which in most circumstances involves an adposition or semantic case functioning as the semantic linker. What is novel about SVCs such as (2.19) and (2.20) is that here it is a verb which is functioning as a semantic linker for a non-structural argument, and as far as I can see, this is something that can only happen in an SVC. The reason why arguments that are lexically marked as oblique have received scant illustration in the LDG literature is that research in this framework has concentrated on the linking of structural arguments to structural cases. Kiparsky (2001), for example, simplifies the binary distinction between structural and non-structural linking with a specification [±sc] ‘structural case’, and from there concentrates on structural linking, or those theta-roles specified to be [+sc]. This dissertation adopts this notational convention, illustrated in (2.22)-(2.23) (=(2.19)-(2.20)), for its compactness (dispensing, for now, with the [±hr] and [±lr] linking specifications). However, here just as much attention will be paid to non-structural ([-sc]) arguments as to structural arguments. This is because Theta Structure Property for SVCs, set out in the next section, accords oblique arguments as much importance as structural arguments. (2.22) a. Tsoda lá egbà ba cigbà. Tsoda take axe cut tree ‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63) b. lába: λz λy λx λs {CUT(x,y) & USE(x,z)}(s) −sc +sc +sc

(2.23) a. Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5) ′ ′ ′ b. wo. fún: λz λy λx λs λs {WEAR(x,y)(s ) & BENEFIT(s ,z)}(s) −sc +sc +sc CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 64

The [-sc] specification for both z arguments in (2.22) and (2.23) signifies that these arguments are lexically marked as oblique,2 but on what grounds are these non- structural statuses determined? At very least, the non-structural status of these z arguments is predictable from the predicates in which they are embedded; USE and BENEFIT are two predicates that feature in the configurational definitions for oblique arguments set out in Section 2.3.3, with USE featuring crucially in the definition of instrumental oblique arguments and BENEFIT in the definition of benefactive oblique arguments. The discussion in Section 2.3.3 also brings out the point that the semantics of these predicates is inherently ‘preposition-like’ (or ‘semantic case-like’) rather than verb-like (where USE is concerned, Chapter 4 furthermore argues that this predicate cannot generally occur alone in a Semantic Form representation, and so cannot be realized as the lone verb of a sentence). The English preposition that most commonly corresponds to USE is ‘with’ and that which most commonly corresponds to BENEFIT is ‘for’. But what ultimately determines the non-structural status of the z arguments in (2.22) and (2.23) are the serialization lexical rules which alter the semantics of the verbs lá ‘take’ and fún ‘give’ to USE and BENEFIT respectively,3 while at the same time specifying that the internal arguments of these predicates are non-structural. These serialization lexical rules will be set out in Section 2.4.1. While the internal (z) arguments of lá ‘take’ in (2.22) and fún ‘give’ in (2.23) are lexically marked as oblique, and so receive [-sc] features, the arguments of verbs (or affixes or particles) with preposition-like semantics need not always be lexically marked as oblique. In fact, the title of Wunderlich (1997a), ‘Argument Extension by Lexical Adjunction’, refers to a process whereby semantic components, including some with preposition-like semantics, can introduce additional structural arguments. These additional structural arguments fill new structural argument positions in the clause; raka ‘reach’ in (2.24), for example, introduces a new structural argument

2A [-sc] specification could alternatively respond to an argument’s failure to satisfy the Structural Argument constraint, or to its demotion to non-structural status through an argument structure operation, although neither of these types of non-structural argument will be encountered in the SVCs of this dissertation. For the most part it can be assumed that a [-sc] argument is lexically marked as oblique, and was thus never in contention for structural status. 3What lá ‘take’ and fún have their semantics altered from are USE and ACT(x) & BE- COME(POSS(z,y)) respectively. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 65

nahu ‘me’, which ﬁlls the second structural argument position directly after the verb complex, while ka.maru ‘make sleep’ in (2.25) introduces a structural argument ana kea ‘baby’, which likewise ﬁlls this second structural argument position:4

(2.24) a. na.mai raka.ku nahu, ... 3.come reach.ass 1p ‘he came to me, ...’ (Bima) b. mai raka: λy λx λs {MOVE-TOWARDS-DEICTIC-CENTER(x) +sc +sc & BECOME(LOC(x,AT(y)))}(s)

(2.25) a. Amu rawa ka.maru.na ana kea. Amu sing caus.sleep.3 baby ‘Amu sang the baby to sleep.’ (Bima) b. rawa ka.maru: λy λx λs {SING(x) & BECOME(ASLEEP(y))}(s) +sc +sc

The internal arguments of lá ‘take’ in (2.22) and fún ‘give’ in (2.23), in contrast, do not fill new structural argument positions in the clause – i.e. they do not call into being a third structural argument position. (2.22) and (2.23) have SVC-specific clause structures, where a third argument is realized, not in a third structural argument position, but as a non-structural, oblique object of a verb with preposition-like semantics. This section has described the essential distinguishing features of LDG, admit- tedly quite briefly. To conclude here, I will reiterate those features of this framework which impact directly upon the Theta Structure Property for SVCs. Firstly, LDG incorporates a level of Theta Structure (TS), where arguments are λ-abstracted from Semantic Form (SF). Crucially, the arguments at Theta Structure are divided between structural and non-structural arguments on the basis of whether they fulfill the Structural Argument constraint, whether they are suppressed by argument structure operations, or whether (as will actually be the case for the non-structural arguments exemplified in this dissertation) they are lexically marked as oblique (so

4A diﬀerence between these two examples is that, while raka ‘reach’ in (2.24) has the preposition-like semantics of BECOME(LOC(x,AT(y))), ka.maru ‘make sleep’ in (2.25) does not have preposition-like semantics. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 66

non-structural) to begin with. I have opted to signal this division at Theta Structure through [+sc] and [-sc] features. This section also stated and demonstrated LDG’s ‘Possible Verbs’ constraints – the constraints on the mapping between Semantic Form and Conceptual Structure. These are of crucial importance to this dissertation because they form the basis for the Possible Verb-Impossible Verb distinction in SVCs to be discussed in Chapter 3. These constraints have no bearing upon the rest of this chapter however, since the current focus is the interface not between Semantic Form and Conceptual Struc- ture, but between Semantic Form and Theta Structure. Further to this, Section 2.4 distinguishes among SVCs in terms of how they project from Theta Structure into Morphology/Syntax, although it is at the levels of Theta Structure and Semantic Form alone that the formulation of the Theta Structure Property for SVCs is situated. The next section explores this formulation.

2.3 The Theta Structure Property for SVCs

This section presents the Theta Structure Property for SVCs, claimed to be the most eﬀective and cross-linguistically comprehensive means of deﬁning SVCs argument structurally. Section 2.3.1 states the property, while Section 2.3.2 illustrates its bearing upon structural arguments and Section 2.3.3 its bearing upon oblique arguments.

2.3.1 The Property

Here is the proposal central to this dissertation:

(2.26) The Theta Structure Property for SVCs: An SVC has the Theta Structure of a Possible Verb structure. That is, an SVC has (i) the same number of structural arguments (one, two or three) as some Possible Verb structure, and (ii) the same list of conﬁgurationally deﬁned, overtly realized non-structural (oblique) arguments as some Possible Verb structure. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 67

The question that (2.26) immediately raises is: what is a ‘Possible Verb structure’? Possible Verb structure is intended to be a purely theoretical semantic construct, although it is one for which I have no clear theoretical definition set in stone at this stage. The essence of a Possible Verb structure is that it has the semantic content of a clause headed by a single verb, although the first, essential defining point is that these structures must satisfy the Possible Verbs constraints described in the previous section. However, satisfaction of the Possible Verbs constraints is really only something that can be demonstrated on a case by case basis, in terms specific to certain examples. Chapter 3 demonstrates how several SVCs satisfy the Possible Verbs constraints, but it also sets out some more general event structural profiles for Possible Verb SVCs, which help us to identify which SVCs have Possible Verb structure (and hence are Possible Verb SVCs). Chapter 5 additionally states some motion-specific constraints on Possible Verb structures. In short, this whole dissertation is working towards defining what a Possible Verb structure is. In the absence of a definition for Possible Verb structure at this stage, I adopt the following diagnostic for an SVC meeting the Theta Structure Property, which focuses on the most salient feature of Possible Verb structures: that they have the semantic content of a clause headed by a single verb:

(2.27) Diagnostic for an SVC meeting the Theta Structure Property: If a multi-verb construction has the same number of structural arguments identified at Theta Structure, and the same list of configurationally defined, overtly realized non-structural (oblique) arguments as some clause headed by a single, simplex verb, then it meets the Theta Structure Property for SVCs.

I apply this diagnostic to numerous multi-verb constructions throughout this dissertation in order to establish if they satisfy the Theta Structure Property and hence are SVCs. By requiring an SVC to have the same Theta Structure configuration as some clause containing a single, simplex verb, the Theta Structure Property is effectively enforcing that SVCs have monoclausal argument structure; SVCs are required to show the same configurations of arguments+adjuncts found for clauses headed by single verbs in non-serializing languages, as Durie (1997) commands. The Theta CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 68

Structure Property for SVCs achieves this eﬀect by utilizing the LDG distinction between structural and non-structural arguments, discussed in Section 2.2. Structural and non-structural arguments, and their realization in SVCs, will now be discussed in turn.

2.3.2 Structural arguments and the Theta Structure Prop- erty for SVCs

Structural arguments are always overtly realized in one of the core structural positions (subject, object, or in a third structural grammatical relation, if there is one). The following SVC has one structural argument, which is realized as the subject:

(2.28) a. J´ımò. ó. sáré lo. . Jimo agr run go ‘Jimo ran away.’ (Yoruba; Déchaine 1993, 220)

b. s´ar´elo. : λx λs {RUN(x) & MOVE-AWAY-FROM-DC(x)}(s) +sc

This SVC fulﬁlls the Theta Structure Property for SVCs, trivially, because it has one structural argument, and no oblique arguments, and there certainly exist clauses containing a single verb which have one structural argument (among which, the English translation of (2.28), Jimo ran away). (2.29) likewise fulﬁlls the Theta Structure Property, since it has two structural arguments, and no oblique arguments, and the English translation of this sentence (in which the BECOME(FLAT(y)) component of meaning happens to be expressed by an adjective rather than a verb) could again serve as an example of one single-verb-headed clause (among millions) which similarly has two structural arguments:

` (2.29) a. E. sósà gbé émá!tó. n pè. rhé. . Esosa hit metal be-flat ‘Esosa hit the metal flat.’ (E`. dó; Stewart 2001, 58) b. gbépè. rhé. : λy λx λs {HIT(x,y) & BECOME(FLAT(y))}(s) +sc +sc CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 69

(2.30) is a further example which satisﬁes the Theta Structure Property since its Theta Structure representation consists of two structural arguments and no oblique arguments:

(2.30) a. Amu rawa ka.maru.na ana kea. Amu sing caus.sleep.3 baby ‘Amu sang the baby to sleep.’ (Bima) b. rawa ka.maru: λy λx λs {SING(x) & BECOME(ASLEEP(y))}(s) +sc +sc

That (2.30) contains a second structural argument can be attributed to the fact that the second verb introduces this new structural argument to the clause, as discussed in Section 2.2. Hal-yak ‘ﬂy away’ in (2.31) likewise introduces a second structural argument, since yip ‘blow’ is an intransitive verb whose sole (structural) argument is ne-len ‘the wind’ (the subject):

(2.31) a. Ne-len mi-yip hal-yak na.kat. art-wind perf-blow(intr.) ﬂy-away art-cards ‘The winds blew the cards away.’ (Mwotlap; Fran¸cois 2004, 119) b. yip hal-yak: λy λx λs {BLOW(x) & FLY(y) & +sc +sc MOVE-AWAY-FROM-DC(y))}(s)

That each verb introduces a diﬀerent structural argument in (2.30) and (2.31) has no bearing upon the manner in which these SVCs satisfy the Theta Structure Property. All that is at issue for the satisfaction of the Theta Structure Property is that there are two structural arguments at Theta Structure and that there are clauses containing single simplex verbs that also have two structural arguments at Theta Structure. The following Jeh SVC has three structural arguments, and it fulﬁlls the Theta Structure Property because there exist clauses containing a single verb in other languages which have three structural arguments (such as English you buy me rice):

(2.32) a. Mi ruat d´oh au phei. 2sg buy give me rice ‘You buy me rice.’ (Jeh; Kroeger 2004, 240) b. ruat d´oh: λz λy λx λs {BUY(x,z) & BECOME(POSS(y,z))}(s) +sc +sc +sc CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 70

Structural arguments, as they stand in the Theta Structure Property for SVCs, are blind to thematic role information. So the fact that the three arguments in (2.32) have the thematic roles of agent (or ‘giver’), benefactive/recipient, and theme (or ‘thing given’), and the fact that a single-verb-headed clause could select the same three thematic roles, has no bearing upon the Theta Structure Property for SVCs; all that is at issue for the satisfaction of the Theta Structure Property is that the arguments be structural, and that some single-verb-headed clause contain the same number of structural arguments. That structural arguments are blind to thematic role information is of crucial signiﬁcance to Impossible Verb SVCs and the way in which they satisfy the Theta Structure Property for SVCs (Impossible Verb SVCs are otherwise avoided until Chapter 3, with the focus here on the basic, core class of Possible Verb SVCs). For while the thematic roles corresponding to structural arguments in a Possible Verb SVC such as (2.32) mirror the thematic roles corresponding to structural arguments in single-verb-headed clauses, thematic role information corresponding to individual structural arguments in Impossible Verb SVCs is of a more complex and composite nature, such as could not be mirrored in a clause headed by a single verb. (2.33) and (2.34) are two Impossible Verb SVCs where the structural arguments bear two separate thematic roles with respect to the diﬀerent verbs in such a way that structural arguments in single-verb-headed clauses could not:

(2.33) a. Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima) b. rumpa weha: λy λx λs [{FIND(x,y)} & {TAKE(x,y)}](s) +sc +sc

(2.34) a. Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10) b. léré: λy λx λs [{COOK(x,y)} & {EAT(x,y)}](s) +sc +sc

x in (2.33), for example, is both a ‘ﬁnder’ and ‘taker’ and y both a ‘ﬁndee’ and a ‘takee’. x in (2.34) is both a ‘cooker’ and an ‘eater’, and while it is true that these CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 71

could both be analyzed as subtypes within the broader thematic role of ‘agent’, it is unlikely that there exists a single verb which would select an argument whose thematic role presents the intersection of these two more narrowly deﬁned thematic roles. But the point is that the Theta Structure Property is not sensitive to this; as long as an argument is structural then it is irrelevant whether it has the thematic role characteristics of a structural argument in a clause headed by a single verb or not. For non-structural arguments it is a diﬀerent story.

2.3.3 Non-structural arguments and the Theta Structure Pro- perty for SVCs

Besides requiring an SVC to have the same number of structural arguments as some clause containing a single verb, the Theta Structure Property requires that if an SVC overtly realizes some list of configurationally defined non-structural (oblique) arguments, then there is some clause containing a single verb which selects an identical list. Overtly realized non-structural arguments are those arguments marked [-sc] at Theta Structure which are not linked to the null symbol ∅ (i.e. they cannot be implicit). They are defined and identified through particular configurations of semantic primitives in Semantic Form representations. These particular configurations could be given unique labels, and these labels could be referred to as ‘thematic roles’. This is consonant with Jackendoff’s (1972, 1983) view of thematic roles as “nothing but particular structural configurations in conceptual structure [(or Semantic Form)]; the names for them are just convenient mnemonics for particularly prominent configurations” (1990, 47). Defining and identifying non-structural arguments through configurations of Semantic Form primitives is necessary for the Theta Structure Property, but it is also envisaged that this kind of thematic role information would be central to a theory of how non-structural arguments are linked to certain oblique semantic cases or adpositions, quite independently of SVCs. In what follows I will lay out some of the configurational definitions I will be using in this dissertation which correspond to the thematic role labels indicated. Here some caveats are in order. Firstly, it should be stressed that there is no intention in this CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 72

work to provide a comprehensive theory of thematic roles, or to enumerate or delimit the number of thematic roles. For the purpose of supporting and illustrating the Theta Structure Property for SVCs, all that is at issue is whether a single-verb- headed clause in some other language has non-structurally linked arguments with the same semantic configuration as those of an SVC under consideration. A second point to stress is that these configurational identifications of thematic roles should not be regarded as conclusive or definitive. The Theta Structure Property for SVCs gains strength from not being tied to a particular theory of thematic roles, but is more flexible in the sense that it is compatible with any theory, and so can benefit from any advances in thematic role theory. A third point is that, in the configurational definitions that follow, I have tried to adhere as closely as possible to the range of Semantic Form primitive predicates used in the LDG literature,5 although there are some exceptions and innovations. The BENEFIT predicate that appears in the benefactive role definition in (2.35) below, for example, does not appear anywhere in the LDG literature; this predicate just seemed like the most straightforward and explicit way of capturing the particular kind of participation common to all structures involving a benefactive role. An additional, small number of predicates intended to encode other types of participation in events will be innovated in this dissertation mostly for the sake of illustration, although there is at least more space available in Chapter 4 to describe the kind of instrument participation I have encoded with USE. Also, I have used certain predicates encoding spatial relational concepts (IN, ON, UNDER, etc.) that differ slightly from Wunderlich’s (1991) set of spatial relational primitives, only because I thought them to be more intuitive and transparent. With these caveats out of the way I will lay out some sample configurational definitions for thematic roles. These will be accompanied by examples of SVCs with non-structural arguments which instantiate these thematic roles, and I will also spell out how these SVCs fulfill the Theta Structure Property for SVCs.

5Some of the semantic primitives that feature most prominently and regularly in the LDG literature are ACT (which signifies a non-specific activity), BECOME (which signifies a change of state), LOC (which signifies that an entity is located at a certain place), and POSS (which signifies a possessive relation). There is no CAUSE primitive, as mentioned in Section 2.2 (and defended in Wunderlich 1997b). CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 73

The ﬁrst thematic role we will consider could be labeled ‘benefactive’. It can be deﬁned thus:

(2.35) A ‘benefactive’ is z in {P(x, ..)(s) & ... & BENEFIT(s,z)}

The use of the primitive ‘BENEFIT’ is meant to convey the idea that the situation described by the main verb positively aﬀects, or beneﬁts, z, the benefactive argument. An SVC containing a non-structural argument which instantiates the ‘benefactive’ role is:

(2.36) a. Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5) ′ ′ ′ b. wo. fún: λz λy λx λs λs {WEAR(x,y)(s ) & BENEFIT(s ,z)}(s) −sc +sc +sc

In the Semantic Form representation for this SVC in (2.36b) there is one overall situation or eventuality, s, but within this situation there is a ‘subsituation’ s′ of x wearing y, and it is this latter situation that benefits the third entity z. (2.36) fulfills the Theta Structure Property because there exist clauses containing single verbs which have two structural arguments and a non-structural ‘benefactive’ argument as defined in (2.35). The English translation for (2.36) is just such a clause containing a single verb. What is left unexplained at this stage is how the ‘benefactive’ argument comes to be realized as the object of fún ‘give’. This will be accounted for with a Benefactive Serialization Lexical Rule in Section 2.4.1 and through more general rules pertaining to argument realization in non-contiguous SVCs. Chapter 4 explores instrumental serialization and related phenomena. There the ‘instrument’ role receives the following configurational definition:

(2.37) An ‘instrument’ is z in {P(x,...) & ... & USE(x,z)}

An SVC containing a non-structural argument which instantiates the ‘instrument’ role is: CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 74

(2.38) a. Tsoda lá egbà ba cigbà. Tsoda take axe cut tree ‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63) b. lába: λz λy λx λs {CUT(x,y) & USE(x,z)}(s) −sc +sc +sc

(2.38) fulfills the Theta Structure Property for SVCs because there exist clauses containing single verbs which have two structural arguments and a non-structural ‘instrument’ argument as defined through the configuration in (2.37). The positioning of the instrument egbà ‘axe’ adjacent to lá ‘take’ will also have to be explained through a serialization lexical rule, combined with the principles of argument realization for non-contiguous SVCs set out in Section 2.4.1. Chapter 5 investigates the expression of motion and location in SVCs, and so it is fitting that a number of spatial thematic roles be configurationally defined here. The most salient aspect of these roles is the very large number of different configurational definitions for them, to the extent that it becomes necessary to ask whether it is really worth attempting to find sensible thematic role labels for all of them. Nonetheless, I will close off this section with some examples of spatial roles and their instantiation in SVCs. The configurational definitions for these roles are (mostly) based on Wunderlich’s (1997a, 1997b, 2000) Semantic Form representations. One of the most basic spatial roles could be called ‘locative’ and defined thus:

(2.39) A ‘locative’ is z in {P(...) & ... & LOC(x, AT(z))}

This role features in the following SVC, which satisﬁes the Theta Structure Property because there exist clauses headed by single verbs in other languages which have two structural arguments and a non-structural locative argument:

(2.40) a. Wo zai chufang qie cai. 1sg be.at kitchen cut food ‘I cut food in the kitchen.’ (Mandarin; D´echaine 1993, 254) b. zai qie: λz λy λx λs {CUT(x,y) & LOC(x, AT(z))}(s) −sc +sc +sc CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 75

Zai in (2.40) is translated by Déchaine as ‘be at’, and it is on this basis that (2.40) is given a Semantic Form representation containing LOC(AT), using the configuration of semantic primitives which defines the locative role in (2.39). The English translation for this sentence would be given a Semantic Form representation which uses IN in place of AT on the basis that (using the preposition in signifies that) English speakers conceptualize the participants in the event being contained within the kitchen (room). We could call this thematic role ‘containing locative’ and define it like so:

(2.41) A ‘containing locative’ is z in {P(...) & ... & LOC(x, IN(z))}

But from here there is an explosion of locative roles, among which z in each of the following:

(2.42) a. Locative1 is z in {P(...) & ... & LOC(x, ON(z))} b. Locative2 is z in {P(...) & ... & LOC(x, UNDER(z))} c. Locative3 is z in {P(...) & ... & LOC(x, BESIDE(z))} d. Locative4 is z in {P(...) & ... & LOC(x, BEHIND(z))} e. Locative5 is z in {P(...) & ... & LOC(x, IN-FRONT-OF(z))}, etc. ...

This group of locative roles will be as complicated and extensive as their systems of expression in the language concerned. One consequence of this explosion of roles is that it may not be practicable to invent thematic role titles for each different configuration, meaning that the most effective means of identifying the ‘role’ could be through the configuration of predicates itself. The semantic representations in (2.39), (2.41) and (2.42a)-(2.42e) encode purely locational meanings; they contain Place-functions in Jackendoff’s (1983) terms, that describe how a Figure is configured in relation to a Ground. The terms Figure, Ground and Place-function will be elaborated upon in Chapter 5, as will (Jackendoff’s (1983)) Path-function. Path-functions describe movement towards or away from a Place, and so have Place-functions embedded within them. In LDG Semantic Form representations, movement towards or away from a Place is encoded by embedding LOC predications within BECOME, as in the following configurational definition for the thematic role of ‘goal’: CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 76

(2.43) A ‘goal’ is z in {P(...) & ... & BECOME(LOC(x, AT(z)))}

(2.44) is an SVC containing a ‘goal’ which fulﬁlls the Theta Structure Property for SVCs because there exist clauses containing single verbs which have two structural arguments and a non-structural goal argument:

(2.44) a. Mieri kyari di pikni go a skuul. Mary carry def child go loc school ‘Mary carried the child (thither) to school.’ (Jamaican Creole; Winford 1992, 184) b. kyari go: λz λy λx λs {CARRY(x,y) & BECOME(LOC(y,AT(z)))}(s) −sc +sc +sc

(2.45) below diﬀers from (2.43) by substituting AT with IN; this introduces the concept of containment into this goal representation. The concept of containment is also present in (2.46), where z is not a goal, but a source, while (2.47) presents the predicate decomposition I will use to deﬁne the most basic kind of source (with no containment involved):

(2.45) A ‘containing goal’ is z in {P(...) & ... & BECOME(LOC(x, IN(z)))}

(2.46) A ‘containing source’ is z in {P(...) & ... & BECOME(LOC(x, ¬IN(z)))}

(2.47) A ‘source’ is z in {P(...) & ... & BECOME(LOC(x, ¬AT(z)))}

(2.48) and (2.49) are SVCs containing the respective roles of ‘containing goal’ and

‘containing source’ which fulﬁll the Theta Structure Property for SVCs:

(2.48) a. Kkú s m tò by xwé gbè. Koku take car def enter house in ‘Koku drove the car into the house.’

(Fongbe; Lefebvre & Brousseau 2002, 410)

b. s by : λz λy λx λs {DRIVE(x,y) & BECOME(LOC(y,IN(z)))}(s) −sc +sc +sc CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 77

(2.49) a. ede mpara karefa rai losa.na d’ei toples. dem then frog run exit.3 in jar ‘thenthefrogranoutofthejar.’ (Bima) b. rai losa: λy λx λs {RUN(x) & BECOME(LOC(x,¬IN(y)))}(s) −sc +sc

‘Goal’, ‘containing goal’ and ‘containing source’ are illustrated here because they are the three most commonly occurring thematic roles in the motion SVCs of this dissertation. But of course there will be other motion event thematic roles (such as the goal object of onto in English, or the source object of off (of)), which cumulatively beg the question of whether a multitude of thematic role labels is really practicable. Again, the most effective means of identifying the ‘role’ could be through the configuration of predicates itself. Broadly speaking, there are numerous other ‘thematic roles’ requiring configurational definitions, but here a limited sample has sufficed for the purpose of demonstrating the part played by non-structural arguments in satisfying the Theta Structure Property for SVCs. The next section moves into the challenging territory of how serial verb arguments – structural and non-structural – are mapped into divergent syntactic structures.

2.4 Projecting the arguments of contiguous and non-contiguous SVCs

The Theta Structure Property for SVCs is defined at the level of Theta Structure (TS); it is at this level that the main claim of this dissertation, concerning how SVCs should be defined and identified, obtains. This dissertation holds no strong claims about the syntax of SVCs; SVCs are allowed to display a range of syntactic structures, in contrast to various accounts which actually seek to define SVCs through the (non-contiguous) syntactic configurations they display themselves. The sense in which my approach is syntactically ‘permissive’ could be seen as a natural consequence of SVC definition focusing upon the argument structure level of Theta CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 78

Structure, leaving open the question of how Theta Structure maps into Morphol- ogy/Syntax, or how any syntax-level operations might apply. No attempt will be made to account for language-speciﬁc SVC structures in any detail. Instead, only broad generalizations concerning surface syntactic properties of SVCs will be drawn, chief among which are the argument realization patterns for non-contiguous SVCs in Section 2.4.1. These patterns relate to the central syntactic distinction between contiguous and non-contiguous SVCs recognized here, for although this dissertation holds no strong claims about the syntax of SVCs, there is one very strong message: SVCs can have contiguous verbs, and not just non-contiguous verbs. Section 1.2.3 ﬁrst drew attention to the fact that although there have been numerous syntactic analyses of SVCs, the vast majority of these have only been concerned with SVCs with non-contiguous verbs, such as (2.50), while SVCs with contiguous verbs, such as (2.51), tend to be disregarded or assumed to be compound verbs:

` (2.50) E. sósà gbé émá!tó. n pè. rhé. . Esosa hit metal be-flat ‘Esosa hit the metal flat.’ (Edo; Stewart 2001, 58)

(2.51) Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima)

(2.51) would be a compound verb were its two verbs to constitute a single phonological word. Because rumpa ‘ﬁnd’ and weha ‘take’ do not constitute a single phonological word, but instead each independently carry the stress of a single phonological word, (2.51) is a contiguous serial verb, according to the criteria for serial verbhood set out in Section 1.2.1. Section 1.2.1 also raised the question of whether there is any necessary justiﬁcation for drawing a sharp theoretical distinction between SVCs and compound verbs, although for the most part the main purpose of this section is to reinforce the point that SVCs can have contiguous, and not just non-contiguous verbs, and to demonstrate that the Theta Structure Property for SVCs unproblematically includes both types of SVC. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 79

This dissertation argues that SVCs are sentences containing two or more verbs which are acting as if there is only one verb. These constructions seem to have con- vinced themselves that they contain only one verb, and, if I can extend this metaphor of self-delusion further, I could venture that SVCs such as (2.51) are acting more convincingly as if there is one verb, by clumping their verbs together as if to pretend that there is only one (especially if there is evidence for adopting Kroeger’s (2004) analysis, whereby verbs appear together within a single V′ constituent, to the exclusion of their objects which are realized within the larger VP constituent). Contiguous SVCs could be said to be trying harder to be monoclausal than non-contiguous SVCs such as (2.50), which, in allowing objects to intervene between their verbs, look more as if they could be biclausal. Multiple object non-contiguous serializations such as (2.52) may have even stronger biclausal resemblances ((2.50) is a single object non- contiguous SVC):

(2.52) Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5)

But although (2.50) and (2.52) may look biclausal, this can only be a case of syntactic appearances; if a non-contiguous SVC is to be classed as an SVC in the first place, then it must be shown to be monoclausal, by satisfying the Theta Structure Property for SVCs, among other properties. Nonetheless, one feature of non-contiguous SVCs like (2.52) that cannot be denied is that they contain a greater number, and different arrangement, of object positions for arguments than in simplex clauses. (2.52) has SVC-specific clause structure. The argument realization patterns for non-contiguous SVCs in Section 2.4.1 capture this SVC-specific clause structure, but they also explain how it is that some non-structural arguments, such as the benefactive argument Olú in (2.52), come to look as if they are structural in being realized as the objects of verbs (rather than as the objects of prepositions, or in oblique semantic cases). Taking this into consideration, Section 2.4.1 ultimately accounts for how non-contiguous SVCs with SVC-specific clause structure satisfy the Theta Structure Property for SVCs. In contrast to non-contiguous SVCs such as (2.52), the contiguous SVC in (2.51) CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 80

has (on the most basic syntactic assumptions) non-SVC-specific clause structure identical to that of a simplex clause (albeit with the addition of an extra verb), with identical configurations of surface structural positions for arguments. This holds for almost every contiguous SVC I have encountered, forming the basis for the key question of Section 2.4.2: can contiguous SVCs ever display SVC-specific clause structure which calls for exceptional patterns of argument realization (analogous to the argument realization patterns for non-contiguous SVCs)?

2.4.1 Non-contiguous SVCs

Non-contiguous SVCs occur mostly in African (especially Kwa) languages (2.53)- (2.54) and African substrate creoles (2.55)-(2.56), although they are also found in Asia (2.57)-(2.58) and Austronesia (2.59)-(2.60), among other linguistic areas.

(2.53) Olú fa o. mo. náà já.de. Olu pulled child the go.out

‘Olu pulled the child out.’ (Yoruba; Bamgbos.e 1974, 23)

(2.54) Kkú s às n wá àx`ı m . Koku take crab come market in ‘Koku brought the crab to the market.’ (Fongbe; Lefebvre & Brousseau 2002, 410)

(2.55) Kofi teki Amba go na a foto. Kofi take Amba go loc def city ‘Kofi took Amba to the city.’ (Sranan; Sebba 1987, 59)

(2.56) S´ei wan ijsie d´a mi! sell det ice-cream give 1sg ‘Sell an ice-cream for me!’ (Saramaccan; Veenstra 1996, 95)

(2.57) Nws muab riam txiav nqiaj qaib. 3sg take knife cut meat chicken ‘She cut some chicken with a knife.’ (White Hmong; Jarkey 1991, 83) CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 81

(2.58) Ta yòng zhèbˇa yaòshi kai-de mén. 3sg use key this open-de door ‘He opened the door with his key.’ (Mandarin; Déchaine 1993, 253)

(2.59) Awa ni-nggo aiya uni. mother ir3s-say 2s dead ‘Mother is going to scold you to death.’ (Numbami, Bradshaw 1993, 138)

(2.60) Sira lori tudik sona malu. they take knife stab each.other ‘They use knives to stab each other.’ (Tetun; Williams-Van Klinken et al. 2002, 6)

The purpose of this section is to account for the argument realization patterns of these kinds of SVCs with the argument realization principles for non-contiguous SVCs set out in Section 2.4.1.1. Section 2.4.1.1 provides a simple demonstration of how the argument realization principles work, as well as summarizing three broad generalizations captured by these principles. Sections 2.4.1.2 and 2.4.1.3 then address two diﬀerent categories of multiple object non-contiguous SVC, how the argument realization principles determine their syntactic structure and how they satisfy the Theta Structure Property for SVCs. Section 2.4.1.2 considers multiple object non-contiguous serializations where only one object is structural. It is in this section that the mechanics of how a verb can take a non-structural argument as its object can be illustrated more fully than in the discussion at the end of Section 2.2, which in part means including the serialization lexical rules alluded to there. Section 2.4.1.3 then considers a proba- ble case of a multiple object non-contiguous SVC with two structural objects (which has the semantics of a clause headed by a simplex ditransitive verb), although this section also evaluates Lefebvre and Brousseau’s (2002) analysis that it has the kind of semantics that more likely means it has only one structural object.

2.4.1.1 Principles of argument realization in non-contiguous SVCs

As mentioned in Section 1.2.3, non-contiguous SVCs have received numerous syntactic analyses, which Good (2003) categorizes into complementation, coordination or CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 82

adjunction analyses. These could be termed VP-sequencing analyses, since they involve VP-complementation, coordination or adjunction. Kroeger (2004), meanwhile, gives a V′-sequencing (or V′-coordinating) analysis for non-contiguous SVCs (in contrast to his V-sequencing analysis for contiguous SVCs, which will be illustrated in Section 2.4.2):

(2.61) S

NP VP

V′ V′ ...

V (NP) V (NP)

Regardless of whether a V′-sequencing, VP-sequencing, or some other phrase structure analysis is adopted for non-contiguous SVCs, there needs to be some accounting for which argument is realized as the complement of which verb in the syntax. There also needs to be some account for how some verbs in non-contiguous SVCs can take non-structural arguments as objects. I propose the following principles as a means of capturing argument realization patterns in non-contiguous SVCs:

(2.62) Argument realization in non-contiguous SVCs:

a. Let Verb1 ... Verbn be a sequence of verbs in a non-contiguous SVC where

Verbi linearly precedes Verbi+1.

b. Let Vi be that component of a Semantic Form representation which

encodes the meaning of Verbi.

c. The initial (non-situational) argument of V1 at Semantic Form, x1, is projected into the syntax in the way [+lr], [-hr] arguments usually are in the languages concerned (typically as subject).

d. If V1 has a next, non-initial (non-situational) argument at Semantic Form

y1, this argument is projected into the syntax as the complement of Verb1 CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 83

(regardless of [±hr] and [±lr] speciﬁcations or whether it is structural or non-structural).6

e. If V2 has a (non-situational) argument at Semantic Form y2, which is

distinct from x1 and y1, then y2 is projected into the syntax as the

complement of Verb2 (regardless of [±hr] and [±lr] speciﬁcations or whether it is structural or non-structural).

f. General case: if Vi has a (non-situational) argument at Semantic Form yi,

which is distinct from each of x1, y1 ... yi−1, then yi is projected into the

syntax as the complement of Verbi (regardless of [±hr] and [±lr] speciﬁcations or whether it is structural or non-structural).

These principles should not be interpreted as an actual mapping protocol for any particular language, but rather as a generalization about the argument realization patterns that all non-contiguous SVCs share. (2.62) captures just about the only thing that the diverse syntactic analyses for non-contiguous SVCs described above have to have in common. The generalizations in (2.62) also conform to Wunderlich’s speciﬁcation that “lexical items project to syntactic phrases in which their arguments are realized by complements” (1997b, 33), and are compatible with any kind of VP- sequencing or V′-sequencing syntactic analysis. In the simplest case, the principles in (2.62) account for patterns of argument realization in single object non-contiguous SVCs such as (2.63):

` (2.63) a. E. sósà gbé émá!tó. n pè. rhé. . Esosa hit metal be-flat ‘Esosa hit the metal flat.’ (E`. dó; Stewart 2001, 58) b. gbépè. rhé. : λy λx λs {HIT(x,y) & BECOME(FLAT(y))}(s) +sc +sc

The initial (non-situational) argument of V1=HIT (the Semantic Form representation for the initial verb) is x in (2.63b), and this argument (E.`s´os`a) is realized in subject 6 At present, I know of no non-contiguous SVCs in which V1 takes a third argument. Verbs with the semantics of ‘give’, for example, never appear initially in any of the SVCs I have encountered. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 84

position in accord with part (c) of (2.62). But HIT also has another argument, y, and so, in accord with part (d) of (2.62), this argument, émá!tó. n ‘metal’, is realized as the complement of the initial verb gbé ‘hit’ in (2.63a). y is also the argument of BECOME(FLAT) in (2.63b), which encodes the semantic contribution of the second verb, pè. rhé. ‘be flat’. However, in accord with part (e) of (2.62), y = émá!tó. n ‘metal’ cannot also be realized as the complement of this second verb, since this argument is not distinct from the second argument of the initial verb.7 If, as one option, we were to adopt Kroeger’s (2004) V′-sequencing analysis of non-contiguous SVCs, then the structure for (2.63) would be:

(2.64) S

NP VP

Esosa V′ V′

V NP V

hit metal be.ﬂat

While single object non-contiguous SVCs such as (2.63) present some of the simplest illustrations of how the argument realization principles for non-contiguous SVCs in (2.62) come into effect, the next two sections will concentrate on multiple object non-contiguous SVCs, since these illustrate the effects of the argument realization principles more comprehensively and explicitly. There are three crucial generalizations about non-contiguous SVCs captured by the principles in (2.62) that multiple object non-contiguous SVCs can in particular serve to illustrate. First, these principles capture the fact that non-contiguous SVCs like (2.65) below (=(2.52)) display SVC-specific clause structure, where there is a greater number and/or a different arrangement of objects than in a simplex clause. With specific reference to (2.65), they

7 Under an alternative analysis of (2.63), y could be projected into the object position of pè. rhé. ‘be flat’ as a kind of null element, coreferenced with the (overtly realized) object of gbé ‘hit’. However, in the absence of any evidence necessitating this more complicated structure, I adopt a simpler analysis under which there is no null object sharing. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 85

account (in ways which will be made explicit in the next section) for how this SVC can have two verb objects (when a simple Yoruba clause can contain at most one), and in this SVC-speciﬁc arrangement where each object is the object of a separate verb:

(2.65) Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5)

Second, the argument realization principles for non-contiguous SVCs ensure that the right arguments are realized as the complements of the right verbs (so as.o. ‘clothing’ is realized as the object of wo. ‘wear’ and Olú as the object of fún ‘give’ in (2.65), and not vice versa). Third, they account for how it is that some non-structural arguments, such as the benefactive argument Olú in (2.65), come to look as if they are structural in being realized as the objects of verbs (rather than as the objects of prepositions, or in oblique semantic cases). For the purpose of demonstrating how an SVC like (2.65) satisfies the Theta Structure Property, it is important to acknowledge that the structural appearance of Olú in this SVC is only superficial, and a product of the generalization in (2.62), which places this argument in the object position of a verb. Olú is still a non-structural, benefactive argument, and it is only through recognizing it as such that (2.65) will satisfy the Theta Structure Property, as Section 2.4.1.2 will now explain.

2.4.1.2 Argument realization in multiple object non-contiguous SVCs where only one object is structural

The first objective of this section is to provide a more detailed and specific account of how the argument realization principles for non-contiguous SVCs deliver the correct word order for (2.65), repeated below in (2.67a). Now any such more detailed account necessarily begins with the following benefactive serialization lexical rule, which accounts for what is apparently the case that that part of the Seman- tic Form representation which encodes the semantic contribution of fún ‘give’ is in CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 86

fact BENEFIT(y), and not ACT(x) & BECOME(POSS(y,z)) (the standard semantic representation for the lexical item ‘give’):

(2.66) Benefactive serialization lexical rule:

′ ′ ′ Verb1 ‘give’: λy ... λx λs λs {V1(x, ...)(s ) & BENEFIT(s ,y) }(s) −sc +sc

Because Aje is not interpreted as actually giving Olu the clothing in (2.67a), the benefactive serialization lexical rule applies to produce the semantic representation in (2.67b):

(2.67) a. Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5) ′ ′ ′ b. wo. fún: λz λy λx λs λs {WEAR(x,y)(s ) & BENEFIT(s ,z)}(s) −sc +sc +sc

Serialization lexical rules, and the conditions under which they apply, will be discussed more generally in Section 3.2.2; the current concern is to demonstrate how the argument realization generalizations for non-contiguous SVCs ensure the correct ordering of verbs and NPs in SVCs such as (2.67). Towards these ends, it can be observed that x (Aje), as the initial argument of V1=WEAR is predictably realized as subject in accord with part (c) of (2.62). y (as.o. ‘clothing’), the next argument of V1=WEAR, is projected into the syntax as the object of wo. ‘wear’, in accord with part (d) of (2.62). In addition, V2=BENEFIT takes a (non-situational) argument in this Semantic Form representation, z, which is lexically marked as oblique ([-sc]). Fún ‘give’ is the verb which signifies this meaning of BENEFIT (as established by the benefactive serialization lexical rule which also establishes the [-sc] status of z), and so Olú, the syntactic realization of z, appears as the complement of this second verb in accord with part (e) of (2.62), since this argument is distinct from both the x and y arguments realized so far. (2.67) satisfies the Theta Structure Property for SVCs because its Theta Structure consists of two structural arguments and an oblique, non-structural argument that can CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 87

be identified through particular predicate configurations as a benefactive, and there do exist single-verb-headed clauses similarly containing two structural arguments and an oblique benefactive argument. Ordinarily, an argument which is lexically marked as oblique (and therefore not a candidate for structural status) will be realized through a semantic case or as the object of a adposition. What is unusual about multiple object serializations such as (2.67) is that the oblique argument (here Olú) is instead realized as the object of a verb, and hence has the surface appearance of a structural argument. The argument realization generalizations for non-contiguous SVCs are the means through which such oblique arguments come to look structural, since they force an oblique argument to be realized as the object of a verb in those cases where the verb is linked, through a serialization lexical rule, to a Semantic Form predicate whose internal argument is lexically marked as oblique. So Olú in (2.67) is not structural, it only looks structural in being required to be realized as the complement of the verb fún ‘give’. As.o. ‘clothing’, in contrast, is a structural argument. The other example of a multiple object non-contiguous SVC with one structural object I will consider in this section is the instrumental SVC in (2.69) below. Central to the case study of instrumental SVCs in Chapter 4 is the instrumental serialization lexical rule in (2.68) which accounts for how lá ‘take’ does not mean TAKE in an SVC such as (2.69), but is instead linked to the semantics of the actor USING the instrument:

(2.68) Instrumental serialization lexical rule:

‘take’ Verb2: λz λy λx λs {V2(x,(y)) & USE(x,z) }(s) −sc +sc +sc

(2.69) a. Tsoda lá egbà ba cigbà. Tsoda take axe cut tree ‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63) b. lába: λz λy λx λs {CUT(x,y) & USE(x,z)}(s) −sc +sc +sc

While part (c) of the argument realization principles for non-contiguous SVCs de-

termines that x (Tsoda), the initial argument of V1=CUT, is realized as subject in CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 88

(2.69), part (d) determines that y (cigb`a ‘tree’) is realized as the object of ba ‘cut’. Part (e) then determines that z (egb`a ‘axe’), as the (non-situational) argument of

V2=USE distinct from both x and y, is realized as the complement of lá ‘take’ (since lá is attributed the semantics of USE through the instrumental serialization lexical rule). The Theta Structure for (2.69) consists of two structural arguments and one non-structural, instrument argument, which may however have a structural appearance by dint of being realized as the complement of the verb lá ‘take’. Since there do exist single-verb-headed clauses whose Theta Structures similarly consist of two structural arguments and one non-structural, instrumental argument, (2.69) satisfies the Theta Structure Property for SVCs.

2.4.1.3 Argument realization in multiple object non-contiguous SVCs where two objects are structural

The types of multiple object non-contiguous SVCs which may have two structural objects are those which describe concepts otherwise commonly expressed in simplex clauses headed by ditransitive verbs. (2.70) below is one such SVC, although it is interesting that Lefebvre and Brousseau’s (2002) semantic analysis of this example is instead more in accord with the position that it has only one structural object (like the SVCs in Section 2.4.1.2). This section considers Lefebvre and Brousseau’s semantic analysis for Fongbe SVCs like (2.70) before positing an alternative analysis

and semantic representation.

(2.70) Kkú s às n ná As´ıbá.` Koku take crab def give Asiba

‘Koku gave the crab to Asiba.’ (Fongbe; Lefebvre & Brousseau 2002, 448)

Section 2.2 contained the first suggestion that As´ıbá` in (2.70), unlike Kkú and às n ‘the crab’, might be a non-structural argument, lexically marked as oblique. This suggestion was in fact based on Lefebvre and Brousseau’s (2002) arguments that As´ıbá` is not a recipient/possessor in (2.70), but a locational goal. Goals are usually realized as the arguments of adpositions or in oblique semantic cases, but in the multiple object, SVC-specific clause structure of (2.70), As´ıbá` , the proposed goal, is instead CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 89

realized as the object of the verb n´a ‘give’. Of course, this can only happen if n´a ‘give’ does not actually have the semantics of ‘give’ (ACT(x) & BECOME(POSS(z,y))) in (2.70a), but instead has its semantics altered, through the following serialization lexical rule:

(2.71) Goal serialization lexical rule:

Verb1 ‘give’: λz λy λx λs {V1(x,y) & BECOME(LOC(y,AT(z))) }(s) −sc +sc +sc

The Semantic Form representation for (2.70) would then be: (2.72) s n´a: λz λy λx λs {TAKE(x,y) & BECOME(LOC(y,AT(z)))}(s) −sc +sc +sc

This SVC would then satisfy the Theta Structure Property for SVCs because its Theta Structure would consist of two structural arguments, and one non-structural, goal argument, and there do exist single verb-headed clauses whose Theta Structures similarly consist of two structural arguments and a non-structural goal argument. But there is something not quite right about (2.72) as a semantic representation for (2.70). (2.72) signifies that x took y to z; entirely absent from this representation is the concept that z in some way comes to possess y, which is what the use of the verb ‘give’ in the translation for (2.70) would seem to entail. This observation calls for closer examination of Lefebvre and Brousseau’s (2002) claims that the object of ná ‘give’ is not a recipient (which is a type of possessor, as well as being a goal (Rappaport Hovav and Levin 2008)) but is instead purely a goal, devoid of any possessional meaning. Lefebvre and Brousseau’s first argument in support of the non-recipient, pure goal status of the object As´ıbá` in (2.70a) turns upon the fact that As´ıbá` need not possess the crab, as the clause following the SVC in (2.73) demonstrates:

CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 90

(2.73) Kkú s às n ná As´ıbá` v às n ny´ı As´ıbá` t n Koku take crab def give Asiba but crab def be asiba gen ˇa. Neg ‘Koku gave the crab to Asiba but the crab is not hers.’ (Fongbe; Lefebvre & Brousseau 2002, 448)

Lefebvre and Brousseau contrast this with the Fongbe double object construction, in which As´ıb´a` is argued to be a true recipient, as demonstrated through the inability

to negate the entailment of possession:

(2.74) #Kkú ná As´ıbá` às n v às n ny´ı As´ıbá` t n ˇa. Koku give Asiba crab but crab be Asiba gen Neg #‘Koku gave Asiba crab but the crab is not hers.’ (Fongbe; Lefebvre & Brousseau 2002, 447)

Lefebvre and Brousseau (2002) also claim that, because the object of n´a ‘give’ in (2.75) is a place, which cannot literally receive or possess an object, this object must

be purely a locational goal:8

(2.75) Kkú s àkw ná Kùt nû. Koku take money give Cotonou ‘Koku gave money to Cotonou.’ (Fongbe; Lefebvre & Brousseau 2002, 449)

However, cities can still metaphorically possess or receive money, while (2.73) may not in reality argue against the recipient/possessor status of the object of n´a ‘give’ either, as we will now consider. The kind of data usually used to argue that an argument has the status of goal as opposed to recipient/possessor (or vice versa) concerns what Rappaport Hovav and Levin (2008) describe as ‘successful transfer inference’. The inability to cancel

8In contrast, places are infelicitous as recipients in the double object construction:

Ç ¡Ç ¡e K kú ná Kùt nû àkw . Koku give Cotonou money [Lit.: ‘Koku gave Cotonou money.’] (Fongbe; Lefebvre & Brousseau 2002, 448) CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 91

out the transfer inference, as in (2.76), is usually taken as conﬁrmation of the recipient/possessor status of the non-theme internal argument (Sally in (2.76)), while the ability to cancel the transfer inference, as in (2.77), is usually linked to the supposed goal status of the non-theme internal argument (Bill in (2.77)):

(2.76) #Jane gave Sally the book, but she never got it.

(2.77) Jim threw Bill the lifejacket, but it didn’t reach him.

Rappaport Hovav and Levin (2008) actually demonstrate that the ability to cancel these transfer inferences may be more closely tied to the individual semantics of verbs than to the question of goal versus recipient/possessor status, but the main point that needs to be made in connection with (2.73) is that Lefebvre and Brousseau do not appear to have canceled the right kind of inference for their intended purpose. The inference they would in theory need to cancel in order to establish goal status is one of successful transfer (so, something more along the lines of ‘Koku gave the crab to Asiba, but she never got it’), not the kind of long-standing possessor status of the theme that (2.73) describes. As´ıb´a` is probably a kind of recipient, that possesses the theme, in the Fongbe SVC (2.78a) (=(2.70)). In that case, I propose (2.78b) as a more accurate semantic

representation for this SVC:

(2.78) a. Kkú s às n ná As´ıbá.` Koku take crab def give Asiba ‘Koku gave the crab to Asiba.’

(Fongbe; Lefebvre & Brousseau 2002, 448) b. s n´a: λz λy λx λs {TAKE(x,y) & BECOME(POSS(z,y))}(s) +sc +sc +sc

(2.78b) signiﬁes that z is a recipient/possessor through its position within the BE-

COME(POSS) predication (this is the predication V2 corresponding to Verb2 = ná ‘give’, where z is realized as the object of ná in accord with part (e) of the argument realization principles for non-contiguous SVCs). (2.78b) also signifies that y, as the second argument of TAKE, undergoes a change of location – something not entailed in CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 92

the standard semantic representation for ‘give’ (ACT(x,y) & BECOME(POSS(z,y))). But perhaps of most signiﬁcance in (2.78b) is the structural status attached to z. This arises out of this argument not looking to have the semantics typically associated with an adposition or oblique semantic case, as well as its fulﬁllment of the the Structural Argument constraint, since z L-commands the lowest argument, y, in the following tree representation:

(2.79) t

<< s,t >< s,t >>

> x & TAKE y BECOME t

POSS y

If z is structural, then SVCs like (2.78a) would present the only cases of multiple object, non-contiguous SVCs with three structural arguments. However, (2.78a) does

not instantiate three distinct structural object positions; it more likely realizes one argument (Kkú) in first structural argument position and the other two structural arguments in two different second structural argument positions. In any case, (2.78) will satisfy the Theta Structure Property for SVCs since it has three structural arguments at Theta Structure, as do many (ditransitive) simple clauses in other languages (and indeed in the same language, since Fongbe has a double object construction). It remains to be seen whether there are any serializing languages in which a ‘give’ verb has its semantics altered through a serialization lexical rule to a purely locational CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 93

goal meaning as Lefebvre and Brousseau suggest. The limited data I have available (from languages such as Yoruba (Awoyale 1988), Saramaccan (Veenstra 1996), Thai

(Thepkanjana 2010) and }Hoan (Collins 2002)) would suggest that the alteration to a ‘give’ verb effected by the benefactive serialization lexical rule in (2.66) is instead more commonplace. This concludes the discussion of how the argument realization generalizations for non-contiguous SVCs described in (2.62) play out in the mapping from Theta Struc- ture/Semantic Form to Morphology/Syntax, resulting in SVC-specific clause structures where objects appear between verbs. The next section asks whether contiguous SVCs may have any SVC-specific clause structures (where objects do not, however, appear between verbs), which may require analogous non-standard mapping principles.

2.4.2 Contiguous SVCs

(2.80)-(2.84) are some examples of contiguous SVCs, in which the verbs appear side by side:

(2.80) A ruene paa hapusu oha-u. art water past ﬂow go.over-imm ‘The water has ﬂowed over.’ (Teop; Reinig 2004, 93)

(2.81) Nekepo akuaku vakao loa i thaug-ona na. part weed encircle emph loc house-3sg.poss dem ‘She weeded around her house.’ (Pileni; Naess 2004, 235)

(2.82) Nek me-yeye mat nek ae egen! 2sg pft-laugh die 2sg ana now ‘You’re laughing yourself to death!’ (Mwotlap; Francois 2004, 118)

(2.83) I thege oxo thaamwa. 3sg run follow woman ‘He runs after the woman.’ (Nˆelˆemwa; Bril 2004, 186) CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 94

(2.84) Amu ngaha ka.mpoi.na oha. Amu eat caus.gone.3 rice ‘Amu ate up all the rice.’ (Bima)

Syntactic analyses for contiguous SVCs do not abound in the same way as syntactic analyses for non-contiguous SVCs on account of contiguous SVCs hardly being a well recognized phenomenon. Kroeger (2004), however, does propose a V-sequencing analysis for contiguous SVCs (in contrast to his V′-sequencing analysis for non-contiguous SVCs), according to which (2.84) would receive the following structure:

(2.85) S

NP VP

Amu V′ NP

V V rice

eat caus.gone

(2.84) satisﬁes the Theta Structure Property for SVCs because its Theta Structure consists of two structural arguments, and there do exist clauses headed by single verbs with two structural arguments ((2.82) and (2.83) likewise contain two structural arguments, while (2.80) contains one structural argument and (2.81) one structural argument plus a non-structural, locative argument). But (2.84) not only has the Theta Structure of a single-verb-headed clause; its arguments are also projected into the same surface syntactic positions as in a single-verb-headed clause in this language. (2.86a) is single-verb-headed Bima clause, and (2.86b) is a syntactic representation for it, with the subject and object projected into the same positions as in (2.85):

(2.86) a. Amu ngaha.na oha. Amu eat.3 rice ‘Amu ate (the) rice.’ (Bima) CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 95

b. S

NP VP

Amu V′ NP

V rice

eat

(2.84) does not involve SVC-specific clause structure, and nor do any other SVCs in Bima. In fact, in each of the SVCs in (2.80)-(2.84), the arguments are projected into syntactic positions in exactly the same way they would be in the simplex clauses of those languages. (2.80)-(2.84) do not call into account any SVC-specific argument projection principles for contiguous SVCs. The main question addressed in this section is whether contiguous SVCs can ever involve SVC-specific clause structure, for which special argument realization principles might have to be formulated (to match the non-contiguous argument realization generalizations in (2.62) of Section 2.4.1), and this naturally requires as wide a survey of contiguous SVCs as possible. (2.80)-(2.84) are all drawn from Oceanic languages, which may reflect a reality that Oceanic languages favor contiguous SVCs more strongly than any other language family. Yet, as mentioned in Chapter 1, the apparent absence of contiguous SVCs in other language families (and in particular in African language families, which favor non-contiguous SVCs) most likely also arises as a consequence of researchers being strongly instilled with the belief that SVCs must be non-contiguous in the first place. Contiguous SVCs in other language families may then be overlooked or analyzed as compounds. Baker (1989), for example, claims that any multi-verb structures with contiguous verbs must be compounds, and Collins (2002) analyzes the following

sentences from }Hoan as compounds, without any indication of whether or not the

contiguous verbs involved constitute a single phonological word: (2.87) Ma a- qhu ’o djo ki kx’u na. 1sg prog pour put.in water part pot in

‘I am pouring water into the pot.’ (}Hoan; Collins 2002, 3)

CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 96

(2.88) Ma oe na ka hoam- hoam tca. 1sg still aux sub jog come

‘while I was still coming jogging’ (}Hoan; Collins 2002, 4)

(2.89) Ma qo k´ı- tsaxo ’am a”e. 1sg fut pl cook eat meat

‘I will cook and eat meat (repeatedly).’ (}Hoan; Collins 2002, 4)

(2.90) Ya i tc’eon o’a tsi. 3sg past make absent 3pl

‘He ﬁnished making them.’ (}Hoan; Collins 2002, 4) (2.91) Gya”m-a’a a- tsaxo cu ’am gye ki a”e. child-dim.pl prog cook give 1sg mother part meat

‘The children are cooking meat for my mother.’ (}Hoan; Collins 2002, 4)

If the verb combinations in (2.87)-(2.91) do not constitute single phonological words (and the absence of hyphens would suggest that they do not), then (2.87)-(2.91) should be analyzed as contiguous SVCs. From this perspective it would be unsurprising that (2.87)-(2.91) display exactly the same semantic functional make-up as SVCs, although supposing (2.87)-(2.91) were to be analyzed as compound verbs, this would reinforce the point that SVCs and compound verbs do share a large amount of

functional overlap. Collins (2002) comments on the similarities between these }Hoan ‘verbal compounds’ and (non-contiguous) SVCs in other African languages, and for

each of the }Hoan sentences in (2.87)-(2.91), he provides a non-contiguous analogue with the same function. The overall point of his paper is to account for the functional similarities he observes through a multiple verb movement analysis which derives

these }Hoan sentences from underlying SVC structures. One commonly instanti-

ated type of SVC for which there is no }Hoan analogue, however, are instrumental

SVCs; whereas (2.92) is an acceptable non-contiguous instrumental SVC, the }Hoan sentence in (2.93) is ungrammatical:

(2.92) Nws muab riam txiav nqiaj qaib. 3sg take knife cut meat chicken ‘She cut some chicken with a knife.’ (White Hmong; Jarkey 1991, 83)

CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 97 (2.93) *Ma a- (k´ı-) !u tcxa a”e ki g ama. 1sg prog pl take cut meat part knife

(}Hoan; Collins 2002, 20)

In Section 2.4.1 I observed that non-contiguous SVCs can have biclausal appearances, often by dint of multiple object serialization which involves a phrase structure with a greater number, and different arrangement, of syntactic positions for arguments than in a simplex clause. (2.92) bears out this observation as an instrumental multiple object serialization where instrument and ‘theme’ arguments are realized in different constituents, as the objects of different verbs. Instrumental SVCs necessitate SVC-specific clause structure for non-contiguous SVCs then, but in reverse it could be

asked whether contiguous SVC languages such as }Hoan disallow instrumental SVCs because they have no SVC-speciﬁc clause structures – because, in other words, SVCs in contiguous SVC languages must have identical conﬁgurations of syntactic positions for arguments as in simplex clauses. The only case of a grammatical contiguous instrumental SVC I have encountered is the following Barai example:

(2.94) Fu burede ije sime abe ufu. he bread def knife take cut ‘He cut the bread with a knife.’ (Barai; Foley & Olson 1985, 44)

Unfortunately, I have been unable to ascertain whether the syntactic conﬁguration of arguments in (2.94) constitutes SVC-speciﬁc clause structure in the absence of any information about the simplex structure of the Barai clause. Burede ije ‘the bread’ and sime ‘knife’ have the surface syntactic appearance of objects (even if sime ‘knife’ is determined not to be structural at Theta Structure), making (2.94) a multiple object serialization. What is unclear at this stage is whether simplex clauses in Barai can contain multiple objects. The benefactive SVCs in (2.95) and (2.96) (= (2.91)) below both instantiate multiple object serialization in contiguous SVCs:

(2.95) Mi ruat d´oh au phei. 2sg buy give me rice ‘You buy rice for me.’ (Jeh; Kroeger 2004, 240)

CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 98 (2.96) Gya”m-a’a a- tsaxo cu ’am gye ki a”e. child-dim.pl prog cook give 1sg mother part meat

‘The children are cooking meat for my mother.’ (}Hoan; Collins 2002, 4)

Here the question is whether }Hoan and Jeh allow simplex double object structures, where the ﬁrst object is a recipient/benefactive. The following sentence demonstrates that Jeh does have simplex double object structures, since d`oh ‘give’ takes two objects in the matrix clause:

(2.97) Au d`oh [mi] [lian] la ruat d`oh au phei. I give you money in-order-to buy give me rice ‘I’ll give you money to buy rice for me.’ (Jeh; Cohen 1976, 158)

This signifies that the Jeh SVC in (2.95) does not instantiate SVC-specific clause } structure. If an analogous example could be found in }Hoan, then the Hoan SVC in (2.96) would also not be said to instantiate SVC-specific clause structure; but if such simplex double object constructions do not exist, then (2.96) would instantiate SVC-specific clause structure. This issue awaits further evidence. SVC-specific clause structure for contiguous SVCs is a theoretical possibility which leaves us on the trail of multiple object serializations in contiguous SVCs. Multiple object serializations involve a sequence of (at least) two transitive verbs, where these verbs do not share the same object. Sequences of transitive verbs in contiguous SVCs almost always (with Barai (2.94) as one exception) share the same object, making them not multiple object, but single object serializations. However, one final observation for this section is that some contiguous SVC languages evidently avoid multiple object serialization by suppressing the internal arguments of all but one transitive verb (meaning that these arguments need not be shared). This happens occasionally in contiguous Bima SVCs; the internal arguments of all but the final verb are suppressed. It is only the internal argument of the final verb in (2.98), heko ‘encircle’, for example, that is projected into the syntax, as asi ede ‘the palace’:

(2.98) pidu kali d’end’e heko asi ede, ... seven time accompany encircle palace dem ‘seven times they accompanied (them) around the palace, ...’ (Bima) CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 99

Suppression of the internal arguments of the non-ﬁnal verbs also occurs in the following two Bima SVCs:

(2.99) Reho hengga ka.losa.na harimau (ku ra beba.na). Reho open caus.exit.3 tiger purpose free.3 ‘Reho opened (the cage door) and brought the tiger out (so he was free).’ (Bima)

(2.100) Ede.ra lao fou b’atu kai.na ra’a ed’i Ama Seho ede, dem.emph go chase follow kai.3 blood leg Ama Seho dem ‘So he chased (Ama Seho), following the (trail of) blood from Ama Seho’s leg, ...’ (Bima)

In (2.99), the internal argument of the first verb hengga ‘open’ is the cage door, yet it is only harimau ‘tiger’, the internal argument of the second verb ka.losa ‘let out’, which is realized as an object in the syntax. In (2.100), the internal argument of the verb fou ‘chase’ is Ama Seho, yet it is ra’a ed’i Ama Seho ede ‘the blood from Ama Seho’s leg’, the internal argument of the second verb b’atu ‘follow’, which is realized as an object in the syntax. Since (2.99) and (2.100), unlike (2.98), are Impossible Verb SVCs, suppression of non-final arguments is effected through the Predicate Combining Operations for Impossible Verb SVCs. These operations will be set out in Chapter 3, where (2.99) and (2.100) will also be revisited, but one final comment concerning the Possible Verb SVC in (2.98) is that it resembles, to some extent, resultative sentences such as Myrtle drank Jim’s fridge empty. Here the internal argument of empty (Jim’s fridge) is realized in place of the suppressed internal argument of drink (which will refer to the actual substance or quantity drunk), which in LDG terms corresponds to the suppression of the argument y in the Semantic Form representation DRINK(x,y) & BECOME(EMPTY(z)), on account of its failure to satisfy the Structural Argument Constraint (Kaufmann and Wunderlich 1998, 7). Similarly in (2.98), the internal argument of d’end’e ‘accompany’ (the ‘them’ that refers to a bride and bridegroom), or y in {ACCOMPANY(x,y) & ENCIRCLE(x+y,z)}, is suppressed, while the internal argument of heko ‘encircle’ (‘the palace’), or z, is structural and overtly realized. More research (or more data) is required to establish the extent to which internal argument suppression in SVCs resembles internal argument suppression in non-SVC CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 100

structures (semantically, or otherwise). However, one general observation that can be drawn from these argument suppression cases in (2.98)-(2.100) is that these signify another dimension of ‘eﬀort’ some SVCs will go to in order to appear argument structurally monoclausal. If SVCs are multi-verb constructions which are acting as if they contain only one verb, the kind of argument suppressions exhibited in (2.98)- (2.100) could be seen as a special kind of strategy for acting argument structurally as if there is only one verb.

2.4.3 Conclusion

This section, on projecting the arguments of contiguous and non-contiguous SVCs, has been extremely general in terms of exactly how the relevant arguments are projected, or what the associated phrase structures look like. But then the most important point to emphasize is that there is a contiguous versus non-contiguous distinction in the ﬁrst place. For too long, contiguous SVCs have been excluded from the SVC literature. Phrase structures, or in fact any type of syntax, will not feature very prominently in this dissertation. Most of the focus is directed towards semantic, event structural properties of SVCs, while more prominent still are the argument structural properties of SVCs, since it is at LDG’s argument structure levels of Theta Structure and Semantic Form that the Theta Structure Property for SVCs is formulated. It is moreover because the Theta Structure Property for SVCs is formulated at these levels that it imposes few restrictions on the syntax, and is unproblematically inclusive of contiguous, as well as non-contiguous, SVCs. In fact, if either type of SVC were held to be more problematic, it would probably be the more traditional, core class of non-contiguous SVCs, based on the kinds of complications involved in projecting non-contiguous arguments into the syntax described in Section 2.4.1. The ‘complications’ involved in projecting non-contiguous arguments into the syntax are captured in the argument realization generalizations for non-contiguous SVCs set out in (2.62) of Section 2.4.1. These generalize across non-contiguous SVCs, and various possible phrase structures for non-contiguous SVCs, to produce the word orders which characterize this type of SVC. But the most crucial aspect of the argument CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 101

realization generalizations for non-contiguous SVCs in terms of supporting the Theta Structure Property for SVCs was the demonstration in Section 2.4.1 of how a multiple object serialization with SVC-speciﬁc clause structure, such as (2.101) (=(2.67)), will still satisfy the Theta Structure Property, as a consequence of the reality that some arguments which might look structural (such as Ol´u in (2.101)) actually are not.

(2.101) a. Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5) ′ ′ ′ b. wo. fún: λz λy λx λs λs {WEAR(x,y)(s ) & BENEFIT(s ,z)}(s) −sc +sc +sc

z in (2.101b) is the benefactive argument of BENEFIT, lexically marked as oblique ([-sc]). BENEFIT would ordinarily be realized through an adposition or semantic case, but it so happens that in (2.101) this predicate is associated with a verb, fún ‘give’, through the application of a (benefactive) serialization lexical rule. The argument realization generalizations for non-contiguous SVCs determine that z will be realized as the object of fún ‘give’, although this argument still retains its non- structural status. Taking this into account, (2.101) satisfies the Theta Structure Property for SVCs because its Theta Structure consists of two structural arguments and one non-structural, benefactive argument, and there do exist single-verb-headed clauses whose Theta Structures consist of two structural arguments and one non- structural, benefactive argument. (2.102a) below also satisfies the Theta Structure Property for SVCs, because its Theta Structure in (2.102b), like the Theta Structure of many single-verb-headed clauses, consists of two structural arguments and one non-structural, instrument argument. It just happens that the type of argument realization which characterizes non-contiguous SVCs determines that the non-structural instrument argument is realized as the complement of the verb lá ‘take’ in the SVC- specific clause structure of (2.102a): CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 102

(2.102) a. Tsoda lá egbà ba cigbà. Tsoda take axe cut tree ‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63) b. lába: λz λy λx λs {CUT(x,y) & USE(x,z)}(s) −sc +sc +sc

The possibility for multiple object realization in the SVC-specific clause structures of non-contiguous SVCs means that this type of SVC has considerable scope for ‘argument introducing’; fún ‘give’ in (2.101) and lá ‘take’ in (2.102), for example, introduce benefactive and instrument arguments respectively, and in this capacity they can moreover be regarded as ‘acting’ like adpositions. The scope for ‘argument introducing’ is much more limited in contiguous SVCs, although it would be expanded by any clear cases of SVC-specific clause structure in contiguous SVCs. The main focus for Section 2.4.2 was on the question of whether contiguous SVCs might ever involve SVC-specific clause structures, with arguments projected into positions that differ from those in a simplex clause, and which would require specific argument realization principles (analogous to the argument realization generalizations for non- contiguous SVCs). One possible case of contiguous SVC-specific clause structure included in Section 2.4.2 was the following Barai instrumental SVC:

(2.103) Fu burede ije sime abe ufu. he bread def knife take cut ‘He cut the bread with a knife.’ (Barai; Foley & Olson 1985, 44)

More research is needed to ascertain whether (2.103) does involve SVC-specific clause structure, whether contiguous SVCs in other languages involve SVC-specific clause structure, how widespread this is, and what the nature of the argument realization principles for such SVC-specific clause structures would be (these could even differ from language to language). It is really only once contiguous SVCs gain more widespread acceptance as a type of SVC that these issues can begin to be addressed. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 103

2.5 Conclusion

This chapter stated and demonstrated the Theta Structure Property for SVCs, which I argue to be the most eﬀective and cross-linguistically comprehensive way of enforcing monoclausal argument structure in SVCs. This property can now be listed with the other properties which enforce monoclausal grammatical structure in SVCs, resulting in what I argue to be the best means of deﬁning and identifying SVCs:

(a) SVCs are monoclausal.

(b) The verbs in an SVC must have the same tense, aspect, mood, polarity, and person-number categories.

(d) The intonational properties of a clause containing a serial verb combination are the same as those of a clause containing a simple predicate in the language concerned.

(e) The Theta Structure Property for SVCs: An SVC has the Theta Structure of a Possible Verb structure. Diagnostic for an SVC meeting the Theta Structure Property: If a multi-verb construction has the same number of structural arguments identified at Theta Structure, and the same list of configurationally defined, overtly realized non-structural (oblique) arguments as some clause headed by a single, simplex verb, then it meets the Theta Structure Property for SVCs.

Section 1.2.2 described how many typologically rich functional approaches to SVCs simply list the argument sharing schemes that SVCs display, without attempting to delimit or deﬁne SVCs through argument structure properties. This could mean the failure to exclude covert coordinations, such as (2.104), from the class of SVCs: CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 104

` (2.104) Ozó gbò. ó. ´ıv`ın bòló ó. là. Ozo planted coconut peel corn ‘Ozo planted coconut and peeled corn.’ (Edo; Stewart 2001, 51)

Internal Argument Sharing is the property that has dominated the formal literature on SVCs over the last 20 years, and this property excludes sentences such as (2.104) from the class of SVCs as a consequence of each verb realizing distinct internal arguments instead of sharing a single internal argument. However, I also argued (in Section 1.2.2), that Internal Argument Sharing has such major flaws as to render it an unsuitable defining property for SVCs. Chief among these flaws is the fact that it excludes commonly occurring SVCs containing intransitive+transitive combinations of verbs, such as (2.105), from the class:

(2.105) Taho mpara lao te’e.ku tamba. ﬁne then go install.1 trap ‘Okay then, I’ll go and install a trap.’ (Bima)

But it is just as problematic that Internal Argument Sharing rules out the following (‘comparative’, ‘resultative’ and ‘manner modifying’) constructions as SVCs, since in each case the verbs do not share an internal argument:

` (2.106) Kkú kló hú As´ıbá. Koku be.big surpass Asiba ‘Koku is bigger than Asiba.’ (Fongbe; Lefebvre & Brousseau 2002, 401)

(2.107) Ne-len mi-yip hal-yak na.kat. art-wind perf-blow ﬂy-away art-cards ‘The winds blew the cards away.’ (Mwotlap; Francois 2004, 119)

(2.108) Wela e qiliano-na taqaa baqu. child 3sg:nfut pile.soil.around-3:obj be.bad banana ‘The child piled the soil around the banana tree badly.’ (Toqabaqita; Aikhenvald 2006, 29)

(2.105)-(2.108) may not involve Internal Argument Sharing, but they do realize monoclausal argument structure, as well as displaying every other monoclausal property CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 105

required of an SVC. In order to prove that the Theta Structure Property for SVCs represents an improvement on Internal Argument Sharing and so should be preferred in its place, then, I need to demonstrate that the Theta Structure Property does include (2.105)-(2.108). (2.105) was one of several SVCs containing a combination of intransitive+transitive verbs included in Section 1.2.2 which does not display Internal Argument Sharing. The intransitive verb in these SVCs usually describes a motion event (in which the actor is the moving entity), while the transitive verb usually describes a subsequent (typically non-motion) event initiated by the actor. As such, these SVCs tend not to satisfy the Possible Verbs constraints, since a clause headed by a single verb cannot usually describe a sequence of two temporally distinct (and causally unconnected) events – (2.105), and the other intransitive+transitive verb SVCs accompanying it in Section 1.2.2 are not Possible Verb, but Impossible Verb SVCs, in other words. Impossible Verb SVCs must undergo certain Predicate Combining Operations in order to fulﬁll the Theta Structure Property for SVCs, but since these will not be set out until the next chapter, the Theta Structure and Semantic Form representations for (2.105) in (2.109b) below simply present the end product of these operations, without being explicit about how they apply:

(2.109) a. Taho mpara lao te’e.ku tamba. ﬁne then go install.1 trap ‘Okay then, I’ll go and install a trap.’ (Bima) b. lao te’e: λy λx λs [{GO(x)} & {INSTALL(x,y)}](s) +sc +sc

From (2.109b) we can see that (2.109a) (=(2.105)) satisﬁes the Theta Structure Prop- erty for SVCs, since its Theta Structure consists of just two structural arguments, and there do exist clauses headed by single verbs which similarly contain two structural arguments (such as I will install a trap). (2.106)-(2.108) likewise all fulﬁll the Theta Structure Property for SVCs by virtue of having two structural arguments at Theta Structure:9

9‘COMPARATIVE’ in (2.110b) is meant to indicate that part of the combined serial verb meaning signified by hú ‘surpass’. A serialization lexical rule would link hú to this predicate, which takes

CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 106 (2.110) a. Kkú kló hú As´ıbá.` Koku be.big surpass Asiba ‘Koku is bigger than Asiba.’ (Fongbe; Lefebvre & Brousseau 2002, 401) b. klóhú: λy λx λs {COMPARATIVE(BIG(x,y))}(s) +sc +sc

(2.111) a. Ne-len mi-yip hal-yak na.kat. art-wind perf-blow ﬂy-away art-cards ‘The winds blew the cards away.’ (Mwotlap; Francois 2004, 119) b. yip hal-yak: λy λx λs {BLOW(x) & FLY(y) & MOVE-AWAY(y)}(s) +sc +sc

(2.112) a. Wela e qiliano-na taqaa baqu. child 3sg:nfut pile.soil.around-3:obj be.bad banana ‘The child piled the soil around the banana tree badly.’ (Toqabaqita; Aikhenvald 2006, 29) b. qiliano taqaa: λy λx λs {PILE-SOIL-AROUND(x,y)(s) & BADLY(s)} +sc +sc

And so the Theta Structure Property for SVCs includes all of the types of SVC which Internal Argument Sharing undesirably excludes from the class, and this is why it should be preferred over Internal Argument Sharing. The essential reason why it is able to be more inclusive in this way is that it focuses on the entire array of arguments combined in an SVC at a designated level of argument structure (Theta Structure in LDG, the level from which λ-abstracted arguments are projected into the syntax). Internal Argument Sharing, in contrast, only focuses on localized combinations of verbs and their internal arguments situated at various positions in syntactic trees. The internal arguments of each verb could display identical syntactic positioning (‘true’ internal argument sharing) or the sharing could be mediated by an empty category; whatever the case, the point is that this mode of operation only focuses on a single argument of each verb in an SVC and the relationship between them, instead of zooming out to the bigger picture of the entire, combined argument structure, and a one-place property concept predicate as its argument and results in a two-place comparative predicate. CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 107

the question of whether it ﬁts the mould of a monoclausal argument structure which could be projected into a pattern of arguments and adjuncts such as is seen for a clause headed by a single verb. The Theta Structure Property for SVCs does ‘pan out’ in this manner, and so gives a more complete and accurate picture of how SVCs are monoclausal in an argument structural respect. (2.109)-(2.112) are monoclausal in an argument structural respect and the Theta Structure Property accordingly rules them into the class of SVCs, while Internal Argument Sharing undesirably rules them out. But what has yet to be shown in this chapter is how the Theta Structure Property for SVCs rules out covert coordinations such as (2.113) (=(2.104)):

` (2.113) Ozó gbò. ó. ´ıv`ın bòló ó. là. Ozo planted coconut peel corn ‘Ozo planted coconut and peeled corn.’ (Edo; Stewart 2001, 51)

Exclusion of covert coordinations from the class of SVCs was the primary motivation for the formulation of Internal Argument Sharing. The Theta Structure Property for SVCs does also exclude covert coordinations from the class of SVCs, although it cannot be shown how it does so until the next chapter, which brings Impossible Verb SVCs into the picture. (2.113), were it to be a genuine SVC, would be an Impos- sible Verb SVC, since it describes two temporally disjoint and causally unconnected events, and cannot satisfy the Possible Verbs constraints. However, (2.113) will be shown not to conform to the Predicate Combining Operations for Impossible Verb SVCs in Chapter 3; specifically, (2.113) does not involve coidentification of the lowest structural arguments for each verb (´ıv`ın ‘coconut’, the lowest structural argument of gbò. ó. ‘plant’ and ó. là ‘corn’, the lowest structural argument of bòló ‘peel’) as the fourth Predicate Combining Operation requires, and for this reason is disqualified as an SVC. The scene is now set for the introduction of Impossible Verb SVCs and their Predicate Combining Operations in Chapter 3. A final comment regarding the Theta Structure Property for SVCs is that there is a sense in which it is really very simple, since all it calls for is the argument structural profile of a single-verb-headed clause at CHAPTER 2. THE THETA STRUCTURE PROPERTY FOR SVCS 108

the levels of Theta Structure and Semantic Form where it is formulated. But Section 2.4 of this chapter demonstrated that it is in the projection from argument structure to the syntax where things can become more complicated, if an SVC instantiates SVC- specific clause structure. SVC-specific clause structure was in this chapter equated with the realization of oblique (non-structural) arguments as the objects of verbs, and not as the objects of adpositions or in oblique semantic cases, as in single-verb-headed clauses. SVC-specific clause structure was shown to be primarily (if not exclusively) a feature of non-contiguous SVCs in Section 2.4. To the extent that SVC-specific clause structures present more complications and challenges to SVC status, there is some irony in their being instantiated in non-contiguous, and not contiguous SVCs, since it is non-contiguous SVCs which constitute the traditional survey of SVCs. Chapter 3

Possible Verb and Impossible Verb SVCs

3.1 Introduction

This chapter discusses the distinction central to this dissertation: that between Pos- sible Verb and Impossible Verb SVCs. Possible Verb SVCs constitute the basic, core class of SVCs and Chapter 2 illustrated this type of SVC almost exclusively, avoiding Impossible Verb SVCs as much as possible. The grounds for distinguishing between Possible Verb and Impossible Verb SVCs are simple: whereas Possible Verb SVCs satisfy LDG’s Possible Verbs constraints (Possible Verbs, Connexion and Coherence, as described in Section 2.2), Impossible Verb SVCs do not satisfy the Possible Verbs constraints, but instead consist of at least two (groupings of) predicates which each satisfy the Possible Verbs constraints independently. Strings of conjoined predicates that jointly satisfy the Possible Verbs constraints are enclosed within a single set of {} brackets in the semantic representations of this dissertation. Adopting this notational convention means that there is a very simple graphical representation of the Possible Verb-Impossible Verb distinction in the Semantic Form representations: whereas the Semantic Form representations of Possible Verb SVCs will have only one set of (conjoined predicates enclosed within) {} brackets, the Semantic Form representations for Impossible Verb SVCs will have at least two sets of (conjoined predicates enclosed

109 CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 110

within) {} brackets. The {} brackets themselves represent the bounds of a possible verb – or rather a possible single-verb-headed complex, since it is as much the kinds of argument extension operations and combinations of peripheral (oblique) arguments expressed at Semantic Form that are subject to the Possible Verbs constraints, besides the characteristics of the single verb itself. Section 3.2 contains a number of demonstrations of how particular Possible Verb SVCs satisfy the Possible Verbs constraints, while Section 3.3 demonstrates how particular Impossible Verb SVCs do not satisfy these constraints. Sections 3.2 and 3.3 also present event structural profiles of Possible Verb and Impossible Verb SVCs, thus providing an additional independent means of identifying these two types of SVC. Since they are specific to Possible Verb SVCs, serialization lexical rules are discussed in Section 3.2. Serialization lexical rules have been mentioned on several occasions already, and were actually central to the analyses of non-contiguous multiple object SVCs in Chapter 2, although it is only now, in this chapter, that some of the issues surrounding these rules will be addressed. However, what is of most consequence in this chapter are the Predicate Combining Operations for Impossible Verb SVCs set out in Section 3.3.2. These operations account for how Impossible Verb SVCs can still satisfy the Theta Structure Property for SVCs even though, in raw form, their verbs correspond to two distinct Theta Structures, making satisfaction of the Theta Struc- ture Property impossible. The Predicate Combining Operations for Impossible Verb SVCs present considerable complication and expense, to the extent that we should really ask whether they are justified – which is really to ask whether Impossible Verb SVCs should be included in the class of SVCs at all. Section 3.4 presents arguments in support of including Impossible Verb SVCs in the class of SVCs. Section 3.5 then discusses the consequences of including Impossible Verb SVCs in the class of SVCs, while Section 3.6 concludes this chapter, looking ahead to the case studies of the following two chapters. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 111

3.2 Possible Verb SVCs

This section explores what it means to be a Possible Verb SVC. Section 3.2.1 first examines some basic cases of how various SVCs do satisfy the Possible Verbs constraints and thus qualify as Possible Verb SVCs. Section 3.2.2 next discusses a device specific to Possible Verb SVCs: serialization lexical rules. These rules, which account for the fact that some verbs in some Possible Verb SVCs have their semantics systematically altered within this SVC context, were first encountered in Chapter 2 with little accompanying comment. Section 3.2.2 addresses the issues of how pervasive they are, and also raises the question of whether Possible Verb SVCs which require serialization lexical rules should still qualify as SVCs. Section 3.2.3 then sets out the three possible event structural profiles for Possible Verb SVCs.

3.2.1 Satisfying the Possible Verbs constraints

The basis for the Possible Verb-Impossible Verb distinction is the question of whether or not an SVC fulfills LDG’s Possible Verbs constraints. These constraints were included in Section 2.2 and demonstrated in relation to the simplex sentences Ein motorrad/ein Fuerwerkskörper/*ein Junge heulte um die Ecke (‘A motorbike/a firework/*a boy howled around the corner’). Here the Possible Verbs constraints are repeated but demonstrated in relation to SVCs, or at least those SVCs which do fulfill the constraints: Possible Verb SVCs. The core component of the Possible Verbs constraints is the Possible Verbs constraint itself:

(3.1) Possible Verbs. In a decomposed SF representation of a verb, every more deeply embedded predicate must specify the higher predicate or sortal properties activated by the higher predicate. (Wunderlich 2000, 25)

A Semantic Form representation which fulﬁlls the Possible Verbs constraint cannot consist of just any string of unrelated predicates; instead there are restrictions on all non-initial predicates, not just that they be related, but in the more extreme form of the requirement that they must further specify the higher predicate, or sortal CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 112

properties activated by the higher predicate. The ultimate result of this requirement is that a Semantic Form representation which fulfills Possible Verbs can be realized as a clause headed by a single verb. Possible Verbs is thus hardly a constraint on what constitutes a possible verb in isolation, but rather it is a constraint on the extent and nature of the information that can be expressed in a single-verb-headed clause. SVCs which fulfill Possible Verbs have single-verb-headed translation equivalents, and in terms of the Semantic Form representations to be set out shortly, they are distinguished by the property that they contain only one set of {} brackets, which signifies that their decomposition predicates have been shown to jointly satisfy the Possible Verb constraints. The Possible Verbs constraint in (3.1) is one of three constraints which together make up the Possible Verbs constraints. The other two are Connexion and Coherence (Wunderlich 2000, 25):

(3.2) Connexion. In a decomposed SF structure, each predicate must share at least one argument with another predicate, either explicitly or implicitly.

(3.3) Coherence. Subevents encoded by the predicates of a decomposed SF structure must be connected contemporaneously or causally.

All SVCs satisfy Connexion; this is true even of Impossible Verb SVCs, where the initial actor (or subject) argument (which is the highest structural argument at Theta Structure) is shared between the predicates. But not all SVCs satisfy Coherence; Section 3.3.1 below includes some event structural representations for Impossible Verb SVCs where the subevents are temporally disjoint and causally unconnected as well. Possible Verb SVCs, however, do always satisfy Coherence, and the event structural representations for Possible Verb SVCs set out in Section 3.2.3 reinforce the fact that subevents in Possible Verb SVCs are always connected contemporaneously or causally. Coherence will not feature as prominently in this section as the Possible Verbs constraint itself, however, since it is the latter constraint which goes straight to the heart of what it is for an SVC to be a Possible Verb SVC. The Possible Verb SVC in (3.4) below presents one simple case of the more deeply embedded predicate in the corresponding Semantic Form representation specifying CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 113

the higher predicate. Taqaa ‘be bad’ functions as a manner modiﬁer in (3.4a) and its corresponding predicate in the Semantic Form representation in (3.4b), BAD, accordingly speciﬁes the higher predicate, PILE.SOIL.AROUND, describing the manner in which this action is performed:

(3.4) a. Wela e qiliano-na taqaa baqu. child 3sg:nfut pile.soil.around-3:obj be.bad banana ‘The child piled the soil around the banana tree badly.’ (Toqabaqita; Aikhenvald 2006, 29) b. qiliano taqaa: λy λx λs {PILE.SOIL.AROUND(x,y)(s) & BAD(s)} +sc +sc

Since BAD, in modifying PILE.SOIL.AROUND, specifies the semantic activation space opened up by this higher predicate in (3.4b), this SVC satisfies Possible Verbs. It satisfies Connexion redundantly, since BAD does not take any arguments (other than the situational argument s, that is shared with PILE.SOIL.AROUND), while it satisfies Coherence by default in that there is not more than one distinct subevent in (3.4) (taqaa ‘be bad’ does not correspond to a subevent; this analysis will be explicated in Section 3.2.3 below). Satisfaction of the three Possible Verbs constraints means that (3.4a) properly qualifies as a Possible Verb SVC, which is why the two predicates in the Semantic Form representation in (3.4b) are appropriately placed within a single set of {} brackets, symbolizing this Possible Verb status. In (3.5b), MOVE-AWAY-FROM-DC is a lower predicate which signifies movement away from the deictic center, and because this predicate supplies more specific information within the interpretation space opened up by the higher predicate RUN, (3.5a) satisfies Possible Verbs:

(3.5) a. J´ımò. ó. sáré lo. . Jimo agr run go ‘Jimo ran away.’ (Yoruba; Déchaine 1993, 220) b. sárélo. : λx λs {RUN(x) & MOVE-AWAY-FROM-DC(x)}(s) +sc

(3.5) moreover satisﬁes Connexion, since both predicates share the argument variable, x, and it satisﬁes Coherence because the two subevents corresponding to the verbs CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 114

– that of running and that of moving away from the deictic center – are contemporaneous. (3.5) thus satisﬁes each Possible Verbs Constraint and is a Possible Verb SVC. (3.6) presents another motion SVC, but where the semantic representation in (3.6b) contains the conjunct BECOME(LOC(x,¬IN(y))), representing the semantics of the verb losa ‘exit’ in combination with the PP d’ei toples ‘in jar’:

(3.6) a. ede mpara karefa rai losa.na d’ei toples. dem then frog run exit.3 in jar ‘thenthefrogranoutofthejar.’ (Bima) b. rai losa: λy λx λs {RUN(x) & BECOME(LOC(x, ¬IN(y)))}(s) −sc +sc

(3.6) satisfies Possible Verbs because the directional predicate BECOME(LOC(x, ¬IN(y))) supplies further information within the interpretation space activated by RUN. (3.6) satisfies Connexion because RUN and BECOME(LOC(x, ¬IN(y))) share the argument variable x. It satisfies Coherence because the subevents of running and exiting the jar are coidentified and hence contemporaneous. (3.7) is similar in many respects to (3.6). BECOME(LOC(y, IN(z))) specifies that the ‘kick-ee’ in the higher predicate KICK(x,y) changes location, and configures this change relative to the ground z (d´ıwósu ‘the house’):

(3.7) a. A sikópu d´ı bálu gó a d´ı wósu. 3sg kick det ball go loc det house ‘He kicked the ball into the house.’ (Saramaccan; Veenstra 1996, 70) b. sikópu gó : λz λy λx λs {KICK(x,y) & BECOME(LOC(y, IN(z)))}(s) −sc +sc +sc

Because the embedded predicate in (3.7b) specifies the higher predicate, (3.7a) satisfies Possible Verbs, and it also satisfies Connexion because each predicate shares the argument variable y. In contrast to (3.6), the subevents in (3.7) are not contemporaneous. The action of kicking an object (in (3.7a) a ball) necessarily terminates before the kicked object moves completely over its path. In Section 3.2.3, (3.7) will be characterized event structurally as a non-coidentified resultative SVC; here all that needs CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 115

to be acknowledged is that, since there is a causal connection between the subevents corresponding to the predicates in (3.7b), this SVC does satisfy Coherence, and in satisfying all three Possible Verbs constraints it is a Possible Verb SVC. The ﬁnal two SVCs which satisfy the Possible Verbs constraints in this section take us away from motion events:

` (3.8) a. E. sósà gbé émá!tó. n pè. rhé. . Esosa hit metal be-flat ‘Esosa hit the metal flat.’ (E`. dó; Stewart 2001, 58) b. gbépè. rhé.: λy λx λs {HIT(x,y) & BECOME(FLAT(y))}(s) +sc +sc

(3.9) a. Amu ngaha ka.mpoi.na oha. Amu eat caus.gone.3 rice ‘Amu ate up all the rice.’ (Bima) b. ngaha ka.mpoi: λy λx λs {EAT(x,y) & BECOME(¬EXIST(y))}(s) +sc +sc

(3.8) satisfies Possible Verbs because the embedded predicate, BECOME(FLAT(y)), specifies further information about the process described by the higher predicate HIT(x,y). In the same fashion, (3.9) satisfies this constraint because BECOME(¬EX- IST(y)) specifies further information about the process described by EAT(x,y). Both SVCs satisfy Connexion because the predicates in each example share a single argument variable (in each case annotated ‘y’), and (3.8) and (3.9) also satisfy Coherence since causal relationships hold between the predicates. (3.8) and (3.9) are thus Possi- ble Verb SVCs, and the single sets of {} braces in their Semantic Form representations are also representative of this status.

3.2.2 Serialization lexical rules and Possible Verb SVCs

A feature of Possible Verb SVCs which stands in contrast to Impossible Verb SVCs is the need for serialization lexical rules. None of the Possible Verb SVCs in Section 3.2.1 required serialization lexical rules because the semantics of the verbs involved appear identical between their SVC and non-SVC contexts. In other Possible Verb CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 116

SVCs, however, there are frequently attested and predictable differences between the meanings of verbs as they appear in and out of SVCs. These differences impact upon Possible Verb status; this section includes several illustrations of how it is only through the application of a serialization lexical rule, and the alteration of verbal semantics that it effects, that an SVC comes to satisfy the Possible Verbs constraints. Demonstrating how serialization lexical rules interact with Possible Verb status is one objective for this section, although a more challenging, long-term objective is to survey how wide-ranging and pervasive these rules actually are. This section takes only the first few steps in the latter direction. Chapter 2 contained three serialization lexical rules, the first of which was the following Benefactive serialization lexical rule:

(3.10) Benefactive serialization lexical rule:

′ ′ ′ Verb1 ‘give’: λy ... λx λs λs {V1(x, ...)(s ) & BENEFIT(s ,y) }(s) −sc +sc

This rule accounts for how in some SVCs, termed benefactive SVCs, a verb which in a non-SVC context has the semantics of ‘give’ (ACT(x) & BECOME(POSS(y,z))) instead has the semantics which Section 2.3.3 proposed to encode with the predicate BENEFIT. BENEFIT takes two arguments, the first typically a situation and the second typically an individual, such that the first argument benefits the second. Thus (3.11b), the Semantic Form representation for the benefactive SVC (3.11a), specifies that the situation of x wearing y positively affects a third entity, z:

(3.11) a. Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5) ′ ′ ′ b. wo. fún: λz λy λx λs λs {WEAR(x,y)(s ) & BENEFIT(s ,z)}(s) −sc +sc +sc

(3.11) satisﬁes the Possible Verbs constraint because within the interpretation space activated by the higher predicate WEAR(x,y)(s′), one possible further speciﬁcation is CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 117

the consequence, or at least the intended consequence, that this action benefits a third entity. The embedded predicate BENEFIT(s′,z) encodes this (intended) consequence. (3.11) satisfies Connexion because both predicates share the situation variable s′ and it satisfies Coherence by default because the embedded predicate does not correspond to a subevent (since there is no independent event of BENEFIT-ing that can stand alone, apart from the action of WEARing). (3.11a) is thus a Possible Verb SVC, although it is only through the application of the benefactive serialization lexical rule that it attains this status, since if fún were to retain the semantics of ‘give’, then neither the Possible Verbs nor the Coherence constraints would be satisfied.1 The benefactive serialization lexical rule in (3.10) was not the only serialization lexical rule included in Chapter 2 which altered the semantics of a verb meaning ‘give’. Section 2.4.1.3 also set out a ‘goal serialization lexical rule’ which altered the semantics of ‘give’ to the locational meaning of BECOME(LOC(y AT(z))) (even if this rule was ultimately not judged to apply to Fongbe SVCs in the way Lefebvre and Brousseau (2002) indicate). ‘Give’ could be associated with multiple serialization lexical rules then, although at this stage I do not have enough data to demonstrate how, or the conditions under which the different rules would apply. There is, however, a little more evidence pertaining to the involvement of verbs with the semantics of ‘take’ in multiple serialization lexical rules. The instrumental serialization lexical rule included in Chapter 2 is one example of a serialization lexical rule which alters the semantic contribution of ‘take’:

(3.12) Instrumental serialization lexical rule:

‘take’ Verb2: λz λy λx λs {V2(x,(y)) & USE(x,z) }(s) −sc +sc +sc

This rule accounts for the fact that there is a systematic alteration, common to scores

1If fún were to retain the semantics of ‘give’ (i.e. ACT(x) & BECOME(POSS(y,z))) then (3.11a) would mean something more like ‘Aje wore an outfit and gave Olu the outfit’. This would not satisfy Possible Verbs because the action of giving Olu the outfit does not further specify the properties opened up within the interpretation space of Aje wearing an outfit. Giving Olu the outfit would instead represent a separate development which is temporally disjoint and causally unconnected to the first action, which is why this structure would also not satisfy Coherence. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 118

of instrumental SVCs in many languages, whereby a verb which would otherwise mean ‘take’ signifies only that the actor uses a particular object – the ‘instrument’ – without necessarily first taking it (i.e. moving it). This object, which must be distinct from the object of the main verb (if there is one), can be configurationally identified with the instrument role through the configurational definition laid out in Section 2.3.3. The following SVC is an instrumental SVC whose initial verb lá ‘take’ has its semantics altered in the Semantic Form representation of the SVC through the instrumental serialization lexical rule:

(3.13) a. Tsoda lá egbà ba cigbà. Tsoda take axe cut tree ‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63) b. lába: λz λy λx λs {CUT(x,y) & USE(x,z)}(s) −sc +sc +sc

The resulting Semantic Form representation in (3.13b) satisfies the Possible Verbs constraints (as the single {} bracketing would suggest) since the predicate USE(x,z) specifies further information for the domain activated by the predicate CUT(x,y). It supplies information about the means by which the main action of cutting is carried out. (3.13) satisfies Connexion because USE(x,z) shares the argument variable x with CUT(x,y). It satisfies Coherence because the action of using the instrument is contemporaneous with the action of cutting. The instrumental serialization lexical rule in (3.12) will be elaborated upon in Chapter 4. The current focus is the multiplicity of serialization lexical rules linked to lexical items indicated to have the semantics of ‘take’, and so we will now turn to ‘take’ SVCs of the following form:

(3.14) Tsoda lá hánkàli ba cigbà. Tsoda took care cut tree ‘Tsoda cut the tree carefully.’ (Nupe; George 1976, 63)

(3.15) Mo fi àgbara gé igi. I take strength cut wood ‘I cut wood energetically.’ (Yoruba; Stahlke 1970, 85) CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 119

(3.16) Mo fi o. gbò. n gé igi. 1sg take cleverness cut tree ‘I cut the tree cleverly.’ (Yoruba; Stahlke 1970, 62)

In (3.14)-(3.16), the complements of the verbs indicated to have the semantics of ‘take’ do not depict concrete objects, as in (3.13), but abstract qualities: care, cleverness and strength. The semantic contribution of ‘take’ combined with its abstract quality complement amounts to manner modiﬁcation, as the translations using the manner adverbs ‘carefully’, ‘cleverly’ and ‘energetically’ indicate. Manner adverbs are given a Davidsonian analysis of predicates of events in LDG2 according to which (3.16) would require a Semantic Form representation of the following form:

(3.17) λy λx λs {CUT(x,y)(s) & CLEVER(s)}

This kind of representation raises several issues. First, for the purposes of formulating a serialization lexical rule for the ‘manner modifying’ uses of ‘take’ in (3.14)-(3.16), exactly which part of these Semantic Form representations is ‘take’ supposed to correspond to? How do we characterize the alteration to the semantics of ‘take’ in (3.14)-(3.16)? The second issue is that there are problems with Davidsonian representations of agent-oriented adverbs like ‘cleverly’ in (3.17); it just doesn’t make sense to say that an event (or a situation) is clever. Cleverness is a property that can hold only of a sentient being, or agent, and in fact all of (3.14)-(3.16) involve agent-oriented manner modification. At this stage it is unclear whether all ‘manner modifying’ ‘take’ SVCs are agent-oriented, although the fact that at least some of these SVCs involve agent-oriented modification is reason enough to reject a Davidso- nian analysis. What is required for (3.14)-(3.16) is a serialization lexical rule which avoids this Davidsonian pitfall, and which captures the altered semantic contribution of ‘take’ in this particular SVC context. (3.18) is one possibility; here the nominal complement of ‘take’ denotes some abstract quality, q, and what ‘take’ encodes is that the actor exercises this quality, resulting in (typically agent-oriented) manner modification semantic effects: 2At least, the only representation of manner adverb semantics I have encountered in the LDG literature (Kaufmann and Wunderlich 1998, 8) uses this Davidsonian analysis, although this is probably intended to be merely illustrative rather than being seriously motivated. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 120

(3.18) Manner modifying serialization lexical rule:

‘take’ Verb2: λq λx λs {V2(x,...) & EXERCISE(x,q) }(s) −sc +sc

This rule determines that the Semantic Form representation for (3.19a) (=(3.16)) is as in (3.19b) (with o. gb`o. n ‘cleverness’, the non-structural abstract quality argument of ﬁ ‘take’, having already been lambda-converted):

(3.19) a. Mo fi o. gbò. n gé igi. 1sg take cleverness cut tree ‘I cut the tree cleverly.’ (Yoruba; Stahlke 1970, 62) b. λy λx λs {CUT(x,y) & EXERCISE(x,cleverness)}(s)

CUT(x,y) in (3.19) activates an interpretation space, and within this interpretation space the embedded predicate further speciﬁes that the agent of the cutting exercises cleverness – in other words, s/he cuts cleverly. In order for (3.16) to satisfy Possible Verbs, the embedded predicate must be interpreted as modifying the main action denoted by the higher predicate. The manner modifying serialization lexical rule in (3.18) is meant to serve as a rough illustration of how ‘take’ functions in a serialization lexical rule distinct from the instrumental serialization lexical rule in (3.12), and is certainly open to improvement (particularly questionable is the use of the predicate EXERCISE). The main point to emphasize is that several languages include both instrumental ‘take’ SVCs to which the instrumental serialization lexical rule in (3.12) must apply, and ‘manner modifying’ ‘take’ SVCs, to which something like the rule in (3.18) must apply. If the complement of ‘take’ is an abstract noun denoting a quality, then the manner modifying serialization lexical rule applies, since an abstract complement of ‘take’ would be uninterpretable in the Semantic Form representation generated by the instrumental serialization lexical rule. If the complement of ‘take’ is a concrete object, then the instrumental serialization lexical rule applies, since a concrete object complement of ‘take’ would be uninterpretable in the Semantic Form representation generated by the manner modifying serialization lexical rule. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 121

The involvement of ‘take’ in more than one serialization lexical rule does not stop at instrumental and manner modifying SVCs. (3.20) and (3.21) illustrate a further serial use of ‘take’ in Nupe and Yatye respectively, distinct from the instrumental and modifying uses in these languages:

(3.20) Y´ıg´ıd´ı l´a eg´o wo. sun take grass dry ‘The sun dried the grass.’ (Nupe; George 1976, 63)

(3.21) Iwyi` aw´a uts`ı ik`u. child took door shut(intr) ‘The child shut the door.’ (Yatye; Stahlke 1970, 74)

The objects of lá ‘take’ in (3.20) and awá ‘took’ in (3.21) are also the internal arguments of DRY and SHUT, the predicates corresponding to the final verbs in (3.20) and (3.21). ‘Take’ has a causativizing function in (3.20) and (3.21). The serialization lexical rule I propose for this ‘causativizing’ use of ‘take’ is,

(3.22) Causative serialization lexical rule:

‘take’ Verb2: λy λx λs { ACT(x,y) & BECOME(V2(y))}(s) +sc +sc and it produces the following Semantic Form representations for (3.20) and (3.21) (recall that LDG does not posit a CAUSE predicate; instead the causal relationship between the two components in each of (3.20) and (3.21) is inferred at Conceptual Structure):

(3.23) λy λx λs {ACT(x,y) & BECOME(DRY(y))}(s)

(3.24) λy λx λs {ACT(x,y) & BECOME(SHUT(y))}(s)

The causative serialization lexical rule applies (in those languages which employ it) to alter the semantics of ‘take’ in all cases where ‘take’ and the other verb share a single internal argument, and the other verb describes a state, and hence can appear as the argument of BECOME. The ﬁnal group of ‘take’ SVCs we will consider is the following: CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 122

(3.25) A teki a fisi seri. s/he take the fish sell ‘S/he sold the fish.’ (Sranan; Sebba 1987, 60)

(3.26) Aj´e wo. as.o. w`o. . Aje took dress wear

‘Aje put on some clothes.’ (Yoruba; Awoyale 1988, 3)

(3.27) Kkú s cáv`ı ny`ı gbé. Koku take key def throw outside ‘Koku threw the key outside.’ (Fongbe; Lefebvre and Brousseau 2002, 410)

But what is ‘take’ doing in these SVCs? The single-verb-headed translations for (3.25)-(3.27) are suggestive of Possible Verb status, but they give no clue as to what the semantic contribution(s) of the respective ‘take’s might be. It is possible that (3.25)-(3.27) are mistranslated and are actually Impossible Verb SVCs, where ‘take’ does signify TAKE (so (3.27), for example, would be translated as ‘Koku took the key and threw it outside’), although there is nothing to corroborate this possibility. Several types of SVC in this section pose considerable difficulty in terms of how to represent the semantic alteration that their verbs undergo in serialization lexical rules. For (3.25)-(3.27) this is not so much a difficulty as an impossibility when the semantic contribution of ‘take’ cannot even be characterized in the first place. Jagacinski (1992, 133) does provide one clue as to what the semantic contribution of ‘take’ could be in

commenting that the object of au ‘take’ in Thai SVCs like (3.28) must be deﬁnite,

as well as having an emphasis placed upon it:

(3.28) D`ek au khanˇom kin m`ot l :u. child take dessert eat all asp ‘A child ate all the dessert.’ (Thai; Jagacinski 1992, 126)

However, this semantic effect may not be the same for all languages with SVCs of this form, and, besides, it is no easy task to try and formulate the semantics of definiteness and emphasis in a serialization lexical rule, since LDG’s Semantic Form representations are not fine-grained enough to encode these notions (but are primarily designed to determine which arguments are projected into the syntax as CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 123

structural arguments). More detailed semantic analysis of (3.25)-(3.27) is required before attempting to formulate any serialization lexical rules they may necessitate. LDG’s Semantic Form representations also do not encode information about tense or aspect, the territory into which formulation of serialization lexical rules for the following SVCs takes us:

(3.29) kora yo ja cheg´a nal´ı eli ja kaba bai when 1sg per arrive there 3sg per ﬁnish go ‘When I arrived there he had gone’ (Kristang; Baxter 1988, 213)

(3.30) ’E ’amanaki mavae te sitima. ns hope/expect leave def steamer ‘The steamer is about to leave the harbour.’ (East Uvean; Moyse-Faurie 2004, 213)

Kristang kaba means ‘finish’ when used independently, but signifies completive aspect in an SVC such as (3.29), while East Uvean ’amanaki means ‘to hope, expect’ in a non-SVC context, but signifies a future tense meaning (‘be about to’) in an SVC context such as (3.30). (3.31) is a crude illustration of what a serialization lexical rule for (3.30) could look like:

(3.31) Future tense serialization lexical rule:

’amanaki Verb2: λx λs {V2(x,...)(s) & FUTURE(s)} +sc

The use of a FUTURE predicate which takes the situation variable as its argument may be far from ideal, although here I will not dwell any further on the diﬃculties for semantic formalization posed by SVCs like (3.29) and (3.30). The only point I will add is that kaba in (3.29) and ’amanaki in (3.30) are the ﬁrst verbs with altered semantics in this section which do not take their own complements, although this is unsurprising from the perspective that all they do is contribute tense and aspect information and do not interact with argument structure. Although this survey of serialization lexical rules has been reasonably diverse, I do not imagine it touches upon anywhere near the number of rules that would CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 124

actually be required. Serialization lexical rules may be required for a far wider range of SVCs, instantiating many different functions, but they may also be required to account for semantic shifts that are much subtler than those described here. (3.32), for example, may require a serialization lexical rule to alter the semantics of dé ‘fall’ to DESCEND, which encodes downward movement without the additional manner component of being accidental or uncontrolled that may be entailed (and not just implicated) by dé:3

(3.32) Ozó` sùá Uy`ı´ dé. Ozo push Uyi fall ‘Ozo pushed Uyi down.’ (E`. dó; Stewart 2001, 8)

More careful semantic analysis is required to ascertain whether (3.32) really does require a serialization lexical rule of this type. More careful semantic analysis across all types of SVC would, I predict, necessitate scores of new serialization lexical rules, and so, as more ﬁne-grained semantic analysis is carried out, it may become more and more evident that Sebba’s (1987) requirement that verbs have identical semantics within an SVC as outside it is far too strict. At the same time, it is not unreasonable to ask whether serialization lexical rules, and the altered verbal semantics they give rise to, are really justiﬁed in SVCs. Arguments for and against this position will be weighed up in Section 3.4.

3.2.3 Event structures for Possible Verb SVCs

The final component in this discussion of Possible Verb SVCs are the event structural profiles for these SVCs. These profiles, which are detailed at the level of Conceptual Structure in LDG, could be viewed simply as a way of characterizing Possible Verb SVCs, or alternatively they could be seen as a means of actually identifying Possible Verb SVCs, without having to apply LDG’s Possible Verbs constraints. The use of event structural terminology also looks ahead to Sections 3.5.2 and 3.5.4; the former concerns the event structure of SVCs and the latter concerns event structural 3Not all languages use verbs meaning ‘fall’ in this way; Winford (1992, 188) comments that Caribbean English Creole does not use faaldong ‘fall’ in the way Sranan uses fadon, to predicate downward movement over an object. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 125

distinctions in SVCs and how they compare to the Possible Verb-Impossible Verb distinction. Looking further ahead still, the temporal and event structural terminology set out here (and in Section 3.3.1 with respect to Impossible Verb SVCs) are also an integral part of the analyses of instrumental and motion SVCs in the case studies of Chapters 3 and 4 respectively. Possible Verb SVCs display three patterns of ‘unitary-event’, ‘coidentified’ or ‘non-coidentified resultative’ event structure, and these correspond loosely to Kaufmann and Wunderlich’s (1998, 38) event structural profiles for clauses which satisfy the Possible Verbs constraints. Unitary-event SVCs will be considered first, but prior to this, some theoretical preliminaries are necessary. The Conceptual Structure descriptions below adhere to the following notational conventions: (i) Vi is that component of a Semantic Form representation which encodes the meaning of Verbi, (ii) si

is a (sub)event (or situation) corresponding to Vi, (iii) tci is the time of commence-

ment of si, (iv) tti is the time of termination of si, (v) the symbol

temporal precedence (‘ti

(vi) the symbol =temporal signiﬁes temporal equality (‘ti =temporal tj’ signiﬁes that ti

and tj identify the same moment in time). Unitary-event SVCs display the following patterns of Semantic Form and Concep- tual Structure representation:

(3.33) SF and CS representations for UNITARY-EVENT SVCs

For a unitary-event SVC consisting of verbs Verb1 and Verb2:

SF: ... λx λs1 {V1(x,...)(s1) & V2(s1)}

CS: V1 corresponds to an event s1 with time of commencement tc1 and time

of termination tt1.

V2 does not correspond to a (sub)event.

The key characteristic of a unitary-event SVC is that one of its verbs is ‘defective’ in the sense that it does not have a corresponding subevent. This verb does not introduce a situation variable into the logical representation of the sentence. The subevent

whose situation variable si is introduced is a unitary event in Talmy’s sense that CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 126

it “consists of components that are considered not to constitute events in their own right” (2000, 215). The ‘defective’ verb, which does not correspond to a subevent, has a modifying function: it modiﬁes the verb which does correspond to a subevent. An example of a modifying verb is found in (3.34), where the verb ka.lai ‘caus.separate’ (or ‘separately’, in this context), modiﬁes the verb lowi ‘cook’:

(3.34) a. Rao lowi ka.lai.na oha lab’o uta. Rao cook caus.separate.3 rice with ﬁsh ‘Rao cooked the rice and the ﬁsh separately.’ (Bima)

b. SF: λy λx λs1 {COOK(x,y)(s1) & SEPARATE(s1)}

c. CS: One subevent, s1, of cooking.

F´un ‘give’ in (3.35) also does not correspond to a subevent; its function is to modify the main situation of Aje wearing an outﬁt:

(3.35) a. Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5)

b. SF: λz λy λx λs1 λs {WEAR(x,y)(s1) & BENEFIT(s1,z)}(s)

c. CS: One subevent, s1, of wearing an outﬁt.

Of course, the reason why f´un does not correspond to a subevent of giving is that it does not actually have the semantics of give (i.e. ACT(x) & BECOME(POSS(y,z)))

in (3.35), but instead has its semantics altered to BENEFIT(s1,z) by the benefactive serialization lexical rule included in the previous section. The second type of temporal and event structurally distinguished Possible Verb SVCs are coidentiﬁed SVCs:

(3.36) SF and CS representations for COIDENTIFIED SVCs

For a coidentiﬁed SVC consisting of verbs Verb1 and Verb2:

SF: ... λx λs {V1(x) & V2(x)} OR CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 127

SF: ... λy λx λs {V1(x,y) & V2(y)}

CS: V1 corresponds to a subevent s1 and V2 to a subevent s2, and these two

subevents are coidentiﬁed (temporally dependent). tc1=temporaltc2 and

tt1=temporaltt2.

Coidentification of subevents is a concept described in Levin and Rappaport Hovav (1999) whereby two subevents hold the property of being temporally dependent: they occupy the same time frame, unfold at the same rate, and are inextricably linked, to the point that they are conceptualized as a single event. Kaufmann and Wun- derlich’s (1998) equivalent of this concept is unification of the temporal structures of the two subsituations (“if the event time of coordinated primitives can be identified, the semantic predicates are interpreted as describing different aspects of one (sub)situation” (1998, 38)). However, coidentification of subevents in sentences with resultative semantics actually involves more than simply sharing the same temporal coordinates; temporal dependence entails that one of the subevents serves to measure out the temporal progress of another subevent. The nature of this measuring out is dictated by a property of a participant in the event which reflects the event’s temporal progress; Levin and Rappaport Hovav (1999, 15) recognize three major categories of (i) a path of motion measuring out the temporal progress of a motion event, (ii) a gradable property measuring out the temporal progress of a change of state event, or (iii) the spatial extent of a consumed or created object measuring out the temporal progress of an event of consumption or creation. The following Bima SVC represents the first case, of a path of motion, lu’u.na d’ei kota ‘enter the box’, measuring out the temporal progress of the crawling subevent:

(3.37) a. Ana kea rama lu’u.na d’ei kota. baby crawl enter.3 in box ‘The baby crawled into the box.’ (Bima) b. SF: λx λs {CRAWL(x) & BECOME(LOC(x, IN(the box)))}(s)

c. CS: Two coidentiﬁed subevents: s1 of crawling and s2 of entering the

box. tc1=temporaltc2 and tt1=temporaltt2. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 128

(3.37) involves an initial intransitive manner of motion verb, but (3.36) also allows for an initial transitive verb, where the subsequent verb predicates of the second argument of this transitive verb. (3.38) has an initial transitive verb, represented as DRAG at Semantic Form, and again the property which measures out the temporal progress in this coidentiﬁed SVC is a path of motion, but this time it is the path of the second argument of DRAG, y (realized as wa’i ‘the old woman’):

(3.38) a. wa’u mpara cumpu, randa ka.lu’u kai.na wa’i already then ﬁnished drag caus.enter kai.3 old-woman ede dem ‘once he’d ﬁnished (digging her grave), he dragged the old woman into it’ (Bima) b. SF: λy λx λs {DRAG(x,y) & BECOME(LOC(y,IN(the grave)))}(s)

c. CS: Two coidentiﬁed subevents: s1 of dragging the old woman and s2 of

the old woman entering the grave. tc1=temporaltc2 and tt1=temporaltt2.

(3.39) also contains an initial transitive verb, EAT, but the property which measures out the temporal progress in this resultative SVC is the diminishing spatial extent of the object of consumption, y, realized as oha ‘rice’:

(3.39) a. Amu ngaha ka.mpoi.na oha. Amu eat caus.gone.3 rice ‘Amu ate up all the rice.’ (Bima) b. SF: λy λx λs {EAT(x,y) & BECOME(¬EXIST(y))}(s)

c. CS: Two coidentiﬁed subevents, s1 of eating rice and s2 of the quantity of

rice lessening until it has gone. tc1=temporaltc2 and tt1=temporaltt2.

The final type of temporal and event structurally distinguished Possible Verb SVC are non-coidentified resultative SVCs. As the title would suggest, these do not hold the property that the subevents corresponding to the respective verbs are coidentified. Rather, the subevents are in a causal relationship:

(3.40) SF and CS representations for NON-COIDENTIFIED RESULTATIVE SVCs CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 129

For a non-coidentiﬁed resultative SVC consisting of verbs Verb1 and Verb2:

SF: ... λy λx λs {V1(x, ...) & (V2(y))}

CS: V1 corresponds to a subevent s1 and V2 to a subevent s2, where s1 causes

s2. Either tc1=temporaltc2 or tc1

Kaufmann and Wunderlich (1998) state that a causal relationship must hold between subevents which do not share identical temporal coordinates. My CS description in (3.40) concurs with this, but it also allows for the possibility that two subevents in a causal relationship could have identical temporal coordinates. Just because two subevents in a resultative SVC are not coidentiﬁed, this does not mean they cannot occupy the same temporal interval; the subevents of the children laughing and Fero’s exit from the room in (3.41), for example, could commence (and terminate) at the same points in time, although it is more likely that the children’s laughing commences before Fero’s exiting:

(3.41) a. Ana dou to’i d’oho hari ka.losa.na Fero watu uma. child person small pl laugh caus.exit.3 Fero from house ‘The children laughed Fero out of the house.’ (Bima) b. SF: λy λx λs {LAUGH(x) & BECOME(LOC(y,¬IN(the house)))}(s)

c. CS: Two subevents, s1 of the children laughing and s2 of Fero exiting

from the room, where s1 causes s2. tc1 ≤temporal tc2.

In (3.42) the subevent of kicking almost certainly commences just slightly before the subevent of the ball beginning its path into the house (under the assumption that drawing the foot back, rather than making contact with the ball, marks the beginning of the kicking action):

(3.42) a. A sikópu d´ı bálu gó a d´ı wósu. 3sg kick det ball go loc det house ‘He kicked the ball into the house.’ (Saramaccan; Veenstra 1996, 70) b. SF: λy λx λs {KICK(x,y) & BECOME(LOC(y, IN(the house)))}

c. CS: Two subevents, s1 of the actor kicking the ball and s2 of the ball

entering the house, where s1 causes s2. tc1

Although the two subevents in (3.42) may share the same time of commencement, they do not share the same times of termination; this underscores the fact that these subevents are not coidentified. These are the three possible event structural profiles for Possible Verb SVCs; two further event structural profiles, but for Impossible Verb SVCs, will be set out in Section 3.3.1 below. But as far as the event structural profiles in this section are concerned, now we have a way of identifying Possible Verb SVCs without having to apply LDG’s Possible Verbs constraints: if only one of the verbs in an SVC corresponds to a subevent (as in unitary-event SVCs), or if the subevents in an SVC are coidentified (as in coidentified SVCs), or if the Conceptual Structure of an SVC specifies that there is a causal connection between the two subevents described (as in non-coidentified resultative SVCs), then the SVC is a Possible Verb SVC.

3.2.4 Conclusion

This section explored the basis of Possible Verb status in SVCs: satisfaction of LDG’s Possible Verbs constraints. It also considered the main complication attached to Possible Verb SVCs: the serialization lexical rules required to alter the semantics of particular verbs in the serial verb context. At this stage it is unclear how wide-ranging serialization lexical rules are and/or how semantically subtle they can be. But Section 3.2.3 then returned to the safer territory of the event structural profiles for Possible Verb SVCs; these can be used as an alternative means of identifying Possible Verb SVCs, for if an SVC can be shown to have the unitary-event structure described in (3.33), the coidentified structure in (3.36), or the non-coidentified resultative structure in (3.40), then the SVC can be concluded to be a Possible Verb SVC without having to apply the Possible Verbs constraints. One thing never stated explicitly in this section, but which is of crucial significance to this dissertation, is the fact that Possible Verb SVCs will always satisfy the Theta Structure Property for SVCs, trivially. It is trivial that Possible Verb SVCs should satisfy the Theta Structure Property since any SVC that satisfies the Possible Verbs constraints will have to have, not just the event structure of a clause headed by a single CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 131

verb, but the argument structure of a clause headed by a single verb too. Possible Verb event structures – the three varieties of which were illustrated in Section 3.2.3 – come hand in hand with Possible Verb argument structures which satisfy the Theta Structure Property. Section 3.3 below discusses Impossible Verb SVCs, their event structures, and their argument structures, which do not so trivially satisfy the Theta Structure Property.

3.3 Impossible Verb SVCs

While it can be taken for granted that Possible Verb SVCs will fulﬁll the Theta Struc- ture Property for SVCs (since a Semantic Form representation fulﬁlling the Possible Verbs constraints will always correspond to monoclausal argument structure), Impos- sible Verb SVCs pose more serious challenges for the Theta Structure Property for SVCs. Impossible Verb SVCs do not satisfy the Possible Verbs constraints. The Se- mantic Form representations for Impossible Verb SVCs, such as (3.43a) and (3.44a), instead consist of at least two ({}-bracketed) subcomponents which satisfy the Possi- ble Verbs constraints independently, each of which has a separate set of λ-abstracted structural and non-structural arguments, as in (3.43b) and (3.44b):

(3.43) a. Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima)

b. rumpa weha: λy λx λs1 {FIND(x,y)}(s1) & λv λu λs2 {TAKE(u,v)}(s2) +sc +sc +sc +sc k k k k Predicate Combining Operations for Impossible Verb SVCs ⇓ ⇓ ⇓ ⇓ c. rumpa weha: λy λx λs [{FIND(x,y)} & {TAKE(x,y)}](s) +sc +sc

(3.44) a. Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10) CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 132

(3.43a) and (3.44a) featured in Section 2.3.2 together with the Theta Structure and Se- mantic Form representations as they appear in (3.43c) and (3.44c). The difference between the respective (b) and (c) representations is that whereas the (b) representations have separate Theta Structures for each Semantic Form representation corresponding to a separate {}-bracketed Possible Verb component, in the (c) representations the arguments are amalgamated into one Theta Structure, and the {}-bracketed Possible Verb components are conjoined in a single Semantic Form representation. (3.43c) and (3.44c) have monoclausal argument structure; their Theta Structures each consist of two structural arguments, as also do the Theta Structures of many clauses headed by single verbs. But how do we get from (b) to (c)? Section 3.3.2 below presents the Predicate Combining Operations for Impossible Verb SVCs which account for this shift. These operations are mostly made up of argument coidentification rules, which account for how Impossible Verb SVCs such as (3.43a) and (3.44a) can still realize monoclausal argument structure – thus satisfying the Theta Structure Property – in spite of not fulfilling the Possible Verbs constraints. The Predicate Combining Oper- ations hold the most consequence in this section, and also in this chapter. But before laying these operations out in Section 3.3.2, Section 3.3.1 first characterizes Impossi- ble Verb SVCs by presenting their event structural profiles and describing how they fail to fulfill the Possible Verbs constraints.

3.3.1 Event structures for Impossible Verb SVCs

Impossible Verb SVCs are divided into the two temporal and event structural types of ‘sequential’ Impossible Verb SVCs, illustrated in (3.45), and ‘coextensive’ Impossible Verb SVCs, illustrated in (3.46), purely on the basis of whether the subevents involved CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 133

commence at the same points in time or not ((3.45) and (3.46) follow the same notational conventions as the event structural templates for Possible Verb SVCs, set out in Section 3.2.3):

(3.45) SF and CS representations for SEQUENTIAL Impossible Verb SVCs

For a sequential Impossible Verb SVC consisting of verbs Verb1 and Verb2:

SF: ... λx λs [{V1(x,...)} & {V2(x,...)}](s)

CS: V1 corresponds to a subevent s1 and V2 to a subevent s2. tc1

(3.46) SF and CS representations for COEXTENSIVE Impossible Verb SVCs

For a coextensive Impossible Verb SVC consisting of verbs Verb1 and Verb2:

SF: ... λx λs [{V1(x,...)} & {V2(x,...)}](s)

CS: V1 corresponds to a subevent s1 and V2 to a subevent s2. tc1=temporaltc2

and tt1=temporaltt2.

Each verb in an Impossible Verb SVC corresponds to a separate subevent in the Conceptual Structure representation. (3.45) and (3.46) limit the number of verbs to two for the sake of simplicity. Absent from either of the Conceptual Structure representations in (3.45) and (3.46) are any indications of coidentification or causality – the properties which hold for coidentified and non-coidentified Possible Verb SVCs respectively. In the absence of these properties it can be concluded that we have an Impossible Verb SVC. This is the event structural perspective of what it means to be an Impossible Verb SVC; the vital determinant of Impossible Verb status can otherwise be found not in the Conceptual Structure representations, but in the Semantic Form representations of (3.45) and (3.46), where the two sets of {} brackets signify that the verbs only satisfy the Possible Verbs constraints separately. In addition to setting out the event structural characteristics of a small sample of Impossible Verb SVCs (in their Conceptual Structure representations), this section describes how the verbs in these SVCs fail to satisfy the Possible Verbs constraints jointly. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 134

The ﬁrst Impossible Verb SVC we will consider is a sequential Impossible Verb SVC, as established by the relation of temporal precedence between the subevents, indicated in the Conceptual Structure representation in (3.47c):

(3.47) a. Fero dula lowi.na janga. Fero go-home cook.3 chicken ‘Fero went home and cooked the chicken.’ (Bima)

b. SF: λy λx λs [{GO-HOME1(x)} & {COOK(x,y)}](s)

c. CS: Two subevents, s1 of Fero going home, and s2 of Fero cooking a

chicken. tc1

(3.48) a. Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10)

b. SF: λy λx λs [{COOK1(x,y)} & {EAT(x,y)}](s)

c. CS: Two subevents, s1 of Ozo cooking food, and s2 of Ozo eating food.

tc1

(3.48) does not satisfy the Possible Verbs constraint because ré ‘eat’ does not further specify lé ‘cook’. Eating food could certainly be seen as a natural development that follows on from cooking food, yet the cooking cannot be seen to cause the eating, and so (3.48) fails Coherence too. This absence of a causal relation (which accounts for why (3.48a) fails Coherence) is reflected in the Conceptual Structure representation in (3.48b). Here each verb corresponds to a subevent where the first temporally precedes the second, and there is otherwise no relation or connection between the two. (3.49) and (3.50) are two examples of coextensive Impossible Verb SVCs:

(3.49) a. A ben e sidon fowru en anu na ondro en kakumbe. he T/A T/A sit fold his arm loc under his chin ‘He sat folding his arms under his chin.’ (Sranan; Sebba 1987, 58) b. SF: λy λx λs [{SIT(x)} & {FOLD(x,y)}](s)

c. CS: Two subevents, s1 of the actor sitting, and s2 of the actor folding

his arms. tc1=temporaltc2 and tt1=temporaltt2.

(3.50) a. Na.mbowo d’onga b’a lako uma mena.na. 3.bark look-up by dog house pl.3 ‘The dogs barked, looking up at their house.’ (Bima) b. SF: λy λx λs [{BARK(x)} & {LOOK-UP-AT(x,y)}](s)

c. CS: Two subevents, s1 of the dogs barking, and s2 of the dogs looking

up at the house. tc1=temporaltc2 and tt1temporal=tt2.

Coextensive Impossible Verb SVCs are far less commonly encountered than sequential Impossible Verb SVCs and are generally limited to combinations of a verb that describes an action with a verb that describes some posture held during this action. The subevents in each of (3.49) and (3.50) are coextensive, as indicated in their respective Conceptual Structure representations, so in this sense do fulfill the Coherence Pos- sible Verbs constraint of being contemporaneously connected. However, in neither case are the subevents coidentified, so neither SVC is a coidentified Possible Verb SVC. Coidentification, or temporal dependence, is the property that entails Possible Verb status; in contrast, the mere coincidence of two subevents happening to share the same temporal coordinates (but nothing else) does not entail this status. (3.49) CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 136

and (3.50) also satisfy Connexion, since the highest structural argument is shared between the predicates in each case; this is what we would expect from the event structural template for coextensive Impossible Verb SVCs set out in (3.46), and it is furthermore what is required by the Predicate Combining Operations for Impossible Verb SVCs set out in the next section. But most pertinent to the demonstration that (3.49) and (3.50) are not Possible Verb, but Impossible Verb SVCs is the fact that these SVCs do not satisfy the Possible Verbs constraint. Fowru ‘fold (his arms under his chin)’ does not specify further properties in the interpretation space opened by sidon ‘sit’ in (3.49), and d’onga ‘look up’ does not specify further properties in the interpretation space opened by mbowo ‘bark’ in (3.50). This is a very brief characterization of Impossible Verb SVCs, their event structural profiles, and how they fail to fulfill the Possible Verbs constraints. The four Impossible Verb SVCs included in this section were each accompanied by Semantic Form representations with a single set of λ-abstracted arguments at Theta Structure which, upon closer examination, would confirm that all four SVCs satisfy the Theta Structure Property for SVCs. The next and most pressing task is to account for how these single Theta Structure representations are arrived at through the Predicate Combining Operations for Impossible Verb SVCs.

3.3.2 Predicate Combining Operations for Impossible Verb SVCs

The Predicate Combining Operations for Impossible Verb SVCs in (3.52) below account for how the separate Theta Structure and Semantic Form representations corresponding to distinct Possible Verb components in an Impossible Verb SVC (as in (3.51b)) are amalgamated into a single Theta Structure and Semantic Form representation (as in (3.51c)):

(3.51) a. Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10) CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 137

(3.52) Predicate Combining Operations for Impossible Verb SVCs:

(i) Event combining: subevents corresponding to separate Possible Verb components are combined into one overall event (with the Possible Verb ({}) components conjoined in the Semantic Form representation). (ii) There is obligatory coidentification of the highest structural arguments of each Possible Verb component, and the resulting coidentified argument retains its position as the highest structural argument in the combined Theta Structure. (iii) Language-specific, verb-specific coidentification rules enforce particular further types of argument coidentification. (iv) The (non-suppressed) lowest structural arguments of each Possible Verb component are coidentified, and the resulting coidentified argument retains its position as the lowest structural argument in the combined Theta Structure. (v) Uncoidentified structural arguments retain their structural status and (lowest) positioning while oblique arguments are concatenated in the combined Theta Structure.

It is through the application of these operations that Impossible Verb SVCs, like Possible Verb SVCs, come to have monoclausal patterns of arguments and adjuncts. (3.52) speciﬁes the operations that must be carried out on an Impossible Verb SVC in order for it to fulﬁll the Theta Structure Property for SVCs. If, for example, the highest structural arguments of the separate Possible Verb components cannot be CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 138

coidentiﬁed, as required under (3.52ii), then the structure cannot fulﬁll the Theta Structure Property and cannot be an SVC. In the simplest case, where two verbs each take a single argument, the Predicate Combining Operations (listed i, ii,.. as in (3.52)) proceed according to the following template (square brackets are placed around conjoined Possible Verb components to preserve the correspondence between {} brackets and Possible Verb structures):

(3.53)Verb1 Verb2: λx λs1 {V1(x)}(s1) & λu λs2 {V2(u)}(s2) +sc +sc

i. Event combining: s1+s2=s ii. Highest structural argument coidentiﬁcation: x=u

Combined predicate:

Verb1 Verb2: λx λs [{V1(x)} & {V2(x)}](s) +sc

An Impossible Verb SVC which conforms to this template is the following:

(3.54) a. ede.ra tu’u reb’o kai nahu, ... dem.emph get-up dance kai 1p ‘so then I got up and danced, ... ’ (Bima)

b. tu’u reb’o: λx λs1 {GET-UP(x)}(s1) & λu λs2 {DANCE(u)}(s2) +sc +sc

i. Event combining: s1 + s2 = s ii. Highest structural argument coidentiﬁcation: x=u

Combined predicate: λx λs [{GET-UP(x)} & {DANCE(x)}](s) +sc

This example puts into eﬀect the ﬁrst two operations of the Predicate Combining Operations: ‘event combining’ in (3.52i) ensures that the SVC forms a single, but CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 139

complex, event, while ‘highest structural argument coidentiﬁcation’ in (3.52ii) is universally required in Impossible Verb SVCs, and produces what is elsewhere referred to as ‘same subject serialization’ (Aikhenvald 2006, Bril et al. 2004, Crowley 2002). In the next simplest case, the Predicate Combining Operations proceed according to the following template:

(3.55)Verb1 Verb2: λy λx λs1 {V1(x,y)}(s1) & λv λu λs2 {V2(u,v)}(s2) +sc +sc +sc +sc

i. Event combining: s1+s2=s ii. Highest structural argument coidentiﬁcation: x=u iv. Lowest structural argument coidentiﬁcation: y=v

Combined predicate:

Verb1 Verb2: λy λx λs [{V1(x,y)} & {V2(x,y)}](s) +sc +sc

This template incorporates not only event combining (3.52i) and highest structural argument coidentification (3.52ii), but also lowest structural argument coidentification (3.52iv) for the internal arguments of the two transitive verbs involved. Lowest structural argument coidentification fulfills a similar function to Internal Argument Sharing in the formal SVC literature. While Internal Argument Sharing requires each verb in an SVC to share (or coidentify) an internal argument, lowest structural argument coidentification guarantees that two distinct internal arguments from distinct Possible Verb components will not be vying for the position of lowest structural argument in the combined SVC. Consider the following Impossible Verb SVC that conforms to the template in (3.55):

(3.56) a. Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10)

b. l´er´e: λy λx λs1 {COOK(x,y)}(s1) & λv λu λs2 {EAT(u,v)}(s2) +sc +sc +sc +sc CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 140

i. Event combining: s1+s2=s ii. Highest structural argument coidentiﬁcation: x=u iv. Lowest structural argument coidentiﬁcation: y=v

Combined predicate: l´er´e: λy λx λs [{COOK(x,y)} & {EAT(x,y)}](s) +sc +sc

Because y, the internal argument of COOK in (3.56b), can be coidentified with v, the internal argument of EAT, this examples fulfills lowest structural argument coidentification (3.52iv). Evbàré` ‘food’ is the single, shared object in (3.56a), projected from the single, coidentified lowest structural argument λ-abstractor in this Impossible Verb SVC. (3.56) satisfies the Theta Structure Property for SVCs because its Theta Structure (in the final, combined predicate representation) consists of two structural arguments, as do the Theta Structures for many clauses headed by single verbs. Were the lowest structural arguments of the two verbs to have distinct identities so that they could not be coidentified, the projected structure would not be an SVC, but a covert coordination with distinct lowest structural arguments projected into the distinct object positions of their respective verbs.4 (3.57) is a contiguous Impossible Verb SVC that fits the template in (3.55), and fulfills the Theta Structure Property for SVCs in the same manner as (3.56):

4The following sentence (=(1.29)), for example, is a covert coordination, since the lowest structural arguments are projected into distinct object positions as ´ıv`ın ‘coconut’ and ó. là ‘corn’: ` Ozó gbò. ó. ´ıv`ın bòló ó. là. Ozo planted coconut peel corn ‘Ozoplantedcoconutandpeeledcorn.’ (Edo;Stewart2001,51) This Verb-Object-Verb-Object syntactic structure can by no means be regarded as indicative of covert coordination, as opposed to SVC, structure, however, since in many instances, one of the verbs’ objects will have a non-structural status (as is argued to be the case for the instrument argument

ati-E ‘the stick’ in (3.58) below). In such instances, lowest structural argument coidentification is not even at issue (since the objects are not both lowest structural arguments), and these Verb-Object- Verb-Object structures will usually be SVCs (with SVC-specific clause structures), so long as their Theta Structure arrays of highest structural argument, lowest structural argument and (some type of) non-structural argument match the Theta Structural profiles of clauses headed by single verbs. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 141

(3.57) a. Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima)

b. rumpa weha: λy λx λs1 {FIND(x,y)}(s1) & λv λu λs2 {TAKE(u,v)}(s2) +sc +sc +sc +sc

i. Event combining: s1+s2=s ii. Highest structural argument coidentiﬁcation: x=u iv. Lowest structural argument coidentiﬁcation: y=v

Combined predicate: rumpa weha: λy λx λs [{FIND(x,y)} & {TAKE(x,y)}](s) +sc +sc

The template in (3.55) (and indeed (3.56) and (3.57), which conform to this template) are representative of the most typical, common-place SVCs containing two transitive verbs. But there are two notable departures from this most typical predicate combining scenario. The first is presented by the predicate combining operation (3.52iii). This operation allows for language-specific, verb-specific coidentification rules to enforce certain types of argument coidentification, and in particular rules that pertain to lowest structural arguments, that work to preclude lowest structural argument coidentification. Consider the following instrumental SVC, which is an Im-

possible Verb SVC with sequential semantics (if its translation is to be taken literally): (3.58) a. Kofi ts ati- fo Yao. Kofi take stick-def hit Yao ‘Kofi took the stick and hit Yao with it.’ (Ewe; Collins 1997, 466)

b. ts fo: λy λx λs1 {TAKE(x,y)}(s1) & λw λv λu λs2 {HIT(u,v) & +sc +sc −sc +sc +sc USE(u,w)}(s2)

Lowest structural argument coidentiﬁcation cannot apply here, because the lowest structural argument of TAKE (y) does not have the same identity as the lowest structural argument of HIT (v). y, the thing taken, cannot be coidentiﬁed with v, the thing hit. Instead, y has the same identity as w, the non-structural instrument CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 142

argument of USE that v is hit with. What is needed here is a special (language- specific, verb-specific) coidentification rule that coidentifies y with w. Chapter 4 will set out an (operation (iii)) instrumental SVC argument coidentification rule which does exactly that, cited in the following continuation of (3.58b):

... (3.58b) resumed:

i. Event combining: s1 + s2 = s ii. Highest role coidentiﬁcation: x=u

iii. Instrumental SVC argument coidentiﬁcation rule: y=w, λy [−sc]

Combined predicate: λy λv λx λs [{TAKE(x,y)} & {HIT(x,v) & USE(x,y)}](s) −sc +sc +sc

In addition to coidentifying y and w, this instrumental SVC argument coidentification rule ensures that the resulting coidentified argument (notated y in the final, combined predicate representation) has non-structural status. This means that (3.58) can satisfy the Theta Structure Property for SVCs, by virtue of having two structural arguments and one non-structural, instrument argument at Theta Structure, as also will many clauses headed by single verbs. As well as presenting the instrumental SVC argument coidentification rule itself, Chapter 4 considers how widespread (or rather, non-widespread) such operation (iii) rules may be, and in this connection, to what extent they are justified. The second departure from the most typical (lowest structural argument coidentification) predicate combining scenario for SVCs containing two transitive verbs is signaled by the (bracketed) reference to non-suppressed lowest structural arguments in (3.52iv), repeated here:

(iv) The (non-suppressed) lowest structural arguments of each Possible Verb component are coidentiﬁed, and the resulting coidentiﬁed argument retains its position as the lowest structural argument in the combined Theta Structure.

Suppression of an argument means that it is linked to ∅, and not projected into the syntax. (3.52iv) signiﬁes that if the lowest structural argument of a Possible Verb CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 143

component is suppressed, then it need not be coidentified with any other lowest structural argument(s). Lowest structural argument coidentification can be obviated by the suppression of a lowest structural argument, that is, and the instances in which it is apparent that this occurs are where a lowest structural argument, were it not suppressed, would not have the same identity as the other lowest structural argument(s) (meaning that it actually could not be coidentified). Bima is one language that displays this behavior. In this language the lowest structural arguments of all verbs but the last in the verb complex can be suppressed, as in (3.59):

(3.59) a. Reho hengga ka.losa.na harimau ... Reho open caus.exit.3 tiger ‘Reho opened (the cage (door)) and brought the tiger out ...’ (Bima)

b. hengga ka.losa: λy λx λs1 {ACT(x) & BECOME(OPEN(y))}(s1) & +sc +sc ∅ λw λv λu λs2 {ACT(u) & BECOME(LOC(v,¬IN(w)))}(s2) −sc +sc +sc ∅ i. Event combining: s1 + s2 = s ii. Highest structural argument coidentiﬁcation: x=u (Since y is suppressed, (iv) lowest structural argument coidentiﬁcation need not apply)

Combined predicate: λw λv λy λx λs [{ACT(x) & BECOME(OPEN(y))} & {ACT(x) & +sc +sc ∅ ∅ BECOME(LOC(v,¬IN(w)))}](s)

The unrealized referent of y (the internal argument of hengga ‘open’) in (3.59a) is a cage (or a cage door); this much is apparent from the context (not included in (3.59a)). Harimau ‘tiger’ cannot be the internal argument of hengga since a tiger cannot usually be opened; harimau is the internal argument of ka.losa ‘make exit’ only. The predicate combining operations for this SVC as set out in (3.59b) mention CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 144

that lowest structural argument coidentification need not apply (and indeed could not apply, since y=6 v) as a consequence of y being suppressed (linked to ∅, as indicated in the first line of (3.59b)). (3.60) is another Bima Impossible Verb SVC in which the lowest structural argument of a non-final verb is suppressed, obviating the need for lowest structural argument coidentification:

(3.60) a. Ede.ra lao fou b’atu kai.na ra’a ed’i Ama Seho ede, dem.emph go chase follow kai.3 blood leg Ama Seho dem ‘So he chased (Ama Seho), following the (trail of) blood from Ama Seho’s leg, ...’ (Bima)

b. lao fou b’atu: λy λx λs1 {CHASE(x,y) & MOVE-AWAY-FROM-DC(x)} +sc +sc ∅ (s1) & λv λu λs2 {FOLLOW(u,v)}(s2) +sc +sc

i. Event combining: s1 + s2 = s ii. Highest structural argument coidentiﬁcation: x=u (Since y is suppressed, (iv) lowest structural argument coidentiﬁcation need not apply)

Combined predicate:

λv λy λx λs [{CHASE(x,y) & MOVE-AWAY-FROM-DC(x)} +sc +sc ∅ & {FOLLOW(x,v}](s) y, the ‘chas-ee’, is suppressed in (3.60), and not realized in the syntax, although it is clear from the context who the ‘chas-ee’ is: it is Ama Seho. The argument of the ﬁnal verb b’atu ‘follow’, which is realized in the syntax, is ra’a ed’i Ama Seho ‘the blood from Ama Seho’s leg’. A trail of blood cannot be chased; the internal argument of CHASE does have to be an entity which is moving. Because the lowest structural argument of CHASE is suppressed, lowest structural argument coidentiﬁcation once again need not apply. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 145

(3.59) and (3.60) both satisfy the Theta Structure Property for SVCs because their Theta Structures each consist of two structural arguments and there clearly do exist clauses headed by single verbs whose Theta Structures consist of two structural arguments. Had the respective y lowest structural arguments in (3.59) and (3.60) not been suppressed, then these examples would not satisfy the Theta Structure Property, because these y arguments would not be able to be coidentified with the respective v lowest structural arguments in these examples. In each containing suppressed arguments, (3.59) and (3.60) illustrate a general point about how the Theta Structure Property for SVCs straddles the levels of argument structure and syntax; they demonstrate how what the Theta Structure Property targets is the projection of arguments into the clause (or the non-projection in the case of suppressed arguments). Only when its combination of projected arguments matches that of a single-verb-headed clause will an SVC satisfy the Theta Structure Property. But if this is the primary significance of these argument suppression examples in the context of this dissertation, I should hasten to add that these conclusions are not drawn on the basis of a wide range of data. I do only have these limited argument suppression SVCs from Bima, and no examples from other languages thus far. This also means that no observations can currently be made about which languages allow argument suppression in their SVCs, which do not, and what might condition this behavior. This section has so far demonstrated the application of items (i) (obligatory) ‘event combining’, (ii) (obligatory) highest structural argument coidentification, and (iv) lowest structural argument coidentification of the Predicate Combining Operations for Impossible Verb SVCs in (3.52), and has at least demonstrated the need for an instrumental SVC argument coidentification operation of type (iii), which will be set out in Chapter 4. The following SVC demonstrates why that part of (3.52v) which states that ‘uncoidentified structural arguments retain their structural status and (lowest) positioning’ is necessary, since v (realized as ngaha ‘food’) is just such an example of an uncoidentified structural argument:

(3.61) a. Sa.nai.nai ina sahe kande na.lao ngupa ngaha. one.day.day mother buﬀalo aforementioned 3.go seek food ‘Every day that mother buﬀalo would go and look for food.’ (Bima) CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 146

b. lao ngupa: λx λs1 {GO(x)}(s1) & λv λu λs2 {SEEK(u,v)}(s2) +sc +sc +sc

i. Event combining: s1+s2=s ii. Highest structural argument coidentiﬁcation: x=u v. Uncoidentiﬁed v retains its structural status and lowest positioning.

Combined predicate: lao ngupa: λv λx λs [{GO(x,y)} & {SEEK(x,v)}](s) +sc +sc

For the sake of completeness, it also seems ﬁtting to posit the following two Pred- icate Combining Operations before (v):

(3.62) (iv) b. The (non-suppressed) medial structural arguments of each Possible Verb components are coidentified, and the resulting coidentified argument retains its position as the medial structural argument in the combined Theta Structure. (iv) c. Oblique (non-suppressed) arguments with identical predicate decomposition configurations in separate Possible Verb components are coidentified.

However, I have yet to encounter any Impossible Verb SVCs in which either medial structural arguments or oblique arguments with identical predicate decomposition conﬁgurations in separate Possible Verb components need to be coidentiﬁed, and so these operations remain untested. These untested Predicate Combining Operations would, as the Predicate Com- bining Operations demonstrated in this section already do, deliver single, amalgamated Theta Structure and Semantic Form representations for various Impossible Verb SVCs. These operations ensure that Impossible Verb SVCs satisfy the Theta Structure Property for SVCs. The Predicate Combining Operations may appear somewhat lengthy, yet they arguably present nothing particularly complex or un- predictable; two subevents are combined into one overall event, highest structural CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 147

arguments are coidentified, language-specific coidentification rules may apply, or lowest and medial structural arguments or oblique arguments may be coidentified, unless any of these arguments are demoted to non-structural status and/or are suppressed. But irrespective of whether these operations are regarded as complex, they undoubt- edly present some level of complication or expense for Impossible Verb SVCs, and this is why it is appropriate to question, in the next section, whether Impossible Verb SVCs might not represent too much of an expense to justify their inclusion in the class of SVCs.

3.4 Arguments for including Impossible Verb SVCs in the class of SVCs

Section 3.3.2 demonstrated how Impossible Verb SVCs call for far more technical machinery (i.e. the Predicate Combining Operations for Impossible Verb SVCs) than Possible Verb SVCs in order to satisfy the Theta Structure Property for SVCs. A question which naturally arises is whether this kind of technical expense is really justified. Possible Verb SVCs require no additional predicate combining operations, so why not simply define and identify SVCs through this technically more basic type of SVC? This calls for a closer examination of Impossible Verb SVCs and their justification for being SVCs, but at the same time we could also turn the tables and query the inclusion of Possible Verb SVCs in the class of SVCs, since this second type of structure does have some quirks of its own, most notably in the form of the serialization lexical rules. The complications attached to Possible Verb SVCs could even be seen to strengthen the case for including Impossible Verb SVCs in the class of SVCs, if the claim to SVC status for the supposedly core type of Possible Verb SVC is not as straightforward as assumed. The most prominent quirk of Possible Verb SVCs is that some of these SVCs require serialization lexical rules in order to account for the fact that one of the verbs in these SVCs will have it semantics altered (or ‘specialized’) within the SVC context. The technical expense of the Predicate Combining Operations for Impossible Verb CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 148

SVCs is thus matched to some degree by serialization lexical rules in Possible Verb SVCs. Some of the Possible Verb SVCs included in Section 3.2.2 of this chapter that did require serialization lexical rules are (3.63) (=(3.11)) (which requires a ‘benefactive’ serialization lexical rule), (3.64) (=(3.13)) (which requires an ‘instrumental’ serialization lexical rule), and (3.65) (=(3.29)) (which requires a ‘completive aspect’ serialization lexical rule):

(3.63) Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5)

(3.64) Tsoda lá egbà ba cigbà. Tsoda take axe cut tree ‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63)

(3.65) kora yo ja cehg´a nal´ı eli ja kaba bai when 1sg per arrive there 3sg per ﬁnish go ‘When I arrived there he had gone’ (Kristang; Baxter 1988, 213)

Some Possible Verb SVCs that are not anticipated to require serialization lexical rules include SVCs expressing location, such as (3.66), or motion, such as (3.67), or non-motion resultative SVCs, such as (3.68)-(3.69):

(3.66) Wo zai chufang qie cai. 1sg be.at kitchen cut food ‘I cut food in the kitchen.’ (Mandarin; D´echaine 1993, 254)

(3.67) ede mpara karefa rai losa.na d’ei toples. dem then frog run exit.3 in jar ‘thenthefrogranoutofthejar.’ (Bima)

` (3.68) E. sósà gbé émá!tó. n pè. rhé. . Esosa hit metal be-flat ‘Esosa hit the metal flat.’ (E`. dó; Stewart 2001, 58)

(3.69) Amu ngaha ka.mpoi.na oha. Amu eat caus.gone.3 rice ‘Amu ate up all the rice.’ (Bima) CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 149

My impression is that, just in terms of frequency (rather than types), Possible Verb SVCs not requiring serialization lexical rules outnumber those that do (although more detailed semantic analysis could uncover the need for more serialization lexical rules, seeing this majority shift). Should sentences like (3.63)-(3.65) then be excluded from the class of SVCs through a requirement that the verbs display identical semantics within an SVC as in a non-SVC context? This would mean that serialization lexical rules could be dispensed with altogether. The first reason for rejecting this stance is that there are at least systematic corre- spondences between the verbs as they appear outside and within SVCs; verbs undergo the same alterations to their semantics within SVCs in language after language, and it is exactly these frequently attested semantic alterations that the serialization lexical rules serve to capture. The second reason for not excluding SVCs requiring serialization lexical rules from the class is both more weighty but possibly less justified; this is that (3.63)-(3.65) make up a core part of the traditional survey of SVCs, and that to exclude them would be to alter the membership of the class to the point of unrecognizability. That being the case, it seems better to include Possible Verb SVCs such as (3.63)-(3.65) in the class, along with the minor technical machinery of serialization lexical rules that accompany them. If the technical machinery of serialization lexical rules is ‘minor’ for the reason that it is only partially applicable to Possible Verb SVCs, the Predicate Combining Operations for Impossible Verb SVCs represent major technical machinery in that they must apply to each and every Impossible Verb SVC. The Predicate Combining Operations, or their cumbersomeness, argue against including Impossible Verb SVCs in the class of SVCs. But one feature of Impossible Verb SVCs which argues in favor of their inclusion in the class is the fact that, in contrast to Possible Verb SVCs, the verbs in an Impossible Verb SVC always have identical semantics between their SVC and non-SVC contexts. Impossible Verb SVCs are more clear-cut than Possible Verb SVCs in not requiring serialization lexical rules, and the complications that accompany serialization lexical rules (concerning questions of how pervasive and fine-grained these rules might be). However, in the end the most compelling reason why Impossible Verb SVCs should CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 150

be included in the class of SVCs is also the ultimate reason why Possible Verb SVCs must be included, and this is that Impossible Verb and Possible Verb SVCs alike display every facet of monoclausal behavior that is required of SVCs. As long as an SVC fulﬁlls every property (a)-(e) in the list set out in Section 2.5, repeated below, its SVC status should not be in question, regardless of whether predicate combining operations (or serialization lexical rules) must take eﬀect.

(a) SVCs are monoclausal.

(b) The verbs in an SVC must have the same tense, aspect, mood, polarity, and person-number categories.

(d) The intonational properties of a clause containing a serial verb combination are the same as those of a clause containing a simple predicate in the language concerned.

Fulﬁllment of monoclausal, SVC-deﬁning properties is the bottom-line, principled reason for including both Impossible Verb and Possible Verb SVCs in the class of SVCs. The instrumental and motion SVC case studies of the next two chapters demonstrate that there can be a practical reason for including both types of SVC, which is that, in practice it can be (almost) impossible to distinguish between Possible Verb and Impossible Verb SVCs in these domains. But prior to this, the remainder of CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 151

this chapter will evaluate the consequences of including Impossible Verb SVCs in the class of SVCs. What does this inclusion signify in general terms, and what impact does it have upon what the class of SVCs looks like overall?

3.5 Consequences of including Impossible Verb SVCs in the class of SVCs

This section discusses the four main consequences of including Impossible Verb SVCs in the class of SVCs, summarized in the headings for the four subsections: Section 3.5.1: SVCs are not just doing the things that would be done with clauses headed by single verbs in other languages; Section 3.5.2: SVCs do not just describe single (non- sequential) events; Section 3.5.3: only by including Impossible Verb SVCs can we see how SVCs can be constrained by the notion of ‘recognizable event type’; and Section 3.5.4: the Possible Verb-Impossible Verb distinction is superior to other two-way distinctions in SVCs. While Sections 3.5.1 and 3.5.2 discuss two fairly straightforward, direct consequences of including Impossible Verb SVCs in the class of SVCs, Section 3.5.3 opens a brief discussion (that will not be taken up again until Chapter 6) on less clear-cut issues of how SVCs are constrained cross-linguistically, and why certain languages constrain SVC formation in the way they do. But it is the message of Section 3.5.4 (that the Possible Verb-Impossible Verb distinction is superior to other two-way distinctions in SVCs) that calls for the most clariﬁcation and argumentation, given that this distinction constitutes the second main proposal of this dissertation (after the Theta Structure Property). Section 3.5.4 compares the Possible Verb- Impossible Verb distinction with some other two-way distinctions, asking what they can tell us and where their faults lie.

3.5.1 SVCs are not just doing the things that would be done with clauses headed by single verbs in other languages

That SVCs are not just doing the things that would be done with clauses headed by single verbs in other languages comes as a direct consequence of including Impossible CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 152

Verb SVCs in the class of SVCs since it is Impossible Verb SVCs, such as (3.70)- (3.72), that constitute the most blatant cases of expressing what single-verb-headed clauses could never express:

(3.70) Er´ı, ban.i mu w´e. ni.-mi.. 3sg run go walk-past ‘He went running and then walked.’ (Ijo; Williamson 1963, 152)

(3.71) Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima)

(3.72) Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10)

The idea that SVCs are doing the things that would be done with clauses headed by single verbs in other languages (so, the idea that inclusion of SVCs such as (3.70)- (3.72) effectively scotches) is widely recognized among linguists. It is related to a common misconception, that the purpose of serialization is always to ‘fill in’ functionally for single-verb-headed clauses. For while it is true that serialization can present a kind of strategy which fills the functional gaps created by the absence of particular grammatical categories (such as instrumental prepositions or ditransitive verbs), it is far from true that serialization responds to impoverished grammatical systems across the board, counter to what many linguists assume. Yoruba and Thai, for example, can express instruments either in prepositional phrases ((3.73a) and (3.74a)), or in SVCs ((3.73b) and (3.74b)), demonstrating that these SVCs are not responding to the absence of an instrumental preposition:5

5I assume that these pairs of sentences do, however, hold some (possibly quite subtle) semantic

differences. The use of Èaw ‘take’ in (3.74b)), for example, may have a definiteness and/or an emphatic effect on its object, in line with what Jagacinski (1992, 133) describes in relation to non-

instrumental Thai Èaw ‘take’ SVCs (mentioned in Section 3.2.2 in connection with example (3.28)). CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 153

(3.73) Yoruba (George 1982, 45)

a. Bádé gún Tolú n´ı ò. be. . Bade stab Tolu with knife ‘Bade stabbed Tolu with a knife.’ b. Bádé fi ò. be. gún Tolú. Bade use knife stab Tolu ‘Bade stabbed Tolu with a knife.’

(3.74) Thai (Foley and Olson 1984, 54) a. Kháw tàt yâa dûay mˆıit. he cut grass with knife ‘He cuts grass with a knife.’

b. Kháw aw mˆıit pay tàt yâa. 1sg take knife go cut grass ‘He cuts grass with a knife.’

Meanwhile, the Fongbe SVC in (3.75b) is clearly not responding to the absence of double object structure in this language, as evidenced by (3.75a):

(3.75) a. Kkú ná As´ıbá às n. Koku give Asiba crab

‘Koku gave Asiba crab.’ (Fongbe; Lefebvre and Brousseau 2002, 445)

b. Kkú s às n ná As´ıbá.` Koku take crab def give Asiba ‘Koku gave the crab to Asiba.’ (Fongbe; Lefebvre and Brousseau 2002, 448)

The key questions these data give rise to are: What diﬀers between the SVC and non-SVC sentences which apparently instantiate the same function? Why does the SVC member of the pair even exist? Answers to these questions may reveal that, as well as SVCs not just doing things done with clauses headed by single verbs in other CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 154

languages (as demonstrated through Impossible Verb SVCs with no possible single- verb-headed translations), SVCs are also not even doing things done with clauses headed by single verbs in the same language. It is true that some SVCs in some languages are filling in for functional voids, doing what would be done with single-verb-headed clauses in other languages. But it should also be stressed that this is just a tendency, however strong. It is not an accurate summary of what serialization is all about. (3.73b), (3.74b) and (3.75b) present three subtle cases of SVCs not filling in for functional voids. Impossible Verb SVCs such as (3.70)-(3.72) present more blatant cases of SVCs not filling in for single-verb-headed clauses.

3.5.2 SVCs do not just describe single (non-sequential) events

Section 1.2.4.1 has already argued that SVCs do not just describe single, non-sequential events, in opposition to a number of accounts which seem to suggest that SVCs are event structurally restricted in this way. It was also suggested in Section 1.2.4.1 that these accounts suffer from the major failing of not being specific about what constitutes a single event in the first place. Claims about events in the SVC literature are seldom (if ever) accompanied by any very clear event structural framework or terminology, and so I hope that the event structural characterizations of Possible Verb and Impossible Verb SVCs set out in Sections 3.2.3 and 3.3.1 of this chapter represent an improvement in this direction. Sections 3.2.3 and 3.3.1 established five different temporal and event structurally distinguished types of SVC. There are three temporal and event structurally distinguished types of Possible Verb SVC (unitary-event, coidentified and non-coidentified resultative SVCs) and two types of Impossible Verb SVC (sequential and coextensive Impossible Verb SVCs). But do these event structural distinctions take us any closer to identifying what a ‘single event’ actually is? Unitary-event SVCs have an uncontroversial claim to ‘single eventness’, since they are not even made up of two separate subevents. Coidentified SVCs could be said to have a very strong claim to single event status, since temporal dependence means that the subevents involved in these CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 155

SVCs are inextricably linked to the point that they are conceptualized as a single event. The claim to single event status is rather strong for non-coidentified resultative SVCs since, although they do not exhibit temporal dependence, the subevents are causally connected, and the second verb delimits the action described by the first verb. Non-coidentified resultative SVCs have as much claim to single event status as any non-coidentified resultative sentence. In comparison to these three cases of Possible Verb SVCs, neither sequential Impossible Verb SVCs nor coextensive Impos- sible Verb SVCs have a strong claim to ‘single eventness’, and this is why inclusion of Impossible Verb SVCs in the class of SVCs means that the idea that SVCs always describe single events can be dismissed. The same three instances of Impossible Verb SVCs that were used in Section 3.5.1 to demonstrate how SVCs are not just doing the things that would be done with single-verb-headed clauses in other languages can be used again here to demonstrate how SVCs do not just describe single (non-sequential) events ((3.76)=(3.70), (3.77)=(3.71), (3.78)=(3.72)):

(3.76) Er´ı, ban.i mu w´e. ni.-mi.. 3sg run go walk-past ‘He went running and then walked.’ (Ijo; Williamson 1963, 152)

(3.77) Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima)

(3.78) Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10)

(3.76)-(3.78) each describle multiple, sequential events. That the same three examples as in Section 3.5.1 can be used again here is indicative of the fact that the consequence of including Impossible Verb SVCs in the class of SVCs discussed in Section 3.5.1 and that discussed in this section are of a similar ilk. But this is to be expected taking into account ﬁrstly that all I have done is identify sequential and coextensive Impossible Verb SVCs as the two event structurally speciﬁed types of CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 156

SVC which categorically do not describe single events, secondly that these types of SVC are simply an event structural classification of the SVCs which do not satisfy the Possible Verbs constraints, and thirdly that SVCs which do not satisfy the Possible Verbs constraints cannot be doing the things done with single-verb-headed clauses in other languages. The consequence of including Impossible Verb SVCs in the class of SVCs discussed in this section is in some sense an event structural recasting of the consequence of Section 3.5.1, in other words, although this is not to claim that the use of event structural terms is purely derivative. Event structural analysis of Possible Verb and Impossible Verb SVCs has already been promoted as a way of making the Possible Verb-Impossible Verb distinction more accessible and testable (instead of having to apply LDG’s Possible Verbs constraints in every case). Here, the value of stating what the inclusion of Impossible Verb SVCs in the class of SVCs means event structurally (i.e. as the consequence that SVCs do not just describe single events) can be found in the way that this meets some claims about events in the SVC literature head-on. But looking beyond these current concerns, the event structural reclassification of SVCs even pulls some further explanatory weight all of its own, one example of which is the ease with which the conditions under which SVCs are temporally iconic can be stated in event structural terms (and only in event structural terms). SVCs are temporally iconic in all cases where they have sequential semantics, where sequential semantics is identified in the concise terms of temporal precedence set out in Sections 3.2.3 and 3.3.1. To be maximally concise and explicit, this generalization can be stated as the ‘No Inverse Temporal Iconicity’ Principle in (3.79), which is framed in the following event structural terms (from Sections 3.2.3 and 3.3.1): (i)

Vi is that component of a Semantic Form representation which encodes the meaning of Verbi, (ii) si is a (sub)event (or situation) corresponding to Vi, (iii) tci is the time of commencement of si, (iv) tti is the time of termination of si, and (v) the symbol

(3.79) No Inverse Temporal Iconicity:

If tci

The truth of (3.79) could be veriﬁed through every SVC included in this dissertation, although there is not the space to pursue this goal here. The point is that the conditions under which SVCs are temporally iconic can be best captured (and can arguably only be captured) in the event structure terms of (3.79), so this is another element of SVC research which recommends event structural analysis.

3.5.3 Only by including Impossible Verb SVCs can we see how SVCs can be constrained by the notion of ‘recognizable event type’

Section 1.2.4.2 observed that SVC formation seems to be constrained diﬀerently in different languages, but that, at least in some languages, SVC formation is constrained by the notion of what constitutes a ‘recognizable event type’ in the society in which that language is spoken. Evidence for this kind of constraint on SVC formation is unfortunately rather limited at present, although Chapter 6 will attempt to formulate some more evidence, mainly based on the Bima data included throughout this dissertation. The three cases discussed in Section 1.2.4.2 were: (i) Yoruba, in which Durie commented that the concept of eating and then sleeping could be expressed in SVC form only because this constitutes a recognizable event type in Yoruba-speaking society (in contrast to eating and then returning home, which does not constitute a recognizable event type); (ii) White Hmong, in which the concept of dancing and playing bamboo pipes can be expressed in SVC form only because this constitutes a recognizable event type in White Hmong society (Jarkey 1991); and (iii) Alamblak, in which the concept of climbing a tree and looking for insects can be expressed in the following SVC, because this constitutes a recognizable event type in this society (in contrast to climbing a tree and looking at the stars): CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 158

(3.80) Myt ritm muh-hambray-an-m. tree insects climb-search.for-1s-3pl ‘I climbed the tree, looking for insects.’/‘I climbed the tree and looked for insects.’ (Alamblak; Bruce 1986, 26)

(3.80) is an Impossible Verb SVC; it does not satisfy the Possible Verbs constraints and there can exist no single-verb-headed clause which expresses the concept of climbing a tree and looking for insects. The Yoruba SVC that expresses the concept of eating and then sleeping and the White Hmong SVC that expresses the concept of dancing and playing bamboo pipes will also both be Impossible Verb SVCs for the same reasons. Taking into account that the three Alamblak, Yoruba and White Hmong SVCs under discussion are all Impossible Verb SVCs, the consequence that heads up this section becomes quite obvious: it is only by including Impossible Verb SVCs in the class of SVCs that we can see how SVCs can be constrained by the notion of ‘recognizable event type’ in these cases. If these Alamblak, Yoruba and White Hmong Impossible Verb SVCs had not been admitted as SVCs in the ﬁrst place, these insights about how event conceptualization can constrain SVC formation would be missed.6 But if this consequence of including Impossible Verb SVCs in the class of SVCs is obvious, the answers to some of the questions connected to it are not. Must the SVCs of any given language be constrained by some type of event conceptualization (not necessarily characterized as a ‘recognizable event type’)? How are these conceptualizations best characterized and formalized? Could SVC formation be completely unconstrained in some languages? These are questions which, as indicated in the closing remarks of Chapter 1, will only be addressed speculatively in Chapter 6.

6This is not to say that SVC formation could not be constrained by Possible Verb event structures – which do naturally constitute recognizable event types – in some languages. But language- particular constraints on SVC formation that extend into the realms of Impossible Verb structure provide more interesting and surprising indications of what a recognizable event type is. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 159

3.5.4 The Possible Verb-Impossible Verb distinction is superior to other two-way distinctions in SVCs

An immediate by-product of recognizing another type of SVC in addition to the core class of Possible Verb SVCs is that we now have a two-way distinction between Pos- sible Verb and Impossible Verb SVCs. Section 1.2.4.3 drew attention to the fact that there has been a long tradition in the SVC literature of dividing the class of SVCs into two types, with these divisions usually drawn up in event structural terms, or in pre- cursors to event structural terms. Christaller (1875) first drew a distinction between ‘accidental’ and ‘essential’ SVCs, where the former express successive actions while the verbs in the latter category are in “an internal or inseparable relation or connection” (1875, 144). Sebba (1987) distinguishes ‘coordinating’ and ‘subordinating’ SVCs, where the former “refer to several actions” while the latter refer “to a single action rather than a series of related actions” (1987, 112). Stewart’s ‘consequential’ SVCs “express a natural sequence of events” with the verbs in a temporally iconic order (2001, 14), while his ‘resultative’ SVCs are non-sequential and are made up of one event (2001, 87). Aikhenvald’s (2006) distinction between ‘symmetrical’ and ‘asymmetrical’ SVCs, although not based in event structural terms, recognizes the former as mostly describing multiple, sequential events and the latter single, unitary events. The basis for the Possible Verb-Impossible Verb distinction is not event structural in nature (but instead concerns the question of whether an SVC satisfies the Possible Verbs constraints), although the analyses of the temporal and event structures of Possible Verb and Impossible Verb SVCs in Sections 3.2.3 and 3.3.1 makes for easy comparison with two-way distinctions in SVCs which do divide the class along event structural lines. But before making any comparisons it seems fitting to ask: why have all these studies of SVCs converged upon the idea that two types of SVC should be distinguished, and so often in such a way that event structure notions are integral? The most likely answer centers upon the fact that the two types are so distinct, event structure descriptions capture the sense in which they are distinct, and the point of drawing the distinction is to emphasize that both of these types are included in the CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 160

class. If one type of inclusion needs to be emphasized in particular, then it is the inclusion of the event structurally ‘looser’, multiple event SVCs, since a number of SVC researchers, such as most of those surveyed in Section 1.2.4.1, apparently seek to exclude multiple event SVCs by stressing a ‘single eventness’ criterion for SVCs. Researchers who draw a distinction between single event and multiple event SVCs have succeeded in recognizing that multiple event SVCs are just as monoclausal in their grammatical behavior as single event SVCs, and so should also be included in the class. This is the justification for why multiple event SVCs should be included, and a two-way distinction in SVCs be drawn in the first place; the next question to ask is why the distinction drawn in this dissertation, between Possible Verb and Impossible Verb SVCs, should be preferred over all the other two-way distinctions in the SVC literature. What does it have to offer that the others do not? Part of the reason for preferring the Possible Verb-Impossible Verb distinction concerns the problems with some of the other two-way distinctions. Christaller’s seminal distinction, although very insightful, is problematic (or at least not very serviceable) in that it is not framed in modern terms. As mentioned in Section 1.2.4.3, the open versus closed class basis for Aikhenvald’s (2006) asymmetrical and symmetrical SVCs renders this distinction inapplicable in many cases. Section 1.2.4.3 did not mention any of the problems with Sebba’s (1987) coordinating-subordinating distinction or Stewart’s (2001) consequential-resultative distinction, but the most questionable feature of the former is the inclusion of non-SVC covert coordinations such as (3.81) in the category of ‘coordinating’ SVC,

(3.81) Kofi sutu Amba kiri Kwaku. Kofi shoot Amba kill Kwaku ‘Kofi shot Amba and (?but) killed Kwaku.’ (Sranan; Sebba 1987, 110) while the most serious failing of the latter is the restricted membership of the ‘resultative’ category. Stewart (2001) specifies that ‘resultative’ SVCs strictly include only those SVCs with resultative semantics;7 instrumental, aspectual, benefactive and

7Stewart (2001, 75) identiﬁes resultative SVCs through the following aspectual properties: a. The ﬁrst subevent in the resultative construction must be either a process or an activity. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 161

comparative SVCs, among many others, are not included in the category of resultative SVCs, and so are not included in the class of SVCs at all. Stewart (2001) claims to have shown that the “core of SVCs” (2001, 272) is much more restricted than was previously thought in being limited to only resultative and consequential SVCs, but his exclusions of certain types of SVC are dubious to say the least,8 and the ultimate result is a misrepresentation of the cross-linguistic class of SVCs. One two-way distinction not discussed in Section 1.2.4.3 is van Staden and Reesink’s (2008) distinction between ‘component’ and ‘narrative’ serialization. At the heart of this distinction is the concept of a ‘macro-event’; while component SVCs describe single macro-events, narrative SVCs describe two or more macro-events, and van Staden and Reesink draw on both the work of Talmy (2000), (who originally coined the term) and Bohnemeyer (1999, 2001) (who amended the term) in order to describe what a macro-event is. Van Staden and Reesink’s description of macro-events, as informed by the work of Talmy (2000) focuses on the idea that a macro-event could be made up of both a ‘framing event’ and a ‘co-event’. While framing events contribute some of the more (temporal, event and argument) structurally signiﬁcant information, such as Path information (Jerry walked IN ), ‘temporal contouring’ (she KEPT ON shouting), or change of state (the kettle boiled DRY ), the co-event “bears a support relation to the framing event” (Talmy 2000, 220), and could express, say, the Manner in which the action is performed. Framing events and co-events constitute distinct subevents within a single macro-event. Van Staden and Reesink’s description of macro-events as informed by Bohnemeyer (1999, 2001) emphasizes that macro-events (which are possibly composed of two (or more) subevents) are uniquely bounded in time and share a single set of semantic arguments. Presumably (had they been writing a little later), van Staden and Reesink’s macro-events would be deﬁned by Bohnemeyer et

b. The event denoted in the resultative construction can only be delimited once. c. A resultative construction can only be modified by an ‘in an hour’ type of temporal adverb, not ‘for an hour’. 8 He excludes E`. dóinstrumental constructions which have the initial verb yá ‘use’ from the class of serial verbs, for example, on the fairly strong grounds that yá does not display most of the language- specific properties of verbs, although this does not entail that every other type of instrumental SVC (including the E`. dóinstrumental construction which has the initial verb rhié ‘take’) should not have serial verb status, especially if they do display the full quota of serial verb properties. CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 162

al.’s (2007) Macro-event Property, which specifies that “temporal operations such as time adverbials, temporal clauses, and tenses necessarily have scope over all subevents encoded by the construction” (the Macro-event Property will be considered further in Chapter 5). The distinctions between Possible Verb and Impossible Verb SVCs on the one hand and component and narrative SVCs on the other are framed in different terms, but do they amount to the same distinction? That they do not is evident from the inclusion of SVCs like the following in the class of component SVCs; (3.82) cannot be a Possible Verb SVC since it does not satisfy the Possible Verbs constraints, it does not fit the unitary-event, coidentified or non-coidentified resultative event structural profiles, and its content could never be expressed in a clause headed by a single verb:

(3.82) K-tone k-´a. 1sg-go 1sg-eat ‘I go (and/to) eat.’ (Tetun; van Klinken 1999, 262)

Component SVCs cannot be equated with Possible Verb SVCs then, and narrative SVCs cannot be equated with Impossible Verb SVCs. Narrative SVCs are actually restricted to Papuan SVCs in van Staden and Reesink’s survey (of a small number of Austronesian and Papuan SVCs), and in fact the main import of the component- narrative distinction as it appears in this paper may be what it can tell us about the SVCs of this family, which in many respects appear more complex than the SVCs of any other language family. The two-way distinctions discussed in this section display diﬀerent focuses and data restrictions, and in some cases they also display some serious faults. The Possible Verb-Impossible Verb distinction emerges as a distinction with a cross-linguistically very wide range of application, which does not display any of the faults described here. However, the ultimate reason for preferring the Possible Verb-Impossible Verb distinction is that it captures an insight which none of the other distinctions do. This insight is the fact that some SVCs (Possible Verb SVCs) correspond to simplex clauses headed by single verbs in other languages, whereas others (Impossible Verb SVCs) cannot. If this insight seems anti-climactic it is only because it has taken us back CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 163

full circle to the consequence of including Impossible Verb SVCs already discussed in Section 3.5.1: SVCs are not just doing the things that would be done with clauses headed by single verbs in other languages. As consequences of including Impossible Verb SVCs in the class of SVCs go, the one discussed in this section could appear somewhat trivial in the sense that, as soon as Impossible Verb SVCs are admitted into the class, a two-way distinction between Possible Verb and Impossible Verb SVCs comes for free. But it is important to spell out why this distinction is to be preferred over all the others. This has meant laying bare the faults of other two-way distinctions, but more importantly, it has meant focusing on the one generalization which other approaches to SVCs have (a) not captured in their two-way distinctions, and (b) somehow failed to recognize entirely, whether because their SVC distinctions do not capture it, or because it is regarded as simply too obvious to mention. It is not too obvious to mention, and so I’ll say it again: some SVCs are doing the things done with single-verb-headed clauses in other languages while others are not.

3.6 Conclusion

While the previous chapter argued for the Theta Structure Property for SVCs as the means of enforcing monoclausal argument structure in SVCs, this chapter has argued for a looser interpretation of monoclausal argument structure. The Predicate Combining Operations for Impossible Verb SVCs are the means through which this loosening occurs, and the result is the admission of Impossible Verb SVCs into the class of SVCs, producing (yet) another two-way distinction in SVCs. Impossible Verb SVCs are constructions which, like Possible Verb SVCs, behave in every grammatical respect as if they are monoclausal, and this is why the Predicate Combining Operations should be permitted to expand the class of SVCs. Something that went unremarked in this chapter is how surprising and interesting it is that Impossible Verb SVCs, such as (3.83)-(3.85), should be trying so hard to be monoclausal, when these describe exactly the kinds of concepts that would otherwise be expected to receive biclausal expression: CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 164

(3.83) Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima)

(3.84) Ozó` lé èvbàré ré. Ozo cook food eat ‘Ozo cooked and ate some food.’ (E`. dó; Stewart 2001, 10)

(3.85) Myt ritm muh-hambray-an-m. tree insects climb-search.for-1s-3pl ‘I climbed the tree, looking for insects.’/‘I climbed the tree and looked for insects.’ (Alamblak; Bruce 1986, 26)

Most interesting of all are those Impossible Verb SVCs such as (3.85) that reveal what constitutes a recognizable event type in the language concerned. The chief insight of the Possible Verb-Impossible Verb distinction presented in this chapter – which is moreover an insight that none of the numerous other two-way distinctions in SVCs captures – is that some SVCs (Possible Verb SVCs) are doing what could be done by clauses headed by single verbs in other languages, while others (Impossible Verb SVCs) are not. The Possible Verb-Impossible Verb distinction captures this insight straightforwardly since the Possible Verbs Constraints upon which it is predicated draw a clean division between possible single-verb-headed semantic content and impossible single-verb-headed semantic content (with the former flagged by a single set of {} brackets at Semantic Form and the latter by multiple {} brack- etings). But the Possible Verbs constraints (or fulfillment or non-fulfillment thereof) are not the only means of identifying Possible Verb or Impossible Verb SVCs; in this chapter event structure came to the fore, with the event structural characterizations of Possible Verb and Impossible Verb SVCs providing an alternative means of identifying these two types of SVC. ‘Event structure’ could mean different things; van Staden and Reesink’s (2008) utilization of Talmy’s (2000) and Bohnemeyer’s (1999, 2001) ‘macro-events’ presents one approach to event structure, for example, while Chapter 4 will further explicate the nature of the ‘event complexity’ approach used in this dissertation, which identifies subevents of an event through their temporal coordinates. This event complexity CHAPTER 3. POSSIBLE VERB AND IMPOSSIBLE VERB SVCS 165

approach is contrasted with ‘causal force’ analyses of (instrumental) events in Chapter 4. Although there is still much discussion of event structural issues to follow, I nevertheless want to claim here that my event complexity approach provides to date the most rigorous treatment of the event structure of SVCs – far more rigorous than all the vague claims that ‘SVCs describe single events’, but also more rigorous (or at least more extensive) than van Staden and Reesink’s (2008) ‘macro-event’ treatment of SVC event structures (apparently the only other concerted attempt at event structural analysis in SVCs). The foundations for this rigorousness are the temporal and event structural profiles for Possible Verb and Impossible Verb SVCs set out in this chapter, and one additional sideline of my event structural analysis is the ease with which the conditions on temporally iconic interpretation in SVCs can be formulated, in the No Inverse Temporal Iconicity Principle set out in Section 3.5.2. The No Inverse Temporal Iconicity Principle will be mentioned again briefly in Chapters 4 and 5, although event structural analysis is otherwise implemented purely for the purpose of distinguishing Possible Verb from Impossible Verb SVCs in the case studies of both chapters. In fact, event structural analysis features very prominently towards the end of distinguishing Possible Verb from Impossible Verb instrumental and motion SVCs, and it is moreover of considerable practical importance that it should do so since, in order to show how an SVC satisfies the Theta Structure Property for SVCs, it is first necessary to determine whether the SVC is a Possible Verb or Impossible Verb SVC. The case studies of the next two chapters demonstrate that this is not as easy as it may sound, for while some SVCs (of the ‘Ozo bought and read LGB’ ilk discussed in Section 1.3.3) display blatant Impossible Verb status, instrument and motion Impossible Verb SVCs are seldom, if ever, so blatant. But if there is difficulty in drawing this distinction in these domains, there is at least considerable value to be derived from attempting to do so, in the form of the more complete picture of Possible Verb (and Impossible Verb) events that emerges in the process. Chapter 4

Instrumental SVCs

4.1 Introduction

The instrumental SVCs of this chapter are drawn from a single domain of serialization that presents one of the ﬁnest lines between Possible Verb and Impossible Verb SVCs – two kinds of SVC that satisfy the Theta Structure Property for SVCs in very diﬀer- ent ways. For while (as remarked in Section 1.3.3) some Impossible Verb SVCs scream out that they could never possibly satisfy the Possible Verbs constraints, Impossible Verb instrumental SVCs barely whimper – if there are any Impossible Verb instrumental SVCs at all. Translations of instrumental SVCs are uniformly suggestive of Possible Verb semantics, although some peculiar things do happen in the translations of instrumental SVCs whose initial verb is indicated (in the interlinear gloss) to have the semantics of ‘take’. Jansen, Koopman and Muysken (1978, 146-7), for example, specify that whereas the Sranan sentence which uses the instrumental preposition nanga ‘with’ in (4.1) is exactly equivalent to the English expression indicated, the instrumental SVC using teki ‘take’ in (4.2) instead “implies that there have been two consecutive actions” – but in that case, why do they not translate (4.2) accordingly, as ‘I took the knife and cut the bread’?:1

1Elsewhere they do use this sequential translation; see Jansen, Koopman and Muysken (1978, 153).

166 CHAPTER 4. INSTRUMENTAL SVCS 167

(4.1) Mi e koti a brede nanga a neﬁ. 1p asp cut the bread with the knife ‘I cut the bread with the knife.’ (Sranan; Jansen, Koopman and Muysken 1978, 146)

(4.2) Mi e teki a neﬁ koti a brede. 1p asp take the knife cut the bread ‘I cut the bread with the knife.’ (Sranan; Jansen, Koopman and Muysken 1978, 146)

The following pair of Mandarin instrumental SVCs presents another anomaly, although in this case the irregularity centers upon the initial verb na, which is indicated to have the semantics not of ‘take’, but of ‘hold’:

(4.3) Ta na dao qie-le rou. he hold knife cut-perf meat ‘He cut the meat with a knife.’ (Mandarin; Law 1996, 200)

(4.4) Ta kuaikuaide na dao qie-le rou. he quickly hold knife cut-perf meat (a) ‘He quickly cut the meat with the knife’; (b) ‘He cut the meat with the knife which he quickly took.’ (Mandarin; Law 1996, 213)

Although it can be inferred from the translation of (4.3) – ‘he cut the meat with a knife’ – that the actor does indeed hold the knife, the semantics of ‘hold’ does not actually feature in this translation (as it does, say, in the sentence he cut the meat, holding the knife). This could reflect that an instrumental serialization lexical rule comes into effect in (4.3), determining that na does not have the semantics of ‘hold’ in the SVC context, but instead contributes a more bleached or functional instrumental meaning, such as could be attributed to the English preposition ‘with’, or the verb ‘use’. However, once na is modified in isolation under the (b)-reading of the instrumental SVC in (4.4), it suddenly takes on a new meaning – that of ‘take’. What is the semantics of na in this instrumental SVC context: ‘hold’, ‘take’, or ‘use’/‘with’? Most peculiar of all is the translation for the following Sranan instrumental SVC: CHAPTER 4. INSTRUMENTAL SVCS 168

(4.5) Kofi teki a nefi langzaam koti a brede eseesi. Kofi take the knife slowly cut the bread quickly ‘Kofi used the knife slowly to cut the bread quickly.’ (Sranan; Sebba 1987, 133)

Is it actually possible to use a knife slowly to cut some bread quickly? Using a knife quickly to cut some bread slowly is a common enough scenario; if the bread is solid the agent could be sawing away at it at speed, only managing to detach a slice after some time. But if the action of using the knife is itself slow, it is hard to see how the cutting of the bread could be anything but slow. (4.5) would make more sense were the semantics of teki ‘take’ to be carried through to the translation: ‘Kofi took the knife slowly and cut the bread quickly’. TAKE, unlike USE, can be felicitously modified by SLOWLY in isolation in this example. The three treatments of instrumental SVCs considered so far would appear to have one thing in common: a reluctance to attribute to the initial verb the semantics of ‘take’ in the English translations provided. The only translation in which ‘take’ surfaces in the SVCs above is the (b)-reading of (4.4); ‘take’ is otherwise avoided in the translations for (4.2), (4.3) and (4.5). The use of ‘take’ in the (b)-reading for (4.4) may be necessitated by the fact that kuaikuaide ‘quickly’ cannot modify atelic ‘hold’, but instead requires its closest telic relative, ‘take’. Modification of the initial verb in (4.5) by langzaam ‘slowly’ (combined with the specification that cutting the bread is achieved quickly) should, I argue, necessitate that this SVC is translated with ‘take’. For (4.2), on the other hand, there is no necessity to translate this SVC with ‘take’; it is rather that using a biclausal English translation with ‘take’ would better fit Jansen, Koopman and Muysken’s (1978) specification that this SVC describes two consecutive actions. If it is true that translation of instrumental SVCs with ‘take’ is avoided, even in cases where the use of ‘take’ would seem to be necessitated, then it must be concluded that translations for instrumental SVCs cannot be relied upon as accurate reflections of their semantic properties. This is hardly an unusual or alarming conclusion to arrive at in a linguistic study, although it is still worth asking what motivates these translational inaccuracies. Here I can only guess that non-sequential translations that make no mention of taking (even when taking is clearly entailed) are favored for CHAPTER 4. INSTRUMENTAL SVCS 169

instrumental ‘take’ SVCs for one or more of the following reasons:

(i) non-sequential, ‘non-taking’ translations may more accurately reﬂect the idea that the SVC is monoclausal, in contrast to English translations using ‘take’ that are biclausal;

(ii) non-sequential, ‘non-taking’ translations contain only one verb, and so may be favored by linguists who believe that SVCs are always doing what would be done with a clause headed by a single verb in another language;

(iii) non-sequential, ‘non-taking’ translations may be perceived as the closest functional equivalent to an instrumental ‘take’ SVC.

Linguists may also supply non-sequential translations for instrumental ‘take’ SVCs simply by default, regarding the distinction between sequential and non-sequential semantics to be of little consequence. In the context of the present study, however, this distinction is of considerable consequence. While non-sequential semantics (in which ‘take’ does not correspond to a subevent) obtains in the Possible Verb structure for instrumental SVCs set out in (4.6), sequential semantics obtains in the Impossible Verb structure in (4.7), and so, for the case of instrumental SVCs, sequential versus non-sequential semantics signals the distinction between Possible Verb and Impossible Verb structure itself.

(4.6) Event structural proﬁle for Possible Verb instrumental SVCs (a type of unitary-event SVC)

SF: ... λz ... λx λs {V1(x,...) & USE(x,z)}(s)

CS: V1 corresponds to an event s1. USE does not correspond to a (sub)event.

(4.7) Event structural proﬁle for Impossible Verb instrumental SVCs

SF: ... λz ... λx λs [{TAKE(x,z)} & {V2(x,...) & USE(x,z)}](s)

CS: TAKE corresponds to a subevent s1 and V2 to a subevent s2. tc1

tc2. (USE does not correspond to a subevent.) CHAPTER 4. INSTRUMENTAL SVCS 170

The main goal for this chapter is to distinguish Possible Verb from Impossible Verb instrumental SVCs, and to then show how these SVCs satisfy the Theta Structure Property for SVCs. If some instrumental ‘take’ SVCs can be verified as Impossible Verb SVCs with sequential semantics, then this will lend support to the claim of this dissertation that there is this additional, non-core type of SVC, which cannot be otherwise (accurately) expressed in a clause headed by a single verb, and which, although displaying monoclausal argument structure, can only satisfy the Theta Struc- ture Property for SVCs through the extra complication of the Predicate Combining Operations. But in practice, distinguishing Possible Verb from Impossible Verb instrumental SVCs is no easy matter; the domain of instrumental serialization presents one of the sternest tests for the Possible Verb-Impossible Verb distinction, as borne out by the discussions of Possible Verb and Impossible Verb instrumental SVCs in Sections 4.3 and 4.4 respectively. While certain instrumental SVCs are identified as either Possible Verb SVCs in Section 4.3 or Impossible Verb SVCs in Section 4.4, of just as much importance in these sections are the arguments as to why (4.6) and (4.7) are the appropriate semantic profiles for these SVCs. Section 4.3 in particular needs to account for the use of a USE predicate in (4.6), while Section 4.4 needs to explicate the instrumental SVC argument coidentification rule (hidden in (4.7)) that ensures that the second arguments of TAKE and USE in (4.7) have the same identity. But prior to this I need to clarify exactly what an instrumental SVC is, which in Section 4.2 means illustrating which verbs are involved in these SVCs, how the instrument participant in these SVCs can be identified, and how the SVCs identified as instrumental SVCs fit the profile that semanticists have attributed to an event involving an instrument.

4.2 Instrumental events and instrumental SVCs

This section examines the proﬁle of an event involving an instrument participant (or just ‘instrumental event’) in Section 4.2.1 before describing how instrumental events are expressed in serial verb form in Section 4.2.2. CHAPTER 4. INSTRUMENTAL SVCS 171

4.2.1 Instrumental events

Towards the end of answering the question ‘what happens in an instrumental event?’, this section considers some of the ways in which instrumental events have been semantically analyzed by different linguists. Croft’s (1991) models of events as causal chains are considered first, since this type of semantic representation provides one of the most explicit profiles of an instrumental event, and this is followed by Jackendoff’s (1990) analysis of instrumental events. Lastly, Koenig et al.’s (2007) wide-ranging survey and analysis of instrumental events is discussed. Croft’s (1991) representation of events as causal chains is grounded in Davidson’s (1969) claim that causal structure defines events, as well as Talmy’s (1976, 1988) work on force dynamics. (4.8b), corresponding to the English expression containing an instrument argument in (4.8a), is one example of a causal chain:

(4.8) a. John shot the bottle to pieces with a gun.

b. John gun bottle (bottle) (bottle) • • • • • VOL Contact Change Result State State

Each vector or “arc” in (4.8b) corresponds to an “atomic event”. Causal chains are decomposed into atomic events according to the following criteria (Croft 1991, 165):

(4.9) Criterion 1. An atomic event must be of only one causation type. Criterion 2. An atomic event must be of a single inherent aspectual type, speciﬁcally a state or process. Criterion 3. An atomic event containing two participants must have those participants aligned in the direction of “transmission of force.” Criterion 4. An atomic event must be a single qualitative unit.

With respect to (4.8b), Criterion 1 distinguishes the ﬁrst arc from the second, since the former involves volitional causation whereas the latter (and all the other remaining CHAPTER 4. INSTRUMENTAL SVCS 172

arcs) involve physical causation. Volitional and physical causation are two of the four causation types Croft distinguishes for the application of Criterion 1, following Talmy (1972, 1976):

Physical causation: physical object acting on physical object; Volitional causation: volitional entity acting on physical object; Aﬀective causation: physical object “acting on” entity with mental states; Inducive causation: volitional entity acting on entity with mental states.

Whereas the second arc of (4.8b) involves physical causation because it involves “the interaction of two nonvolitional, nonsentient entities such that one affects the other” (Croft 1991, 167) (i.e. the gun acts on the bottle), the first arc involves volitional causation because it “involves an initiator possessing and exercising his mental capacity (through planning, intending, etc.) acting on a physical object” (Croft 1991, 167). That the first part of this instrumental event from John to the gun involves volitional causation, and nothing more, is crucial to the distinction between Possible Verb and Impossible Verb semantics in instrumental SVCs. I will return to this point shortly. While Criterion 1 of (4.9) distinguishes the first arc of (4.8b) from the second, Criterion 2 distinguishes the third arc from the fourth, since the former corresponds to a process and the latter to a state. Criterion 3 finally distinguishes the second and third arcs in this causal chain, since these align different pairs of individuals in the direction of “transmission of force”. Under the second arc force is transmitted from the gun to the bottle, while under the third arc there is a shift in the transmission of force in that the bottle changes state on its own. This is how the criteria listed in (4.9) can be used to decompose a causal chain into atomic events in the particular case of (4.8b), which happens to correspond to an instrumental event. We could also generalize away from (4.8b) to characterize an instrumental event as follows: at the beginning of the chain is a volitional actor (or initiator, if not sentient), in an atomic event of volitional causation the will of this actor results in a transmission of force (via the actor’s body) onto the instrument, CHAPTER 4. INSTRUMENTAL SVCS 173

the instrument then transmits force onto the next participant, and any transmissions of force that correspond to the changes in state of this participant, possibly culminating in a result state, complete the chain. This description is more specific about what happens at the beginning of the causal chain than at the end, but then it is what does or does not happen at the beginning of the chain that impacts upon the Possible Verb-Impossible Verb distinction in instrumental SVCs. The feature of (4.8b) (or any other causal chain representing an instrumental event) which must be stressed, and which plays directly into the Possible Verb-Impossible Verb distinction in instrumental SVCs, is that VOL in the first arc does not entail that the volitional initiator TAKES the instrument. VOL only signifies that the will of the actor results in some type of transmission of force onto the instrument; this could mean that the actor takes the instrument before acting upon it (and in fact in most situations it is very likely that the actor will take the instrument), but it certainly does not have to mean this. The causal chain in (4.8b) (and indeed the sentence in (4.8a)) is consistent with the actor taking the instrument, in other words, but it certainly does not entail this. John could, for example, be spending every hour of the day and night clutching the gun in a state of psychosis, before shooting the bottle in (4.8a); in that case no initial TAKING event would be involved for this sentence. The following sentence, in contrast to (4.8a), does entail that the actor takes the instrument:

(4.10) John took a gun and shot the bottle to pieces (with it).

Took in (4.10) entails that John moves the gun from one location to another, and it may also entail (depending on the semantics of take, for which see Section 4.3.3) that John does not have the gun to begin with. These semantic properties will be linked to Impossible Verb instrumental SVCs in Section 4.4. In contrast, Possible Verb instrumental SVCs, it will be argued in Section 4.3, entail nothing about the movement or prior possession (or non-possession) of the instrument. One further aspect of Croft (1991) plays directly into instrumental events and how we can expect them to be realized in serial verb form. This is that Croft (1991) draws a distinction between antecedent and subsequent roles based on whether the role appears before or after the object in the causal chain. Benefactive, recipient CHAPTER 4. INSTRUMENTAL SVCS 174

and result are all types of subsequent role, while manner, means, comitative, passive agent, cause and instrument are all types of antecedent role. As antecedent roles, instruments could then be expected to linearly precede patient/theme objects (e.g. the bottle in (4.8b)) in the syntax of the SVC, reflecting their prior position in the causal chain. The data set out in Section 4.2.2 will reveal whether instrumental SVCs are iconic in this way. While Croft (1991) provides some of the most explicit profiles of particular instrumental events, Jackendoff (1990) provides the most explicit general profile of an instrumental event, by focusing on what an instrument is, and the part it plays in an event:

(4.11) “In general, the characteristics of an Instrument are: (1) it plays a role in the means by which the Actor accomplishes the action (notice that Instrumental with can often be paraphrased with by means of NP); (2) the Actor acts on the Instrument; (3) the Instrument acts on the Patient. More subtly, it is the fact that the Actor acts on the Instrument that results in the Instrument acting on the Patient. However, it is left open to pragmatics to determine exactly how the Actor uses the Instrument and exactly what the Instrument does to the Patient.” Jackendoﬀ (1990, 142)

The first part of this quote brings out a point I will emphasize in the event structural analysis of instrumental events in Section 4.3 below, that instrument phrases provide what is essentially means or manner information about the activity described. Parts (2) and (3) form the core of Jackendoff’s generalization about instrumental events (as well as being the basis for his semantic representation of instrumental events): the Actor acts on the Instrument and the Instrument acts on the Patient (or ‘object’). But if that is the case, then it seems that the following two sentences containing (apparent) instrument phrases would fall outside Jackendoff’s generalization:

(4.12) a. Olly watched the birds with binoculars. b. The doctor heard Rosie’s heart with the stethoscope. CHAPTER 4. INSTRUMENTAL SVCS 175

While Olly clearly acts on the binoculars in (4.12a) and the doctor on the stethoscope in (4.12b), it is less clear whether the binoculars could be said to ‘act on’ the birds in (4.12a), or the stethoscope on Rosie’s heart in (4.12b). However, the last part of Jackendoff’s quote in (4.11) could offer some kind of resolution to these issues; in stating that “it is left open to pragmatics to determine ... exactly what the Instru- ment does to the Patient”, Jackendoff could be offering more leeway for the nature of the relationship between the respective instruments and objects (patients). The following examples from Koenig et al. (2007) may stretch this leeway even further, for while it is easy enough to pragmatically construe what the binoculars in (4.12a) or the stethoscope in (4.12b) actually do to their respective objects (the binoculars presumably magnify the birds, while the stethoscope amplifies Rosie’s heart), the kinds of relationships between the respective instruments and objects in (4.13a)-(4.13d) are much more difficult to construe or characterize:

(4.13) a. Susan always practices the piano with a metronome. (Koenig et al. 2007, 199) b. He plays volleyball with gloves. (Koenig et al. 2007, 199) c. Joan hunted the turkey with a bow and arrows. (Koenig et al. 2007, 198) d. The juggler performed a trick with ten balls. (Schlesinger 1979, 322)

In (4.13a) the metronome does something like regulating the tempo of the output of the piano (although if Susan’s sense of rhythm is found wanting it may not even do that). In (4.13b) the gloves feature (minimally) in a game of volleyball whenever the actor’s hands come into contact with the ball. Joan probably shot at (or at least positioned herself to shoot at) the turkey with a bow and arrows in (4.13c). Lastly, although ten balls must be involved in the trick the juggler performs in (4.13d), we know nothing about the nature of this involvement (the balls might not even be juggled). Jackendoﬀ’s characterization of an instrumental event as an Actor acting on an Instrument and an Instrument acting on a Patient may stretch to characterize (4.13a)-(4.13d) taking into account his position that “it is left open to pragmatics to determine ... exactly what the Instrument does to the Patient”, although it is more CHAPTER 4. INSTRUMENTAL SVCS 176

difficult to see how (4.13a)-(4.13d) could be effectively modeled through transmissions of force in one of Croft’s causal chains. But even Jackendoff’s analysis will not stretch to intransitive-based sentences with instrument phrases such as the following, since the existence of a Patient upon which the Instrument acts is central to his characterization:

(4.14) a. Dennis walked with a cane. b. Maddie breathed with a ventilator. c. Timmy ﬂoated with water wings.

Should (4.14a)-(4.14c) be judged to contain true instrument phrases, then Jackend- oﬀ’s analysis will be too narrow to accommodate them. This underscores the importance of establishing a reliable and theory-independent means of deciding what an instrument phrase (or an instrumental event) is, in connection to which I now turn to Koenig et al. (2007). Koenig at al. (2007) present an analysis of instrumental events which encompasses a much broader range of data than any other study of instrumental events (as their examples (4.13a)-(4.13c) would suggest). But before considering their analysis the most pressing matter to address is: on what basis do they determine that a phrase is an instrument phrase? On what basis does this chapter identify instruments? The broad aim of Koenig et al. (2007) is to “elucidate the semantic space of verbs that describe situations that may or must include instruments” (179), and in order to identify which verbs describe situations that may or must include instruments, they employ linguistically trained raters to answer the following questions of 4200 verbs:

(4.15) a. Does one of the verb’s sense describe situations in which one participant must or can use another participant to perform an action (e.g. does Marc cut the chicken require or allow Marc to have used something)? b. Does one of the verb’s senses describe situations in which one participant must or can perform the action described by the verb with something (e.g. does Marc cut the chicken require or allow Marc to cut it with something?) (Koenig et al. 2007, 179) CHAPTER 4. INSTRUMENTAL SVCS 177

Lakoﬀ (1968) observes that sentences of the form X VERB-ed Y with Z can have semantically identical paraphrases of the form X used Z to VERB Y. More speciﬁcally, it is only instrumental uses of with that can be paraphrased in this way, as in (4.16); that Barbara in (4.17) cannot be expressed as the object of use can be attributed to (4.17a) containing a comitative, rather than an instrumental, use of with:

(4.16) a. Sarah cut the bread with the knife. b. Sarah used the knife to cut the bread.

(4.17) a. Sarah cut the bread with Barbara. b. ?Sarah used Barbara to cut the bread.

Because instruments can always be expressed in with-phrases in English, and because the ‘use test’ can be used to reliably exclude comitative arguments, I will use these with and use tests as the strongest indicators of instrument status.2 For the purpose of determining whether an SVC expresses an event containing a true instrument participant in a given language, tests for instrument status specific to that language would ideally be used. One indication of why this is desirable is Schlesinger’s (1979) argument for the existence of a continuum from instrument to comitative status, at different points along which languages cut off one of these roles from the other. However, in the absence of language-specific tests for instrument status, I will instead use the English with and use tests as a kind of guide for instrument status. But before illustrating the ways in which instrumental events are expressed in serial verb form in Section 4.2.2, I will first summarize some of the main points of Koenig et al. (2007), since this paper presents some significant advances in the semantic analysis of instrumental events. Koenig et al. (2007, 176) characterize an event involving an instrument as follows: “an agent of some sort acts on the instrument and the result of this action directly or indirectly causes a change of state in a patient or theme”. Koenig et al. (2007) claim

2This is not strictly the approach that Koenig et al. (2007) take. They specify that they are using these morphosyntactic tests as “proxies for a semantically based selection of verbs [which must or can involve instruments]” (180) (emphasis mine). CHAPTER 4. INSTRUMENTAL SVCS 178

to be simply following previous work on instrumental events (such as Croft 1991 and Talmy 2000) in this characterization, which is accompanied by the following formula:

(4.18) cause(s1, s2) & Pred(s1, A, I) & Pred(s2, I, P) & cause(s2, s3) & Pred(s3, P)

One of the main contributions of Koenig et al. (2007), as they see it, is the demonstration that at least some verbs which semantically require instrument arguments involve two causal relations in a sequence of three subsituations (as in (4.18)), breaching the bounds of a possible verb meaning in a semantic framework such as Rappaport Hovav and Levin (1998). Rappaport Hovav and Levin (1998) claim that verb meanings can include at most one causal relation. In the description of the event structural profile for instrumental events in Section 4.3.2 I will argue against decomposing these verb meanings into two causal relations. The main contribution of Koenig et al. (2007), as I see it, is the breadth of their survey, combined with their attention to the relationships that hold between different pairings of participants in the meanings of verbs that require or allow instruments. One of the key observations Koenig et al. make about the meanings of these verbs is that in them the nature of the relationship between the Agent and the Instrument is less strongly determined and features less prominently than the relationship between the Instrument and the Patient. In (4.19), for example, it is assumed that Bill ‘acts on’ the knife in some manner, presumably applying force, yet it is the action of the knife cutting the bread that is more specific and integral to the meaning of cut:

(4.19) Bill cut the bread with the knife.

The nature of the relationship between the Instrument and the Patient (and not the Agent and Instrument) also plays a primary part in the organization of the semantic space (or “micro-geography”) of verbs which require or allow instrument participants, according to Koenig et al. (2007). In their attempt to represent this space, Koenig et al. (2007, 213) distinguish among ﬁve main types of relationship between Instrument and Patient in instrumental events, ranking them according to their causal force within the event. I will not include each type here, but will instead simply state that CHAPTER 4. INSTRUMENTAL SVCS 179

they range from cases where the instrument’s action upon the patient is perceived as a ‘true cause’, as is the case for the knife’s cutting of the bread within the event described in (4.19), to cases where the instrument’s action upon the patient plays a much weaker causal role of ‘helping’. ‘Helping’ is characterized as “caus[ing] the event/action to lead to a better resulting state or to be performed better” (Koenig et al. 2007, 213). (4.20) presents a standard example of ‘helping’; the manner in which the spoon manipulated by Bill traps or contains the noodles – hardly a ‘true cause’ for the eating – does ‘help’ in the accomplishment of this activity:

(4.20) Bill ate the noodles with a spoon.

(4.13a)-(4.13b) above, repeated here, also present cases of ‘helping’, underscoring the point that it is not even necessary to determine the speciﬁcs of what an Instrument does to a Patient in order for it to qualify as ‘helping’:

(4.21) a. Susan always practices the piano with a metronome. (Koenig et al. 2007, 199) b. He plays volleyball with gloves. (Koenig et al. 2007, 199)

Besides distinguishing among five causally ranked types of instrumental event, the relationship between the Instrument and the Patient also figures prominently in Koenig et al.’s (2007) more fine-grained distinction into 30 types of verb meaning (18 for which the instrument is obligatory, 12 for which the instrument is optional). (4.22), for example, is the semantic formula for verbs of the ‘Cut class’ (amputate, bone, cut, dissect, guillotine, hack, incise, knife, core, castrate, gore, cleave, sever, eviscerate), where the latter part of this formula denoting the relationship between the Instrument and the Patient (contact(s2, I, P) & cause(s2, s3) & incised(s3, P)) determines membership in this class:

(4.22) cause(s1, s2) & act(s1, A, I) & contact(s2, I, P) & cause(s2, s3) &

incised(s3, P)

These types of representations will be revisited in Section 4.3.2, but for now two summary points concerning Koenig et al.’s analysis are that: (i) they do not provide CHAPTER 4. INSTRUMENTAL SVCS 180

representations for individual verbs, and (ii) their data survey seems to be devoid of intransitive-based instrumental events. With respect to the last point, I have already stated that the means of identifying instrument participants that I have opted to use in this chapter is the ability for the participant to appear both within a with- phrase and as the object of use in a X used Z to ... clause. But in that case, the following instransitive-based sentences (= (4.14a)-(4.14c)) must all be said to contain instrument arguments:

(4.23) a. Dennis walked with a cane. Dennis used the cane to walk. b. Maddie breathed with a ventilator. Maddie used the ventilator to breathe. c. Timmy ﬂoated with water wings. Timmy used the water wings to ﬂoat.

It is unclear why Koenig et al. (2007) exclude intransitive verb meanings which require or allow an instrument participant, given that the with and use tests are also their means of identifying instruments. Perhaps this omission is influenced by their prior prescription that in an instrumental event “an agent of some sort acts on the instrument and the result of this action directly or indirectly causes a change of state in a patient or theme” (Koenig et al. 2007, 176). Koenig et al. (2007), and the event structural analysis it presents, will be discussed further in Section 4.3.2. This section has so far explored (a) what happens in a typical instrumental event, as characterized under Croft’s (1991) and Jackendoff’s (1990) analyses, (b) the kinds of things that can happen in less typical instrumental events, such as the ‘helping’ contribution of the instrument that Koenig et al. (2007) describe, and (c) how to identify a participant as an instrument in the first place. With respect to (c), the with and use tests described above were elected as the most reliable means of establishing instrument status in English, and this English-specific method of identifying instruments can serve as a guide for identifying instrument participants in other languages. Now it is time to see how situations involving instrument participants are expressed in other languages, and specifically in serial verb CHAPTER 4. INSTRUMENTAL SVCS 181

form.

4.2.2 Instrumental SVCs

SVCs which express situations that involve instrument participants consist of (at least) two verbs, one of which can be termed the ‘handling’ verb. The handling verb takes the instrument as its object. Based on the limited data I have been able to collect, the handling verb is (indicated in the interlinear gloss to be) one of the three following verbs: ‘use’, ‘hold’ or ‘take’. (4.24)-(4.28) are some examples of instrumental SVCs whose handling verb is translated as ‘use’ in the interlinear gloss:

(4.24) Isò` . kè. n yá ábé. fián èmió!wó. Isoken use knife cut meat ‘Isoken used the knife to cut the meat.’ (Edo; Stewart 2001, 3)

(4.25) Olú fi ´ıs.ó gún o. mo. náà. Olu use nail pierce boy the ‘Olu pierced the boy with a nail.’ (Yoruba; Bamgbos.e 1974, 46)

(4.26) Mo ﬁ `ıbo. n pa e.k`un. 1sg use gun kill leopard ‘I killed the leopard with a gun.’ (Yoruba; Carstens 2002, 11)

(4.27) Ta yòng zhèbâ yaòshi kai-de mén. 3sg use key this open-de door ‘He opened the door with his key.’ (Mandarin; Déchaine 1993, 253)

(4.28) Zhang-san yòng kuàizi chi-fàn. Zhang-san use chopsticks eat ‘Zhang-san eats with chopsticks.’ (Mandarin; Li and Thompson 1973, 96)

‘Use’ and its instrument complement almost always precede the other verb in an instrumental ‘use’ SVC, reﬂecting the fact that the instrument is one of Croft’s ‘antecedent’ roles, occupying a prior position in the causal chain. The only exceptions I have encountered are the following: CHAPTER 4. INSTRUMENTAL SVCS 182

(4.29) Npun bobay npake sandal. 3.sg-kill mosquito 3.sg-use thong ‘He killed the mosquito with a thong.’ (Taba; Bowden 2001, 312)

(4.30) Ku-taku uhu w`angu huru. 1sg:n-scoop rice use spoon ‘Iscoopricewithaspoon.’ (Kambera;Klamer1998,287)

While all of the examples included so far contain participants with credible claims to instrument status (as the acceptability of semantically equivalent English with- INSTR and use-INSTR-to expressions would suggest), the SVCs in (4.31) and (4.33), which have the same X ‘use’ Z VERB (Y) structure as (4.24)-(4.28), are not felicitously expressed in the semantically equivalent English with-INSTR and use-INSTR- to clauses in (4.32) and (4.34) (provided by me):

(4.31) Ajé fi àga jókòó. Aje use chair sit-down ‘Aje sat on a chair.’ (Yoruba; Awóyalé1988, 17)

(4.32) a. ?Aje sat down with a chair b. ?Aje used a chair to sit down

(4.33) J´ımò. ó. fi o. mo. rè. rán is.é.. Jimo agr use son 3s.gen send job ‘Jimo sent a message via his son.’ (Yoruba; Abraham 1958, 576)

(4.34) a. ?Jimo sent a message with his son. (Good under an accompanying, but not an instrumental, reading) b. ?Jimo used his son to send a message.

While (4.32a)-(4.32b) and (4.34a)-(4.34b) indicate that a chair sat on and a son via whom a message is sent fall outside an English-specific category of instrument, Yoruba-specific tests may rule these participants within a Yoruba-specific category of instrument. These examples thus push the boundaries of what an instrument can be CHAPTER 4. INSTRUMENTAL SVCS 183

in a particular language, as well as emphasizing that the English with and use tests can only be used as a guide to instrument status across languages. The second member of the handling verb class identiﬁed above is ‘hold’. So far, I have only encountered four instrumental SVCs where the handling verb is indicated to be ‘hold’, including (4.38), for which the translation of na as ‘hold’ was actually suggested in Section 4.1 to be incorrect:

´ (4.35) O j`ı-ri ´mmà bh. á j´ı. 3s hold knife peel-A yam.gen ‘S/he peeled yam[s] with [a] knife.’ (Igbo; Déchaine 1993, 241)

(4.36) Og`u j`ır`ı mm`a gbuo agwo. . Ogo hold-rV(past) knife kill-A snake ‘Ogu killed a snake with a knife.’ (Igbo; Uwalaka 1988, 28)

(4.37) Na.ambi wosa na.wosa waga. 1sg:irreal.hold oar 1sg:irreal.paddle canoe ‘I will paddle the canoe with the oar.’ (Numbami; Bradshaw 1993, 148)

(4.38) Ta kuaikuaide na dao qie-le rou. he quickly hold knife cut-perf meat (a) ‘He quickly cut the meat with the knife’; (b) ‘He cut the meat with the knife which he quickly took.’ (Mandarin; Law 1996, 213)

Because of the small number of examples collected, instrumental ‘hold’ SVCs will not be considered any further in this chapter. The signiﬁcance of (4.38) will, however, be discussed further in Section 4.4. Finally, (4.39)-(4.44) are some examples of instrumental ‘take’ SVCs:

´ (4.39) Aràú. zuye àk`ı. buru teri-m´ı. she basket take yam cover-simple-past ‘She used a basket to cover a yam.’ (Ijo; Williamson 1965, 53)

` (4.40) Ozó rhié ùghànmwàn ghuò. ghó. ówá. Ozo take axe break stall ‘Ozo broke the stall with an axe.’ (Edo; Stewart 2001, 207) CHAPTER 4. INSTRUMENTAL SVCS 184

(4.41) No teki baskita tyari watra. neg take basket carry water ‘Don’t carry water with a basket.’ (Sranan; Sebba 1987, 59)

(4.42) Iywi awa otsi iku utsi. child take stick shut door

‘The child shut the door with a stick.’ (Yatye; Stahlke 1970, 74)

(4.43) Kkú s àt´ın xò As´ıbá.` Koku take stick hit Asiba

‘Koku hit Asiba with a stick.’ (Fongbe; Lefebvre 1991, 61)

(4.44) Kk´u s j`ıv´ı s n l`an. Koku take knife cut meat ‘Koku cut meat with a knife.’ (Fongbe; Lefebvre and Brousseau 2002, 415)

As was the case for ‘use’ SVCs, some ‘take’ SVCs, such as (4.45) and (4.47), involve participants that are not felicitously instantiated in both of the equivalent English with-INSTR and use-INSTR-to clauses (provided by me in (4.46) and (4.48), where again I stress that these English expressions are only meant to serve as a guide for the likelihood of instrument status):

′ (4.45) ∅ ˇaw khón pˇaj khút-h¯et kh ng.mˇuang. take people go dig-do/make canal ‘They got the people to dig the canals.’ (Lao; Enfield 2002, 19)

(4.46) a. ?They dug the canals with people. b. (?)They used people to dig the canals.

(4.47) Kofi teki en sisa go bay krosi. Kofi take his sister go buy clothes ‘Kofi bought clothes with the help of his sister.’ (Sranan; Jansen, Koopman and Muysken 1978, 130)

(4.48) a. Koﬁ bought clothes with his sister. (Good only under a comitative reading.) b. ?Koﬁ used his sister to buy clothes. CHAPTER 4. INSTRUMENTAL SVCS 185

(4.46) and (4.48) suggest that instrument status in Lao and Sranan may extend beyond what we ﬁnd in English, although one key observation is that something (4.45) and (4.47) have in common is that their (putative) instrument arguments are human. English allows human instruments more reluctantly; (4.49a) and (4.50a) are two acceptable instantiations of human instruments (as the acceptable ‘use’ paraphrases in (4.49b) and (4.50b) would suggest):

(4.49) a. The prisoner won the appeal with a highly paid lawyer. (Schlesinger 1979, 310) b. The prisoner used a highly paid lawyer to win the appeal.

(4.50) a. Miriam bought Christmas presents with a personal shopper. b. Miriam used a personal shopper to buy Christmas presents.

It should be stressed the objects of ‘take’ in (4.45) and (4.47) are only tentatively classed as instruments based on the fact that these SVCs are structurally identical to instrumental SVCs in the languages concerned. Khón ‘the people’ in (4.45) is actually analyzed as a causee in a causative construction by Enfield (2002), while en sisa ‘his sister’ in (4.47) is perhaps more straightforwardly interpreted as a comitative argument. But it is interesting that the translation for (4.47) refers to the helping function of Kofi’s sister, echoing the ‘helping’ function that Koenig et al. (2007) attribute to many types of instrument. Kofi’s sister (whether she knows it or not) may sit somewhere uncomfortably between a comitative and an instrument argument. The final feature of instrumental ‘take’ SVCs I will draw attention to is the fact that ‘take’ and its instrument complement always linearly precede the other verb. This was true also of the instrumental ‘hold’ SVCs; the instrumental ‘use’ SVCs in (4.29)-(4.30) above are the only cases I have encountered where the handling verb follows the other verb. All of the instrumental SVCs included in this section have non-sequential translations suggestive of Possible Verb semantics. This is true of the instrumental ‘take’ SVCs (4.39)-(4.44), which were recognized in Section 4.1 to at least have the potential CHAPTER 4. INSTRUMENTAL SVCS 186

for sequential, Impossible Verb semantics, in the case that ‘take’ means TAKE. Sec- tion 4.4 will investigate whether there really are any Impossible Verb instrumental ‘take’ SVCs, but ﬁrst Possible Verb instrumental SVCs are considered, along with their proposed event structural proﬁle, which may seem to be at odds with some of the analyses of instrumental events discussed in Section 4.2.1. The claim that these SVCs have unitary-event structure, in which USE does not correspond to a subevent, may appear particularly surprising from the perspective of Koenig et al.’s (2007) event structural decomposition of instrumental events. Section 4.3 now turns to these issues.

4.3 Possible Verb instrumental SVCs

4.3.1 Introduction

Everything that follows in this section is built around the claim that Possible Verb instrumental SVCs (whatever the semantics of their handling verb is indicated to be) have the following unitary-event proﬁle (=(4.6)):

(4.51) Event structural proﬁle for Possible Verb instrumental SVCs (a type of unitary-event SVC)

SF: ... λz ... λx λs {V1(x,...) & USE(x,z)}(s)

CS: V1 corresponds to an event s1. USE does not correspond to a (sub)event.

Instrumental SVCs that fit this profile fulfill the Theta Structure Property for SVCs (trivially, as all Possible Verb SVCs do) because their Theta Structures consist of either one or two structural arguments, plus a non-structural, configurationally-defined instrument argument, and there do exist clauses headed by single verbs whose Theta Structures also consist of one or two structural arguments and one non-structural, instrument argument. Fulfillment of the Theta Structure Property is, in a sense, a less interesting feature of (4.51); more interesting (or more in need of an explanation and/or motivation) are the predicates that feature in this profile, and the event structural claims attached to them. CHAPTER 4. INSTRUMENTAL SVCS 187

Where the predicates of (4.51) are concerned, it is USE that plays the most important part in this profile. USE is at the heart of instrumental status; the instrument role itself was defined configurationally in Chapter 2 as the second argument of (non- initial) USE, or z in {V(x,...) & USE(x,z)}(s). USE(x,z) does not signify that x moves z or that x possesses z; all it signifies is mere utilization of z for some purpose, where this purpose is specified in the initial, ‘main’ predicate in the Semantic Form representation. This is to say that USE is semantically dependent upon another predicate, a quality that will be emphasized in the analysis of instrumental ‘use’ SVCs in Section 4.3.2. For Possible Verb instrumental ‘take’ SVCs, a serialization lexical rule is required to alter the semantics of ‘take’ to the USE we see in (4.51). This rule will be included in the discussion of this type of instrumental SVC in Section 4.3.3. Section 4.3.3 also evaluates the grounds for claiming that ‘take’ has its semantics altered to USE in these SVCs, or what grounds there are for calling these SVCs Possible Verb SVCs (similar comments and analysis pertain also to the instrumental ‘hold’ SVCs with Possible Verb semantics not considered further in this chapter). In contrast, no serialization lexical rule is required for Possible Verb instrumental ‘use’ SVCs, since ‘use’ corresponds straightforwardly to the predicate USE. Section 4.3.2, on instrumental ‘use’ SVCs, nonetheless constitutes the bulk of this section on Possible Verb instrumental SVCs, because it is given over to the mechanically rather weighty matter of arguing for the event semantic profile for Possible Verb instrumental SVCs set out in (4.51). It is to these arguments that I now turn.

4.3.2 Instrumental ‘use’ SVCs

‘Use’ corresponds to the Semantic Form predicate USE in an instrumental ‘use’ SVC, and this section presents arguments for why the Possible Verb structure in (4.51) is the only event structural proﬁle applicable to these SVCs. Section 4.3.2.1 argues that instrumental ‘use’ SVCs cannot be Impossible Verb SVCs, in which USE is situated

in a distinct Possible Verb component from the predicate (V1) that represents the ‘main’ action. Section 4.3.2.2 argues that USE cannot correspond to a subevent, CHAPTER 4. INSTRUMENTAL SVCS 188

and that hence these structures must be unitary-event SVCs. Finally Section 4.3.2.3 explains why, under the type of event structural approach I am using, instrumental events are not composed of three subevents, as Koenig et al. (2007) claim. If, as I claim, all instrumental ‘use’ SVCs fit the profile in (4.51), then all of these SVCs are composed of single, unitary events, and it is this event structural status which moreover confirms the Possible Verb status of these SVCs, independent of having to apply the Possible Verbs constraints.

4.3.2.1 Instrumental ‘use’ SVCs are always Possible Verb SVCs and never Impossible Verb SVCs

Instrumental ‘use’ SVCs, such as (4.52a)-(4.54a) (=(4.25)-(4.27)), are always, I claim, Possible Verb SVCs that ﬁt the semantic proﬁle in (4.51) above:

(4.52) a. Olú fi ´ıs.ó gún o. mo. náà. Olu use nail pierce boy the ‘Olu pierced the boy with a nail.’ (Yoruba; Bamgbos.e 1974, 46) b. λz λy λx λs {ACT(x) & USE(x,z) & BECOME(PIERCED(y))}(s)

(4.53) a. Mo ﬁ `ıbo. n pa e. k`un. 1sg use gun kill leopard ‘I killed the leopard with a gun.’ (Yoruba; Carstens 2002, 11) b. λz λy λx λs {ACT(x) & USE(x,z) & BECOME(DEAD(y))}(s)

(4.54) a. Ta yòng zhèbâ yaòshi kai-de mén. 3sg use key this open-de door ‘He opened the door with his key.’ (Mandarin; Déchaine 1993, 253) b. λz λy λx λs {ACT(x) & USE(x,z) & BECOME(OPEN(y))}(s)

What I am claiming in particular is that instrumental ‘use’ SVCs do not have either of the following coextensive (4.55) or sequential (4.56) Impossible Verb SVC proﬁles, with two separate Possible Verb components:

(4.55) SF: ... λz ... λx λs [{V1(x,...)} & {USE(x,z)}](s) CHAPTER 4. INSTRUMENTAL SVCS 189

(or: SF: ... λz ... λx λs [{USE(x,z)} & {V1(x,...)}](s), where the conjuncts corresponding to contemporaneous subevents are simply reversed.)

CS: V1 corresponds to a subevent s1 and USE to a subevent s2. tc1=temporaltc2

and tt1=temporaltt2.

(4.56) SF: ... λz ... λx λs [{USE(x,z)} & {V2(x,...)}](s)

CS: USE corresponds to a subevent s1 and V2 to a subevent s2. tc1

tc2.

This is because in (4.55) and (4.56) the two actions (the ‘main’ action and the action of using the instrument) are left unrelated. (4.55), for example, may be the appropriate template for a sentence such as Johanna built an igloo and used a gun (concurrently), the oddness of which can be attributed to the absence of any information concerning exactly what Johanna may be using a gun for. But Possible Verb instrumental SVCs are never (I claim) of this odd ﬂavor, and instead ﬁt the template in (4.51). Having

USE(x,z) occur after V1, within the same Possible Verb bracketing as V1 in (4.51) signiﬁes that USE(x,z) is further specifying the interpretation space opened up by

V1 – it signifies that using the instrument has to be related to the action described in V1, in other words. In (4.54b), for example, USE(x,z) signifies that the actor uses z (the key), and this further specifies the interpretation space opened up by ACT(x), the result of which is specified in BECOME(OPEN(y)). USE(x,z) supplies more information about the means through which the actor does something that ultimately causes the door to be open. The ability for USE to further specify an initial predicate will be contrasted with the inability for TAKE to further specify an initial predicate in Section 4.4, for whereas USE supplies only a kinds of means information that focuses upon the instrument over which the actor exerts control for the purpose specified in the initial predicate, TAKE (even as it pertains to the taking of an instrument) describes a distinct action outside the interpretation space of the initial predicate. In essence, what an English with+INSTR combination, or a serial verb ‘handling verb’+INSTR combination, or (in semantic representation terms) what a USE(x,z) component is doing in an instrumental event is contributing some additional (and CHAPTER 4. INSTRUMENTAL SVCS 190

grammatically optional) means or manner information about the way in which the action described by the main verb, V1, is performed. This much is recognized by Jackendoff (1990) in his characterization of instrumental events as “play[ing] a role in the means by which the Actor accomplishes the action” (1990, 142), although means or manner characterizations feature less strongly in Croft’s (1991) or Koenig et al.’s (2007) analyses of instrumental events. The ramifications of Koenig et al. (2007) not attributing what is essentially a modifying function to a constituent containing an instrument will in Section 4.3.2.3 be brought to bear upon the arguments in support of my event structural analysis for instrumental SVCs. The approach I take identifies subevents through their temporal coordinates, in contrast to Koenig et al. (2007), where subevents are derived from the relationships that hold between participants in a causal chain. It is to my event structural analysis that Section 4.3.2.2 now turns, by first highlighting the most salient event structural feature of the Possible Verb instrumental SVC profile in (4.51), which is that USE does not correspond to a separate subevent.

4.3.2.2 USE does not correspond to a subevent in an instrumental ‘use’ SVC

USE cannot correspond to a subevent under my event structural analysis because USE cannot correspond to a well-formed event on its own. It cannot appear independently in a Semantic Form representation such as (4.57a), and any (attempted) realizations of this Semantic Form, such as (4.57b), will be unacceptable:

(4.57) a. *λz λx λs {USE(x,z)}(s) b. ?Caroline used a stick.

The natural response to (4.57b), assuming it is uttered independently, separate from any other discourse, would be: (Caroline used a stick) to do what? This request for the speciﬁcation of a particular action reﬂects the fact that USE has to pertain to, or further specify, some other predicate – a predicate that does correspond to an independent subevent. Thus (4.58a), with some possible English realizations in CHAPTER 4. INSTRUMENTAL SVCS 191

(4.58b), is an acceptable Semantic Form, since here USE further speciﬁes the action of beating:

(4.58) a. λz λy λx λs {BEAT(x,y) & USE(x,z)}(s) b. Caroline used a stick to beat the rug. Caroline beat the rug with a stick.

The complicating factor attached to this argument that USE cannot correspond to a well-formed event on its own is that USE can be realized as use, which is, of course, a full English verb with all the syntactic qualiﬁcations to head a sentence on its own, and which in certain contexts may even do so:

(4.59) a. Tiger Woods uses Nike golf clubs. b. Caroline wiped the table. She used a cloth.

(4.59a) and the second sentence of (4.59b) would, on the face of it, appear to argue against (4.57a); they would appear to argue that USE can be the sole predicate in a Semantic Form representation, without being required to further specify some preceding predicate. But what I would argue is that, while use may be syntactically independent in these sentences, USE is still semantically dependent on a preceding predicate that it further speciﬁes. Use can only appear independently in (4.59a) because we know that golf clubs are quite obviously used to play golf, while use can only be used in an independent sentence in (4.59b) because we know from the previous sentence that the actor uses the cloth to wipe table. These sentences could be represented as (4.60a)-(4.60b), where the dotted lines signify that the initial predicates, which specify what kinds of events USE modiﬁes, are not realized syntactically:

(4.60) a. λz λx λs { PLAY-GOLF(x) & USE(x,z)}(s)

b. λz λy λx λs { WIPE(x,y) & USE(x,z)}(s)

Because only the initial predicate in the Semantic Form for an instrumental event corresponds to a subevent, instrumental events have unitary-event status. This CHAPTER 4. INSTRUMENTAL SVCS 192

subevent corresponding to the initial predicate has distinct temporal coordinates (times of commencement and termination between which the property denoted by the predicate holds continuously), but USE, in not corresponding to a subevent, has no such temporal coordinates. I assume that USE falls under Kaufmann and Wunderlich’s category (presumably intended to capture modiﬁcation in general) of “components that do not provide temporal information by themselves [but] have to be associated with the temporal structure of other components, in order to be temporally integrated into the event” (1998, 38). Under these assumptions, USE has to be associated with the temporal structure of the initial predicate in order to be temporally integrated into the event. However, for the sake of argument it is still worth asking: what if USE were to correspond to a well-formed subevent with its own temporal coordinates? What kind of (non-unitary-event) Possible Verb event structure would we have then? Consider (4.61):

(4.61) Jeremy paddled with a ping pong bat. λz λx λs {PADDLE(x) & USE(x,z)}(s)

If USE were to correspond to a subevent in (4.61), then it would have the same times of commencement and termination as the PADDLE subevent. In fact, so integral is the act of using the ping pong bat to the act of paddling in (4.61) that these two acts could even be said to be temporally dependent, in the sense that they occupy the same time frame, unfold at the same rate, and are inextricably linked. However, (4.61), were USE adjudged to correspond to a subevent, would not fit the profile of coidentification set out for coidentified Possible Verb SVCs in Section 3.2.3, for the reason that there is no scale attached to either subevent. Coidentified Possible Verb SVCs were specified in Section 3.2.3 to incorporate a (result) subevent attached to which is a scale that measures out the other (manner) subevent. This feature is absent from (4.61), but not from (4.62), through the addition of a path phrase into the pool, which does provide a scale: CHAPTER 4. INSTRUMENTAL SVCS 193

(4.62) Jeremy paddled into the pool with a ping pong bat. λz λx λs {PADDLE(x) & USE(x,z) & BECOME(LOC(x,IN(y)))}(s)

(4.62), then, has coidentified semantics where the path encoded by BECOME(LOC(x, IN(y))) measures out the paddling and the use of the ping pong bat coextensive with the paddling. While (4.62) has coidentified semantics, (4.63) has non-coidentified resultative semantics, since the scale attached to the result predicate down does not measure out the subevent of chopping in this sentence:

(4.63) Adam chopped the tree down with an axe. λz λy λx λs {CHOP(x,y) & USE(x,z) & BECOME(DOWN(y))}(s)

In the usual scenario (where the tree is reasonably large) the chopping subevent will extend for some period of time only at the termination of which will the tree begin its descent. Because the chopping and the falling need not even be coextensive, (4.63) cannot have coidentified semantics, but must have non-coidentified resultative semantics. Now were USE to correspond to a subevent in (4.63), this subevent would have the same temporal coordinates as the manner activity CHOP subevent, just as USE would have the same temporal coordinates as the manner activity PADDLE subevent in (4.62). The use of the axe terminates the moment the chopping terminates in (4.63), and together these are temporally distinct from, and prior to, the concluding result subevent of the tree falling. And, as a variation on (4.63), when no manner activity is specified, as in (4.64), the USE subevent still has the temporal coordinates of the manner activity part of the event (corresponding to non-specific ACT in (4.64)), temporally prior to the result subevent of the tree falling:

(4.64) Adam felled the tree with an axe. λz λy λx λs {ACT(x) & USE(x,z) & BECOME(DOWN(y))}(s)

In short, these examples point to the conclusion that, were USE to correspond to a subevent, it would have the same temporal coordinates as the manner activity CHAPTER 4. INSTRUMENTAL SVCS 194

subevent, or the temporal coordinates this subevent would have in the case that no manner activity is overtly speciﬁed. (4.65) and (4.66), however, present some of the rarer cases where the USE ‘subevent’ might not have the same temporal coordinates as the manner activity component of the event:

(4.65) Clayton killed the rat with the sledgehammer. λz λy λx λs {ACT(x) & USE(x,z) & BECOME(DEAD(y))}(s)

(4.66) Jim gutted the ﬁsh with the knife. λz λy λx λs {ACT(x) & USE(x,z) & BECOME(GUTTED(y))}(s)

(4.65) and (4.66) contain the change of state verbs kill and gut, which specify the result state of the event, but not the activity or process leading to this state. But in each case the ACT(x) predicate (which signifies that the actor has to have been involved in some kind of activity or process leading up to the result state) is further specified by USE(x,z). This makes the manner activity information slightly more specific by referring to an instrument used in the process concerned, but what is less clear is whether this USE ‘subevent’ should necessarily have the same temporal coordinates that the ACT subevent would have. In (4.65), for example, we do not know exactly how the actor goes about killing the rat, or what he does with the sledgehammer; we do not know if the use of the sledgehammer is continuous and unrelenting, contributing entirely to the death of the rat, or else singular or partial, as only part of the method of killing the rat. In the latter case, the USE ‘subevent’ would not have the same temporal coordinates the ACT subevent would have. The USE ‘subevent’ in (4.66), meanwhile, will almost certainly not have the same temporal coordinates the ACT subevent would have, because this USE ‘subevent’ would only be coextensive with the first half of the activity of gutting a fish, which involves making an incision, and not the second half, which usually involves removing the guts by hand. (4.65) and (4.66) suggest that the temporal coordinates for a USE ‘subevent’ cannot be equated with those of the manner activity component of the event across the board. The temporal coordinates for the USE subevent could either be indeterminate, CHAPTER 4. INSTRUMENTAL SVCS 195

as in (4.65), or else, to the extent that they are determined, what they are determined by is the complex internal event structure of the cooccurring predicate, which in (4.66) is gut. While there is a great deal of scope for the investigation of further predicates with complex internal structures and how USE ‘subevents’ play into them, what I will come back to now is my bottom-line argument that USE does not correspond to a subevent in instrumental events. My main argument for this is the semantic dependence of USE, demonstrated above, but (4.65) and (4.66) may also support this position in the sense that, if the temporal coordinates of USE ‘subevents’ are either indeterminate or wholly determined by the properties of another predicate, then this is tantamount to saying that USE ‘subevents’ have no independent, deﬁnite temporal coordinates of their own. This in itself is tantamount to saying that USE does not correspond to a subevent. With no independent temporal information of their own, USE components do have to be associated with the temporal structure of other components in order to be temporally integrated into the event, as Kaufmann and Wunderlich (1998) describe. USE components do, after all, only provide a kind of means or manner modiﬁcation.

4.3.2.3 Instrumental events are not composed of three subevents under the present event semantic approach

At this stage of this section I have provided arguments for why USE should not correspond to a subevent under the event semantic approach I am adopting, which identiﬁes subeventual components through their temporal coordinates. But it must be emphasized that this is not to claim that USE cannot correspond to subevents under event semantic approaches which are based upon diﬀerent principles, where temporal properties are not at the heart of subevent status. Koenig et al. (2007) was noted in the previous section to derive subevents from relationships between participants in causal chains. I would not wish to claim that there is anything wrong with this approach (on the contrary, it delivers some interesting results, as discussed in Section 4.2.1), or with their observation that an instrumental event can, in that case, consist of three subevents and two causal relations, as in the following formula CHAPTER 4. INSTRUMENTAL SVCS 196

(where ‘A’ refers to Agent, ‘I’ to Instrument, and ‘P’ to Patient):

(4.67) cause(s1, s2) & Pred(s1, A, I) & Pred(s2, I, P) & cause(s2, s3) & Pred(s3, P)

But it is less convincing for Koenig et al. (2007) to claim that their three-subevent analysis of instrumental events disproves Rappaport Hovav and Levin’s (1998) claim that a possible verb meaning can, in Rappaport Hovav and Levin’s (1998) event semantic framework, consist of at most two subevents and at most one causal relation. Rappaport Hovav and Levin’s (1998) claim about the event structural limits on a clause headed by a single verb is built up in a semantic framework in which ‘event complexity’ is the most salient foundational property. ‘Event complexity’ refers to the subeventual composition of an event or situation; if an event is complex, then it will consist of more than one subevent. Levin and Rappaport Hovav (2005,112) specify that a subeventual analysis indicates “(i) the number and type of constituent subevents; (ii) the number and identity of the arguments participating in the particular subevent; and (iii) the nature of the temporal relations between the subevents”. They identify Engelberg (1994, 2000), Grimshaw and Vikner (1993), McClure (1994), Pustejovsky (1995), van Hout (1996, 2000a, 2000b) and Wunderlich (1997a, 1997b, 2000) among some other work which uses this kind of subeventual analysis. This is also the kind of analysis I am using, and I can further emphasize why I want to use it, which is that this type of event complexity approach can formulate the event structural bounds on clauses headed by a single verbs, as Rappaport Hovav and Levin (1998) have done. The whole point of setting out the event structural profiles for Possible Verb SVCs in Chapter 3 is that these corroborate LDG’s Possible Verbs constraints; they provide an alternative means of identifying what a clause headed by a single verb can be, which in the terms set out in Chapter 3 means holding one of three Possible Verb SVC statuses: unitary-event, coidentified, or non-coidentified resultative status. Non-coidentified resultative SVCs are the event structurally ‘loosest’ type of SVC because here the verbs correspond to subevents that are in a causal relationship and which are not (usually or necessarily) temporally aligned. As such, non-coidentified resultative event structure represents the upper bound on clauses CHAPTER 4. INSTRUMENTAL SVCS 197

headed by a single verbs (equivalent to Rappaport Hovav and Levin’s (1998) upper bound). Possible Verb instrumental SVCs do not breach this bound because they are unitary-event SVCs where USE does not correspond to a subevent. So, this dissertation benefits from using an ‘event complexity’ approach to event structure since this can capture the event structural bounds of a clause headed by a single verb, and hence what a Possible Verb SVC is. Koenig et al.’s (2007) analysis, on the other hand, does not, or cannot, capture this bound. But Koenig et al. (2007) do argue against an event complexity-derived bound on possible verb meanings – specifically that proposed in Rappaport Hovav and Levin (1998). Because this dissertation does recognize an event complexity-derived bound on possible verb meanings, some more comments are in order as to why Koenig et al.’s (2007) arguments do not stand up. As mentioned above, Koenig et al.’s (2007) and Rappaport Hovav and Levin’s (1998) analyses present two different approaches to events. They are based on different conceptualizations and principles; whereas Rappaport Hovav and Levin (1998) take an ‘event complexity’ approach, one of the main focuses of which are the temporal relations between subevents, Koenig et al.’s (2007) causal approach derives subevents through relations between participants in a causal chain. The foundations upon which these analyses are built are different, and the materials out of which they are built are different. It is also significant that they have different aims. The aim constant throughout most of the work of Levin and Rappaport Hovav is to discover grammatically relevant aspects of lexical meaning. This lends itself to the methodological prominence of diathesis alternations, among other grammatical phenomena. Koenig et al. (2007), on the other hand, are unwilling to attach serious semantic significance to grammatical phenomena, only reluctantly consenting to identify instrument participants through the use and with tests, for example. If Koenig et al.’s (2007) aim is not to discover grammatically significant aspects of meaning, then what is their aim? It seems to be to discover something about meaning in isolation; they want to “map out the space of possible stem meanings: its informational boundaries and its organization, its geography, so to speak” (Koenig et al. 2007, 175). In practice this amounts to drawing divisions among verb meanings on the basis of what they judge CHAPTER 4. INSTRUMENTAL SVCS 198

to be semantically related, and as their semantic formulae will allow. (4.68)-(4.69) are some examples of their semantic formulae for the categories of verb indicated that require instruments:

(4.68) Cut verbs, including: amputate, bone, cut, dissect, guillotine, hack, incise, knife, core, castrate, gore, cleave, sever, eviscerate

cause(s1, s2) & act(s1, A, I) & contact(s2, I, P) & cause(s2, s3) &

incised(s3, P)

(4.69) Doodle verbs, including doodle, draw, ink, inscribe, dot, pencil, sketch, scribble, transcribe, write, underline, underscore, typewrite, type, print, sign

cause(s1, s2) & act(s1, A, I) & mark(s2, I, M, P) & cause(s2, s3) & 3 represent (s3, M, T)

These semantic formulae do not accord very well with my earlier characterization of English with+INSTR combinations, or serial verb ‘handling verb’+INSTR combinations, as constituents that contribute to the clause some additional and optional means or manner information that modiﬁes the ‘main’ action (i.e. the action described by the verb in an English sentence with a with+INSTR phrase, or by the other (‘non-handling’) verb in an instrumental SVC). For one thing, the fact that instruments are always grammatically optional (since no verb will ever require an instrumental with-phrase) is immaterial to Koenig et al. (2007), because their focus is on their concept of the semantic obligatoriness of instruments for particular verbs. Koenig et al. (2007) actually regard the participation of an instrument as part of the meaning of certain verbs, and so the only diﬀerence between a clause headed by an obligatory instrument verb that does not contain an instrumental with-phrase

3Verbs like doodle and scribble could be argued not to represent anything. Koenig et al. have this to say in response to this issue: “The notion of representation involved is intended to be quite broad and to cover the meaningless marks one leaves when one doodles or scribbles aimlessly. But even in these limiting cases, we would claim that the marks are representational in that they are deﬁned negatively as abnormally meaningless. To use Fillmore’s (1982) terminology, such uses of doodle or scribble are deﬁned relative to a frame in which marks represent something” (2007, p194 fn. 14). CHAPTER 4. INSTRUMENTAL SVCS 199

and a clause headed by an obligatory instrument verb that does contain an instrumental with-phrase is that the latter is more speciﬁc in actually stating what the instrument is. The Koenig et al.(2007)-style representations in (4.70a) and (4.71a) illustrate this point; because (4.70) includes the instrument phrase with the knife, (4.70a) can accordingly specify what the instrument is (knife), whereas, because the instrumental with-phrase is absent in (4.71), (4.71a) cannot specify what the instrument is, although this formula does still specify that an instrument of some sort (I) is involved:

(4.70) James cut the bread with the knife.

a. cause(s1, s2) & act(s1, james, knife) & contact(s2, knife, bread) &

cause(s2, s3) & cut(s3, bread) b. [ [ james ACT ] CAUSE [ BECOME [ bread ] ] ] c. SF: λz λy λx λs {CUT(james,bread) & USE(james,knife)}(s)

(4.71) James cut the bread.

a. cause(s1, s2) & act(s1, james, I) & contact(s2, I, bread) & cause(s2,

s3) & cut(s3, bread) b. [ [ james ACT ] CAUSE [ BECOME [ bread ] ] ] c. SF: λz λy λx λs {CUT(james,bread)}(s)

In contrast to the Koenig et al. (2007)-style representations for (4.70) and (4.71), which incorporate instrument participation regardless of whether an instrument is grammatically realized, the Levin and Rappaport Hovav (2005)-style representations in (4.70b) and (4.71b) do not represent instrument participation, even in correspondence to the clause containing an instrumental with-phrase in (4.70). This is because Levin and Rappaport Hovav’s event structure representations are designed to account for the realization of obligatory arguments, and not optional, non-core arguments like instruments. The LDG Semantic Form representations I am using in (4.70c) and (4.71c), on the other hand, include a USE(james, knife) component in correspondence CHAPTER 4. INSTRUMENTAL SVCS 200

to the realization of the instrumental with-phrase in (4.70), but omit it when there is no instrumental with-phrase, as in (4.71). It is these LDG representations, then, that best capture the sense in which an instrument phrase constitutes grammatically optional, detachable information. Besides not capturing the idea that instrument participation constitutes grammatically optional information, Koenig et al.’s (2007) semantic representations do not obviously signify that instrument participation constitutes manner or means information. If, on the other hand, means or manner information is understood to precede result (or change of state) information, then the conclusion that the instrument represents manner information may be easily arrived at, since the instrument

participates in subevents s1 and s2, but not s3, which follows these two subevents and 4 encodes a result or change of state. Because of its involvement in subevents s1 and

s2, the instrument is inextricably tied into the manner component of the meaning of a verb, but this also means that it is impossible to divorce the manner information contributed solely by the instrument from any other manner information that may belong to the verb in isolation, separate from the instrument. Some verbs, such as run, nibble, poke, prod, groom and scribble (among hundreds of others), are noted for supplying manner, as opposed to result, or change of state, information (see Rap- paport Hovav and Levin 2008). I raise this point because it may signify a problem for some of Koenig et al.’s (2007) representations for individual verb meanings, such as scribble, a manner verb which was included in the Doodle class of verbs above. The following general semantic formula Koenig et al. (2007) provide for the Doodle verbs is accompanied by Koenig et al.’s description that “verbs in this class describe situations in which putting marks M on the patient P using the instrument I makes M symbolize or represent something T, however imperfectly” (2007, 194):

(4.72) cause(s1, s2) & act(s1, A, I) & mark(s2, I, M, P) & cause(s2, s3) &

represent(s3, M, T)

But what is the formula speciﬁc to the individual verb meaning of scribble supposed

4Koenig et al. (2007) analyze any and every event in which an instrument participates as involving a change of state. Their category of change of state thus admits a far wider range of situations than, say, Rappaport Hovav and Levin (1998). CHAPTER 4. INSTRUMENTAL SVCS 201

to look like? As a manner verb, scribble describes a particular way in which an actor can make marks to symbolize something, in which case should this component of manner meaning be incorporated into a more speciﬁc version of the ACT predicate

of subevent s1, ‘ACT-SCRIBBLY’?:

(4.73) cause(s1, s2) & act-scribbly(s1, A, I) & mark(s2, I, M, P) & cause(s2,

s3) & represent(s3, M, T)

This looks strange, but I cannot see any other way to incorporate the manner meaning

of scribble into this representation (other than somehow stating that s1 and s2 jointly constitute a situation of scribbling). Under the LDG Semantic Form representations I am using, in contrast, the meaning of scribble is clearly represented in a predicate SCRIBBLE, which signiﬁes certain manner information distinct from the manner information signiﬁed by the (optional) instrumental USE(x,z) component:

(4.74) SF: λz λy λx λs {SCRIBBLE(x,y) & USE(x,z)}(s)

That both SCRIBBLE and USE signify manner information would have to be explicated at the level of Conceptual Structure (CS) in LDG, rather than in these Semantic Form representations. It is at Conceptual Structure that a Possible Verb SVC’s status as unitary-event, coidentiﬁed or non-coidentiﬁed resultative holds, and this is also where the upper event structural bound on a clause headed by a single verb can be located. This upper event structural bound is actually more clearly formulated and articulated in the event complexity framework of Levin and Rappa- port Hovav (which does not split semantic representation between two levels), but because LDG’s event structure approach is based in similar principles, most of what Levin and Rappaport Hovav describe can be easily transferred into LDG. The event complexity approach is probably expounded most extensively in the work of Levin and Rappaport Hovav, and, without going into further detail, I will simply observe that the reason why Rappaport Hovav and Levin (1998) can posit an upper bound on verb meanings (consisting of two subevents and one causal relation) is exactly because of their strict delimitations on what the parts of a possible event are – which are manner parts, which are result (or change of state) parts, what their temporal CHAPTER 4. INSTRUMENTAL SVCS 202

properties and scales are, and how they fit together. This brings us back to the point that Rappaport Hovav and Levin’s (1998) event structural framework is based on very different principles and is constructed from very different materials than Koenig et al.’s (2007) event structural framework. Taking this into account, it simply does not make sense for Koenig et al. (2007) to claim that Rappaport Hovav and Levin’s (1998) upper event structural bound on verb meanings (and indeed the upper bound I want to use) is erroneous.

4.3.2.4 Conclusion

This section has argued that Possible Verb instrumental SVCs should, under the event semantic approach I am using, have the following unitary-event semantic proﬁle (=(4.51)):

(4.75) Event structural proﬁle for Possible Verb instrumental SVCs

SF: ... λz ... λx λs {V1(x,...) & USE(x,z)}(s)

CS: V1 corresponds to an event s1. USE does not correspond to a (sub)event.

Specifically, this section argued that USE cannot correspond to a subevent because it is semantically dependent upon the initial predicate it follows, and because it does not have the independent temporal coordinates required of a subevent. As unitary- event SVCs, Possible Verb instrumental SVCs do not breach Rappaport Hovav and Levin’s (1998) upper bound on possible verb meanings, counter to Koenig et al.’s (2007) claim that verb meanings that involve instruments do exactly that. This is not to deny that an instrument is a part of the causal chain in the ways indicated in the situations concerned; it is more a matter of taking a different perspective on event structure. Koenig et al.’s (2007) perspective on instrumental events is helpful in terms of distinguishing among the kinds of relationships that hold between different participants, and from this characterizing the range of variation in instrumental events. It is an event complexity perspective that identifies subevents through their temporal coordinates, however, that is more valuable in the context of this dissertation, because it can accurately formulate the event structural limits on clauses headed by single verbs, or, what is a Possible Verb SVC. CHAPTER 4. INSTRUMENTAL SVCS 203

4.3.3 Possible Verb instrumental ‘take’ SVCs

(4.76)-(4.78) are some examples of instrumental SVCs in which the handling verb is indicated to mean ‘take’, but which nonetheless look to have Possible Verb, non- sequential semantics in which ‘take’ means USE, if the translations for these SVCs are to be strictly adhered to:

´ (4.76) Aràú. zuye àk`ı. buru teri-m´ı. she basket take yam cover-simple-past ‘She used a basket to cover a yam.’ (Ijo; Williamson 1965, 53)

(4.77) Tsoda lá egbà ba cigbà. Tsoda take axe cut tree

‘Tsoda cut the tree with an axe.’ (Nupe; George 1976, 63)

(4.78) Kk´u s j`ıv´ı s n l`an. Koku take knife cut meat ‘Koku cut meat with a knife.’ (Fongbe; Lefebvre and Brousseau 2002, 415)

This section brings us back to the issues raised in the introduction to this chapter, then, of whether ‘take’ can ever mean TAKE in an Impossible Verb sequential instrumental SVC, or if instrumental SVCs always have Possible Verb, non-sequential semantics. If an instrumental ‘take’ SVC is a Possible Verb SVC, then the serialization lexical rule in (4.79) must apply to link ‘take’ to the semantics of USE, and the SVC has the same event structural proﬁle as for Possible Verb instrumental ‘use’ SVCs in (4.80):

(4.79) Instrumental serialization lexical rule:

‘take’ Verb2: λz λy λx λs {V2(x,(y)) & USE(x,z) }(s) −sc +sc +sc

(4.80) Event structural proﬁle for Possible Verb instrumental SVCs

SF: ... λz ... λx λs {V1(x,...) & USE(x,z)}(s)

CS: V1 corresponds to an event s1. USE does not correspond to a (sub)event. CHAPTER 4. INSTRUMENTAL SVCS 204

The basis for positing the existence of the category of Possible Verb non-sequential instrumental SVCs where ‘take’ means USE is the preponderance of instrumental ‘take’ SVCs with non-sequential translations (and the likelihood that, for at least some of these, non-sequential translations do accurately reflect their temporal structure), assertions of non-sequential semantics for these SVCs, and in addition, Durie’s (1997) claim that “serialized ‘take’ does not typically imply the movement of the instrument to the agent” (1997, 346). However, actual evidence that these SVCs have non-sequential semantics is hard to come by, possibly because it is quite difficult to formulate. In fact, it may even be more difficult to establish non-sequential semantics for an instrumental ‘take’ SVC than to establish sequential semantics for an instrumental ‘take’ SVC, in spite of the impression carried in the literature that the latter category is a far rarer commodity. But Impossible Verb, sequential instrumental ‘take’ SVCs are the topic for Section 4.4; this section will consider the evidence for the existence of Possible Verb, non-sequential instrumental ‘take’ SVCs, starting by establishing what TAKE means, or rather what it doesn’t mean in a Possible Verb instrumental ‘take’ SVC. Durie’s (1997, 346) statement that “serialized ‘take’ does not typically imply the movement of the instrument to the agent” is informed by Jackendoff’s (1990) formulation for the semantics of TAKE:

+ α (4.81) TAKE1: CS ([ ]A, [GOPOSS([β]A, [TO[α]])]) (Jackendoﬀ 1990) ‘α acts on β causing β to come into α’s possession’

In order to explain the ﬁrst part of (4.81), Jackendoﬀ’s (1990) framework and notational conventions would need to be described in detail, but instead I will bypass this and simply suggest that we rely upon the second part of (4.81) – the description in quote marks. This description makes it clear that actor-controlled movement is entailed under this semantic representation for TAKE, but the other key component of its meaning is possession. The actor’s possession of the object is the goal for this actor-controlled movement. In contrast, possession is absent from Lefebvre’s (1991) semantic representation for TAKE, which entails actor-controlled movement only:

(4.82) TAKE2: [x cause [y undergo change of location]] (Lefebvre 1991) CHAPTER 4. INSTRUMENTAL SVCS 205

The position taken here is that both of these representations are correct – there are two TAKEs, in other words. TAKE1, represented in (4.81), is more akin to get or obtain; it entails that the actor does not possess the object at the starting point, but s/he does at the endpoint, as in Jim took Sally’s marble collection from the safe. The actor-controlled movement of TAKE2 represented in (4.82), in contrast, carries no entailments about whether the actor possesses the object, at the starting point, the endpoint, or in between. Although the actor must exert some kind of control over the movement of the object under TAKE2, s/he need not possess it; Jan does not possess her neighbor (physically, or in any other sense) in the sentence Jan took her neighbor to the zoo. Also, whereas actor possession is the goal for the actor-controlled movement of TAKE1, there is no such goal inherent in the meaning of TAKE2, and this is why goal phrases – such as to the zoo – can be predicated of TAKE2 (but not

TAKE1). A number of instrumental ‘take’ SVCs describe situations in which ‘take’ cannot mean TAKE (either TAKE1 or TAKE2, or both), and so argue for the existence of Possible Verb, non-sequential instrumental SVCs where ‘take’ instead means USE.

‘Take’ is unlikely to mean TAKE1 or TAKE2 in SVCs where the object of ‘take’ is

abstract, such as (4.83)-(4.84), for example:

(4.83) ∅ ’ˇaw n w-visáa m¯aj máa kh ng.khˇan káp háw naa. take manner-plan new come compete with 1p pcl ‘They will fight us with a new strategy, you know.’ (Lao; Enfield 2002, 18)

(4.84) Kofi teki en dede mama kosi Amba. Kofi take his dead mother curse Amba ‘Kofi cursed Amba by his dead mother’s name.’ (Sranan; Sebba 1987, 59)

As far as TAKE1 is concerned, ‘a new strategy’ as in (4.83), or a ‘dead mother’s name’ as in (4.84) cannot be physically possessed, and neither of these objects can be

physically moved either, with respect to TAKE2. Unless the semantics of TAKE is somehow extended into a non-spatial, metaphorical dimension (which is actually the approach Lefebvre (1991) adopts), then ‘take’ cannot mean TAKE in (4.83)-(4.84). ‘Take’ would also be unlikely to signify TAKE in an SVC where a concrete object of ‘take’ is unable to be moved, or, more weakly, if it is extremely unlikely to be moved. CHAPTER 4. INSTRUMENTAL SVCS 206

Unfortunately, I have not encountered any SVCs of this nature (which would express situations such as William washed the dishes with a dishwasher or Sally shortened the plank with an industrial table saw); the closest is the following SVC:

′

(4.85) t ∅ ˇaw kh¯uang-cák pˇaj h¯et náa. ng. but take machinery go do/make irrigated.rice.field ‘but they used machinery to make the irrigated rice fields.’ (Lao; Enfield 2002, 19)

I assume the machinery referred to in (4.85) can be moved; the crucial question to be asked of Lao speakers is: (in the situation described in (4.85)) is it? Instrumental ‘take’ SVCs in which the object of ‘take’ is a body part of the actor also beg the question of whether ‘take’ can mean TAKE:

(4.86) Ajé mú o. wó. ti `ılèkù. mó. o. mo. . Aje take hand lock door press-on child ‘Aje manually locked the door on the child.’ (Yoruba; Awóyalé1988, 17)

(4.87) Ko-i toa na noko ko-i mue-ia i te paua. ta-3sg take 3sg.poss bottom ta-3sg slam-tr loc art clam ‘He slammed into the clam with his bottom.’ (Pileni; Naess 2004, 243)

Certainly, TAKE1 cannot take a body part as its object, because it is almost impossible to imagine a situation in which the actor does not possess his/her body part at the

starting point. The pure controlled movement interpretation of TAKE2, on the other hand, is possible only if the actor’s will and the actor’s body part are distinguished as two different entities. The significance of (4.83)-(4.87) is that these SVCs present some borderline cases where ‘take’ is unlikely to mean TAKE in instrumental ‘take’ SVCs. As such they suggest a kind of semantic extension, or grammaticalization of ‘take’. If ‘take’ is found not to mean TAKE, but has instead grammaticalized in other, related SVCs in the language concerned, this would fit into a broader picture of widespread and pervasive grammaticalization which could (at most) support the position that ‘take’ does not mean TAKE in the instrumental SVCs of that language. The abstract noun and body CHAPTER 4. INSTRUMENTAL SVCS 207

part objects of ‘take’ in the following (internal argument-sharing) non-instrumental SVCs of (4.88) and (4.89), for example, echo the abstract noun and body part objects

of the instrumental SVCs in (4.83)-(4.84) and (4.86)-(4.87) above:

(4.88) Kh´au au khwa:mkh´ıt n´ı: san khan´akammaka:m pai s/he take idea cl this propose committee dr(away)

l´:u. asp ‘S/he already proposed this idea to the committee.’ (Thai; Jagacinski 1992, 130)

′ (4.89) b p ˇaw hˇua-lâan són kˇan. ... style take head-bald make.collide rcp ‘ ... in the manner of butting each other [with bald heads].’ (Lao; Enfield 2002, 18)

‘Take’ also cannot mean TAKE in the ‘manner modifying’/adverbial SVC of (4.90),

the ‘causativizing’ example in (4.91), or in (4.92), where au ‘take’, according to Jagacinski (1992), signiﬁes that its object is deﬁnite (all of these languages also instantiate their respective ‘take’s in instrumental SVCs):

(4.90) Mo fi o. gbò. n gé igi. 1sg take cleverness cut tree ‘I cut the tree cleverly.’ (Yoruba; Stahlke 1970, 62)

(4.91) Y´ıg´ıd´ı l´a eg´o wo. sun take grass dry

‘The sun dried the grass.’ (Nupe; George 1976, 63)

(4.92) D`ek au khanˇom kin m`ot l :u. child take dessert eat all asp ‘A child ate all the dessert.’ (Thai; Jagacinski 1992, 126)

Extending the survey of ‘take’ SVCs in this way could ultimately reveal quite a lot about the ways in which ‘take’ grammaticalizes in (and outside) SVCs, although all I would suggest at present is that these examples could be at most symptomatic of ‘take’ not meaning TAKE in the instrumental ‘take’ SVCs of the languages concerned. CHAPTER 4. INSTRUMENTAL SVCS 208

An alternative to the possibility that ‘take’ either never means TAKE or else always means TAKE in an instrumental ‘take’ SVC is that the semantics of ‘take’ could, in the instrumental SVCs of some languages, be vague. ‘Take’ could either mean or not mean TAKE in some contexts. But if proving that ‘take’ cannot mean TAKE in an instrumental SVC is diﬃcult, then proving vagueness would be even more complicated. At present all I can do is conclude with the general observation that it is striking how little evidence there apparently is for instrumental ‘take’ SVCs having Possible Verb, non-sequential status. This would seem to be at odds with the fact that these SVCs receive Possible Verb, non-sequential translations in the literature, almost without exception.

4.3.4 Conclusion

This section has discussed Possible Verb instrumental SVCs in which the handling verbs are translated in the interlinear glosses as ‘use’ and ‘take’. Both types of SVC were argued to have the following event structural proﬁle:

(4.93) Event structural proﬁle for Possible Verb instrumental SVCs

SF: ... λz ... λx λs {V1(x,...) & USE(x,z)}(s)

CS: V1 corresponds to an event s1. USE does not correspond to a (sub)event.

Possible Verb instrumental ‘take’ SVCs were argued to undergo serialization lexical rules that alter the semantics of ‘take’ to USE in Section 4.3.3. But it was Section 4.3.2, on instrumental ‘use’ SVCs, that contained the most formal substance, in arguing that (4.93) was the correct event structural representation for Possible Verb instrumental SVCs, under the type of event complexity approach I am using. In contrast, the claim that Possible Verb instrumental SVCs fulfill the Theta Structure Property for SVCs requires less (if any) argumentation; these SVCs fulfill the Theta Structure Property because their Theta Structures contain one or two structural arguments and one non-structural, instrument argument, and there do exist clauses headed by single verbs with identical Theta Structures. Because semantic structures that fulfill the Possible Verbs constraints always have Theta Structures that CHAPTER 4. INSTRUMENTAL SVCS 209

single-verb-headed clauses could have, Possible Verb SVCs fulfill the Theta Structure Property for SVCs trivially. Impossible Verb SVCs, in contrast, always involve predicate combining operations, and so never fulfill the Theta Structure Property trivially. Section 4.4 now considers Impossible Verb instrumental (‘take’) SVCs, which, in necessarily involving a special instrumental SVC argument coidentification rule, fulfill the Theta Structure Property in one of the least trivial ways of all.

4.4 Impossible Verb instrumental SVCs

In Section 4.3.3 I claimed that TAKE could mean one of two things: either actor- controlled movement of an object that results in the actor’s possession of the object, as in TAKE1 of (4.94), or actor-controlled movement of an object that entails nothing about the actor’s possession of the object, or about the goal of the movement, as in

TAKE2:

(4.94) TAKE1: ‘α acts on β causing β to come into α’s possession’

+ α (Simpliﬁcation of Jackendoﬀ (1990): CS ([ ]A, [GOPOSS([β]A, [TO[α]])]))

(4.95) TAKE2: [x cause [y undergo change of location]] (Lefebvre 1991)

If ‘take’ can be shown to signify either of these TAKEs in an instrumental ‘take’ SVC then the SVC will have to be an Impossible Verb SVC with sequential semantics. Specifically, it will have the following event structural profile (where TAKE signifies

either TAKE1 or TAKE2):

(4.96) Event structural proﬁle for Impossible Verb instrumental SVCs

SF: λz ... λx λs [{TAKE(x,z)} & {V2(x,...) & USE(x,z)}](s)

CS: TAKE corresponds to a subevent s1 and V2 to a subevent s2. tc1

tc2.

This Impossible Verb proﬁle is necessitated if ‘take’ means TAKE because speciﬁca- tion of an (instrument-)taking action cannot be interpreted as further specifying the CHAPTER 4. INSTRUMENTAL SVCS 210

situation described by V2 (the ‘main’ verb) in the following (putative) Possible Verb structure:

(4.97) SF: *λz ... λx λs {V2(x,...) & TAKE(x,z)}(s)

TAKE cannot further specify V2 because a prior act of taking the instrument falls outside the interpretation space of V2, and this is what is entailed when ‘take’ means 5 TAKE1 or TAKE2 in a sequential Impossible Verb instrumental ‘take’ SVC. In contrast, USE can further specify V2; it supplies further information about the way in

5 Ç

In contrast, Lefebvre (1991) provides an analysis for Fongbe s¡/z´e ‘take’ SVCs according to Ç which s¡/z´e can signify TAKE2 within an instrumental SVC with non-sequential, Possible Verb

semantics. The following two SVCs, for example, have the non-sequential semantics indicated in Ç their translations, yet s¡ ‘take’ is still allowed to have the semantics of TAKE2 in Lefebvre’s combined Lexical Conceptual Structures for these examples:

Ç ¡Ó (a) K kú s àt´ın xò As´ıbá. Koku take stick hit Asiba ‘Koku hit Asiba with a stick.’ (Fongbe; Lefebvre 1991, 61)

Lefebvre’s (1991) Lexical Conceptual Structures for (a): Ç s¡ ‘take’: [x cause [y undergo change of location]] (TAKE2)

x`o ‘hit’: [y undergo change of location (come into contact with) z] Ç

s¡ x`o: [x cause [y undergo change of location (come into contact with) z]]

Ç ¡Ç ¡E (b) K k´u s j`ıv´ı s n l`an. Koku take knife cut meat ‘Kokucutmeatwithaknife.’ (Fongbe;LefebvreandBrousseau 2002, 415)

Lefebvre’s (1991) Lexical Conceptual Structures for (b): Ç

s¡ ‘take’: [x cause [y undergo change of location]] (TAKE2) E

s¡n ‘cut’: [y undergo change of location (come through) z]

Ç ¡E

s¡ s n: [x cause [y undergo change of location (come through) z]] E What Lefebvre (1991) is doing is casting the semantics of x`o ‘hit’ and s¡n ‘cut’ as verbs of location or movement into which the meaning of TAKE2 is easily integrated to preserve non-sequential, Possible Verb semantics in the combined Lexical Conceptual Structures. But it is only because these verbs are recast as verbs of location or movement that TAKE2 can be interpreted as further specifying the interpretation spaces they open up. Localist approaches to semantics construe, or recast, every type of event as a spatial motion or location event, and Lefebvre’s analysis of Fongbe ‘take’ SVCs is localist in the extreme. Even Jackendoﬀ (1972, 1983, 1990), whose work presents the prime exemplar of localist approaches, would capture in his Lexical Conceptual Structures what it is for an object to be cut, with the change-of-state predicate BE.CUT. In short, the inaccuracies that arise from

recasting any and every verb that cooccurs with ‘take’ in an instrumental ‘take’ SVC as a verb of E location or movement (so, the inaccuracy of not attributing s¡n ‘cut’ in (b) any of the semantics of CUTting or incision, for example) constitute the strongest arguments for not adopting Lefebvre’s (1991) approach. CHAPTER 4. INSTRUMENTAL SVCS 211

which the actor performs the action described by V2, by specifying the instrument used. USE entails that the actor exerts control over the instrument in a relation of volitional causation, but it does not entail that the actor takes the instrument. If ‘take’ means TAKE in an Impossible Verb instrumental SVC, the event structural proﬁle in (4.96) speciﬁes that the subevent of taking temporally precedes the

other, ‘main’ event (tc1

preceding the other verb (Verb2) in the syntax:

(4.98) No Inverse Temporal Iconicity:

If tci

From the fact that ‘take’ appears initially in all of the instrumental ‘take’ SVCs I have encountered it can be concluded that there is no SVC, at least in this survey, that will violate the No Inverse Temporal Iconicity Principle. Further research will reveal whether there are any such No Inverse Temporal Iconicity-violating instrumental ‘take’ SVCs, where the issue of utmost importance is whether ‘take’ really does mean TAKE in a sequential structure. Even in the absence of such hypothetical No Inverse Temporal Iconicity-violating instrumental ‘take’ SVCs, proof of sequential, Impossible Verb status remains a focus here. The first main goal of this section is to demonstrate that there are some instrumental ‘take’ SVCs for which ‘take’ means TAKE and which have the Impossible Verb sequential profile in (4.96). The other main goal is show how these Impossible Verb sequential instrumental ‘take’ SVCs satisfy the Theta Structure Property for SVCs, which for the most part means explicating the instrumental SVC argument coidentification rule covertly responsible for ensuring that the second argument of TAKE and the second argument of USE are identical in (4.96). It is to the first of these goals that I now turn. One type of evidence that could be used to demonstrate that ‘take’ means TAKE in an Impossible Verb instrumental ‘take’ SVC turns upon the modification properties of a language. If certain words or phrases can be shown to modify the verb ‘take’ CHAPTER 4. INSTRUMENTAL SVCS 212

in isolation, without also having scope over the other verb, then this could provide proof that ‘take’ occupies a temporally prior space of its own, and thus that the SVC as a whole has sequential, Impossible Verb semantics. However, the workability of such tests is entirely dependent on the positions that particular modiﬁers can occupy, and how they modify (so, are they pre-modifying or post-modifying, and what kind of syntactic constituent can they modify?). But one indication that such a test can work is found in the following Mandarin data (=(4.3)-(4.4) from Section 4.1):

(4.99) Ta na dao qie-le rou. he hold knife cut-perf meat ‘He cut the meat with a knife.’ (Mandarin; Law 1996, 200)

(4.100) Ta kuaikuaide na dao qie-le rou. he quickly hold knife cut-perf meat (a) ‘He quickly cut the meat with the knife’; (b) ‘He cut the meat with the knife which he quickly took.’ (Mandarin; Law 1996, 213)

The Mandarin adverb kuaikuaide ‘quickly’ functions as a premodifier, modifying words or phrases to its right, and in (4.100) this modifier is placed before the serial verb combination of na ‘hold’ and qie ‘cut’. (4.100) is ambiguous. Under the (a) reading, kuaikuaide ‘quickly’ scopes over the combined serial combination, while under the (b) reading it scopes over the first verb (and its instrument object) only. Law translates na as ‘hold’ in the interlinear glosses of (4.99) and (4.100), but a reading where kuaikuaide ‘quickly’ scopes over a verb with the (atelic and non-dynamic) semantics of HOLD does not seem very likely (?He cut the meat with the knife which he quickly held).6 Kuaikuaide ‘quickly’ can felicitously modify a verb with the (telic and dynamic) semantics of TAKE, however, and this is exactly what we see in (4.100b): na is translated with ‘take’ (or ‘took’), revealing that it actually does have the semantics of TAKE, and probably should have been translated as such in the first place. Now for na ‘take’ to be able to be modified by kuaikuaide ‘quickly’ in isolation, na ‘take’ does have to signify a subevent of taking which occupies its own temporal

6At best, an adverb with the semantics of ‘quickly’ could modify atelic, non-dynamic HOLD with a kind of ‘onset repair’ reading, so that he quickly held the knife really means he quickly came to hold the knife. CHAPTER 4. INSTRUMENTAL SVCS 213

space. Semantically, the constituent that kuaikuaide ‘quickly’ modifies must be an event (or subevent) which has a time scale. Furthermore, the only reading allowed for (4.100) is the temporally iconic one where this space is temporally prior to the temporal space of the ‘main’ subevent. Putting all this together, the second reading of (4.100) shows that instrumental SVCs in Mandarin can have sequential semantics where ‘take’ means TAKE (even though na is indicated to mean ‘hold’). Some other indications of ‘take’ meaning TAKE in an Impossible Verb sequential instrumental ‘take’ SVC are, simply, statements to the effect that a taking subevent has to occur in the SVCs concerned. Such statements can be found in Saramaccan serial verb research. So, while Byrne (1985) submits that Saramaccan instrumental SVCs may have either sequential or non-sequential semantics, and provides alternative translations that reflect this, as in (4.101), Veenstra (1996) and McWhorter (2004) reject Byrne’s claim that Saramaccan instrumental SVCs can have non-sequential interpretations, asserting that only sequential semantics, where ‘take’ means TAKE, may obtain in these constructions.

(4.101) A tei goni suti di pingo. he take gun shoot the pig ‘He shot the pig with a gun’, or ‘He took the gun and (then) shot the pig.’ (Saramaccan; Byrne 1985, 281-2)

These conflicting claims could reflect dialectal differences; whatever the case, the upshot is still that for at least some speakers of Saramaccan, téi ‘take’ always means TAKE in instrumental SVCs, and these SVCs are uniformly sequential. But although McWhorter states that “téi [instrumental] serials are only grammatical in cases where the taking is an immediately salient component of the event” (2004, 91), it is notable that he still gives these SVCs non-sequential translations from which the concept of taking is absent, as in (4.102) (Veenstra (1996), in contrast, always gives Saramaccan

instrumental SVCs biclausal sequential translations that specify a taking subevent):

(4.102) Kob´ı téi d´ı matjáu kóti d´ı b . Kobi take def axe cut def bread ‘Kobi cut the bread with an axe.’ (Saramaccan; McWhorter 2004, 91) CHAPTER 4. INSTRUMENTAL SVCS 214

This further confirms the unreliability of SVC translations as indicators of temporal structure; even linguists who address the temporal structure of instrumental SVCs directly do not feel compelled to make their translations reflect this temporal structure. All issues of translation aside, McWhorter’s (2004) comments about Saramaccan instrumental SVCs form just one part of an interesting thesis about the Saramac- can language. The broader argument of McWhorter’s paper is that Saramaccan is a ‘young’ grammar, and one exemplification of this ‘youth’ is the sequential semantics of Saramaccan instrumental SVCs, in which téi means TAKE. The object of Sara- maccan téi ‘take’ is not extended beyond being a prototypical concrete (and movable) instrument. Extension of the object of ‘take’, such as has apparently not happened in Saramaccan, is understood to signify grammaticalization of ‘take’ itself. Section 4.3.3 included some instrumental SVCs from other languages for which the object of ‘take’ is not concrete or movable, such as (4.103), as well some non-instrumental SVCs such

as (4.104)-(4.105) in which ‘take’ has grammaticalized in diﬀerent directions:

(4.103) ∅ ’ˇaw n w-visáa m¯aj máa kh ng.khˇan káp háw naa. take manner-plan new come compete with 1p pcl ‘The will fight us with a new strategy, you know.’ (Lao; Enfield 2002, 18)

(4.104) Mo fi o. gbò. n gé igi. 1sg take cleverness cut tree ‘I cut the tree cleverly.’ (Yoruba; Stahlke 1970, 62)

(4.105) Y´ıg´ıd´ı l´a eg´o wo. sun take grass dry ‘The sun dried the grass.’ (Nupe; George 1976, 63)

But the point is that Saramaccan téi does not seem to have grammaticalized in this way; téi always takes a concrete object in the (instrumental and non-instrumental) SVCs Byrne (1985), Veenstra (1996) and McWhorter (2004) provide, as well as in the SVCs drawn from a small corpus of stories (Savon (1974a, 1974b), Amoida (1974a,1974b,1974c,1974d), Amoida and Kuse (1974), Beejën (1982), Amoida, Amoida and Amoida (1978)). The only exceptions I found in this corpus are the following, where téi ‘take’ appears to be used idiomatically: CHAPTER 4. INSTRUMENTAL SVCS 215

(4.106) A kandë a sa tei faja tjuma i tu. 3sg maybe 3sg may take fire burn 2sg also ‘It (the gasoline) might also catch alight (lit. take fire) and burn you.’ (Saramaccan; Beejën 1982, 22 (translation mine))

(4.107) N¨o¨o ju o-toona tei woto go saka na Amana so 2sg.emph fut-return take bus go descend loc Amana Lio River ‘So you will come back taking a bus down to the Amana River’ (Saramaccan; Amoida 1974d, 9 (translation mine))

But these idiomatic uses of téi ‘take’ are quite limited (‘taking’ a bus may even have come directly from English); what we apparently don’t see in Saramaccan is the kind of extensive, systematic grammaticalization of ‘take’ that characterizes ‘take’ SVCs in other languages. McWhorter (2004) seizes upon the ungrammaticalized nature of téi to argue for Saramaccan as a ‘young’ grammar, with Saramaccan sequential instrumental SVCs in which téi ‘take’ means TAKE as one thrust of this argument. These Mandarin and Saramaccan data are the strongest indications of which I am aware that ‘take’ can mean TAKE in an Impossible Verb sequential instrumental ‘take’ SVC. It would seem that ‘take’ means TAKE all of the time in Saramaccan instrumental téi SVCs, and at least some (if not all) of the time in Mandarin instrumental na SVCs. The consequence of ‘take’ being able to mean TAKE in an instrumental ‘take’ SVC is that these SVCs will not fulfill the Possible Verbs constraints, and will have the following Impossible Verb sequential event structural profile (=(4.96)), as claimed above:

(4.108) Event structural proﬁle for Impossible Verb instrumental SVCs

SF: λz ... λx λs [{TAKE(x,z)} & {V2(x,...) & USE(x,z)}](s)

CS: TAKE corresponds to a subevent s1 and V2 to a subevent s2. tc1

tc2.

But what went unmentioned above is that this additional event structural proﬁle for instrumental SVCs comes in tow with its own complications. For while the Semantic CHAPTER 4. INSTRUMENTAL SVCS 216

Form representation in (4.108) is the final Semantic Form representation, to which the predicate combining operations have already applied to produce a single Theta Structure that abstracts over two conjoined Possible Verb ({}) components, it is not the case that the predicate combining operations that have to have applied are particularly typical or trivial. The predicate combining operations that have to have applied do not consist of just, say, event combining (operation (i)), highest structural argument coidentification (operation (ii)), and lowest structural argument coidentification (operation (iv)) (this being the typical set of predicate combining operations to apply in an Impossible Verb SVC). Instead, the following instrumental SVC argument coidentification rule is required to ensure that the second argument of TAKE in the first Possible Verb component is coidentified with the second argument of USE in the second Possible Verb component:

(4.109) Instrumental (‘take’) SVC argument coidentiﬁcation rule: In an instrumental Impossible Verb SVC of the form,

‘take’ Verb2: λy λx λs1 {TAKE(x,y) ... }(s1) +sc +sc

& λw ... λu λs2 {V2(u,...) ... & USE(u,w) ... }(s2) −sc +sc y = w, λy −sc

The second argument of a (non-initial) USE predicate was identified as an instrument argument in Section 2.3.3. Linking the object of TAKE to this second argument of USE means that the object of ‘take’ is also recognized as an instrument, and the rule in (4.109) additionally specifies that this coidentified argument has the non- structural status expected of an instrument (through the [-sc] specification). The instrumental SVC argument coidentification rule in (4.109) falls under item (iii) of the Predicate Combining Operations in Chapter 3; it is a ‘language-specific, verb- specific coidentification rule’. It precludes lowest structural argument coidentification

(rule (iv)) in the case that Verb2 is transitive, and it is moreover important that it should do so, considering that instrumental SVCs (Impossible Verb or Possible Verb) CHAPTER 4. INSTRUMENTAL SVCS 217

present a type of multiple object serialization, where the verbs do not share a single object, but each selects a diﬀerent one. In (4.110a) (=(4.101)), for example, di pingo

‘the pig’ is the object of (Verb2) suti ‘shoot’, and this is distinct from goni ‘gun’, the object of tei ‘take’:

(4.110) a. A tei goni suti di pingo. he take gun shoot the pig ‘He shot the pig with a gun’, or ‘He took the gun and (then) shot the pig.’ (Saramaccan; Byrne 1985, 281-2)

b. tei suti: λy λx λs1 {TAKE(x,y)}(s1) & λw λv λu λs2 {SHOOT(u,v) & +sc +sc −sc +sc +sc USE(u,w)}(s2)

i. Event combining: s1+s2=s ii. Highest structural argument coidentiﬁcation: x=u iii. Instrumental SVC argument coidentiﬁcation rule: y = w, λy −sc Combined predicate: tei suti: λy λv λx λs [{TAKE(x,y)} & {SHOOT(x,v) & USE(x,y)}](s) −sc +sc +sc

Di pingo ‘the pig’ is the second structural argument of (4.110a), while goni ‘gun’ is a non-structural instrument argument. This statement is true in the case that (4.110a) is a Possible Verb instrumental ‘take’ SVC (where a serialization lexical rule alters the semantics of tei ‘take’ to USE), but it is also true in the case that (4.110a) is an Impossible Verb instrumental SVC where ‘take’ means TAKE. (4.110a) is analyzed as an Impossible Verb SVC in (4.110b) (in line with McWhorter’s (2004) and Veenstra’s (1996) claims that Saramaccan instrumental SVCs have sequential semantics where tei means TAKE). (4.110b) specifies the predicate combining operations (4.110a) must undergo: there is event combining, as in all Impossible Verb SVCs, highest structural argument coidentification, as in all Impossible Verb SVCs, but then the instrumental SVC argument coidentification rule takes effect to coidentify the internal argument of TAKE (y) in the first Possible Verb component with the non-structural instrument argument of USE (w) in the second Possible Verb component. The instrumental SVC argument coidentification rule additionally specifies that this coidentified argument CHAPTER 4. INSTRUMENTAL SVCS 218

is non-structural, and so (4.110a) satisfies the Theta Structure Property for SVCs because its (post-predicate combining operations) Theta Structure consists of two structural arguments and one non-structural, instrument argument, as will the Theta Structure of a single-verb-headed clause that describes an instrumental event. If the instrumental SVC argument coidentification rule had not applied in (4.110b), lowest structural argument coidentification would require the internal argument of TAKE (y) to be coidentified with the internal argument of SHOOT (v), resulting in a single object serialization where TAKE does not have an instrument object. So far I have not encountered any instrumental ‘take’ SVCs whose other, ‘main’ 7 verb (Verb2) is intransitive. But were such an intransitive-based instrumental ‘take’ SVC to be encountered, and were it moreover shown to have Impossible Verb semantics where ‘take’ means TAKE, it would be just as important as in the transitive-based case that the instrumental SVC argument coidentification rule apply in its predicate combining operations. This is because this rule is the only means through which the internal argument of TAKE is matched with its (non-structural) instrument status. The Theta Structure resulting from the operation of the predicate combining operations would satisfy the Theta Structure Property for SVCs in consisting of one structural argument and one non-structural, instrument argument – a Theta Struc- ture combination that many single-verb-headed clauses share. Now that the instrumental SVC argument coidentification rule in (4.109) has been explained and illustrated, the remainder of this section argues for its very existence. For if the Predicate Combining Operations for Impossible Verb SVCs present a considerable complication or expense attached to including Impossible Verb SVCs in the class of SVCs, the instrumental SVC argument coidentification rule presents an additional complication within the Predicate Combining Operations themselves. On top of this, the reality that this instrumental SVC argument coidentification rule is apparently the only (item (iii)) “language-specific, verb-specific coidentification rule”

7The only intransitive-based instrumental SVC I have encountered is the following Mandarin instrumental ‘use’ SVC (=(4.28)): Zhang-san yòng kuàizi chi-fàn. Zhang-san use chopsticks eat ‘Zhang-san eats with chopsticks.’ ((Mandarin; Li and Thompson 1973, 96) CHAPTER 4. INSTRUMENTAL SVCS 219

that seems to be required across all types of SVC verges on making its inclusion in the Predicate Combining Operations look even quite dubious in its isolation. Why would no other types of SVC – such as benefactive, exceptive or locative SVCs – require analogous “language-specific, verb-specific” coidentification rules? Benefactive, exceptive and locative SVCs are, like instrumental SVCs, ‘argument- introducing’ SVCs. While the type of argument introduced in an instrumental SVC is an instrument, the types of arguments introduced in benefactive, exceptive and locative SVCs, such as (4.111)-(4.113), are benefactive, exceptive and locative arguments (Olú, mi ‘me’, and chufang ‘kitchen’ respectively):

(4.111) Benefactive SVC:

a. Ajé wo. as.o. fún Olú. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awóyalé1988, 5)

′ ′ ′ b. wo. f´un: λz λy λx λs λs {WEAR(x,y)(s ) & BENEFIT(s ,z)}(s) −sc +sc +sc

(4.112) Exceptive SVC:

a. B`e di b´e kp mi. they+cons eat they+cons cut me ‘They eat without me.’ (or: they eat without giving me food) (Baule; N’Guessan 2000, 85)

b. di kp: λy λx λs {EAT(x) & EXCLUDE(x,y)}(s) −sc +sc

(4.113) Locative SVC:

a. Wo zai chufang qie cai. 1sg be.at kitchen cut food ‘I cut food in the kitchen.’ (Mandarin; D´echaine 1993, 254)

b. zai qie: λz λy λx λs {CUT(x,y) & LOC(x, AT(z))}(s) −sc +sc +sc CHAPTER 4. INSTRUMENTAL SVCS 220

Fún ‘give’, kp ‘cut’ and zai ‘be at’ are the ‘introducing verbs’ in (4.111)-(4.113) respectively. Now the reason why Impossible Verb instrumental ‘take’ SVCs call for a specific coidentification rule is that the ‘introduced’ instrument argument of ‘take’ can only attain its non-structural, instrument status by being coidentified with a non- structural, configurationally-defined instrument argument, in a Possible Verb component distinct from that which represents the semantics of ‘take’. Were benefactive, exceptive or locative SVCs to require analogous argument coidentification rules, they would coidentify the internal arguments of the ‘introducing verbs’ with benefactive, exceptive or locative arguments in separate Possible Verb components. But none of the verb representations in (4.111)-(4.113) are separated into distinct Possible Verb components, in reflection of the fact that none of (4.111)-(4.113) are Impossible Verb SVCs. (4.111), for example, is not an Impossible Verb SVC with the semantics of ‘Aje wore an outfit and gave it to Olu’, which would be represented with separate {WEAR(x,y)} and {ACT(x) & BECOME(POSS(v,y))} Possible Verb components. (4.111) is a Possible Verb SVC, since the semantics of fún ‘give’ is altered through a serialization lexical rule to BENEFIT, a (benefactive-introducing) predicate that necessarily appears non-initially in a Possible Verb component (further specifying the activation space opened up by the initial predicate, which in (4.111b) is WEAR). Similarly, (4.112) is not an Impossible Verb SVC with the semantics of ‘they eat and they cut me’, and (4.113) is not an Impossible Verb SVC whose semantic components that represent being in the kitchen and cutting food are separated into distinct Pos- sible Verb components, with no relationship (of further specification of the latter by the former) between them. The short answer to the question of why no other types of SVC require language- specific, verb-specific coidentification rules analogous to the instrumental SVC argument coidentification rule is that no other type of argument-introducing SVC that I know of has been shown to have Impossible Verb semantics.8 For an SVC to require a specific argument coidentification predicate combining operation, it is obviously

8The following Cantonese SVC is the only possible case of an argument-introducing Impossible Verb SVC that I am aware of (although Section 3.3.2 also observes that the Impossible Verb status of this SVC is questionable from the perspective that the combination lo2 bei2 ‘take give’ also appears in an SVC with the Possible Verb, benefactive translation ‘I have taken a book for him/her’): CHAPTER 4. INSTRUMENTAL SVCS 221

going to have to be an Impossible Verb SVC that undergoes predicate combining operations. A longer answer to the question of why no other types of SVC require language-specific, verb-specific coidentification rules awaits further data in the form of Impossible Verb argument-introducing SVCs then, yet the ultimate conclusion could still be that Impossible Verb instrumental ‘take’ SVCs are the only kinds of SVC that warrant a specific coidentification rule, as a consequence of the way in which the introducing verb ‘take’ can have the full, verbal semantics of TAKE, in a separate Possible Verb component, without, however, conferring instrument status upon the internal argument that it introduces.9 There may be something genuinely special or

a. Keoi5 lo2-zo2 bun2 syul bei2 ngo5. 3sg take-perf cl book give 1sg ‘S/he took the book and gave it to me.’ (Cantonese; Bodomo et al. 2003, 64) 2 2 b. lo bei : λy λx λs1 {TAKE(x,y)}(s1)& λw λv λu λs2 {ACT(u) & BECOME(POSS(v,w))}(s2) +sc +sc +sc +sc +sc i. Event combining: s1 + s2 = s ii. Highest structural argument coidentification: x=u iv. Lowest structural argument coidentification: y=w vii. Uncoidentified v retains its medial structural argument status. Combined predicate: λy λv λx λs [{TAKE(x,y)} & {ACT(x) & BECOME(POSS(v,y))}](s) +sc +sc +sc No verb-specific coidentification rule is required here, however, because all of the arguments are structural, and the predicate combining operations listed in (b) account for all of the coidentification operations required. 9Impossible Verb comitative ‘take’ SVCs are the only other kinds of SVC I could imagine requiring a specific coidentification rule. However, I have no examples of Possible Verb, let alone Impossible Verb, comitative ‘take’ SVCs. The closest thing is the following (=(4.47a)), although I remarked in Section 4.2.2 that it is not clear whether en sisa ‘his sister’ should be regarded as an instrument or a comitative argument in the absence of any Sranan-specific instrument (or comitative) tests. Kofi teki en sisa go bay krosi. Kofi take his sister go buy clothes ‘Kofi bought clothes with the help of his sister.’ (Sranan; Jansen, Koopman and Muysken 1978, 130) Suppose en sisa ‘his sister’ was a comitative argument (distinct from instrument arguments in this language) and this SVC had the Impossible Verb semantics of ‘Kofi took his sister and bought clothes (with her)’. Then a comitative SVC argument coidentification rule would be required to coidentify the argument of ‘take’ with a non-structural, configurationally-defined comitative argument in a separate Possible Verb component (whose initial predicate is BUY). But this coidentification rule would be a very close relative of the instrumental SVC argument coidentification rule, because of the use of a verb ‘take’ in each, which itself may reflect that instrument and comitative arguments are close relatives, placed along a continuum. CHAPTER 4. INSTRUMENTAL SVCS 222

exceptional about the way in which ‘take’ can mean TAKE in an Impossible Verb instrumental ‘take’ SVC, for while the action of taking is not in itself instrument- defining, it is nonetheless a typical (if not necessary) extrapolation of what tends to happen in the initial stages of an instrumental event. Of course, there is no point in further pursuing the issue of whether an instrumental SVC argument coidentification rule may be exceptionally warranted without being sure that Impossible Verb instrumental ‘take’ SVCs do exist. The Mandarin and Saramaccan data and accompanying comments presented in this section provide the first indications that Impossible Verb instrumental ‘take’ SVCs do exist, and these indications may be strengthened by further observations. If the evidence for the existence of Impossible Verb instrumental ‘take’ SVCs does seem somewhat insubstantial, this is only because there has been no motivation in the serial verb literature to distinguish Possible Verb from Impossible Verb SVCs, with researchers tending to provide Possible Verb translations for instrumental SVCs either for the sake of compactness, or to reflect the perception that SVCs universally describe single, non-sequential events. Impossible Verb instrumental ‘take’ SVCs provide one of the subtler demonstrations of how SVCs do not always describe single, non-sequential events.

4.5 Conclusion

The main result of this chapter was the demonstration in Section 4.4 that ‘take’ apparently can mean TAKE in an instrumental ‘take’ SVC and that there is thus a category of Impossible Verb instrumental SVC. This presents one source of support for the inclusion of Impossible Verb SVCs in the class of SVCs more generally. Up until now, the recognition of a category of Impossible Verb SVC has mainly been supported with reference to structures which quite blatantly could not be expressed in a clause headed by a single verb (and so should not be Possible Verb SVCs), which nonetheless display every facet of monoclausal behavior that argues for their inclusion in the class of SVCs. Impossible Verb instrumental ‘take’ SVCs also display every facet of monoclausal behavior, yet their failure to exhibit Possible Verb semantics is far CHAPTER 4. INSTRUMENTAL SVCS 223

from blatant. Indeed, Impossible Verb instrumental SVCs are routinely given single- verb-headed translations suggestive of Possible Verb semantics. The significance of recognizing Impossible Verb instrumental SVCs is that Impossible Verb SVCs can no longer be perceived as a category that stands out, far apart from Possible Verb SVCs. Impossible Verb SVCs may instead be more pervasive, lurking around the fringes of what might be perceived as the more typical or ‘solid’ cases of Possible Verb SVCs, such as instrumental SVCs. This chapter (and the next) uncover the very fine line between Possible Verb and Impossible Verb SVCs, through close inspection within a single, unified semantic domain. While Possible Verb instrumental SVCs satisfy the Theta Structure Property in exactly the straightforward manner to be expected of any Possible Verb SVC, Im- possible Verb instrumental SVCs pose special challenges for the satisfaction of the Theta Structure Property. If (4.114) is an Impossible Verb SVC, then the Predi- cate Combining Operations for Impossible Verb SVCs must ensure that the internal argument of TAKE (y) is coidentified, not with the lowest structural argument of SHOOT (v), but with the instrument argument of USE (w) in this second Possible Verb component:

(4.114) a. A tei goni suti di pingo. he take gun shoot the pig ‘He shot the pig with a gun’, or ‘He took the gun and (then) shot the pig.’ (Saramaccan; Byrne 1985, 281-2)

b. tei suti: λy λx λs1 {TAKE(x,y)}(s1) & λw λv λu λs2 {SHOOT(u,v) & +sc +sc −sc +sc +sc USE(u,w)}(s2)

This necessary effect is achieved through the following instrumental SVC argument coidentification rule, which precludes lowest structural argument coidentification:

(4.115) Instrumental (‘take’) SVC argument coidentiﬁcation rule: In an instrumental Impossible Verb SVC of the form,

‘take’ Verb2: λy λx λs1 {TAKE(x,y) ... }(s1) +sc +sc

& λw ... λu λs2 {V2(u,...) ... & USE(u,w) ... }(s2) −sc +sc CHAPTER 4. INSTRUMENTAL SVCS 224

y = w, λy −sc ... (4.114b) resumed:

The instrumental SVC argument coidentification rule is one of the ‘language-specific, verb-specific coidentification rules’ referred to in operation (iii) of the Predicate Com- bining Operations set out in Chapter 3. In fact, it may be the only such rule required across Impossible Verb SVCs (with comitative ‘take’ SVCs, or other such ‘take’ SVC relatives, as the only potential additional cases), which in itself appears questionable. But what I would suggest is that there really may be something special or exceptional about the way in which the instrument-introducing verb ‘take’ can mean TAKE in an Impossible Verb instrumental SVC without, however, conferring instrument status upon its internal argument. This exceptional property is what necessitates an operation (iii) rule for Impossible Verb instrumental SVCs, but for no other type of argument-introducing Impossible Verb SVC. So far this conclusion has outlined the main imports of this chapter as they pertain to the Theta Structure Property and the Possible Verb-Impossible Verb distinction, but the other main import of this chapter was the explication of the ‘event complexity’ approach to event structure I am using here, and how it is in this event structural framework that the upper bound on Possible Verb structure (or the semantic content of a clause headed by a single verb) can be formulated more perspicuously than in any other. ‘Event complexity’ approaches identify the subevents of an event through their temporal coordinates, and in Section 4.3.2 I analyzed instrumental events as unitary- event structures in which the initial predicate (describing the core action) corresponds to a subevent with independent temporal coordinates, but USE (which introduces CHAPTER 4. INSTRUMENTAL SVCS 225

the instrument) does not correspond to a subevent. Instead the USE component only signifies a kind of means or manner modification, where the act of using the instrument is usually (but not always) coextensive with the manner component of the overall event (as in he chopped the tree down with an axe). As unitary-event structures, instrumental events fall well within the upper bound of Possible Verb event structure, which I identify as non-coidentified resultative structure – the loosest of the three Possible Verb event structures I recognize, in which there is a causal connection between two temporally distinct subevents. Much of the explication of the ‘event complexity’ approach in this chapter was set against the contrast of the ‘causal chain’ approach to event structure, rooted in the conceptualization of events as asymmetrical transmissions of force from participant to participant. This conceptualization of events informs Croft’s (1991) analysis of instrumental events, discussed in Section 4.2.1, and Koenig et al.’s (2007) analysis, discussed in Sections 4.2.1 and 4.3.2.3. But while these analyses may put some aspects of instrumental events into clear perspective, it is less clear how, or whether, the upper bound on Possible Verb structure can be formulated under these approaches. Because the ‘event complexity’ approach to event structure does easily lend itself to the formulation of an upper bound on Possible Verb structure, it is the tool best suited to the purposes of this discussion, and in particular to the claim that there is one type of SVC (Possible Verb SVCs) that has the semantic content of a clause headed by a single verb, and another type (Impossible Verb SVCs) which do not. Chapter 5

Motion SVCs

5.1 Introduction

Like the previous chapter, this chapter examines a unified body of SVC data from a single semantic domain, in this case the domain of motion. This chapter looks at how motion SVCs satisfy the Theta Structure Property for SVCs, but, once again, a precondition to this is the need to establish which motion SVCs are Possible Verb SVCs and which are Impossible Verb SVCs. Once again, Possible Verb and Impossible Verb SVCs are prised apart along a very fine line, but as determined by factors some of which are wholly exclusive to the domain of motion. Other factors that constrain Possible Verb motion SVCs are, in contrast, recognizable as motion-specific instantiations of more general, universal constraints on Possible Verb structures. (5.1)-(5.4) exemplify the category of motion SVC examined in this chapter, in which two or more verbs impart information about an event of translational motion – an event in which “an object’s base location shifts from one point to another in space” (Talmy 2000, 35), and where this object, or figure, follows a trajectory, or path, which furthermore may be defined with reference to particular landmarks, or grounds:1

1SVCs such as the following, where only one verb imparts motion information, are mostly ignored in this chapter:

226 CHAPTER 5. MOTION SVCS 227

(5.1) ndadi na.lampa.ra, ne’e lu’u d’i ade kampo. thus 3.go-along.emph ascend enter loc liver village ‘he went along for a while, and then went up into the village.’ (Bima)

(5.2) Er´ı wári. dùo bó-mi. 3sg house go-through come-past

‘He came from the house.’ (Ijo., Williamson 1963, 154)

(5.3) Piti dn kh n paj. Piti walk ascend go ‘Piti walked up, away from the speaker.’ (Thai; Muansuwan 2002, 56)

(5.4) nggara ta.loja [lao ari].sa, ta.mbi’a sambura ... if 2p.sail go out.cond 2p.break scatter ‘if you sail out, then you will break and scatter ...’ (Bima)

The figures are third person entities in (5.1) and (5.2), Piti in (5.3), and the addressee in the conditional structure in (5.4). Some examples of grounds that fix the paths traversed to particular locations in these SVCs are the village up into which the actor enters in (5.1), and the house from which the actor comes through in (5.2). No grounds are referred to in (5.3) and (5.4), although the speaker in (5.3) could be seen as “a special type of ground” (Levinson 2003, 71), in as much as paj ‘go’ is a deictic verb that signifies movement away from this entity. (5.3) and (5.4) additionally contain an extra component of motion meaning: the manner of motion. The verbs

imparting this type of motion information in (5.3) and (5.4) are dn ‘walk’ and loja ‘sail’ respectively. (5.1)-(5.4) illustrate the essential ingredients of the kinds of motion SVCs I will be investigating in this chapter, and they are also indicative of the wide range of motion information that can be contained in a single motion expression. Perhaps the most striking feature of motion SVCs is the sheer number of motion components that could, in theory, be encoded by motion verbs in a single motion SVC. It is

Er´ı, indi fé bó-m´ı. 3sg fish buy come-past ‘He bought fish and came.’ (Ijo; Williamson 1963, 153) CHAPTER 5. MOTION SVCS 228

particularly striking how many motion components could theoretically be encoded by verbs in a single Possible Verb motion SVC. Section 5.2 divides motion information into three broad categories of manner of motion, direction of motion and ground- fixture information, and then formulates a number of constraints on Possible Verb motion events that pertain to these three types of information. These constraints, although not formulated for motion SVCs in particular (but for Possible Verb motion events in general), jointly impose a kind of theoretical upper bound on the number of verbs that could express different motion components in a Possible Verb motion SVC. But Section 5.2 rarely ventures beyond the realms of the theoretical; it is not until the language-specific case studies of Sections 5.3, 5.4, and 5.5 that I begin to show what motion SVCs actually look like, and to what extent they meet the theoretical possibility of realizing several verbs. Nevertheless, Section 5.2 is still at the heart of this chapter, because it holds the key to distinguishing Possible Verb from Impossible Verb motion SVCs – a necessary prerequisite for demonstrating how certain motion SVCs satisfy the Theta Structure Property for SVCs – and, in more general terms, it provides another (albeit motion-specific) illustration of Possible Verb structure that contributes to a more comprehensive picture of what the bounds on single-verb-headed clauses are. While the case studies of Ijo, Thai and Bima motion SVCs in Sections 5.3, 5.4, and 5.5 provide a chance to test the limits on the number of verbs that can be expressed in a motion SVC, their broader concern is with motion SVC structure in general. This basically amounts to questions of permissible verb orderings, for while the instrumental SVCs of Chapter 4 contained only two verbs (the ‘handling’ verb and the ‘main’ verb), giving rise to two possible verb orders (but where the ‘handling’-verb- initial order almost always prevails), the commonplace occurrence of motion SVCs with three or more verbs means there are many more verb order possibilities. The case studies of Sections 5.3, 5.4, and 5.5 proceed from the observation that the verbs in the motion SVCs of different languages are not just freely ordered, but are instead fixed by language-specific motion SVC templates. Another observation is that the SVCs that conform to these templates have a strong tendency towards Possible Verb semantics. And where the search for the elusive Impossible Verb motion SVCs is CHAPTER 5. MOTION SVCS 229

concerned, a further tendency is uncovered: while single instantiations of a motion SVC template do tend to signify Possible Verb semantics, Impossible Verb SVCs tend to comprise structurally of an iteration of the template. This trend is borne out by the languages considered in this chapter, although its range of application needs to be tested further. In any case, it is in these observed tendencies that the examination of motion SVC structure meets the line of investigation whose object is to distinguish Possible Verb from Impossible Verb motion SVCs. Section 5.2 now characterizes and analyzes these two types of motion SVC.

5.2 The semantic composition of Possible Verb and Impossible Verb motion events

This section dissects motion events into components for the purpose of drawing a firm theoretical division between Possible Verb and Impossible Verb motion SVCs. Sec- tion 5.2.1 categorizes motion components into manner of motion, direction of motion and ground-fixture information, as mentioned in Section 5.1, because these are the components targeted by the constraints on Possible Verb motion events, formulated in Section 5.2.2. But Section 5.2.1 also makes a key observation about the apparent freedom of further specification possibilities between motion components in a LDG Semantic Form representation. If manner, direction and ground-fixture components can further specify each other in just about any order in a Semantic Form representation, then this means there is no semantic determination or guidance with respect to the order in which we should expect to find the verbs of a motion SVC. Motion SVCs could then be expected to display various verb orders cross-linguistically, and the language-specific studies of Sections 5.3, 5.4, and 5.5 confirm that this is indeed the case. While one of the main objectives for the case studies of Sections 5.3, 5.4, and 5.5 is to establish language-specific motion SVC templates, another is to distinguish Possible Verb from Impossible Verb motion SVCs, and it is in relation to this latter objective that the constraints on Possible Verb motion events in Section 5.2.2 set CHAPTER 5. MOTION SVCS 230

much of the groundwork. With the main exception of the Unique Vector Constraint of Section 5.2.2.2, these constraints mostly present motion-speciﬁc versions of more general constraints on Possible Verb events. Recognition of this fact can enhance a fuller understanding of the composition of Possible Verb structure in general, as can the by-now-familiar event structural proﬁles for Possible Verb and Impossible Verb motion events set out in Section 5.2.3. Finally, Section 5.2.4 concludes this section and again raises the issues of verb ordering discussed at the end of Section 5.2.1, but here with direct reference to the motion SVCs of Nigerian Pidgin English. Section 5.2.4 thus acts as a bridge into the case studies of Sections 5.3, 5.4, and 5.5.

5.2.1 Three types of motion information

This section distinguishes three broad categories of motion information and illustrates how they are encoded in a LDG Semantic Form representation, with an eye towards formulating the constraints on Possible Verb motion events in Section 5.2.2. These categories are manner of motion, direction and ground-fixture information. As will be elaborated upon shortly, ground-fixture information basically just includes any information that refers to a ground, minus what I term ‘ground-set directions’ such as towards Australia or away from Mary, where the grounds Australia and Mary ‘set’ the directions and do not necessarily fall upon the path. Direction and ground- fixture information are both components of path, but while the division between manner and path information is universally recognized and absolutely fundamental, this division I am drawing within the path component is much less so. However, it is essential that direction information be identified and isolated for the application of the Unique Vector Constraint, described in Section 5.2.2.2, while the Grounds Constraint described in Section 5.2.2.3 and the Single Endpoint Constraint in Section 5.2.2.4 target ground-fixture information. After describing these three types of motion information, the question posed is: what further specification relations can obtain between manner of motion, direction and ground-fixture predicates at Semantic Form? Probing these further specification possibilities clarifies how the components of motion information come together in the description of a single motion event. But first, the CHAPTER 5. MOTION SVCS 231

properties and characteristics of these three diﬀerent types of motion information are discussed, starting with manner of motion. Out of Talmy’s four main components of a motion event, Figure and Path are obligatory, while Manner (and Ground) are not. (5.5) provides a sample of manner of motion predicates, any one of which could be found in a Semantic Form representation:

(5.5) Manner of motion predicates: WALK, RUN, SKIP, AMBLE, SPRINT, CANTER, CREEP, MARCH, STOMP, JUMP, TIPTOE, TROT, SLITHER

These predicates all happen to be based upon verbs drawn from the English inventory of manner of motion verbs – an inventory which is very large, as this language displays a high level of manner salience (Slobin 2006). Now it may seem reasonable to assume that a clear definition of manner of motion could be formulated based on such plentiful data sources (such as English and other languages provide), in which the finest of distinctions seem to be drawn, but that is not the case. Slobin comments that Talmy provides no clear definition, and does not offer one himself, remarking only that manner of motion “covers an ill-defined set of dimensions that modulate motion, including motor pattern, rate, rhythm, posture, affect, and evaluative factors”(2004, 5 fn. 5). Ill-defined as the dimensions of manner of motion may be, for the most part nobody seems to have any trouble identifying a component of a motion event as a manner component. The most prominent exceptions to the ease of identification of manner of motion semantics are verbs such as climb in English, which are often claimed to combine both manner and path components (but see Rappaport Hovav and Levin (2008) for a qualification of this claim). Zlatev and Yangklang (2004) also claim that Thai exhibits a class of ‘manner+path’ verbs that contain aspects of both types of motion information. Section 5.4 will consider these verbs further. Direction of motion information, a type of path information, is exemplified in the alternative sentence-final phrases of (5.6), while (5.7) lists the predicates that correspond to these phrases (where ‘DC’ in the final predicate stands for ‘deictic center’): CHAPTER 5. MOTION SVCS 232

(5.6) Bobby ran northwards/seawards/downwards/towards the gate/away from the house/away.

(5.7) Direction predicates: MOVE-NORTH, MOVE-SEAWARDS, DESCEND, MOVE-TOWARDS(x,the gate), MOVE-AWAY-FROM(x,the house), MOVE-AWAY-FROM-DC

Within the category of direction are three subcategories: absolute direction information, exemplified by MOVE-NORTH, MOVE-SEAWARDS and DESCEND; ground- set direction information, exemplified by MOVE-TOWARDS(x,the gate) and MOVE- AWAY-FROM(x,the house), where the directions are ‘set’ by particular ground argument points or locations (either towards or away from); and deictic information, exemplified by MOVE-AWAY-FROM-DC. Directions are one of Jackendoff’s (1983) three types of path-function (the other two will be discussed shortly). Jackendoff (1983) states that directions involve reference to a ground that does not fall upon the path, but would were the path to be extended. This description best fits what I refer to as ground-set directions, although Jackendoff does also include absolute directions, not set by any particular point or location, in his category of directions. As for deictic components such as MOVE-AWAY-FROM-DC, and its converse, MOVE- TOWARDS-DC, these simply present special cases of ground-set directions, where the ground (which is the deictic center) is implicit. Talmy (2000) recognizes deictic information as one of the components of Path, while the ground-set direction predicates MOVE-TOWARDS and MOVE-AWAY-FROM feature among the ‘Motion- aspect formulas’ that comprise his Vector component of Path. But Talmy does not appear to figure absolute direction information into his decomposition of Path. Stud- ies such as Levinson (2003) and Levinson and Wilkins (2006) have demonstrated how prominently absolute direction information can feature in cross-linguistic descriptions of space and motion, sometimes in ways that are surprising from a European perspective (such as a Guugu Yimithirr speaker telling the researcher to look out for an ant north of his foot (Levinson 2003, 4)). Absolute direction information takes on some importance in Section 5.2.2.2, where it features prominently in Bohnemeyer’s (2003) Unique Vector Constraint. CHAPTER 5. MOTION SVCS 233

The last type of motion information discussed here is ground-fixture information, which is basically anything that fixes the path traversed to any particular point or location. Jackendoff’s (1983) two other types of path-function (besides directions) are bounded paths and routes, and these include ground-fixture information. Bounded paths, exemplified by to the house, from the garden, out of the box or off the tram- poline, are defined by the existence of a ‘reference object’, or ground, which is a boundary; either a source (FROM which the figure moves) or a goal (TO which the figure moves). The reference object for a route, in contrast, is related to some point within the interior of the path, and some example route expressions are along the river, through the tunnel, past the house and via the shops. These bounded path and route expressions all describe ground-fixture information because they fix the path traversed to particular locations (or near particular locations in the case of route expressions like along the river and past the house). Just as Jackendoff (1983) distinguishes sources from goals within his category of bounded paths, and distinguishes these again from routes, Talmy (2000) divides his Motion-aspect formulas into three categories of departure from a source (as in the Motion-aspect formula ‘a point MOVE FROM a point, at a point of time’), arrival at a goal (as in ‘a point MOVE TO a point, at a point in time’), and traversal of a route (as in ‘a point MOVE VIA a point, at a point of time’). This fundamental three-way distinction is seized upon in the formulation of the Grounds Constraint in Section 5.2.2.3. But any analysis of motion events will also have to account for the capacity that languages have for expressing various kinds of geometric complexity that go beyond the basic distinction between sources, routes and goals; the English PPs into the house, onto the house and to the left of the house, for example, describe different geometric relationships between a figure and a ground which is in each case a goal. Jackendoff (1983) accounts for this kind of complexity by embedding place-functions, exemplified by in the room, on the table or under the house, within path-functions, as in (5.8):2

(5.8) The dog went into the room.

2[EVENT], [PATH], [PLACE] and [THING] are some of the ontological categories Jackendoﬀ recognizes, and these can be alternatively notated with italicized subscripts, as in (5.8). CHAPTER 5. MOTION SVCS 234

[Event GO ([Thing DOG],[P ath TO ([Place IN ([Thing ROOM ])])])]

Talmy (2000), meanwhile, posits a Conformation component of Path that expands upon the information within the Motion-aspect formula of the Vector component by encoding more detailed geometric information about the ground and how the figure relates to it. Languages lexicalize different Conformation notions. English, for example, lexicalizes the Conformation notion ‘which is of the inside of [an enclosure]’ in the prepositions in (to) and out of. In fact, there is mind-boggling complexity in terms of the ways in which languages express the geometric relationships that hold between figures and grounds in motion events (as the cross-linguistic studies of Levinson and Wilkins (2006) have confirmed), although the following, very small sample of LDG ground-fixture predicates is only at the tip of the iceberg:

(5.9) Ground-ﬁxture predicates: y is goal (‘to y’, ‘into y’, ‘onto y’): BECOME(LOC(x,AT(y)), BECOME(LOC(x,IN(y))), BECOME(LOC(x,ON(y)) y is route (‘across y’, ‘via y’, ‘through y’): MOVE-ACROSS(x,y), MOVE-VIA(x,y), MOVE-THROUGH(x,y) y is source (‘from y’, ‘out of y’, ‘oﬀ of y’): BECOME(LOC(x,¬AT(y))), BECOME(LOC(x,¬IN(y))), BECOME(LOC(x,¬ON(y)))

These are some of the predicates that are pressed into service in the representations of motion SVCs in the second half of this chapter. Now that we have a description of the three types of motion event information, each with sample predicates, we can consider how these are put together in a LDG Semantic Form representation. (5.10) is a typical LDG Semantic Form representation for a motion event:

(5.10) John ran onto the street. run onto: λy λx λs {RUN(x) & BECOME(LOC(x,ON(y)))}(s) (Kaufmann and Wunderlich 1998, 7) CHAPTER 5. MOTION SVCS 235

(5.10) demonstrates how manner of motion predicates precede ground-fixture, path predicates in LDG representations, but does anything necessitate this ordering? The crucial question for a LDG representation of a motion event is: must the predicates be specified in a particular order in a Semantic Form representation in order for the structure to satisfy the Possible Verbs constraints? With specific reference to (5.10), this is to ask whether BECOME(LOC(x,ON(y))) necessarily further specifies RUN(x), or is it otherwise possible for RUN(x) to further specify BECOME(LOC(x,ON(y)))? RUN encodes information about the manner in which the figure moves in a motion event, and it can be further specified by BECOME(LOC(x,ON(y))), because this encodes the added information that the event is one of translational motion (i.e. it has a path) that culminates in the figure’s position on the street. But this path component could alternatively be further specified by RUN, because this manner predicate adds further information about the way in which the figure traverses the path. Either order of further specification is possible. Consider now an example containing a direction component: in the Semantic Form representation for ‘go north to London’ should MOVE-NORTH(x) precede or follow BECOME(LOC(x,AT(London)))? While the latter component may be more specific with respect to the location of the goal, the former component is more specific with respect to absolute direction, so it may be possible for either component to further specify the other. In fact, there may be multiple possible orderings for many combinations of motion components in a LDG Semantic Form representation, wherever it is the case that each motion component contributes a distinct piece of motion information which can further specify just about any other component, but which is in no way dependent upon any other piece of motion information. In this sense motion components contrast with, say, the USE(x,z) predicate that introduces an instrument participant (z), which is wholly dependent upon there being a main predicate that this component further specifies. If any manner, direction or ground-fixture component can further specify any other of these components, then there can be no real expectations in terms of how the verbs encoding these components should be ordered in a motion SVC. However, the case CHAPTER 5. MOTION SVCS 236

studies in the second half of this chapter do suggest a strong tendency towards ordering manner of motion verbs before direction and ground-fixture verbs. This ordering happens to match the convention for ordering these components in a Semantic Form representation, and it may also reflect a conceptualization that the manner of motion somehow causes the path to be traversed, as in Croft’s (1998) causal force analysis of motion events. Another tendency (but perhaps weaker) is for verbs which describe deictic information to appear last in motion SVC templates, although this chapter will also illustrate some counterexamples to this tendency. In short, because this analytical section makes no predictions about the ordering of manner, direction and ground-fixture components within motion SVCs, most of the interest lies in seeing exactly how they are ordered in the motion SVC templates presented for different languages in the second half of this chapter. But while further specification possibilities do not necessitate any particular ordering of manner, direction or ground-fixture information, there are restrictions within these three motion types, in terms of which (if any) manner components can further specify manner components, which direction components can further specify other direction components, and which ground-fixture components can further specify other ground-fixture components. Section 5.2.2 now considers these restrictions, as constraints on Possible Verb motion events.

5.2.2 Constraints on Possible Verb motion events

This section, which constitutes the bulk of Section 5.2, contains an in-depth analysis of the division between Possible Verb and Impossible Verb structure in the sphere of motion events, detailed in five constraints on Possible Verb motion events in Sections 5.2.2.1-5.2.2.5. Besides stating the constraints (both in general terms and in terms of further specification in a LDG Semantic Form representation), each of these sections also describes certain ‘red flags’ for Impossible Verb motion event status. There is thus a practical side to this section, in addition to the theoretical side that demarcates Possible Verb structure, since the purpose of these red flags is to set up the language- specific motion SVC case studies in Sections 5.3, 5.4 and 5.5 with some practical means for identifying Impossible Verb motion SVCs. CHAPTER 5. MOTION SVCS 237

5.2.2.1 The Manner of Motion Constraint on Possible Verb motion events

The ﬁrst constraint on Possible Verb motion events targets manner information:

(5.11) The Manner of Motion Constraint on Possible Verb motion events: a Possible Verb motion event can contain at most one manner of motion component.

(5.12) presents a simple recasting of (5.11) in the LDG terms of further speciﬁcation within a Semantic Form representation:

(5.12) The Manner of Motion Constraint in terms of further speciﬁcation in a Semantic Form representation: a manner of motion predicate cannot be further speciﬁed by another manner of motion predicate.

One illustration of how a manner of motion component cannot be further specified by another in a Possible Verb motion event is the fact that WALK cannot further specify the activation space opened up by RUN, because in intuitive terms it is not possible to run and walk at the same time. At best, attempting to further specify one manner of motion verb by another can produce some strange, composite manner of motion verbs, such as ‘bounce-walk’, ‘skippity-run’ or ‘hop-jump’. However, none of these, or anything else evoking the Ministry of Silly Walks, is particularly acceptable. There is a deeper, non-motion-specific basis for the constraints in (5.11) and (5.12). This is that Possible Verb events, of any kind, can specify at most one manner activity, such as is lexicalized in one verb (see, e.g., Levin and Rappaport Hovav2010). There can of course be manner modification of this manner activity, involving further specification of the activation space opened up by the manner verb, but what Possi- ble Verb events do not allow are distinct manner components, such as are lexicalized by different verbs, that cannot specify each other but instead each open up separate activation spaces. Rappaport Hovav and Levin’s (1998) predicate decompositions, in being more templatic in nature than LDG Semantic Form representations, better put across the point that there is only one manner ‘slot’ in a Possible Verb event. (5.13) are their semantic representations for the four possible (Aktionsart) event types: CHAPTER 5. MOTION SVCS 238

(5.13) [x ACT] (activity) [x ] (state) [BECOME [x ] ] (achievement) [x CAUSE [BECOME [x ] ] ] (accomplishment)

[ [x ACT] CAUSE [BECOME [x ] ] ] (accomplishment)

The italicized predicates in angle brackets in (5.13) represent idiosyncratic ‘constant’ (Rappaport Hovav and Levin 1998) or ‘root’ (Pesetsky 1995, Levin and Rappaport Hovav 2005) information speciﬁc to particular verbs or verb complexes (squaring with Grimshaw’s (1994) notion of semantic content), while the other, ‘primitive’ predicates represent the structural aspects common to many verbs or verb complexes (equivalent to Grimshaw’s (1994) semantic structure). The predicate that modiﬁes

the activity predicate ACT in the ACT components in (5.13) represents one type of idiosyncratic, root information. While states and achievements do not

encode manner information, the presence of this ACT component in the first and last event structures in (5.13) signifies that activities do and accomplishments can, but the crucial point is that it is only possible to encode one of these manner components in an activity or accomplishment. In no event structure can two manner components be included, and so these representations of every possible event type in (5.13) provide a kind of templatic reinforcement of the idea that no Possible Verb event (motion or otherwise) can contain more than one manner component. While the manner of motion constraints in (5.11) and (5.12) present motion- specific reflexes of the more general constraint on Possible Verb events that these allow at most one manner component or ‘slot’, the event structure representations in (5.13) point towards another general constraint on Possible Verb events: that these allow at most one result component or ‘slot’, as encoded by the primitive predicate BECOME. No Possible Verb event type can encode more than one result component, with more than one BECOME. Section 5.2.2.4 formulates a motion-specific version of this general result constraint on Possible Verb events, but the final point to mention here is that there is a rather clear red flag for Impossible Verb motion event status attached to CHAPTER 5. MOTION SVCS 239

the manner of motion constraints above: Impossible Verb status is signaled simply by the instantiation of more than one manner of motion verb.

5.2.2.2 A constraint on direction information in Possible Verb motion events: The Unique Vector Constraint

This section describes a constraint on Possible Verb motion events that is more speciﬁc to the domain of motion than any other of the constraints discussed here, for the reason that it concerns direction information unique to this domain. Bohnemeyer’s (2003) Unique Vector Constraint determines how direction speciﬁcations may combine in a Possible Verb motion event:

(5.14) Unique Vector Constraint (UVC): all direction speciﬁcations in a single simple clause referring to a single continuous motion event must denote the same ‘unbounded’ direction vector, i.e. the same direction vector after abstraction from head and tail coordinates. This direction vector is interpreted as holding for the entire motion event. (Bohnemeyer 2003, 102)

This constraint targets the absolute direction, ground-set direction and deictic components described in Section 5.2.1, requiring that if more than one of these components is instantiated, then they must encode collinear direction vectors of the same polarity. Multiple direction speciﬁcations may only obtain within a single simple clause, that is to say, if they point in the same direction and can be mapped onto a single line. The Unique Vector Constraint, as stated in (5.14), has scope over ‘single simple clause[s]’, although the relevance to this section is that it also scopes over Possible Verb structures – structures that satisfy the Possible Verbs constraints at Semantic Form, and so can be expressed in clauses headed by single verbs. (5.15) recasts this constraint in terms of further speciﬁcation possibilities in a Possible Verb structure:

(5.15) Unique Vector Constraint in terms of further specification in a Semantic Form representation: a direction predicate3 may only be further specified by another direction predicate if this second direction specification

3‘Direction predicates’ are as listed in Section 5.2.1: MOVE-TOWARDS-DC(x), MOVE-AWAY- FROM-DC(x), MOVE-TOWARDS(x,y), MOVE-NORTH(x), MOVE-SOUTH(x), etc. CHAPTER 5. MOTION SVCS 240

can be collinear and of the same polarity as the direction speciﬁcation(s) already encoded.

In Bohnemeyer et al. (2007), the scope of the Unique Vector Constraint is extended beyond both ‘single simple clauses’ and Possible Verb structures to the larger structural unit of ‘macro-event expressions’, although the consequences of this extension, and macro-event status itself, will not be discussed until the end of this section. The Unique Vector Constraint explains why the following expressions are unacceptable:

(5.16) *Sally went up (the hill) down (to the river).

(5.17) *The ﬁgure moved north via B east to C. (Bohnemeyer 2003, 98)

(5.16) and (5.17) violate the Unique Vector Constraint because they are structured as simple, single-verb-headed clauses (Possible Verb structures), yet in each case the multiple direction specifications they contain cannot in combination denote a single collinear direction vector. Up and down in (5.16) cannot jointly denote a single vector because they have opposing polarity, while north and east in (5.17) cannot be collinear because they describe different directions (north-east, on the other hand, does denote a single direction, but this counts as just one direction specification and one which is not, moreover, specified in (5.17)). (5.16) and (5.17) do, however, become acceptable if altered to become more complex structures where coordinators separate the direction specifications:

(5.18) Sally went up (the hill) and (then) down (to the river).

(5.19) The ﬁgure moved north via B and east to C.

(5.18) and (5.19) are what Bohnemeyer (2003) would identify as underlying biclausal ‘gapping’ constructions, in which the verbs (as well as subjects) are omitted from the second conjuncts. Bohnemeyer (2003) does also acknowledge that such sentences could instead involve PP-coordination or VP-coordination, but the important point CHAPTER 5. MOTION SVCS 241

is that they are not single simple clauses, or Possible Verb structures, and it is to these structures only that the Unique Vector Constraint applies. Another way of illustrating how the Unique Vector Constraint works is through the fact that the simple clause containing two direction speciﬁcations (MOVE-AWAY- FROM(F,A) and MOVE-TOWARDS(F,B)) in (5.20) must describe a single vector motion event such as depicted in (5.21a), but never a multi-vector motion event such as depicted in (5.21b):

(5.20) The Figure moved away from A towards B. (Bohnemeyer 2003, 99)

(5.21)a. b. f

A• f •B

A• • B (Bohnemeyer 2003, 99)

However, one feature of the Unique Vector Constraint emphasized in Bohnemeyer (2003) is that this is a constraint only on what direction information can be coded in a single clause. What it does not require is that any and every single-verb-headed motion event description have a single-vector denotation; (5.22), for example, is a single-verb-headed, Possible Verb structure, yet it does not map onto a single vector:

(5.22) Jim went from Westminster via the London Eye to St Paul’s Cathedral.

But the crucial point regarding (5.22) is that this motion event description does not contain any direction specifications; instead it only describes the path in terms of location changes. Only once direction specifications are introduced into a Possible Verb motion event description does the Unique Vector Constraint take effect. If the Unique Vector Constraint were to scope over Possible Verb structures, and nothing larger, then multiple, non-collinear direction specifications within a single motion SVC could be interpreted as a sure sign of Impossible Verb status – this would be a clear red flag for Impossible Verb status, in other words. However, this issue is complicated by the fact that Bohnemeyer et al. (2007) extend the scope of the Unique CHAPTER 5. MOTION SVCS 242

Vector Constraint to ‘macro-event expressions’: structures that hold the Macro-Event Property, formulated as a measure of the tightness of packaging of subevents in a construction. Bohnemeyer et al. (2007) state that “a construction has the Macro- Event Property if temporal operations such as time adverbials, temporal clauses, and tenses necessarily have scope over all subevents encoded by the construction” (2007, 7). So, (5.23a) has the Macro-Event Property because the time adverbial in the morning in (5.23b) scopes over all three subevents of departure, traversal and arrival, while (5.23c) demonstrates how distinct time adverbials cannot scope over these subevents separately:

(5.23) a. Floyd went from Rochester via Batavia to Buffalo. b. Floyd went from Rochester via Batavia to Buffalo in the morning. c. *Floyd went from Rochester at seven via Batavia at seven forty-five to Buffalo at eight thirty. (Bohnemeyer et al. 2007, 15,16)

The ‘gapping’ construction in (5.24a), on the other hand, does not have the Macro- Event Property, because distinct time adverbials can scope over these same subevents of departure, traversal and arrival separately, as in (5.24b):

(5.24) a. Floyd left Rochester, passed through Batavia, and arrived at Buffalo. b. Floyd left Rochester at seven, passed through Batavia at seven forty-five, and arrived at Buffalo at eight thirty. (Bohnemeyer et al. 2007, 16)

Possible Verb SVCs will always have the Macro-Event Property because it is not possible for one time positional operator within a Possible Verb structure to be further speciﬁed by another, distinct time positional operator.4 Each Possible Verb component within an Impossible Verb SVC could, in theory, have its own distinct time positional operator, meaning that the SVC as a whole is not a macro-event expression. All that is required of an SVC, after all, is that the verbs have identical

4Unless the operators are thematic disconstituents in Brunson’s (1992) sense to be discussed in Section 5.2.2.3. This morning and at 8am are thematic disconstituents in the sentence Jilly caught the bus this morning at 8am (where the former is more coarse-grained and the latter more ﬁne-grained). CHAPTER 5. MOTION SVCS 243

tense specifications; in theory these verbs could still be modified by separate time adverbials – if the syntax of the SVC were to allow it. However, given that SVCs do tend to express closely (temporally) connected events, it does seem more likely that all SVCs will have the Macro-Event Property. Until I encounter an Impossible Verb SVC that does not have the Macro-Event Property, I will assume Impossible Verb SVCs are Macro-Event Expressions too. But under this assumption, all Impossible Verb motion SVCs should satisfy the Unique Vector Constraint, meaning that this constraint cannot be used to distinguish Impossible Verb from Possible Verb motion SVCs. In conclusion to this section, the first point to stress is that the Unique Vector Constraint is a constraint on Possible Verb motion events, that can be formulated in terms of further specification (im)possibilities in a Possible Verb Semantic Form representation. It just so happens that it may constrain other, more complex structures, i.e. Macro-Event Expressions. If there are any Impossible Verb motion SVCs that are not Macro-Event Expressions, then violation of the Unique Vector Constraint would be one indication of Impossible Verb status in these cases. For although doubts were raised here about the existence of non-Macro-event Impossible Verb SVCs, the fact remains that there is not, at present, a wealth of evidence in the SVC literature in direct support of these doubts.

5.2.2.3 The Ground Arguments Constraint on Possible Verb motion events

The topic of this section is the following constraint:

(5.25) The Ground Arguments Constraint on Possible Verb motion events (simple version): a Possible Verb motion event can express at most one departure from a source, one traversal through a route, and one arrival at a goal.

This section addresses the question of how many ground-ﬁxture components are permitted in a Possible Verb motion event, to which the simple answer turns out to be maximally three (source + route + goal), although there do abound what appear to be some rather blatant exceptions to this rule, which call for a more complicated CHAPTER 5. MOTION SVCS 244

formulation of this constraint. The more complex reformulation of (5.25) in (5.35) later on makes crucial use of Brunson’s (1992) notion of thematic disconstituents, as will be explained shortly, but ﬁrst, (5.25) is transposed into LDG’s terms of further speciﬁcation in (5.26):

(5.26) The Ground Arguments Constraint in terms of further specification in a Semantic Form representation (simple version): a source predicate may not be further specified by another source predicate, a route predicate may not be further specified by another route predicate, and a goal predicate may not be further specified by another goal predicate, where source, route and goal predicates are as listed in (5.27).

(5.27) Source predicates: BECOME(LOC(x,¬AT(y))) (‘from y’) BECOME(LOC(x,¬IN(y))) (‘out of y’) BECOME(LOC(x,¬ON(y))) (‘oﬀ of y’) Route predicates: MOVE-VIA(x,y) (‘via y’) MOVE-ALONG(x,y) (‘along y’) BECOME(LOC(x,PAST(y))) (‘past y’) MOVE-ACROSS(x,y), BECOME(LOC(x,ACROSS(y)))5(‘across y’)

5The difference between MOVE-ACROSS(x,y) and BECOME(LOC(x,ACROSS(y))) is that whereas the figure does not traverse the whole extent of the route y in the first case, but still ends up located within y, as in (a) below, the figure does traverse the whole route in the second case, ending up at a point beyond the boundary of y, as in (b). Analogous comments pertain to MOVE-THROUGH(x,y) and BECOME(LOC(x,THROUGH(y))) in (c) and (d):

(a) Johnny ran across the paddock for 10 seconds. (MOVE-ACROSS(x,y)) (b) Kim crossed the road in 20 seconds (and entered the shop). (BECOME(LOC(x,ACROSS(y)))) (c) Jilly drove through the tunnel for two minutes (and broke down). (MOVE-THROUGH(x,y)) (d) Jim drove through the tunnel in two minutes. (BECOME(LOC(x,THROUGH(y))))

The other diﬀerence between MOVE-ACROSS(x,y)/MOVE-THROUGH(x,y) and BE- COME(LOC(x,ACROSS(y)))/BECOME(LOC(x,THROUGH(y))) is that whereas the former CHAPTER 5. MOTION SVCS 245

MOVE-THROUGH(x,y), BECOME(LOC(x,THROUGH(y))) (‘through y’) Goal Predicates: BECOME(LOC(x,AT(y)) (‘to y’) BECOME(LOC(x,IN(y))) (‘into y’) BECOME(LOC(x,ON(y))) (‘onto y’)

(5.27) contains the most complete collection of ground-fixture predicates listed so far, although it is still far from complete, as well as being based entirely upon English expressions of space and motion. These are deficiencies that can be improved upon at a later stage; at present I will focus on (5.25) or (5.26), and how they constrain Possible Verb motion events. At the heart of the constraints in (5.25) or (5.26) is the contention that there are no other path roles than source, route, or goal – a position with which the analyses of motion events discussed in Section 5.2.1 concur. The constraint that a Possible Verb motion event can express no more than one each of these path roles has a well-founded basis in one of the main tenets of Fillmore’s (1968) case grammar, that “each case relationship [or thematic role] occurs only once in a simple sentence” (1968, 21). That no thematic role (path role or not) should be realized more than once, in more than one NP or adjunct, is also implemented under Bresnan’s (1982) ‘Biuniqueness Condition’ or Chomsky’s (1981) ‘Theta-Criterion’, which require one-to-one mappings between NPs and thematic roles. Bohnemeyer (2003) calls this the ‘Argument Uniqueness Constraint’. The Argument Uniqueness Constraint scopes over single simple clauses, which here I equate with Possible Verb structures (i.e. anything that can be realized in a clause headed by a single verb). The Grounds Constraints in (5.25) and (5.26) above present motion-specific applications of the Argument Uniqueness Constraint, and (5.28)-(5.30) demonstrate how they work:

are atelic, the latter have telic semantics – this much is apparent from the use of a BECOME predicate that signiﬁes a telic culmination or change of state in LDG Semantic Form representations. Being able to distinguish telic from atelic instantiations of ‘across’ and ‘through’ assumes some importance in Section 5.2.2.4, since the Single Endpoint Constraint discussed there only applies to the former (telic) cases whose semantics are represented with BECOME. CHAPTER 5. MOTION SVCS 246

(5.28) Julie drove from Southampton via London to Cambridge. λx λs {DRIVE(x,(y)) & BECOME(LOC(x,¬AT(Southampton))) & MOVE-VIA(x,London) & BECOME(LOC(x,AT(Cambridge)))}(s)

(5.29) *Julie drove from Southampton from London. *λx λs {DRIVE(x,(y)) & BECOME(LOC(x,¬AT(Southampton))) & BECOME(LOC(x,¬AT(London)))}(s)

(5.30) *Julie drove to London to Cambridge. *λx λs {DRIVE(x,(y)) & BECOME(LOC(x,AT(London))) & BECOME(LOC(x,AT(Cambridge)))}(s)

While (5.28) is acceptable because it realizes a source (from Southampton), a route (via London), agoal(to Cambridge) and nothing more, (5.29) is unacceptable because it realizes two sources and (5.30) because it realizes two goals. But in that case why are (5.31)-(5.33) all acceptable?; (5.31) realizes three route expressions, (5.32a) and (5.32b) two goal expressions, and (5.33a) and (5.33b) two source expressions:

(5.31) Our ﬁnal leg in the Across America North tour will take us [across upper New York State] [thru the rolling farm country] [past Rochester]. (Bohnemeyer et al. 2007, fn. 14)

(5.32) a. Jim drove [into Guildford] [to Bob’s house]. b. Jane went [to London] [to the London Eye].

(5.33) a. Jane went [from her house] [out of Southampton]. b. Jane came [from Cambridge] [from her college].

Do (5.31)-(5.33) constitute genuine violations of the Argument Uniqueness Con- straint? Bohnemeyer et al. (2007) observe that the path segments in the sentence in (5.31) must overlap (which also holds for (5.32)-(5.33)), and ultimately argue that the three route phrases in this sentence form a single superordinate path phrase, and CHAPTER 5. MOTION SVCS 247

thus do not violate Argument Uniqueness. Likewise, I would argue that the multiple goal phrases in (5.32) form single superordinate goal phrases, while the multiple source phrases in (5.33) form single superordinate source phrases. The basic idea behind the claim that the combinations of path phrases in (5.31)-(5.33) in each case constitute single superordinate phrases is that these combinations of phrases are related or connected in such a way that they can jointly specify a single cohesive path. Brunson (1992) explicates the nature of the relationship that must hold between the path phrases (and indeed many other types of phrase) in order for them to form single superordinate phrases; in her terms they are thematic disconstituents. Brunson’s (1992) notion of thematic disconstituent is set out in the following “Interpretive Constraint on Thematic Discontinuity”:

(5.34) Interpretive Constraint on Thematic Discontinuity X1 ... Xn are disconstituents of a thematic discontinuity expressing the role θi iﬀ they cumulatively identify a single referent and

X1θi ⊃θi X2θi ⊃θi ... ⊃θi Xnθi

where X1θi ⊃θi X2θi is true iﬀ X1 is coarser-grained with respect to the role θi than X2. (Brunson 1992, 120)

Central to this formulation of the concept of thematic discontinuity is the idea that disconstituents supply information of varying grains. In (5.32b) above, to the London Eye is more fine-grained than to London in the sense that it is more localized, or more specific with respect to the location of the path. To the London Eye hones in more accurately on the exact location of the figure, in contrast to to London, which provides more coarse-grained information about the figure’s location. The following (slightly) longer restatement of the original Grounds Constraint in (5.25) allows for multiple path phrases of the same type – but of differing grain sizes – to form thematic discontinuities, the disconstituents forming single superordinate roles: CHAPTER 5. MOTION SVCS 248

(5.35) The Ground Arguments Constraint on Possible Verb motion events (longer version): a Possible Verb motion event can express at most one departure from a source, one traversal through a route, and one arrival at a goal. Each of these path roles could, however, be composed of multiple ‘disconstituent’ roles in Brunson’s (1992) sense.

(5.36), meanwhile, reformulates (5.26) above in terms of the spatial property of containment that must hold between path role disconstituents in a Possible Verb motion description:

(5.36) The Ground Arguments Constraint in terms of further specification in a Semantic Form representation (longer version): A source predicate may not be further specified by another source predicate unless the source argument of the former spatially contains that of the latter. A route predicate may not be further specified by another route predicate unless the route argument of the former spatially contains that of the latter. A goal predicate may not be further specified by another goal predicate unless the goal argument of the former spatially contains that of the latter.

The ground arguments of coarser-grained path role disconstituents must spatially contain those of ﬁner-grained path role disconstituents. Thus London, in coarser- grained to London, spatially contains the London Eye in ﬁner-grained to the London Eye, and so the goal predicate corresponding to the latter can further specify that corresponding to the former in the Semantic Form representation of (5.37):

(5.37) Jane went [to London] [to the London Eye]. λx λs {GO(x) & BECOME(LOC(x,AT(London))) & BECOME(LOC(x,AT(London Eye)))}(s) Spatially contains The property of spatial containment also holds between the source disconstituent arguments Cambridge and Jane’s college in (5.38), allowing the predicate containing the former source to be further speciﬁed by that containing the latter, while Wiltshire CHAPTER 5. MOTION SVCS 249

spatially contains Salisbury in (5.39), accounting for the possibility for the route predicate corresponding to the former to be further speciﬁed by that corresponding to the latter:

(5.38) Jane drove [from Cambridge] [from her college]. λx λs {DRIVE(x,(y)) & BECOME(LOC(x,¬AT(Cambridge))) & BECOME(LOC(x,¬AT(college)))}(s) Spatially contains (5.39) Jake drove [across Wiltshire] [via Salisbury]. λx λs {DRIVE(x,(y)) & BECOME(LOC(x,(ACROSS(Wiltshire)))) & MOVE-VIA(x,Salisbury)}(s) Spatially contains Now that the Grounds Constraint (in its longer forms) has been stated and explained, this section turns to the question of whether this constraint can serve to distinguish Possible Verb motion SVCs from Impossible Verb motion SVCs. The answer to this question is yes, because an Impossible Verb motion SVC could potentially describe more than one departure, one traversal or one arrival. The discussion up until now has focused on how Possible Verb motion events are limited to realizing at most one source, one route and one goal (each possibly composed of disconstituents), for the reason that the Argument Uniqueness Constraint that scopes over Possible Verb structures is stated in terms of these thematic roles. But the Grounds Constraint as it appears in (5.25) and (5.35) refers not just to path roles, but to departures, traversals and arrivals. The Semantic Form predicates to which the Grounds Con- straints in (5.26) and (5.36) refer encode information not just about sources, routes and goals, but about departures, traversals and arrivals as well. The predicate BE- COME(LOC(x,AT(London))), for example, describes an arrival, with the argument London as the goal, while BECOME(LOC(x,¬AT(Cambridge))) describes a departure, with Cambridge as the source argument. Once these event-denoting terms of departure, traversal and arrival are drawn into the picture, a possible means of distinguishing Impossible Verb motion SVCs from Possible Verb motion SVCs comes into view, for the reason that Impossible Verb motion SVCs, unlike Possible Verb SVCs, could describe more than one departure, one traversal or one arrival (with CHAPTER 5. MOTION SVCS 250

no relationship of thematic discontinuity holding between the respective ground arguments). However, the Theta Structure Property for SVCs does dictate that an Impossible Verb SVC may still only overtly realize at most one source, one route and one goal, which means that if an Impossible Verb SVC describes two departures, one of the sources will have to be suppressed, or if it describes two traversals, one of the routes will have to be suppressed, or if it describes two arrivals, one of the goals will have to be suppressed. An example of the kind of thing an Impossible Verb motion SVC could describe, then, is illustrated in (5.40), where the suppression of the ﬁrst of the two goal arguments is illustrated in an accompanying Semantic Form representation:

(5.40) (Bo got to the stream.) He ran in (to the stream), and (then) into the jungle. λx λs [{RUN(x) & BECOME(LOC(x,(IN(stream)))) & {RUN(x) & BECOME (LOC(x,(IN(jungle))))](s) ∅

(5.40) encodes two arrivals (with in (to the stream) and into the jungle). Where SVCs are concerned, the red ﬂag for Impossible Verb status will be the instantiation of more than one departure-encoding verb, or more than one traversal-encoding verb, or more than one arrival-encoding verb, where the ground arguments concerned are unlikely to be disconstituents in a relationship of spatial containment. It remains to be seen whether there do exist any Impossible Verb motion SVCs of this ilk, although Section 5.5 does present one possible Bima example. If there is one thing that these inquiries serve to underscore, however, it is that, of all the constraints on Possible Verb motion event status discussed in this chapter, it is the Ground Arguments Constraint that pertains most directly to the Theta Structure Property for SVCs. It is this constraint that requires a motion SVC to overtly realize at most one source, one route and one goal (each role possibly composed of disconstituents), in exact semblance to the Theta Structure lists of motion event descriptions headed by single verbs.

5.2.2.4 The Single Endpoint Constraint on Possible Verb motion events

This section discusses a constraint which, like the Ground Arguments Constraint, is concerned with ground-ﬁxture information, and which is in some sense a complement CHAPTER 5. MOTION SVCS 251

to the Manner of Motion Constraint on Possible Verb motion events in Section 5.2.2.1:

(5.41) The Single Endpoint Constraint on Possible Verb motion events: a Possible Verb motion event must have a single, unique endpoint, or result location.6

Just as the Manner of Motion Constraint presents a motion-specific version of the limitation on Possible Verb events to one manner component, the constraint in (5.41) presents a motion-specific version of the limitation on Possible Verb events to one result component. Because the endpoint of a path in a motion event signifies a motion- specific type of result state or telic culmination, the limitation to a single, unique endpoint in (5.41) puts into effect the broader principle that a Possible Verb event should have only one result state or telic culmination. Telic culminations are encoded by BECOME in LDG Semantic Form representations, facilitating a straightforward recasting of (5.41) in LDG terms:

(5.42) The Single Endpoint Constraint in terms of further speciﬁcation in a Semantic Form representation (incomplete version): a BECOME(LOC) predicate may not be further speciﬁed by another BECOME(LOC) predicate.

(5.42) lays bare just how non-motion-speciﬁc the core of this constraint really is – take away the ‘LOC’ parts of the predicates in (5.42) and we have the most general of constraints on Possible Verb events, that these contain at most one telic culmination, as encoded by BECOME. But in motion and non-motion events alike, single telic culminations will sometimes be encoded by two BECOME predicates in LDG representations, in contradiction to (5.42):

(5.43) Jane went [to London] [to the London Eye]. λx λs {GO(x) & BECOME(LOC(x,(AT(London)))) & BECOME(LOC(x,(AT(London Eye))))}(s)

6This constraint could be matched with a Single Starting Point Constraint on Possible Verb motion events, that a Possible Verb motion event have a single, unique starting point. This section concentrates on the Single Endpoint Constraint, however, because endpoints, or result locations, align with the motion-speciﬁc result component of an event. CHAPTER 5. MOTION SVCS 252

(5.44) The earth froze hard. λx λs {BECOME(FROZEN(x)) & BECOME(HARD(x))}(s)

This reflects the fact that (5.42) is only an incomplete version of the Single Endpoint Constraint in terms of further specification; what is further required is a qualification of the conditions under which one BECOME(LOC) predicate can further specify another. With respect to the non-motion example in (5.44), the reason why the first BE- COME predication can be further specified by a second BECOME predication is that this second predication does not encode a distinct telic culmination or change, but instead only provides additional information about the change encoded by the first BECOME predication. The information that the earth becomes hard further elabo- rates upon the main change of state information that it becomes frozen. For (5.43) it is much the same story; BECOME(LOC(x,AT(London Eye))) provides additional, more fine-grained information that further specifies the result location encoded in BECOME(LOC(x,AT(London))). The (longer version of the) Ground Arguments Constraint in terms of further specification in (5.36) moreover specifies the exact conditions under which one such BECOME goal predicate can further specify another BECOME goal predicate: when the region denoted by the goal argument of the former spatially contains the point or region denoted by the goal argument of the latter. So, this goal-specific subconstraint of the Ground Arguments Constraint in (5.36) constitutes a first qualification to the Single Endpoint Constraint in terms of further specification in (5.42); now the other qualifications to this constraint have to be formulated, which pertain to the possible further specifying combinations of BECOME predicates which are not both goal predicates, and so are not covered by the Ground Arguments Constraint. (5.45) lists all of the predicates containing BECOME from the (incomplete) list in (5.27); there are two goal predicates, three route predicates, and one, final source predicate:

(5.45) (y is goal) BECOME(LOC(x,AT(y))), BECOME(LOC(x,IN(y))) CHAPTER 5. MOTION SVCS 253

(y is route) BECOME(LOC(x,PAST(y))), BECOME(LOC(x,ACROSS(y))), BECOME(LOC(x,THROUGH(y))) (y is source) BECOME(LOC(x,¬AT(y))), BECOME(LOC(x,¬IN(y)))

Of these predicates, the goal predicate BECOME(LOC(x,AT(y))) is distinguished by the fact that it is the only one to actually specify exactly what (or where) the endpoint or result location is. In the following sentence, London is the endpoint (even though it may only be conceptualized as a point at quite coarse a grain):

(5.46) Jane went to London. λx λs {GO(x) & BECOME(LOC(x,(AT(London))))}(s)

BECOME(LOC(x,IN(y)) is the other predicate in (5.45) whose internal argument specifies a goal, although this argument does not actually specify where the endpoint is – it only specifies the region, y, that contains the endpoint. Similarly, the route predicates BECOME(LOC(x,PAST(y))), BECOME(LOC(x,ACROSS(y))) and BECOME(LOC(x,THROUGH(y))) also do not specify exactly where the endpoint is. Instead, BECOME(LOC(x,PAST(y))) specifies that the endpoint lies in the region beyond the route argument y that is passed, BECOME(LOC(x,ACROSS(y))) specifies that the endpoint lies in the region beyond the route argument y that is crossed, and BECOME(LOC(x,THROUGH(y))) specifies that the endpoint lies in the region beyond the route argument y that is gone through. Of the source predicates, BECOME(LOC(x,¬IN(y))) does not specify what or where the endpoint is, but it does specify that it lies in the region outside y, the source argument. BECOME(LOC(x,¬AT(y))), in contrast specifies nothing about the location of the endpoint (other than it is not at y), although it does specify the exact location of the starting point, y. Because BECOME(LOC(x,IN(y))), BECOME(LOC(x,PAST(y))), BECOME (LOC(x,ACROSS(y))), BECOME(LOC(x,THROUGH(y))) and BECOME(LOC(x, ¬IN(y))) do not specify where the endpoint is, but only specify the region within CHAPTER 5. MOTION SVCS 254

which the endpoint is located, these predicates can all be further specified by a BE- COME(LOC(x,AT(y))) predicate, which does specify where the endpoint is.7 This presents one of the conditions under which one BECOME(LOC) predicate can further specify another. In (5.47), for example, the (‘exit’) predicate BECOME(LOC(x,¬IN (house))) specifies that the endpoint lies within the region outside the house, and this is further specified by BECOME(LOC(x,AT(church))), which specifies that the endpoint within this region is located at the church:

(5.47) Gideon went out of the house to the church. λx λs {GO(x) & BECOME(LOC(x,¬IN(house))) & BECOME(LOC(x,AT(church)))}(s)

• •

In (5.48), BECOME(LOC(x,ACROSS(road))) specifies that the endpoint lies within the region across the road, and this is further specified by BECOME(LOC(x,AT(mountain))), which specifies that the endpoint within this region is located at (the foot of) the mountain:

(5.48) Harry ran across the road to the mountain. λx λs {RUN(x) & BECOME(LOC(x,ACROSS(road))) & BECOME(LOC(x,AT(mountain)))}(s)

•

The church in (5.47) and the mountain in (5.48) must be located not very far from the house and the road respectively. This reﬂects the fact that the regions within which

7BECOME(LOC(x,¬AT(y))), which speciﬁes nothing about the endpoint, can also be further speciﬁed by a BECOME(LOC(x,AT(y))) predicate. CHAPTER 5. MOTION SVCS 255

these endpoints are located are pragmatically construed to encompass, not the entire universe outside the house or some vast area beyond the road, but instead more limited areas within the immediate vicinity of the house or the road. Similar comments pertain to the regions within which the starting points for these predicates must be located, for BECOME(LOC(x,¬IN(y))), BECOME(LOC(x,ACROSS(y))), BE- COME(LOC(x,IN(y))), BECOME(LOC(x,PAST(y))) and BECOME(LOC(x, THROUGH(y))) are all what could be called ‘region-to-region’ predicates, which denote paths from starting points to endpoints within these respective regions. While the region that contains the starting point in a BECOME(LOC(x,¬IN(y))) (‘exit’) predicate is straightforwardly interpreted as the containing argument y (which in (5.47) is the house), the region that contains the starting point in a BECOME(LOC(x, ACROSS(y))) predicate is pragmatically construed as that area within the immediate vicinity of y, that falls directly before the boundary of y in terms of temporal and spatial progress. (5.49a) indicates the location of this region (along with the region that contains the endpoint) for the predicate BECOME(LOC(x,ACROSS(road))), while (5.49b) indicates the locations of the regions containing the starting point and endpoint for the predicate BECOME(LOC(x,THROUGH(tunnel))):

(5.49) a. Harry crossed the road. b. Frank walked through the tunnel. λx λs {GO(x) & BECOME λx λs {WALK(x) & BECOME (LOC(x,ACROSS(road)))}(s) (LOC(x,THROUGH(tunnel)))}(s)

REGION CONTAINING REGION CONTAINING STARTING POINT ENDPOINT

• • REGION CONTAINING REGION CONTAINING STARTING POINT tunnel ENDPOINT • •

Being able to draw a path from somewhere within the starting point-containing region to somewhere within the endpoint-containing region for a region-to-region predicate (as I have done in (5.49a) and (5.49b)) is a useful illustrative aid for explaining the conditions under which one region-to-region predicate can further specify another. I now turn to these conditions. (5.50a) is an example that involves further speciﬁcation of one region-to-region predicate by another; as a clause headed by a single verb this is a Possible Verb CHAPTER 5. MOTION SVCS 256

structure, although the Possible Verb interpretation can only obtain under the condition that the garden lies directly outside the door leading out of the house (as do most gardens at most houses):

(5.50) a. Gideon went out of the house into the garden. λx λs {GO(x) & BECOME(LOC(x,(¬IN(the house)))) & BECOME(LOC(x,IN(garden))))}(s) b. c. BECOME(LOC(x,(¬IN(the house)))) BECOME(LOC(x,(¬IN(the house)))) & BECOME(LOC(x,IN(garden))))

• • GARDEN • • BECOME(LOC(x,IN(garden)))) GARDEN

If the garden lies directly outside the house in (5.50a), then the region-to-region BECOME(LOC(x,(¬IN(the house)))) (‘exit’) path out of the house spatially contains or wholly overlaps the region-to-region BECOME(LOC(x,IN(garden)))) (‘enter’) path into the garden, as diagrammed in (5.50b). And because the BECOME(LOC(x,¬IN (house))) path spatially contains or wholly overlaps the BECOME(LOC(x,IN(garden)))) path, the former can be further speciﬁed by the latter in a Possible Verb structure, for the reason that this second path does not cover any new territory or provide any new locational information (but instead just provides a diﬀerent perspective on the same locational information). (5.50c) illustrates how these two pieces of path information are amalgamated into a single path in a Possible Verb structure with a single starting point and a single endpoint. In contrast to (5.50a), the Possible Verb structure in (5.51a) is far from acceptable, and this turns upon the fact that, in most possible worlds, where churches do not lie directly outside houses, the region-to-region BECOME(LOC(x,(¬IN(house))))) (‘exit’) path out of the house is unlikely to spatially contain or wholly overlap the region-to-region BECOME(LOC(x,(IN(church)))) (‘enter’) path into the church, as depicted in (5.51c): CHAPTER 5. MOTION SVCS 257

(5.51) a. ?Gideon went out of the house into the church. ?λx λs {GO(x) & BECOME(LOC(x,(¬IN(house)))) & BECOME(LOC(x,(IN(church))))}(s) (Possible Verb structure) b. Gideon went out of the house and into the church. λx λs [{GO(x) & BECOME(LOC(x,¬IN(house)))} & {MOVE(x) & BECOME(LOC(x,IN(church)))}](s) (Impossible Verb structure) c. BECOME(LOC(x,(¬IN(house))))

BECOME(LOC(x,IN(church))) • • • • (Endpoint 1) (Endpoint 2)

The path denoted by BECOME(LOC(x,IN(church))) cannot further specify that denoted by BECOME(LOC(x,(¬IN(house)))) if, as is most likely and as (5.51c) depicts, these paths occupy different locations, meaning that the BECOME(LOC(x,(¬IN (house)))) (‘exit’) path does not spatially contain or wholly overlap the BECOME (LOC(x,IN(church))) (‘enter’) path. (5.51c) depicts two spatially disjoint paths instead of a single, continuous path, and this diagram also draws attention to the fact that these paths have two distinct endpoints (as well as two distinct starting points). This gives a clear visual demonstration of how any expression for which (5.51c) represents a possible or likely extension is not going to satisfy the Single Endpoint Constraint on Possible Verb motion events. The claim that the path denoted by a region-to-region (BECOME) predicate has to spatially contain that denoted by a region-to-region (BECOME) predicate that further specifies it is included as the third qualification (c) in the (for now) complete version of the Single Endpoint Constraint in terms of further specification:

(5.52) The Single Endpoint Constraint in terms of further speciﬁcation in a Semantic Form representation (complete version):

BECOME(LOC(x,PRED1(y)) may not be further speciﬁed by BECOME(LOC(x, PRED2(z)) unless: CHAPTER 5. MOTION SVCS 258

(a) BECOME(LOC(x,PRED1(y))) and BECOME(LOC(x,PRED2(z))) are both goal predicates and the region denoted by the goal argument y spatially contains that denoted by z. (Goal-speciﬁc subconstraint of the Ground Arguments Constraint)

(b) BECOME(LOC(x,PRED1(y))) is a ‘region-to-region’ source/route predicate and the further specifying predicate is BECOME(LOC(x,AT(z))) (entailing that the ‘endpoint region’ connected to

BECOME(LOC(x,PRED1(y))) spatially contains the endpoint, z).

(c) BECOME(LOC(x,PRED1(y))) and BECOME(LOC(x,PRED2(z))) are both ‘region-to-region’ predicates, where the path denoted by the former spatially contains or wholly overlaps that denoted by the latter.

Although this qualification in (c) needs to be tested with more extensive data and possibly refined, one thing that is clear by now is that the instantiation of multiple verbs that have the semantics of region-to-region predicates is a red flag for Impossible Verb status in motion SVCs. If the grounds to which the predicates refer are most likely situated far apart, meaning that the two region-to-region paths to which the predicates refer do not overlap, or are not in a relationship of spatial containment, this almost certainly signifies Impossible Verb status. Section 5.4 will consider two Thai SVCs containing two region-to-region BECOME predicates which most likely have Impossible Verb semantics.

5.2.2.5 The Continuity Constraint on Possible Verb motion events

The ﬁnal constraint on Possible Verb motion events discussed here is:

(5.53) The Continuity Constraint on Possible Verb motion events: the path in a Possible Verb motion event must be spatially continuous.

Simple, and even quite obvious, this constraint rules that a Possible Verb motion event cannot consist of spatially disjoint paths, as in Joe arrived at Waterloo Station and took a train north from Kings Cross, where Joe has to have travelled from Waterloo to Kings Cross in some way not speciﬁed in this sentence. CHAPTER 5. MOTION SVCS 259

The Continuity Constraint is similar to Matsumoto’s (1996) Path Condition:

(5.54) The Path Condition: Path arguments of a verb (source, route and goal) must represent aspects of a single path or a spatially continuous sequence of paths. (Matsumoto 1996, 255)

The main respect in which the Continuity Constraint diﬀers is that it is more general in not referring to the arguments of a verb. The constraint in (5.53) is not easily recast in terms of further speciﬁcation in a LDG Semantic Form representation. This is because it is not really a constraint on what is coded, but is instead a more general extensional condition on Possible Verb motion events. The extension of a motion event (or at least of its non-manner, path component) can be visualized or represented as a line on a two-dimensional plane. Quite simply, what the Continuity Constraint on Possible Verb motion events requires is that this line be unbroken, as in (5.55), and not ‘in pieces’, as in (5.56):

(5.55) • •

(5.56)

• •

The concept of path continuity could be described in the following, Semantic Form- independent terms:

(5.57) Deﬁnition of a continuous path:

If P0 is the startpoint for a path and Pn is the endpoint, the path is

continuous iﬀ for any points on the path Px, Pz where 0 ≤ x < z ≤ n, there

exists a point on the path Py, where x < y < z.

While there may be no obvious way of constraining a Semantic Form encoding of a motion event in such a way that the property of path continuity described in (5.57) CHAPTER 5. MOTION SVCS 260

is guaranteed, the final point to mention in this section is that this may not be wholly undesirable from the perspective that not all motion events need be of the translocational, durative type. For while the focus of this chapter has been stated to be events of translational or translocational motion, which involve continuous, durative displacement of a figure along a trajectory, Levinson and Wilkins (2006) draw attention to the fact that some languages appear to describe motion events with non-durative temporal shapes and/or discontinuous spatial shapes. Yukatek, for example, describes motion events with what Levinson and Wilkins (2006) (and Bohnemeyer and Stolz (2006)) refer to as change of location semantics. Instead of involving the transition through an indefinite series of points in space over time, change of location motion events involve a single change of (locative) state, with no transitional phases. Because change of location motion events are characterized by the absence of a trajectory which is itself characterized by the property of spatial continuity, these types of motion events fall outside the scope of the Conti- nuity Constraint on Possible Verb motion events. However, the only further point I would add is that it does seem more appropriate for a work on motion SVCs to focus on translocational, durative motion events, with their continuous spatial trajectories. This is because motion SVCs necessarily involve the conflation of more than one type of motion information (whether it be manner+ground-fixture, manner+direction or direction+ground-fixture information), whereas it is not clear at this stage whether non-durative, change of location motion events can conflate more than one type of motion information. As Levinson and Wilkins remark, in change of location motion events “details of the trajectory, the manner of motion, the medium and the instruments involved are out of focus” (2006, 531), and Yukatek is once again used as an example, in this case of how manner of motion cannot be fused with change of location semantics. Future research will no doubt reveal the extent and range of motion information that can be conflated in a change of location motion event. CHAPTER 5. MOTION SVCS 261

5.2.3 Event structural proﬁles of Possible Verb and Impos- sible Verb motion SVCs

This section turns the spotlight directly onto motion SVCs, setting out the event structural profiles for all types of SVC that describe events of translocational motion. The first two profiles are for types of Possible Verb motion events:

(5.58) Event structural proﬁle for unitary-event motion SVCs:

SF: ... λs {V1 & V2 ( ... & Vn)}(s)

CS: V1 & V2 ( ... & Vn) corresponds to a (path) event s. There are no additional subevents.

(5.59) Event structural proﬁle for coidentiﬁed motion SVCs:

V1 represents a manner of motion verb, others are path verbs.

SF: ... λs {V1 & V2 ( ... & Vn)}(s)

CS: V1 corresponds to a (manner) subevent s1 and V2 ( ... & Vn) to a (path)

subevent s2, and these two subevents are coidentiﬁed (temporally dependent).

tc1=temporaltc2 and tt1=temporaltt2.

V1, V2, ( ... Vn) in (5.58) all correspond to verbs with path of motion semantics – either direction semantics (MOVE-NORTH, MOVE-SEAWARDS, DESCEND, MOVE- AWAY-FROM-DC, etc.) or ground-fixture semantics (BECOME(LOC(x,AT(y)), BECOME(LOC(x,IN(y))), MOVE-VIA(x,y), BECOME(LOC(x,¬AT(y))), etc.). These describe different aspects of a single, continuous path, and must jointly satisfy the Unique Vector Constraint of Section 5.2.2.2, the Ground Arguments Constraint of Section 5.2.2.3, and the Single Endpoint Constraint of 5.2.2.4. (5.60) is a unitary- event motion SVC that fits the template in (5.58):

(5.60) a. Er´ı, p´a mu-mi. 3sg exit go-past ‘He went out.’ (Ijo; Williamson 1963, 151) b. λy λx λs {MOVE-AWAY-FROM-DC(x) & BECOME(LOC(x,¬IN(y)))}(s) CHAPTER 5. MOTION SVCS 262

The event structural proﬁle in (5.59) adds an extra element to a Possible Verb motion event: a verb with the manner of motion semantics represented by V1, of which there can be only one, in fulﬁllment of the Manner of Motion Constraint of Section 5.2.2.1. Because this manner of motion component corresponds to a distinct

subevent, ‘measured out’ by the path subevent denoted by V2 ( ... & Vn), motion SVCs of this type have coidentiﬁed semantics. An example is:

(5.61) a. Ana kea rama lu’u.na d’ei kota. baby crawl enter.3 in box ‘The baby crawled into the box.’ (Bima) b. λx λs {CRAWL(x) & BECOME(LOC(x,IN(the box)))}(s)

While the manner of motion verb in (5.61) is intransitive, (5.62) contains a transitive manner of motion verb:

(5.62) a. wa’u mpara cumpu, randa ka.lu’u kai.na wa’i already then ﬁnished drag caus.enter kai.3 old-woman ede dem ‘once he’d ﬁnished (digging her grave), he dragged the old woman into it’ (Bima) b. SF: λy λx λs {DRAG(x,y) & BECOME(LOC(y,IN(the grave)))}(s)

c. CS: Two coidentiﬁed subevents: s1 of dragging the old woman and s2 of

the old woman entering the grave. tc1=temporaltc2 and tt1=temporaltt2. Because randa ‘drag’ denotes continuously imparted force (Levin 1993), (5.62) has the coidentified semantics profiled in (5.59), but if a transitive manner of motion verb in a motion SVC denotes instantaneously caused ballistic motion, as in (5.63), then the SVC will fit the non-coidentified resultative profile of (5.64):

(5.63) a. A sikópu d´ı bálu gó a d´ı wósu. 3sg kick det ball go loc det house ‘He kicked the ball into the house.’ (Saramaccan; Veenstra 1996, 70) b. SF: λy λx λs {KICK(x,y) & BECOME(LOC(y, IN(the house)))}

c. CS: Two subevents, s1 of the actor kicking the ball and s2 of the ball

entering the house, where s1 causes s2. tc1

(5.64) Event structural proﬁle for non-coidentiﬁed resultative motion SVCs:

SF: ... λy λx λs {V1(x,y) & V2(y, ...) ( ... & Vn(y, ...))}(s)

CS: V1 corresponds to a (manner) subevent s1 and V2 ( ... & Vn) to a (path)

subevent s2, where s1 causes s2. Either tc1=temporaltc2 or tc1

Finally, (5.65) presents an event structural proﬁle for Impossible Verb motion SVCs:

(5.65) Event structural proﬁle for Impossible Verb motion SVCs:

V1 and W1 represent manner of motion verbs, others are path verbs.

SF: ... λs [{(V1 &) V2 ( ... & Vn)} & {(W1 &) W2 ( ... & Wn)}](s)

CS: (V1 &) V2 (... & Vn) corresponds to a Possible Verb subevent s1 (itself

possibly composed of subevents) and (W1 &) W2 (... & Wn) corresponds to a

Possible Verb subevent s2 (also possibly composed of subevents).

tc1

Hidden from (5.65) are the Predicate Combining Operations an Impossible Verb motion SVC has to have undergone. Predicate combining operations will be set out for some of the Impossible Verb motion SVCs in the case studies of Sections 5.3, 5.4 and 5.5. An example of an Impossible Verb motion SVC is:

(5.66) a. Er´ı, [ban.i mu] [w´e. ni.]-mi.. 3sg run go walk-past ‘He went running and then walked.’ (Ijo.; Williamson 1963, 152, fn.2) b. SF: λx λs [{RUN(x) & MOVE-AWAY-FROM-DC(x)} & {WALK(x) & MOVE(x)}](s)

c. CS: Two subevents, s1 of the actor running away and s2 of the actor

walking. tc1

5.2.4 Conclusion, and Nigerian Pidgin English example case

While the event structures of Section 5.2.3 provided the first tangible illustrations of motion SVCs and how they fit the Possible Verb and Impossible Verb structures with which we are already familiar, the constraints on Possible Verb motion events in Section 5.2.2 supplied the gritty details of how a motion SVC might be identified as Possible Verb or Impossible Verb in practice. These constraints jointly determine that a Possible Verb motion event could include at most one manner of motion component, some indefinite number of direction components (providing they jointly satisfy the Unique Vector Constraint), one source ground-fixture component (possibly composed of ‘disconstituents’), one route ground-fixture component (possibly composed of ‘disconstituents’), and one goal ground-fixture component (possibly composed of ‘disconstituents’). Do serializing languages express all of these components with verbs in a motion SVC? Hardly. Quite apart from the question of whether each of these components can actually be expressed in verbal form, the fact of the matter is that languages tend to be highly restrictive in terms of how many, and what type, of motion components can be expressed in their motion SVCs. The verbs expressing these components also appear in a variety of different orders. Consider the case of Nigerian Pidgin English (Dillon 2004). This language can express at most three verbs in its motion SVCs: a manner of motion verb, a deictic verb and ground-fixture verb (in this order only), as in (5.67)-(5.68):

(5.67) Di pickin waka go pass di house. def child walk go pass def house ‘The child walked past the house [moving away from the speaker].’ (Nigerian Pidgin English; Dillon 2004, 13)

(5.68) Di pickin waka come pass di house. def child walk come pass def house ‘The child walked past the house [moving towards the speaker].’ (Nigerian Pidgin English; Dillon 2004, 13)

(5.69) is a template for Nigerian Pidgin English motion SVCs, where each component is optional: CHAPTER 5. MOTION SVCS 265

Manner of motion Deictic Path (Ground-ﬁxture) (5.69) waka, run, jump go, come pass, comot, reach

The order of the components in this template happens to match the order of the motion components discussed in Section 5.2.1: manner of motion, direction of motion, then ground-fixture information. These components also appear in this order in all of the Semantic Form representations so far, although nothing necessitates this ordering. Sections 5.3, 5.4 and 5.5 establish whether the motion SVC templates for Ijo, Thai and Bima are similarly ordered, as well as investigating how these templates organize and restrict the information these languages can convey in their motion SVCs. But the case studies of these sections also investigate whether these languages can do something with their motion SVCs that Nigerian Pidgin English apparently cannot: describe Impossible Verb motion events. This section identified some of the ‘red flags’ for Impossible Verb status in motion SVCs: multiple manner of motion verbs, direction specifications that do not conform to the Unique Vector Constraint, realization of more than one (non-disconstituent) source, route or goal, and specification of more than one endpoint. The Ijo, Thai and Bima case studies seek out these red flags, as well as testing the observation put forward in Section 5.1, that Impossible Verb motion SVCs tend to result through iterations of the motion SVC template in the language concerned.

5.3 Ijo motion SVCs

While the Nigerian Pidgin English motion SVC template presented in Section 5.2.4 was of a rather uncomplicated and tidy nature,8 the Ijo motion SVC template set out in this section is anything but. Ijo has one of the most idiosyncratic and eccentric motion SVC templates, one facet of which is the possibility of path+manner word order in very restricted circumstances. Path+manner word order is relatively rare in motion SVCs, although these restricted Ijo path+manner SVCs sport the further distinction of having been used to argue against temporally iconic ordering of verbs

8A tidiness that could, however, reﬂect the limited amount of data collected. CHAPTER 5. MOTION SVCS 266

within an SVC by Carstens (2002). In this section I demonstrate why these examples do not actually constitute violations of what was formulated as the No Inverse Iconic- ity Principle in Section 3.5.2 (repeated later here), and I also suggest that this kind of data can only be properly explained through the kind of event complexity analysis I am adopting here. But the main focus of this section is the Ijo motion template, culminating in the first, although very limited, illustration of how Impossible Verb motion semantics results through its iteration. The case studies of this section and the next two sections present some of the first SVC data in support of the tendency, first described in Section 5.1, for Impossible Verb semantics to result through the iteration of language-specific motion SVC templates. (5.70) presents the core facts of the Ijo motion SVC template, extracted and compacted from Williamson (1963) (the rare instances of path+manner word order alluded to above – in which the path verb dùo. ‘go through’ features – clearly do not fit this template, but will be illustrated later):

(5.70) MANNER PATH DEIXIS

iy´eri ‘descend’

bà.i ‘run’ dèsi ‘jump’ bó ‘come’ pá ‘exit’ ok´ı ‘swim’ mú ‘go’ wé.ni. ‘walk’ yò. u. ‘paddle’ bó. ‘pass’ (intr.) wai.´ı ‘turn’ té.´ı ‘escape’ bó. ‘pass’ (tr.) akana ‘encircle’ bé.in ‘cross’

The ﬁrst, general observation arising from (5.70) is that, in contrast to the Nigerian Pidgin English motion SVC template, the deictic component is ordered after the path component (mainly comprised of ground-ﬁxture verbs). As mentioned in Section CHAPTER 5. MOTION SVCS 267

5.2.1, template-ﬁnal deictic positioning is cross-linguistically very common, and will in fact be encountered again in the Thai case study of the next section. Some examples that conform to the template in (5.70), or at least to the left- and bottom-most boxes that contain manner and particular path verbs, are:

(5.71) Er´ı, yo. u. te..i-m´ı.. MANNER + PATH (intr.) 3sg paddle escape-past ‘He escaped by paddling.’ (Ijo; Williamson 1963, 151)

(5.72) Er´ı, w´e. ni. b´o. -mi.. MANNER + PATH (intr.) 3sg walk pass-past ‘He passed walking.’ (Ijo; Williamson 1963, 151)

The path verbs in (5.71) and (5.72) are intransitive according to Williamson (1963). They appear to be ground-ﬁxture verbs, but where the ground arguments are covert. (P´a ‘exit’, the path verb in the center of (5.70), has the same status.) The path verbs in (5.73)-(5.74), on the other hand, are transitive, and their ground arguments are realized as objects before the manner+path verb combination, in these contiguous SVCs:9

(5.73) Er´ı, w´ari. w`e. ni.-ni., akana-m´ı.. object + MANNER + PATH (tr.) 3sg house walk-lnk encircle-past ‘He walked round the house.’ (Ijo; Williamson 1965, 51)

(5.74) To. bo. u. -b´ı., k´ı.mi. b`a.i, b´o. -yemi. object + MANNER + PATH (tr.) child-def man run pass-progr ‘The child is running past the man.’ (Ijo; Williamson 1965, 51)

As the template in (5.70) indicates, none of (5.71)-(5.74) can realize one of the deictic verbs, b´o ‘come’ or m´u ‘go’, after the path verb; these SVCs are limited to two verbs. The top- and right-most boxes in (5.70) also represent SVCs limited to two verbs, but here it is an initial manner of motion verb that is not permitted. (5.75) is one (of the two) possible verb combinations permitted:

9 The linker ni. that appears in (5.73) (and also (5.85) below) is semantically empty and is introduced after a verb or m´a ‘then’ when the next unit begins with a vowel (Williamson 1965, 53). CHAPTER 5. MOTION SVCS 268

(5.75) Er´ı, iy´eri mu-mi. PATH (intr.) + DEIXIS 3sg descend go-past ‘He went down.’ (Ijo; Williamson 1963, 151)

Lastly, taking the ‘middle route’ through (5.70) can produce combinations of three verbs in an Ijo motion SVC, as in (5.76)-(5.77), although something not indicated in (5.70) is the fact that any one of these verbs is optional, so that (5.78), (5.79) and (5.80) are all possible:

(5.76) Er´ı, ba.i pa b´o-mi. MANNER + PATH (intr.) + DEIXIS 3sg run exit come-past ‘He came running out.’ (Ijo; Williamson 1963, 152)

(5.77) Er´ı, yo. u. pa m´u-mi. MANNER + PATH (intr.) + DEIXIS 3sg paddle exit go-past ‘He went paddling out.’ (Ijo; Williamson 1963, 152)

(5.78) Er´ı, ba.i pa-m´ı.. MANNER + PATH (intr.) 3sg run exit-past ‘He came/went out running.’ (Ijo; Williamson 1963, 151)

(5.79) Er´ı, desi bo-m´ı. MANNER + DEIXIS 3sg jump come-past ‘He came jumping.’ (Ijo; Williamson 1963, 151)

(5.80) Er´ı, p´a mu-mi. PATH (intr.) + DEIXIS 3sg exit go-past ‘He went out.’ (Ijo; Williamson 1963, 151)

All of the SVCs in (5.71)-(5.80) have Possible Verb semantics. Or at least, all of the translations for (5.71)-(5.80) are suggestive of Possible Verb semantics, where the manner of motion activity is measured out by the path subevent (and multiple path verbs describe aspects of a single path). None of the translations suggests the Impos- sible Verb semantics of one motion subevent temporally succeeded by another. So (5.79) and (5.80) directly above, for example, will have the Semantic Form representations in (5.81a) and (5.82a) respectively, and not the Impossible Verb representations in (5.81b) and (5.82b): CHAPTER 5. MOTION SVCS 269

(5.81) a. λx λs {JUMP(x) & MOVE-TOWARDS-DC(x)}(s) +sc NOT: b. λx λs [{JUMP(x)} & {MOVE-TOWARDS-DC(x)}](s) +sc (‘He jumped and (then) came.’) (Impossible Verb structure)

(5.82) a. λx λs {MOVE-AWAY-FROM-DC(x) & BECOME(LOC(x,¬IN(y)))}(s) +sc NOT: b. λx λs[{BECOME(LOC(x,¬IN(y)))}&{MOVE-AWAY-FROM-DC(x)}](s) +sc (‘He went out and (then) went’) (Impossible Verb structure)

These SVCs satisfy the Theta Structure Property for SVCs because their Theta Struc- tures each consist of a single structural argument, as do the Theta Structures of many clauses headed by single verbs. The issue of whether particular Ijo SVCs can be confirmed to have coidentified Possible Verb semantics bears special relevance to the idiosyncratic behavior of the path verb dúo ‘go through’, and the question of whether SVCs containing this verb, such as (5.83), violate the No Inverse Temporal Iconicity Principle in (5.84):

(5.83) Er´ı ben´ı d`uo yo. u. -m´ı.. he water go-through paddle.past ‘He paddled through the water.’ (Ijo., Williamson 1965, 49) λx λs {PADDLE(x) & MOVE-THROUGH(x,water)}(s) NOT: λx λs [{PADDLE(x)} & {MOVE-THROUGH(x,water)}](s) (‘He paddled and (then) went through the water.’) (Impossible Verb structure)

(5.84) No Inverse Temporal Iconicity:

If tci

Because d´uo and its preceding locative object appear before the manner of motion verb, this path verb does not conform to the template in (5.70). Carstens (2002) further claims that (5.83) proves that SVCs are not temporally iconic since, according to her, the paddling has to temporally precede going through the water, counter to CHAPTER 5. MOTION SVCS 270

the path+manner word ordering. But this is a causal force interpretation of (5.83) (comparable to the causal force interpretations of instrumental events described in Chapter 4), where the actor is understood to initiate the action of paddling, the force of which then results in movement through the water in the causal chain. As mentioned many times before, this dissertation utilizes an ‘event complexity’ semantic approach which identifies subevents through their temporal coordinates. Under this approach, the subevents of paddling and going through the water have identical temporal coordinates; they are coidentified in such a way that the path through the water measures out the paddling. Because the two subevents have identical times of commencement, (5.83) will not violate the No Inverse Temporal Iconicity Principle, since this principle claims nothing about the word orders for SVCs whose subevents share the same time of commencement. Although there is not the space to dwell on this issue further, I would argue that adopting this event complexity approach, along with the No Inverse Temporal Iconicity Principle, can better account for how temporal iconicity comes into play in SVCs than any approach to event semantics based on interpretations of causal force. While the initial position of the path verb dúo constitutes one unpredicted type of complication to the Ijo motion template in (5.70), another complication is supplied by SVCs such as (5.85) and (5.86):10

(5.85) [MANNER + DEIXIS] + [object + PATH (tr.)]

Er´ı, w´e. ni. bo-ni, am´a la-mi.. 3sg walk come-lnk town reach-past ‘He came walking and reached a town.’/‘He came walking into town.’ (Ijo; Williamson 1963, 152)

10Williamson gives (5.85) and (5.86) each two translations; the first with what looks like Impossible Verb semantics, the second with Possible Verb semantics. It is difficult to know what significance could or should be attached to this. CHAPTER 5. MOTION SVCS 271

(5.86) [MANNER + DEIXIS] + [object + PATH (tr.)]

Er´ı, ok´ı mu to. ru. be..in-mi.. 3sg swim go river cross-past ‘He went swimming and crossed a river.’/‘He swam across a river.’ (Ijo; Williamson 1963, 152)

These examples instantiate the ﬁrst and last (manner and deictic) ‘slots’ (or boxes) of the template, but following these (predictable) verb combinations are transitive path verbs, preceded by their objects, resulting in non-contiguous SVC structure. Of these two transitive path verbs, be..in ‘cross’ already appeared in the large group of path verbs (bottom-most box) in (5.70), and it can alternatively be instantiated in a contiguous SVC, with its object realized before the manner verb:

(5.87) Ofóni-bi, tó. ru. f`ın, bé. in-mi.. object + MANNER + PATH (tr.) bird-def river fly cross-past ‘The bird flew across the river.’ (Ijo; Williamson 1965, 51)

(5.70) speciﬁes that a deictic verb cannot be instantiated after the combination of a manner verb and a path verb such as be..in ‘cross’, so, from this perspective it is possible to view (5.86) as a kind of solution to the problem of realizing all three (manner, path and deictic) verbs in one sentence. But that is probably not the end of the story, and there may be further complications from other verbs displaying idiosyncratic behaviors besides. Clearly, more data is required to give a more complete picture – Williamson (1963) is after all only a four and half page paper (supplemented by a small amount of data from Williamson (1965)). However, the template in (5.70) does demonstrate that there is at least some degree of systematicity in the Ijo motion SVC template, and Williamson’s (1963, 1965) data do also suggest that the Ijo motion template encapsulates Possible Verb semantics. In contrast, (5.88), the one SVC Williamson (1963) supplies that deﬁnitely does not conform to the template, but must represent a sequence of two templates, has Impossible Verb semantics, supporting the idea put forward in Section 5.1 that iteration of motion SVC templates is a standard structural pattern for Impossible Verb motion SVCs: CHAPTER 5. MOTION SVCS 272

(5.88) Er´ı, [ban.i mu] [w´e. ni.]-mi.. 3sg run go walk-past ‘He went running and then walked.’ (Ijo.; Williamson 1963, 152, fn.2)

Ban.i mu ‘run go’ in (5.88) presents one instantiation of the Ijo motion template, and w´e. ni. ‘walk’ one further, single-verb instantiation (all components of the template being optional). (5.88) must have Impossible Verb semantics because, in containing two manner of motion verbs, it violates the Manner of Motion Constraint on Pos- sible Verb Motion Events, a ‘red ﬂag’ for Impossible Verb status. (5.89) provides the Semantic Form representation for this SVC, including the Predicate Combining Operations that it must, as an Impossible Verb SVC, undergo:

(5.89) ban.i mu w´e.ni.: λx λs1 {RUN(x) & MOVE-AWAY-FROM-DC(x)}(s1) +sc

& λu λs2 {WALK(u) & MOVE(u)}(s2) +sc

i. Event combining: s1 + s2 = s ii. Highest structural argument coidentiﬁcation: x=u

Combined predicate: λx λs [{RUN(x) & MOVE-TOWARDS-DC(x)} & {WALK(x) & MOVE(x)}](s) +sc

This SVC satisfies the Theta Structure Property for SVCs straightforwardly, because its Theta Structure consists of just a single structural argument. But the more significant aspect of (5.88) is that it bears out the tendency described in Section 5.1 for Impossible Verb semantics to result through the iteration of motion SVC templates. Sections 5.4 and 5.5 next investigate whether this tendency is also borne out in Thai and Bima motion SVCs respectively, although as a precondition to these investiga- tions it is necessary to establish, first, what the Thai and Bima motion SVC templates actually look like, and second, whether these languages really do have any Impossible Verb motion SVCs. CHAPTER 5. MOTION SVCS 273

5.4 Thai motion SVCs

Thai motion SVCs have received considerably more coverage than those of Ijo in the linguistics literature, and this section draws on information from Thepkanjana (1986), Muansuwan (2002), Wechsler (2003), and Zlatev and Yangklang (2004), among other sources. The picture that emerges is of a less idiosyncratic template than Ijo, and one that mainly corresponds to Possible Verb semantics, although it may also be the case that Impossible Verb semantics are ‘let in through the back door’, through the possibility of instantiating multiple region-to-region BECOME predicates that do not satisfy the Single Endpoint Constraint on Possible Verb motion events described in Section 5.2.2.4. Multiple iterations of the template also result in Impossible Verb semantics, although it is suggested by some that these structures are no longer monoclausal, meaning that they are not SVCs. As in the previous case study, the ﬁrst object must be to establish the motion SVC template, starting with Thepkanjana’s (1986, 136) presentation:

(5.90) Thepkanjana (1986)’s Thai motion SVC template:

1 2 3 4 5 Manner of Geometric Direction with Direction with Direction with motion shape of respect to the respect to the respect to speech the path previous path outside world act participants

(Deictic)

dn ‘walk’ tro ‘go j n ‘reverse’ l j ‘pass’ paj ‘go’

wˆı ‘run’ straight’ th j ‘retreat, phaàn ‘pass’ maa ‘come’ lˇaj ‘flow’ won ‘circle’ back up’ klàp ‘return’

khàp ‘drive’ khót ‘zigzag’ khâw ‘enter’

bin ‘ﬂy’ chˇee ‘veer’ k ‘exit

wˆaaj ‘swim’ lo ‘descend’

(5.90) contains examples of each verb type (not exhaustive for the ﬁrst and fourth categories), while (5.91) is an SVC that realizes each type of verb (where the subscripts CHAPTER 5. MOTION SVCS 274

on the second line indicate the verb type):

(5.91) Malii wˆı tro j n kl`ap khˆaw paj. Malii run1 go-straight2 reverse3 return4a enter4b go5 ‘Malii ran straight back in, away from the speaker.’ (Muansuwan 2002, 43)

The fourth component of (5.90), ‘Direction with respect to the outside world’, is actually realized by two verbs in (5.91), klàp ‘return’ and khâw ‘enter’,11 taking the total number of verbs up to six. Six is the maximum number of verbs permitted in a Thai motion SVC according to Thepkanjana (1986), although all of the components presented in (5.90) are optional, and most Thai motion SVCs contain fewer verbs (typically two, as in most languages). While the fourth component of (5.90) corresponds roughly to the categories of path in the Nigerian Pidgin English and Ijo motion SVC templates, this Thai template is interesting for its inclusion of two further types of path component: the second, ‘Geometric shape of the path’ component, and the third, ‘Direction with respect to the previous path’ component. The inclusion of these components demonstrates that, although the description of the components of a motion event in Section 5.2.1 was intended to be quite comprehensive, it cannot have been entirely so; specifically, Section 5.2.1 was built upon a covert, default assumption that the geometric shape of the path in a translational motion event was of the ‘go straight’ variety, while components encoding ‘direction with respect to the previous path’ were never even alluded to. The omission of this latter category in particular may be justified from the perspective that ‘direction with respect to the previous path’ is quite a rare and little understood kind of component; further explication and exemplification (from Thai, or other languages) is needed to clarify its status. Thepkanjana’s (1986) template in (5.90) is impressive as the first comprehensive representation of the Thai motion SVC template, although linguists have subsequently made amendments to it. The main point of contention is the rigidity of the word

11Thepkanjana (1986) claims that these verbs belong to subcategories of the ‘Direction with respect to the outside world’ category: (4a) ‘Direction with respect to an object located in the outside world’ and (4b) ‘Direction resulting from interaction between the path and the outside world’. However, the basis for this distinction is unclear. CHAPTER 5. MOTION SVCS 275

ordering; Thepkanjana (1986) claims that the verbs in a Thai SVC must appear in exactly the order presented in (5.90), while Muansuwan (2002) claims that any of the path components may be reordered, with the exception of the ﬁnal, deictic verb. (5.93)-(5.95), with the second, third and fourth components reordered, are all claimed

by Muansuwan (2002, 55) to be acceptable variants of (5.92):

(5.92) Piti tro j n khˆaam s`a haan k paj. Piti go-straight2 reverse3 [cross bridge]4a exit4b go5

‘Piti went back straight, crossing the bridge, out away from the speaker.’

(5.93) Piti j n khˆaam s`a haan tro k paj. Piti reverse3 [cross bridge]4a go-straight2 exit4b go5

‘Piti went back straight, crossing the bridge, out away from the speaker.’

(5.94) Piti tro k j n khˆaam s`a haan paj. Piti go-straight2 exit4b reverse3 [cross bridge]4a go5

‘Piti went back straight, crossing the bridge, out away from the speaker.’

(5.95) Piti k tro khˆaam s`a haan j n paj. Piti exit4b go-straight2 [cross bridge]4a reverse3 go5 ‘Piti went back straight, crossing the bridge, out away from the speaker.’

Zlatev and Yangklang (2004, 164) take an intermediate position on this issue, claiming that the order of the verbs in a Thai motion SVC is neither as strict as Thepkanjana claims nor as variable as Muansuwan claims. The template they present is:

(5.96) [Manner-V > Manner+Path-V > Path-V* > Deictic-V]Clause (Zlatev and Yangklang 2004, 168)

By allowing multiple path verbs with no speciﬁed ordering, this is in eﬀect, if not intent, in full agreement with Muansuwan’s (2002) claims about variable word ordering, although the more novel or notable feature of (5.96) is the inclusion of a new component, realized by a new category of verb: the ‘Manner+Path verbs’. ‘Manner+Path’ verbs are claimed by Zlatev and Yangklang (2004) to display characteristics of both manner and path verbs, from which their position (situated CHAPTER 5. MOTION SVCS 276

between these two types of verb) is predictable. The examples of manner+path verbs that Zlatev and Yangklang (2004) provide can be divided into three groups on the basis of the type of path information they provide: (i) phlòo ‘pop out’, thalú ‘pierce’, thˆım ‘puncture’ include ‘a semantic aspect of going through a landmark’ (Zlatev and Yangklang 2004, 167) in addition to their manner information; (ii) tók ‘fall’, lòn ‘fall’, lóm ‘collapse’, hòklóm ‘trip and fall’, com ‘sink’ express downward motion, and (iii) lâj ‘chase’ describes motion in the direction of a moving object. An SVC containing

the ﬁrst of these verbs is:

(5.97) Ch´an dn phl`oo k paj. I walk pop.out exit go ‘I popped out, walking.’ (Thai; Zlatev and Yangklang 2004, 167)

Unfortunately, Zlatev and Yangklang (2004) do not provide any further examples or clarification in relation to manner+path verbs. It could be asked: Are they all really motion verbs? (Thalú ‘pierce’, thˆım ‘puncture’, lóm ‘collapse’, hòklóm ‘trip and fall’ appear particularly suspicious in this regard.) Can they really appear in a motion SVC? And can they really appear after a manner verb in a motion SVC, like phlòo ‘pop out’ in (5.97)? And, for the matter, what does phlòo actually mean? Because it does not seem possible to literally pop out of somewhere or something while concurrently walking, the more colloquial interpretation of a quick, casual departure (‘I’ll just pop out to the shops’) seems more likely. If this is true, then it could be observed that these manner elements of quickness and casualness are somewhat removed from the more typical manner elements (the “ill-defined set of dimensions that modulate motion, including motor pattern, rate, rhythm, posture, affect, and evaluative factors”), which focus more on how the figure is using its body. In any case, in the absence of any more information about these manner+path verbs (as well as any proof that they are actually motion verbs), I assume they – and in particular their potential cooccurrence with manner of motion verbs – pose little threat to the Manner of Motion Constraint on Possible Verb motion events. One final feature of the Thai motion SVC template not indicated in (5.90) or (5.96) is the fact that any ground arguments of the verbs are realized directly after CHAPTER 5. MOTION SVCS 277

these verbs, as in (5.98):12

(5.98) Piti dn khâam (sà haan) klàp (càak talàad) paj (roo rian). Piti walk cross (bridge) return (from market) go (school) ‘Piti walked, crossing (the bridge), back (from the market), away (to school).’ (Thai; Muansuwan 2002, 51)

This signifies that Thai has non-contiguous motion SVCs, whereas the Ijo motion SVCs of Section 5.3 were in some instances contiguous and in other instances non- contiguous. (5.98) also indicates that these ground arguments are optional, and in fact Muansuwan (2002) considers this to be definitive of SVC status: only if the complements of the verbs in a SVOVO string can be optionally omitted can this structure legitimately be considered to be an SVC according to her. The capacity for Thai SVCs to realize multiple path verbs and their object grounds could potentially mean more challenges to the satisfaction of the Theta Structure Property for SVCs (i.e. if the list of ground arguments realized does not resemble that of any clause headed by a single verb). (5.98), however, will satisfy the Theta Structure Property (with complements realized), because its Theta Structure consists of one structural argument (Piti), one non-structural route argument (the bridge crossed), one non- structural source argument (the market departed from), and one non-structural goal argument (the school), and some single-verb-headed clauses will likewise have Theta Structures composed of one structural argument plus non-structural source, route and goal arguments. The capacity for Thai SVCs to realize multiple path verbs – specifically those denoting region-to-region paths – could also challenge the Possible Verb status of certain SVCs, in ways that will be examined shortly. But first we can observe that most SVCs conforming to the Thai motion SVC template apparently do have Possible Verb semantics. All of the SVCs included in this section so far, which all conform to the template in (5.96), have Possible Verb semantics. At very least, none of them have translations suggestive of sequential Impossible Verb semantics, and certainly none of them violate

12This SVC also demonstrates how paj ‘go’, besides signifying motion away from the deictic center, can also take a goal argument. Verbs translated as ‘go’ display this dual behavior in many serializing languages. CHAPTER 5. MOTION SVCS 278

any of the constraints on Possible Verb motion events. Wechsler (2003) provides an overt demonstration of the Possible Verb semantics of (5.99):

(5.99) Piti den khˆaw (pay nay) roorian. Piti walk enter go in school ‘Piti walked into the school.’ (Thai; Wechsler 2003, 5)

One of the ways Wechsler shows that (5.99) has Possible Verb semantics is through a time adverbial test; because nay welaa s`ıp naatii ‘in ten minutes’ in (5.100) measures the entering event, and the manner event of walking occupies no (temporally prior) independent temporal space of its own (as in Piti walked and (then) entered the school), he concludes that the two subevents are coextensive (or coidentiﬁed, under the analysis adopted here):

(5.100) Piti den khˆaw (pay nay) roorian nay welaa s`ıp naatii. Piti walk enter go in school in time ten minute ‘Piti spent 10 minutes walking into the school.’ (Pragmatically odd reading where the entering lasts ten minutes, ‘perhaps because the door to the school was diﬃcult to open or the entryway was crowded’) (Thai; Wechsler 2003, 7)

Under the event structural analysis used in this dissertation, nay welaa s`ıp naatii ‘in ten minutes’ attaches a time scale to the entering event, and this scale applies equally to the walking subevent, for the reason that the entering subevent measures out this walking subevent in this coidentiﬁed structure. Ideally, this kind of time adverbial test would be applied to every motion SVC claimed to have Possible Verb status in this chapter, but in reality this has not been possible. On the other hand, it is arguable that many of the SVCs included in this chapter could not have anything but Possible Verb semantics. This is not something that could be argued for the following two Thai motion SVCs, however:

(5.101) a. Chán dn khâam thanˇon khâw paj naj sˇuan. I walk cross road enter go in park ‘I walked across the road and into the park.’ (Thai; Zlatev and Yangklang 2004, 160) CHAPTER 5. MOTION SVCS 279

b. c.

• • park • • • • park

(5.102) a. Piti j n khâam sà haan tro khâw bâan paj. Piti reverse cross bridge go-straight enter house go ‘Piti went back, crossing the bridge, straight into the house, away from the speaker.’ (Thai; Muansuwan 2002, 54) b. c.

• • • • • • bridge bridge

These are the Thai SVCs containing multiple region-to-region path verbs with object grounds alluded to above. In both of these SVCs khâam ‘cross’ (BECOME(LOC(x, ACROSS(y)))) is the initial region-to-region path verb and khâw ‘enter’ (BECOME (LOC(x,IN(y)))) is the second region-to-region path verb. The realization of multiple region-to-region path verbs was cited as a ‘red flag’ for Impossible Verb status in Sec- tion 5.2.2.4, which described the Single Endpoint Constraint on Possible Verb motion events. The Single Endpoint Constraint on Possible Verb motion events stated that one BECOME(LOC(x,PRED1(y)) predicate could not be further specified by another

BECOME(LOC(x,PRED2(z)) predicate, although one of the conditions under which this could be possible would be if these were both region-to-region predicates, and the path denoted by the former spatially contained that denoted by the latter. Now if the grounds to which the two region-to-region path predicates refer are situated far apart, then the path denoted by the former is unlikely to spatially contain the latter, meaning that the Single Endpoint Constraint on Possible Verb motion events will not be sat- isﬁed. What this means in relation to (5.101a) is that, if the park entered is situated directly adjacent to the road crossed as depicted in (5.101b), then the ‘crossing’ path can contain the ‘entering’ path, and the two paths can be amalgamated into a single BECOME(LOC(x,ACROSS(road))) & BECOME(LOC(x,IN(park))) path speciﬁca- tion in a Possible Verb SVC with the following Semantic Form representation: CHAPTER 5. MOTION SVCS 280

(5.103) λx λs {WALK(x) & BECOME(LOC(x,ACROSS(road))) & +sc BECOME(LOC(x,IN(park)))}(s) (Possible Verb structure)

If, on the other hand, the park is situated at some distance from the road, as depicted in (5.101c), then Possible Verb structure is not possible, and (5.101a) is an Impossible Verb SVC with the following Semantic Form representation (which is exactly what the translation for (5.101a) would suggest):

(5.104) λx λs [{WALK(x) & BECOME(LOC(x,ACROSS(road)))} &{WALK(x) & +sc BECOME(LOC(x,IN(park)))}](s) (Impossible Verb structure)

The same analysis applies to (5.102a), although Impossible Verb status is, if anything, more likely for this SVC, because a bridge is less likely to be directly adjacent to the entrance to a house. Regardless of whether these SVCs have Possible Verb or Impossible Verb semantics, they will both satisfy the Theta Structure Property for SVCs, since in either case their Theta Structures will consist of a single structural argument (the ﬁgure), one non-structural route argument (the landmark crossed) and one non-structural goal argument, and there do exist single-verb-headed clauses with identical Theta Structures. To summarize, we cannot conclusively know if (5.101a) and (5.102a) are Impossible Verb motion SVCs or not, because we do not know enough about the landmarks referred to. But if, as seems likely, the landmarks are in each case situated far enough apart to render these Impossible Verb SVCs, the importance of these examples is that they signify that the motion SVC template – a syntactic structure – can no longer be equated with Possible Verb semantics – at least in the case of Thai. No Ijo motion SVCs conforming to the Ijo template set out in the previous section were found to have Impossible Verb semantics, and neither will any Bima motion SVCs conforming to the Bima template in the next section have Impossible Verb semantics. The greater likelihood of Impossible Verb semantics obtaining within the Thai motion SVC template can be attributed to the possibility of instantiating multiple region-to-region path predicates and their arguments. But we can also ask if Impossible Verb semantics result through a diﬀerent and more typical mechanism, CHAPTER 5. MOTION SVCS 281

which is the iteration of the template itself, as would appear to be the case for Ijo. The short answer to the question of whether multiple iterations of the Thai motion SVC template deliver Impossible Verb SVCs is yes, but only if the resulting structures are really SVCs. Zlatev and Yangklang (2004) claim the Thai motion template corresponds to a clause, and that (5.105), in which maa ‘come’ signals the boundary of the ﬁrst

instantiation of the template, thus contains two clauses: (5.105) Piti [wˆı maa]T emplate1 [lo bandaj]T emplate2. Piti run come descend stairs ‘Piti ran towards the speaker to come down the stairs.’ (Thai; Muansuwan 2002, 57)

If it does consist of two clauses then (5.105) cannot be an SVC, in contrast to (5.106),

which, with the verbs reordered, contains a single template and is thus monoclausal: (5.106) Piti [wˆı lo bandaj maa]. Piti run descend stairs come ‘Piti ran down the stairs towards the speaker.’ (Thai; Muansuwan 2002, 57)

According to Zlatev and Yangklang (2004), two templates means two clauses, although it is striking that most other Thai linguists (such as Thepkanjana (1986), Muansuwan (2002), Sudmuk (2005) and Iwasaki and Ingkaphirom (2005)) characterize multi-templatic sentences such as (5.105) as SVCs. Muansuwan’s semantic characterization of (5.105) is that “while [(5.106)] encodes a single event, [(5.105)] denotes a sequence of two separate events, the second of which is not entailed to occur” (2002, 57). Iwasaki and Ingkaphirom (2005) describe sentences like (5.105) as ‘purposive SVCs’. The SVC status of (5.105) may appear questionable from the perspective that the two subevents do not share the same tense/aspect/mood values semantically. But in the absence of any grammatical markers of the tense/aspect/mood values for these subcomponents, there are no grounds for ruling (5.105) out as an SVC under the deﬁnition for SVCs set out in Chapter 1. True enough, these Thai sentences would perhaps never even have been expected to display tense/aspect/mood values, given the paucity of this type of CHAPTER 5. MOTION SVCS 282

marking in this language, but this does not change the fact that there are no grounds for disqualiﬁcation from SVC status – both under my deﬁnition for SVCs set out in Chapter 1, and apparently also according to all of the Thai linguists mentioned above. It may well be the case that some grammatical test could be devised that proves that sentences like (5.105) are biclausal, but until such proof becomes available, it should be assumed that (5.105) is an Impossible Verb SVC, whose Impossible Verb semantics results through the iteration of the Thai motion SVC template.

5.5 Bima motion SVCs

5.5.1 Introduction

This ﬁnal case study of Bima motion SVCs is divided into three subsections that address the three types of form-to-meaning correspondence illustrated in (5.107)- (5.108):

(5.107) Fero lampa lu’u.na Fero walk enter.3 ‘Fero walked in.’13 (Bima)

(5.108) Fero lu’u lampa.na. Fero enter walk.3 (a) ‘Fero came in and walked’ or (b) ‘Fero came in walking.’14 (Bima)

(5.107) exempliﬁes the most frequently instantiated type of Bima motion SVC. It conforms to the Bima motion SVC template that appears at the beginning of Sec- tion 5.5.2, and it has Possible Verb semantics. Section 5.5.2 considers whether all Bima motion SVCs that conform to the Bima motion template have Possible Verb semantics. On the face of it, (5.108) looks like that rare thing: a motion SVC with path+manner word order. But Section 5.5.3 will show that, at least under the (a) interpretation

13Original Indonesian translation from consultant: Fero jalan masuk (Fero go walk). 14The original Indonesian translations from my consultant are: (a) Fero masuk dan berjalan (Fero enter and walk) and (b) Fero masuk berjalan (Fero enter walk). CHAPTER 5. MOTION SVCS 283

for (5.108), this is not a word order that arises out of any (alternative) Bima motion SVC template. Instead, the form-to-meaning correspondence that holds for the (a) reading of (5.108) is argued to involve two (single verb) iterations of the Bima motion SVC template presented in Section 5.5.2. Multi-templatic Bima SVCs of this kind have sequential, Impossible Verb semantics. They thus bear out the tendency recognized in this chapter for Impossible Verb motion semantics to be derived through iterations of a motion SVC template. Finally, Section 5.5.4, considers the type of form-to-meaning correspondence illustrated through the (b) reading of (5.108). This (b) reading has Possible Verb semantics. Certain Bima manner of motion verbs, it would seem, ‘prefer’ the path+manner word order of (5.108), and Section 5.5.4 contains only a suggestion (and not necessarily a solution) as to why they should do so.

5.5.2 The Bima motion SVC template, and SVCs that conform to the Bima motion SVC template

By far the most common type of Bima motion SVCs are those that conform to the following template, where every item in the template is optional:

(5.109) Bima motion SVC template:

Deictic Manner of motion Path (direction) Path (ground-ﬁxture)

lao ‘go’ lampa ‘walk’/‘go ne’e ‘ascend’ lu’u ‘enter’ mai ‘come’ normal speed’ londo ‘descend’ losa ‘exit’ rai ‘run’/‘go quickly’ lao d’ei ‘go dula ‘go home’ liwa ‘swim’ inwards’ heko ‘encircle’ loja ‘sail’ lao ari ‘go rama ‘crawl’ outwards’

Most linguists will view the initial placement of the deictic verb as the most striking and unusual feature of this template, given the preponderance of deictic-ﬁnal motion SVCs in the serial verb literature. This feature will be discussed shortly, but CHAPTER 5. MOTION SVCS 284

ﬁrst (5.110)-(5.115) are some examples of Bima motion SVCs which conform to the template in (5.109):

(5.110) Deictic + Manner of motion:

... mod’a.ku loa lao rai wa’u. easy.1 able go run ﬁrst ‘I could easily run away ﬁrst.’ (Bima)

(5.111) Deictic + Path (direction):

... ede.ra lao ne’e kai.na aka tando ncai .. dem.emph go ascend kai.3 loc(dist) front door ‘ ... so he went up to their front door ...’ (Bima)

(5.112) Deictic + Path (ground-ﬁxture):

au mai lu’u.na [d’i tamba nahu] binata ake, lao losa mena! what come enter.3 loc trap 1p animal dem go exit pl ‘what are these animals doing coming into my trap, get out!’ (Bima)

(5.113) Deictic + Manner of motion + Path (ground-ﬁxture):

Ede.ra lao lampa heko kai b’a Maharaja Kurma [lewi dem.emph go walk encircle kai by Maharaja Kurma garden ede] ... dem ‘Then Maharaja Kurma went walking around the garden ...’ (Bima)

(5.114) Path (direction) + Path (ground-ﬁxture):

ndadi na.lampa.ra, ne’e lu’u [d’i ade kampo]. thus 3.go-along.emph ascend enter loc liver village ‘he went along for a while, and then went up into the village.’ (Bima) CHAPTER 5. MOTION SVCS 285

(5.115) Manner of motion + Path (direction) +Path (ground-ﬁxture):

Amu rai londo losa.na [d’i uma]. Amu run descend exit.3 loc house ‘Amu ran down out of the house.’ (Bima)

These Bima motion SVCs, like all Bima SVCs, have contiguous structure, in which verbs that take objects must appear in final position, adjacent to their (verb complex- external) objects. From this it is predictable that ground-fixture Path verbs should appear template-finally, since these are the only verbs in (5.109) that take objects. That said, heko ‘encircle’ is actually the only ground-fixture verb in (5.109) whose ground argument is realized as a direct object, as in (5.113); dula ‘go home’ does not realize an object, since the ground arrived at (home) is specified as part of the meaning of the verb, while lu’u ‘enter’ and losa ‘exit’, the most commonly instantiated ground-fixture verbs, that appear in (5.112), (5.114) and (5.115), realize their ground arguments in prepositional phrases headed by d’i. D’i is a locative preposition that does not encode movement (in any direction) and also does not encode any geometric information more specific than location (‘at’ would probably be the best English equivalent).15 The feature of the Bima motion template in (5.109) already observed to be unusual is the initial placement of the deictic verbs. On the other hand, this word order may not be so surprising from the perspective that there are many Bima SVCs, like (5.116), in which an initial lao ‘go’ or mai ‘come’ is followed by some non-motion verb:

(5.116) ... wara udi ma.mai mpanga janga Ompu Ico. exist lizard rel.come steal chicken Ompu Ico ‘... there was a lizard who came and stole some of Ompu Ico’s chickens. (Bima)

SVCs like (5.116) recall Durie’s (1997, 310) comment from Section 1.2.2 of Chapter 1 that “every serializing language [he has] encountered includes a category of motion serialization, where a verb of motion is combined with some other verb in such a

15The ground arguments of lu’u ‘enter’ and losa ‘exit’ are sometimes instead realized in prepositional phrases headed by d’ei ‘in’. CHAPTER 5. MOTION SVCS 286

way that the motion verb comes first and the moving argument is the Agent of the second verb”. These SVCs appear to be very common in Oceanic languages; Bril and Ozanne-Rivierre’s (2004) and van Staden and Reesink’s (2008) surveys of Oceanic SVCs contain many such SVCs. But the question we really have to ask about Bima deictic-initial motion SVCs like (5.110)-(5.113) is: does lao/mai actually signify a temporally prior motion subevent that precedes another motion subevent (i.e. in the same way the lao/mai motion subevent has to temporally precede the non-motion subevent in SVCs like (5.116)), or does lao/mai signify motion away from/towards the deictic center that is coextensive (or coidentified) with the other motion information? While the former case involves sequential, Impossible Verb semantics, the latter involves Possible Verb semantics. In fact, Bima deictic-initial motion SVCs appear to always instantiate Possible Verb, coidentified semantics. The English Possible Verb sentence ‘Fero ran away’, for example, will always be translated as Fero lao rai.na (Fero go run), although I do have to admit that in some cases, like (5.117), I have no actual proof that this SVC has the Possible Verb semantics indicated by the first translation, and not the Impossible Verb semantics indicated by the second translation:

(5.117) au mai lu’u.na [d’i tamba nahu] binata ake, lao losa mena! what come enter.3 loc trap 1p animal dem go exit pl ‘what are these animals doing coming into my trap, get out!’/‘what are these animals doing coming and (then) entering my trap, get out!’ (Bima)

This uncertainty cuts to the central concern of this section, which is the question of whether all Bima motion SVCs conforming to the template in (5.109) have Possible Verb semantics. (5.118) is one mannner+path SVC conforming to the template that can be shown to have coidentiﬁed Possible Verb semantics, since lima menit mpa measures out the entering subevent, with the walking subevent occupying no temporally prior space of its own:

(5.118) Reho lampa lu’u.na d’i sekolah lima menit mpa. Reho walk enter.3 loc school five minute just ‘Reho walked into the school in five minutes.’ (i.e. took five minutes to enter the school.) (Bima) CHAPTER 5. MOTION SVCS 287

This is Wechsler’s (2003) test for coextensive semantics, demonstrated in relation to Thai motion SVCs in the previous section. Ideally, tests of this kind would be carried out on all Bima motion SVCs conforming to the template, although in reality this has not been possible. The most I can claim, then, is that all Bima motion SVCs conforming to the template look to have Possible Verb semantics (and certainly none of them are incompatible with Possible Verb semantics). Section 5.5.3 now considers some Bima motion SVCs that have clear and unambiguous Impossible Verb semantics, and whose structural correlate is the iteration of the Bima motion SVC template.

5.5.3 Multi-templatic Bima motion SVCs

Multi-templatic sequential Bima motion SVCs were exempliﬁed in Section 5.5.1 through the (a) reading of the following SVC (= (5.108)):

(5.119) Fero [lu’u] [lampa].na. Fero enter walk.3 (a) ‘Fero came in and walked’ or (b) Fero came in walking.’ (Bima)

The claim is that each of these verbs corresponds to a separate instantiation of the template in (5.109): [lu’u ‘enter’] is one instantiation (with all other elements of the template absent), and [lampa ‘walk’] is another instantiation of the template. Multi- templatic structure corresponds to sequential, Impossible Verb semantics in the same way outlined for Ijo in Section 5.3; the (a) reading of (5.119) involves an entering subevent temporally succeeded by a walking subevent. Multi-templatic structure can perhaps be seen more clearly when instantiations of the template hold more than one verb, as do at least one of the (bracketed) instantiations of the template in each of the following:

(5.120) Rai b’atu kai b’a Ompu Ico [rai lu’u] [liwa].na d’ei run follow kai by Ompu Ico go-fast enter swim.3 in karombo udi. cave lizard ‘Ompu Ico ran after (the lizard), dived into (the river), and swam into the lizard’s cave.’ (Bima) CHAPTER 5. MOTION SVCS 288

(5.121) Sia [rai lu’u] [liwa kabei lao ese].na. 3p go-fast enter swim pass go up-at.3 ‘He dived in (to the river) and swam past (the boat), going upstream.’ (Bima)

(5.122) Harimau b’i’a.na kuru, [losa] [lao rai lu’u].na d’ei wub’a. tiger break.3 cage exit go run enter.3 in forest ‘The tiger broke the cage, went out and ran into the forest.’ (Bima)

(5.123) Nasi [losa] [lao ngemo].na. bird exit go ﬂy.3 ‘The bird went out (of the cage), and ﬂew away.’ (Bima)

(5.120)-(5.123) have sequential, Impossible Verb semantics. (5.120) and (5.121) violate the Manner of Motion Constraint on Possible Verb motion events by each realizing two manner of motion verbs, and so must be Impossible Verb SVCs. (5.122) may violate the Single Endpoint Constraint on Possible Verb motion events if the cage is situated some distance from the forest, so that the region-to-region ‘exit’ path out of the cage does not contain the region-to-region ‘enter’ path into the forest (this would also render the description the tiger went out of the cage into the forest inapplicable). There are otherwise no ‘red ﬂags’ for Impossible Verb status for (5.122), or for (5.123), which leaves the sequential translations for these SVCs as the strongest indicators of their Impossible Verb status (these sentences were accepted by native speakers as descriptions of sequential situations depicted in a series of pictures). As Impossible Verb SVCs, (5.120)-(5.123) must undergo predicate combining operations. (5.124) demonstrates these operations for (5.120):

(5.124) Rai b’atu kai b’a Ompu Ico [rai lu’u] [liwa].na d’ei run follow kai by Ompu Ico go-fast enter swim.3 in karombo udi. cave lizard ‘Ompu Ico ran after (the lizard), dived into (the river), and swam into the lizard’s cave.’ (Bima) CHAPTER 5. MOTION SVCS 289

rai lu’u liwa: λx λs1 {GO-FAST(x) & BECOME(LOC(x,IN(river)))}(s1) +sc

& λu λs2 {SWIM(u) & BECOME(LOC(u,IN(cave)))}(s2) +sc

i. Event combining: s1 + s2 = s ii. Highest structural argument coidentiﬁcation: x=u

Combined predicate: λx λs [{GO-FAST(x) & BECOME(LOC(x,IN(river)))} & {SWIM(x) & +sc ∅ BECOME(LOC(x,IN(cave)))}](s)

(5.124) is interesting for the reason that it encodes two arrivals at (or into) two distinct goals; there is the diving into the river and the swimming into the lizard’s cave. If the river does not spatially contain the cave, then these two goals will not be thematic role disconstituents (in Brunson’s (1992) sense), and this SVC will violate the Ground Arguments Constraint on Possible Verb motion events, in addition to the Manner of Motion constraint. But (5.124) does satisfy the Theta Structure Property for SVCs, since only one goal (karombo udi ‘the lizard’s cave’) is realized overtly; the Theta Structure for this SVC thus consists of one structural argument and one non-structural, overtly realized goal argument, and there do exist clauses headed by single verbs with identical Theta Structures. In each of (5.120)-(5.123), multi-templatic structure corresponds directly with sequential semantics: the ﬁrst instantiation of the template corresponds to a (Possible

Verb) motion subevent that commences at time t1, the next instantiation of the template corresponds to a subevent that commences at time t2 (later than t1), and so forth. Because (5.120)-(5.123) are all multi-templatic SVCs with Impossible Verb semantics, these (along with (5.119)) lend weight to the observation that one of the main ways – if not the main way – in which Impossible Verb semantics is derived in motion SVCs is through the iteration of language-speciﬁc motion SVC templates. It remains to be seen whether this observation, about the way in which syntactic, templatic organization interacts directly with temporal and event structure, is borne out in other languages. But while this issue awaits further research, some additional CHAPTER 5. MOTION SVCS 290

observations can be made that are speciﬁc to multi-templatic, Impossible Verb Bima motion SVCs. While (5.120)-(5.123) are all good illustrations of multi-templatic SVCs (for the reason that they contain multi-verb single instantiations of the template), they are not the most typical kind of Bima Impossible Verb motion SVC encountered. If there is a most typical or commonplace kind of Bima Impossible Verb motion SVC, it is comprised of the verbs londo ‘descend’ followed by lao ‘go’, as in:

(5.125) Ede.ra [londo] [lao] kai.na d’ua.d’ua.na ede, ... dem.emph descend go kai.3 two.two.3 dem ‘So the two of them went down and left, ... ’ (Bima)

(5.126) ede mpara londo lao kai.na d’i uma ompu, ... dem then descend go kai.3 loc house old-man ‘then he went down and went to the old man’s house, ...’ (Bima)

(5.127) ... b’onto.na tare ede, londo lao.na aka doro. close.3 tray dem descend go.3 dist mountain ‘(He decided not to eat,) closed up his tray, went down (out of the house) and went (up?) to the mountain.’ (Bima)

Londo in each case refers to the descent from a traditional Bima house, which has a steeply inclined set of stairs leading up to the front entrance, while lao ‘go’ describes a subsequent motion subevent. Alternative motion verbs, such as lampa ‘walk’ or rai ‘run’, also appear after londo, although lao is most common. I encountered londo lao SVCs time after time in Bima text, and from them it is possible to extrapolate something about the general nature of what Impossible Verb Bima SVCs describe. Londo lao SVCs typify how a Bima Impossible Verb motion SVC will describe an initial, comparatively insignificant motion subevent which seems to do nothing more than clarify the figure’s position or location prior to the second, ‘main’ motion subevent. Thus londo in (5.125)-(5.127) informs us that the figure was located within a house before embarking on a more significant event of relocation. In a discourse context, the information about the figure’s initial location may even be redundant. This characterization of an initial, relatively insignificant motion subevent that states the figure’s CHAPTER 5. MOTION SVCS 291

position or location, followed by a motion subevent of more narrative substance can be extended across all Impossible Verb Bima motion SVCs. What Impossible Verb Bima motion SVCs do not appear to be able to describe are series of motion subevents of equal substance, new information content or narrative weight. Chapter 6 will consider whether Bima Impossible Verb SVCs from other domains can be similarly characterized, what this tells us about how Bima Impossible Verb SVCs are constrained, and how they compare with Impossible Verb SVCs from other languages.

5.5.4 Bima motion SVCs which do not ﬁt the template

This section discusses the small, but signiﬁcant, number of path+manner Bima motion SVCs which do not conform to the template in (5.109) of Section 5.5.2, which are not multi-templatic, and which have Possible Verb, non-sequential semantics. Of the hundreds of motion SVCs encountered in Bimanese texts,16 three contained non-sequential path+manner combinations of londo ‘descend’ and nggoncu ‘jump’, of which (5.128) is one example, while the non-sequential combination of losa ‘exit’ and ngemo ‘ﬂy’ in (5.129) was also encountered:

(5.128) Ede.ra tu’u rai kai ruma sangaji, na.londo nggoncu watu dem.emph get-up run kai ruma sangaji 3.descend jump from ncai riha, ... door kitchen ‘so Ruma Sangaji got up and ran, and jumped down from the kitchen door, ...’ (Bima)

(5.129) Ede.ra hengga kai.na ncai tamba.na ede, losa ngemo mena dem.emph open kai.3 door trap.3 dem exit ﬂy pl kai janga peo ede. kai chicken wild dem ‘Then he opened the door of his trap, and out ﬂew the wild chickens.’ (Bima)

In fact, speakers were about as likely to accept manner+path alternatives to these path+manner word orders when presented with similar examples; 12 speakers preferred the manner+path word order of (5.130) to the path+manner order of (5.131),

16I have about 12,000 words of Bimanese text. CHAPTER 5. MOTION SVCS 292

which nine speakers preferred, for example:

(5.130) Amu nggoncu londo.na awa dana. Amu jump descend.3 from down-at ground ‘Amu jumped down to the ground.’ (Bima)

(5.131) Amu londo nggoncu.na awa dana. Amu descend jump.3 from down-at ground ‘Amu jumped down to the ground.’ (Bima)

Even so, the anomalous path+manner word order of (5.128), (5.129) and (5.131) needs to be accounted for, and towards this end I will suggest two possible explanations. One possibility is that the manner of motion verb is semantically an adverb in these examples. This analysis sits well with the fact that Bima is a uniformly postmodifying language, and with the free instantiation of verbs, such as hura-hara ‘hurry’ from (5.132), with a postmodifying function in other, non-motion SVCs, such as (5.133):

(5.132) Fero hura-hara. Fero hurry ‘Fero is hurrying.’ (Bima)

(5.133) Fero kab’oro hura-hara.na haju ka’a. Fero gather hurry.3 wood ﬁre ‘Fero gathered ﬁrewood in a hurry.’ (Bima)

If nggoncu ‘jump’ in (5.128) and (5.131) and ngemo ‘fly’ in (5.129) are semantically adverbs, then they will not correspond to a subevent, and these structures will be unitary-event SVCs. But if the semantic function of these words is considered enough to disqualify them as verbs (even in the absence of any grammatical or morphological evidence of adverbial status) then (5.128), (5.129) and (5.131) will not even qualify as SVCs. Another possible explanation for the path+manner word order of (5.128), (5.129) and (5.131) is that it represents a stylistic preference. However, this is highly speculative, and difficult to prove. It is difficult enough to account for the stylistic preference CHAPTER 5. MOTION SVCS 293

for rendering the English translation of (5.129) as ‘out flew the wild chickens’ over the less marked ‘the wild chickens flew out’. These path+manner SVCs present a quirk that goes against the grain of what we usually see in Bima motion SVCs, comparable to the limited, verb-specific path+manner quirks of the Ijo motion SVC case study in Section 5.3. While these quirks should not be ignored, more significance should nonetheless be attached to the stronger tendencies these languages display: deictic+manner+direction+ground-fixture verb order in Bima, manner+path+deictic verb order in Ijo, Possible Verb semantics for single template motion SVCs, and Impossible Verb semantics for multi-templatic SVCs. The next, concluding section places these observations within the overall context of everything this chapter has observed about motion SVCs: how their surface structures vary cross-linguistically, how to distinguish Possible Verb from Impossible Verb motion SVCs, and how they satisfy the Theta Structure Property for SVCs.

5.6 Conclusion

The introduction to this chapter first drew attention to the fact that motion SVCs, in contrast to instrumental SVCs, could – at least in theory – contain numerous verbs, each expressing different components of a motion event. Section 5.2 went about deconstructing motion events into distinct components, and then built up a picture of the bounds on the number and type of components that could be expressed in a Possible Verb motion event, through five constraints on Possible Verb motion events. These constraints also contribute to a more general (non-motion-specific) picture of what a Possible Verb event is, or what the bounds on a clause headed by a single verb are, since it is only the Unique Vector Constraint of Section 5.2.2.2 that applies exclusively to the domain of motion. Of the other constraints, the Manner of Motion Constraint on Possible Verb motion events presents a motion-specific version of the more general constraint that a Possible Verb event contain at most one manner component or ‘slot’, while the Single Endpoint Constraint presents a motion-specific version of the more general constraint that a Possible Verb event contain at most one result component or ‘slot’. The Ground Arguments Constraint on Possible Verb CHAPTER 5. MOTION SVCS 294

motion events puts into effect the requirement of one-to-one mapping between theta roles and syntactic positions (Chomsky’s (1981) Theta-Criterion, Bresnan’s (1982) Biuniqueness or Bohnemeyer’s (2003) Argument Uniqueness), which in the domain of motion means a limitation to the realization of at most one source, one route and one goal. The Ground Arguments Constraint also cuts to the core of what the Theta Structure Property requires of SVCs in the sphere of motion events, by honing in on motion-specific argument structure configurations that satisfy this property. Possible Verb SVCs, because they have the semantic content of a clause headed by a single verb, will always display argument structure configurations that satisfy the Theta Structure Property. But a main aim of this chapter was to determine if there are any Impossible Verb motion SVCs, which do not satisfy the Theta Structure Property so trivially. Motion SVCs, like instrumental SVCs, present a semantically unified domain, in which SVCs expressing Impossible Verb events – if they exist – may not stand apart blatantly from the SVCs that express Possible Verb events. Certainly, an Impossible Verb motion (or instrumental) SVC will show none of the qualities of a purely ‘accidental’ Impossible Verb SVC, in which two lexical verbs from distinct domains seem to be randomly ‘thrown together’ (such as the Edo SVC from Section 1.3.3 that expresses the concept of buying LGB and reading it). But if there is nothing in the appearance of Impossible Verb motion SVCs to make them stand apart blatantly from Possible Verb motion SVCs, at least the constraints on Possible Verb motion events formulated in Section 5.2.2, and the ‘red flags’ for Impossible Verb status associated with them provide a workable means of identifying Impossible Verb motion SVCs in the case studies of Sections 5.3, 5.4 and 5.5. Moreover, these Ijo, Thai and Bima case studies do, in fact, turn out some limited cases of Impossible Verb motion SVCs through these means, and these SVCs reinforce everything that Impossible Verb SVCs have already been claimed to signify: that SVCs are not just doing what would be done with clauses headed by single verbs in other languages, that SVCs do not just describe single, non-sequential events, and that there is this fundamental division between two types of SVC. This reinforcement is particularly strong because it comes from a domain in which a considerable amount of work was required to distinguish Impossible Verb from Possible Verb SVCs, rather than CHAPTER 5. MOTION SVCS 295

from ‘SVCs’ that look so accidentally thrown together, and so unlikely to describe recognizable event types, that it seems legitimate to at least suspect that they might not even be SVCs. Chapter 6 takes up some of these suspicions. While the identification of Impossible Verb motion SVCs was a main aim for this chapter, a secondary aim was to explore what cross-linguistic variation there is in the surface structures of motion SVCs. For although motion SVCs seldom live up to the promise of containing the heavy numbers of verbs that the decomposition of motion events into numerous components might suggest, they do quite regularly contain three or more verbs, and this lends itself to many ordering possibilities. The case studies of Ijo, Thai and Bima motion SVCs, and also the brief illustration of Nigerian Pidgin English motion SVCs, demonstrate that there is considerable diversity in this respect. One of the findings was that deictic verbs might not always appear template-finally, as the Ijo, Thai, and most other commonly cited SVC data might suggest; deictic verbs were instead realized in a medial position in Nigerian Pidgin English (after manner but before other path verbs), and template-initially in Bima. Another finding was that, while manner verbs almost always precede path verbs, there were some limited instances of path+manner word order in Ijo and Bima motion SVCs. But perhaps the most significant structural observation in the context of this dissertation was the tendency for single-template structure to correlate with Possible Verb motion semantics and multi-templatic structure with Impossible Verb motion semantics. However, these are only incipient observations (and Thai moreover instantiates single-template motion SVCs that may have Impossible Verb semantics, if the two region-to-region path predicates involve do violate the Single Endpoint Constraint in Possible Verb motion events); more motion SVC data, from a wide range of languages is required to establish how strong these tendencies are, or if there are any alternative structural patterns or correlates to the semantic distinction between Possible Verb and Impossible Verb motion SVCs. Chapter 6

Conclusion

6.1 The Theta Structure Property for SVCs: taking SVC deﬁnition below the surface

The driving idea behind this dissertation is that SVCs should be understood as constructions which, although they contain two or more verbs, are acting as if they contain only one. ‘Acting’ can be taken to mean showing all outward, surface level signs, and accordingly, the standard SVC-defining properties of no coordination/subordination marking, uniform tense, aspect and/or mood marking, and monoclausal phonological behavior are all properties that can be gleaned from the surface structure of the clause that are consistent with the presence of only one verb. But the Theta Structure Property for SVCs – the property proposed here that enforces monoclausal argument structure in SVCs – takes us below the surface level to a level of argument structure representation. Because the Theta Structure Property takes SVC definition below the surface, a fair evaluation could be that fulfillment of this property does not amount to one of the more blatant, strong ways of acting as if there is only one verb. In this concluding chapter, however, I will describe a sense in which the Theta Structure Property need not always be understood as applying below the surface, as well as explaining ultimately why this property is entirely necessary. The sense in which the Theta Structure Property need not always be understood

296 CHAPTER 6. CONCLUSION 297

as applying below the surface emerges from the observation that SVCs do actually display a surface syntactic monoclausal pattern of arguments, with the one major exception being when non-structural arguments are realized as direct objects of verbs. The realization of non-structural arguments as direct objects of verbs is the one circumstance that forces the Theta Structure Property below the surface, that is to say – at least based on the data presented in the previous chapters. (6.1)-(6.4) are some examples of SVCs which can be regarded as displaying a surface monoclausal pattern of arguments in that they realize combinations of grammatical relations (NPSUBJ ,

NPOBJ ) and obliques (usually PPs, as in (6.1), (6.3)-(6.4)) that could also be realized in clauses headed by single verbs:1

(6.1) [NPSUBJ Ana to’i] b’ale ka.lu’u.na [NPOBJ bola] [PP d’ei uma]. child small throw caus.enter.3 ball in house ‘The child threw the ball into the house.’ (Bima)

(6.2) [NPSUBJ Mi] ruat d´oh [NPOBJ au] [NPOBJ phei]. 2sg buy give me rice ‘You buy me rice.’ (Jeh; Kroeger 2004, 240)

` ` (6.3) [NPSUBJ Ozó] sùá [NPOBJ Esósà] dé [PP vbè ò. wàró. kpá]. Ozo push Esosa fall in one minute ‘Ozo pushed Esosa down in one minute.’ (Edo; Stewart 2001, 90)

` (6.4) [NPSUBJ Ozó] lé [NPOBJ èvbàré] khié. n [PP là úzó. lá èvá]. Ozo cook food sell for week two ‘Ozo cooked the food and sold it for two weeks.’ (Edo; Stewart 2001, 80)

(6.5) and (6.6), in contrast, display patterns of grammatical relations that clauses headed by single verbs in these languages could not; their argument realizations are SVC-speciﬁc:

1I analyze the following SVC as displaying a surface monoclausal pattern of arguments since the non-structural argument `ax`ı ‘market’ is realized in a PP, and not as the direct object of w´a ‘come’

(even though it is a complement of this verb):

Ç ¡Ç ¡Ç E [NPSUBJ K kú] s [NPOBJ às n] wá [P P àx`ı m ]. Koku take crab come market in ‘Koku brought the crab to the market.’ (Fongbe; Lefebvre & Brousseau 2002, 410) CHAPTER 6. CONCLUSION 298

` (6.5) [NPSUBJ Ay´ı] kè. [NPOBJ àtó] tsò. [NPOBJ wónù]. Ayi take ladle pour soup ‘Ayi poured soup with a ladle.’ (Ewe; McWhorter 1992, 20)

` (6.6) [NPSUBJ Ajé] wo. [NPOBJ as.o. ] fún [NPOBJ Olú]. Aje wear clothing give Olu ‘Aje wore an outfit for Olu.’ (Yoruba; Awoyale 1988, 5)

On the basis of the data presented in earlier chapters, SVCs of this type would appear to be in the minority; most SVCs do display the surface monoclausal argument structure exemplified by (6.1)-(6.4), although this bias could be attributed to the inclusion of contiguous SVCs, which rarely – if ever – contain verbs that take direct objects with non-structural status. SVCs like (6.5) and (6.6) nonetheless represent a significant strand of what SVCs are about, in the way that the verbs in these SVCs perform the functions more typical of prepositions. It is because of these SVCs that SVC definition must delve below the surface to an argument structure level that recognizes the arguments of kè. ‘take’ in (6.5) and fún ‘give’ in (6.6) as (instrumental and benefactive) non-structural roles. But it is also at this argument structure level (Theta Structure and Semantic Form in LDG) that sentences like (6.7) must be identified not as SVCs, but as covert coordinations, as a consequence of the fact that ´ıv`ın ‘coconut’ and ó. là ‘corn’ cannot be identified with particular thematic role configurations such as could both be contained within a clause headed by a single verb:

` (6.7) [NPSUBJ Ozó] gbò. ó. [NPOBJ´ıv`ın] bòló [NPOBJ ó. là]. Ozo planted coconut peel corn ‘Ozo planted coconut and peeled corn.’ (Edo; Stewart 2001, 51)

The exclusion of sentences like (6.7) from the class of SVCs was the main motivation behind the formulation of Internal Argument Sharing (Baker 1989, Collins 1997), which signals the first time SVC definition was taken below the surface to argument structure. In order to include sentences like (6.5) and (6.6) in the class of SVCs but exclude sentences like (6.7), taking SVC definition below the surface in this way becomes a necessity. What this dissertation has argued for is the necessity of the CHAPTER 6. CONCLUSION 299

Theta Structure Property for SVCs, repeated in (6.8), as the means of taking SVC deﬁnition to the level of argument structure:

(6.8) The Theta Structure Property for SVCs: An SVC has the Theta Structure of a Possible Verb structure. Diagnostic for an SVC meeting the Theta Structure Property: If a multi-verb construction has the same number of structural arguments identified at Theta Structure, and the same list of configurationally defined, overtly realized non-structural (oblique) arguments as some clause headed by a single, simplex verb, then it meets the Theta Structure Property for SVCs.

The reasons for preferring the Theta Structure Property over other argument structural criteria were mentioned in Chapter 2, but can be summarized briefly here: Internal Argument Sharing excludes several types of sentence which under any other reasonable criteria would properly qualify as SVCs (‘go’/‘come’+Agentive Verb combinations featuring prominently here), while functional approaches, although tending to be in keeping with the idea that SVCs should display monoclausal argument structure, do not actually set in place any kind of definition towards these ends (but instead just tend to list off all the possible argument structural patterns). And so the Theta Structure Property is entirely necessary for the purpose of defining SVCs – for including (6.5) and (6.6), and excluding covert coordinations like (6.7); the main question that remains is: why hasn’t it been done before? But before addressing this issue, I want to emphasize one of the strongest points in favor of the Theta Structure Prop- erty, which is that it accounts equally well for both contiguous and non-contiguous SVCs. One of the main features of this dissertation has been the way in which it has turned the tables on what should qualify as an SVC, by including contiguous SVCs like (6.9) which many SVC researchers would exclude from the class, while at the same time questioning the inclusion of some of the more traditional members of the class like (6.10): CHAPTER 6. CONCLUSION 300

(6.9) Fero rumpa weha.na fo’o. Fero ﬁnd take.3 mango ‘Fero found and took a mango.’ (Bima)

` (6.10) Ozó kpèé èmà dé. . Ozo beat drum buy ‘Ozo beat the drum (and then) bought it.’ (Edo; Stewart 2001, 49)

The inclusion or exclusion of SVCs like (6.10) is a topic that will be taken up in the next section (where it will also be noted that the languages containing such SVCs also tend to be the same languages from whose classes of SVCs covert coordinations need to be excluded). But the observation I will make here and now is that non- contiguous SVCs like (6.10) seem to be the only type of SVC that occur in African (and African substrate creole) languages,2 and since for many decades SVCs were only known of in African languages, this may have created a historical bias whereby only non-contiguous SVCs were seen to ‘count’ as cases of serialization. The recognition of contiguous SVCs like (6.9) represents a break from this tradition that arises out of looking at a more balanced cross-linguistic survey of SVCs, I would argue, although it may also be true that this approach runs the risk of including certain compound verbs in the class of SVCs, in the absence of any reliable evidence that the verbs concerned each constitute distinct phonological words. On the other hand, it seems perfectly valid to me to regard compound verbs as a type of SVC, as Margetts (1999) has done (in “propos[ing] that the term ‘compound’ does not by definition contradict an analysis as serialization” (1999, 101)). But whatever stance is taken, the important point is that not all sentences with contiguous verbs can be dismissed from the class of SVCs as compound verbs, as in Baker (1989); there must be some contiguous SVCs. The Theta Structure Property for SVCs accounts equally well for both contiguous and non-contiguous SVCs for the very reason referred to in the title for this section: purely because it takes SVC definition below the surface. Because the Theta Structure Property is defined at the argument structure level of Theta Structure and Semantic Form in LDG, this leaves structures fulfilling the property free to project into whatever

With Collins’ (2002) }Hoan ‘verb compounds’, presented in Section 2.4.2, as some possible exceptions. CHAPTER 6. CONCLUSION 301

(contiguous or non-contiguous) syntactic structures the language dictates, in contrast to many other approaches to SVCs (such as Good (2003) surveys and Section 1.2.3 summarizes) that define SVCs through various (non-contiguous) surface syntactic configurations. So, the act of taking SVC definition below the surface is not just necessary for the purposes of including multiple object serializations like (6.5)-(6.6) and excluding covert coordinations like (6.7), it is also decidedly advantageous in its inclusion of both contiguous and non-contiguous structures. When we begin to look at the reasons why the Theta Structure Property has not been done before, an initial observation to make is that approaches that define SVCs through particular non-contiguous syntactic configurations appear farthest from formulating anything like the Theta Structure Property. This can be attributed to the simple absence of any recognition in this research that SVCs display the argument structure of a clause headed by a single verb, and this is usually accompanied by the more general absence of any kind of recognition that SVCs are constructions which are behaving as if they contain only one verb. Many approaches that recognize only non-contiguous SVCs appoint Internal Argument Sharing as a central SVC-defining property. But while Internal Argument Sharing does take SVC definition to the level of argument structure, it is not the case that this property enforces monoclausal argument structure. The purpose of Internal Argument Sharing is to exclude covert coordinations, which only happen not to display monoclausal argument structure; Internal Argument Sharing otherwise excludes numerous other constructions that were argued in Chapter 2 to display monoclausal argument structure. Now there is no necessary reason why approaches that recognize only non-contiguous SVCs should be farthest from attributing SVCs monoclausal argument structure. But it is nonetheless unsurprising that approaches that additionally recognize contiguous SVCs are more likely to identify monoclausal argument structure as a property of SVCs, given that, on the whole, the surface argument structural appearance of contiguous SVCs (i.e. their arrangements of arguments+adjuncts) more resembles that of a clause headed by a single verb. Bril and Ozanne-Rivierre (2004), Aikhen- vald (2006) and Senft (2008) are some examples of SVC research that recognize both the existence of contiguous SVCs and the fact that SVCs have monoclausal argument CHAPTER 6. CONCLUSION 302

structure. These studies do not begin to formulate anything like the Theta Structure Property though, since, in being more functionally and typologically-oriented, they have no formal theory in place to do so. Durie (1997) stands out as the one study to recognize that SVCs have monoclausal argument structure and to analyze serialization in a formal semantic framework (that of Jackendoff (1990)), although his analysis consists of only a brief preliminary sketch (of one SVC in isolation) that has never been extended. Supposing some SVC study were to both recognize the monoclausal argument structural nature of SVCs and adopt a theoretical framework within which an equivalent of the Theta Structure Property could be formulated, then is there anything intrinsically difficult or challenging for this formulation? I would say not in essence, although probably the most challenging key feature is to draw a division between structural and non-structural arguments. The Theta Structure Property, or any equivalent to it, needs to treat these two types of argument in very different ways. Whereas all thematic role information pertaining to a structural argument must be ignored, since this information could be of a complex and composite nature such as could never hold of a single argument in a clause headed by a single verb, the thematic role status of a non-structural argument, in contrast, figures crucially in an SVC’s satisfaction of the Theta Structure Property. The thematic role statuses of non-structural arguments figure crucially in how we understand certain SVCs to have the argument structure of a clause headed by a single verb, often in spite of the fact that the surface structural appearance of these SVCs (especially if they are multiple object non-contiguous serializations) look to be at odds with this status. Of course, another requirement for the formulation of something like the Theta Structure Property is the need to have clearly defined thematic roles. It so happens that LDG draws the key distinction between structural and non- structural arguments at Theta Structure, while also laying out the kinds of predicate decompositions at Semantic Form from which thematic role information can easily be configurationally defined and identified. This explains the ease with which the Theta Structure Property can be formulated in this framework, which was in fact one motivation behind implementing this property in LDG in the first place. But at the same CHAPTER 6. CONCLUSION 303

time I would stress that this property is completely adaptable into other frameworks. An implementation in Lexical Functional Grammar (LFG) for example, would need to hone in on those arguments not marked with the features [+r(estricted),+o(bjective)] at a-structure for the purposes of isolating the structural arguments (corresponding to SUBJ, OBJ and OBJθ at F-structure), leaving all other [+r,+o] arguments as non-structural arguments (OBLθ at F-structure, where θ specifies the theta role values). The nearest analogues to structural arguments in Role and Reference Grammar (RRG), meanwhile, are the generalized semantic roles, or macroroles, of actor and undergoer. The thematic roles of all non-macrorole arguments (i.e. the analogues of non-structural arguments) are easily configurationally identified in RRG’s Logical Structures – lexical decompositions that differ in only a few details from the Semantic Form representations presented throughout this dissertation. So long as a theoretical framework incorporates some level of argument structure representation that encompasses both syntactically relevant information about which arguments are projected into core grammatical relation positions, along with semantic, thematic role information about all the remaining arguments, this framework should be able to capture which SVCs have the monoclausal argument structure the Theta Structure Property prescribes.

6.2 SVCs: substitutes for clauses headed by single verbs, and what they tell us when they’re not substituting clauses headed by single verbs

In addition to proposing the Theta Structure Property for SVCs as a core deﬁning property for SVCs, this dissertation argues for the recognition of an important distinction: that between Possible Verb and Impossible Verb SVCs. In practical terms, Impossible Verb SVCs need to be distinguished from Possible Verb SVCs for the reason that the former have the argument structure of more than one clause headed by a single verb, and so must undergo the Predicate Combining Operations for Impossible CHAPTER 6. CONCLUSION 304

Verb SVCs in order to fulfill the Theta Structure Property. But the broader significance of the Possible Verb-Impossible Verb distinction is that it tells us something about SVCs that the numerous other two-way distinctions in research on SVCs do not – it tells us that some SVCs are doing things done with single-verb-headed clauses in other languages while others are not. The Possible Verb-Impossible Verb distinction supplies a precise semantic formulation for which SVCs could be regarded as substitutes for clauses headed by single verbs. Possible Verb SVCs can be regarded as substitutes for clauses headed by single verbs since they fulfill LDG’s Possible Verbs constraints, and since they have one of three event structural profiles (unitary-event, coidentified or non-coidentified resultative) that a clause headed by a single verb could have. Chapters 4 and 5 provided further details of what a Possible Verb structure is, in the domains of instrumental and motion expressions respectively. Chapter 5 in particular formulated several constraints on Possible Verb motion events, all but one of which can be reformulated as constraints on Possible Verb events in general. Being able to apply LDG’s Possible Verbs constraints, event structural profiling, and the constraints on Possible Verb events derived in Chapter 5 means being able to draw a clear line between Possible Verb and Impossible Verb status in SVCs, even if the Possible Verb or Impossible Verb status of many SVCs appears ambiguous, and even if establishing these statuses can take some amount of work. Being able to apply these constraints and event structural profiles also means having a precise semantic formulation of which SVCs (i.e. Possible Verb SVCs) can be regarded as substitutes for clauses headed by single verbs, as mentioned above, although this precise semantic formulation is by no means the only, necessary interpretation of when an SVC might be substituting a clause headed by a single verb. Suppose, for example, a language were to have Impossible Verb instrumental SVCs (where ‘take’ means TAKE in a sequential structure, as described in Chapter 4), and no other means of expressing instrumental events, then it would seem only natural to regard these Impossible Verb instrumental SVCs as substitutes for clauses headed by single verbs. Impossible Verb instrumental SVCs would be regarded as fulfilling exactly the same function as simplex clause expressions of instrumental events in other languages. At the same time, certain Possible Verb SVCs may not even be felicitously regarded as substitutes CHAPTER 6. CONCLUSION 305

for clauses headed by single verbs, as in the case of Thai (6.11), which, as (6.12) demonstrates, does not respond to the absence of simplex instrumental expression in this language (but may simply entail an additional semantic element of deﬁniteness

for the instrument object of aw ‘take’, as Jagacinksi (1992) suggests):

(6.11) Kháw aw mˆıit pay tàt yâa. 1sg take knife go cut grass ‘He cuts grass with a knife.’ (Thai; Foley and Olson 1984, 54)

(6.12) Kháw tàt yâa dûay mˆıit. he cut grass with knife ‘He cuts grass with a knife.’ (Thai; Foley and Olson 1984, 54)

However, the focus for the rest of this section will be on Impossible Verb SVCs – speciﬁcally those Impossible Verb SVCs which would never be regarded as substitutes for clauses headed by single verbs, such as (6.13)-(6.15):

(6.13) Myt ritm muh-hambray-an-m. tree insects climb-search.for-1s-3pl ‘I climbed the tree, looking for insects.’/‘I climbed the tree and looked for insects.’ (Alamblak; Bruce 1986, 26)

´ (6.14) O je. un s`un. he ate sleep ‘He ate before he slept.’ (Yoruba; Bamgbos.e 1974, 30)

(6.15) Nws dhia tshov qeej. 3sg dance blow bamboo.pipes ‘He dances playing the pipes.’ (White Hmong; Jarkey 1991, 169)

If the driving idea behind this dissertation is that SVCs are constructions which are acting as if they contain only one verb even though they contain at least two, then the driving question behind this dissertation is: why are SVCs acting as if they contain only one verb? In particular, why are SVCs like (6.13)-(6.15), whose semantic content could never even be expressed in a clause headed by a single verb, acting as if they contain only one verb? (6.13)-(6.15) were chosen because they provide the CHAPTER 6. CONCLUSION 306

most interesting answer to this question: (6.13)-(6.15) can be expressed in SVC form, with all its monoclausal facets, only because the actions described in these sentences can be jointly conceptualized as a single event, or ‘recognizable event type’, by the speakers of Alamblak, Yoruba and White Hmong respectively. Climbing a tree and looking for insects can only be expressed in SVC form in (6.13) because these two actions are commonly performed together by Alamblak speakers, sleeping then eating is a common course of events for speakers of Yoruba, explaining the acceptability of the SVC in (6.14), while dancing and playing bamboo pipes is expressed in an SVC in (6.15) because these are typical joint activities in White Hmong society. In each of (6.13)-(6.15), the actions described ‘go together’ in culturally speciﬁc ways. (6.16)- (6.18), in contrast, cannot be expressed in SVC form, since the concepts described cannot be conceptualized as recognizable event types by the speakers of Alamblak,

Yoruba and White Hmong: (6.16) *Myt gu˜nm muh-h ti-an-m. tree stars climb-see-1s-3pl (‘I climbed the tree, looking for stars.’) (Bruce 1986, 26)

´ (6.17) *O je. un pad`a. he ate return (‘He ate and came back.’) (Yoruba; Bamgbos.e 1974, 31)

(6.18) *Nws dhia mloog nkauj. 3sg dance listen song (‘He dances and listens to music.’) (White Hmong; Jarkey 1991, 170)

This kind of ‘recognizable event type’ data speaks directly to the presupposition carried in the heading for this section, that SVCs that cannot be substitutes for clauses headed by single verbs might actually tell us something. The SVCs in (6.13)- (6.15), in each case impossible as substitutes for clauses headed by single verbs, tell us something about what the speakers of the languages concerned conceptualize as single events. But are Impossible Verb SVCs which cannot be substitutes for single-verb-headed clauses always telling us something about what the speakers of serializing languages conceptualize as recognizable event types? In Section 1.2.4.2 I CHAPTER 6. CONCLUSION 307

suggested that this cannot be the case, since languages display variable limits on SVC formation, many of which appear not to align with any notion of recognizable event type. Some languages (such as Edo and Saramaccan, which I will consider shortly) may not even display any limits on their SVCs, in which case the Impossible Verb SVC non-substitutes for clauses headed by single verbs in these languages might not even tell us anything. What is required to get to the bottom of this issue of what SVC non-substitutes for clauses headed by single verbs may or may not tell us are more comprehensive surveys of how SVC formation is limited in different languages. Mostly, this has to be a direction for future research, although I can at least say a little more here about Bima SVCs, and how they are constrained, since I have more SVC data from Bima than any other language. Chapter 5 already described how Bima Impossible Verb motion SVCs tend to describe an initial, relatively insignificant motion subevent that states the figure’s position or location, followed by a ‘main’ motion subevent of more narrative substance. The ubiquitous londo lao SVCs (where londo refers to the descent from a traditional Bima house) such as (6.19) fit this mould, as does (6.20), where losa refers to the exit from a more modern house that does not have a steep descent:

(6.19) ede mpara londo lao kai.na d’i uma ompu, ... dem then descend go kai.3 loc house old-man ‘then he went down and went to the old man’s house, ...’ (Bima)

(6.20) Sia d’oho na.losa lampa-lampa kai Ben Hur. 3p pl 3.exit go-around instr Ben Hur ‘They’ll go out (of the house) and go around in a Ben Hur.’ (Bima)

(6.21) and (6.22) also contain initial motion verbs that clarify matters of position or location (which may in any case already be obvious from the context), but these are followed by non-motion verbs that describe ‘main’ events of more narrative substance:

(6.21) Amu lao ne’e maru.na. Amu go ascend sleep.3 ‘Amu went up (to her room/into the house) and slept.’ (Bima) CHAPTER 6. CONCLUSION 308

(6.22) Fero dula lowi.na janga kai roa na’e. Fero go-home cook.3 chicken instr pot big ‘Fero went home and cooked the chicken in a big pot.’ (Bima)

(6.23)-(6.25) also describe relatively insigniﬁcant and contextually recoverable initial subevents; in (6.23) one that pertains to the actor’s position or posture before com- mencing the main action of dancing, and in (6.24)-(6.25) ones that pertain to the object’s position or state of control or possession in relation to the actor:

(6.23) ede.ra tu’u rebo kai nahu, ... dem.emph get-up dance kai 1p ‘and I got up and danced, ...’ (Bima)

(6.24) Amu hanta wa’a.na meja ese uma. Amu lift-up carry.3 table up-to house ‘Amu carried the table up to the house.’ (Bima)

(6.25) Reho weha cili.na piti. Reho take hide.3 money ‘Reho took and hid the money.’ (Bima)

The initial subevents in all of (6.19)-(6.25) may be best described as ‘inceptive’ events; the verbs concerned all describe actions that occur at the inception of the main events described in the verbs that follow. Van Staden and Narasimhan (2004) describe a similar tendency in the SVCs of Tidore; the initial verb in (6.26), for example, describes the inceptive action of fetching a machete, which speakers included in spite of the fact that no action of fetching the machete was portrayed in the video stimulus for this example:

(6.26) Nau=ge oro peda tola luto. boy=there fetch machete chop ﬁre.wood ‘The boy fetches a machete (and) chops ﬁre wood.’ (Tidore; van Staden and Narasimhan 2004, 7)

Van Staden and Narasimhan (2004) claim that (6.26) demonstrates how Tidore speakers segment the continuum of experience and perception into events, in such a way CHAPTER 6. CONCLUSION 309

that the starting point of an event is placed at its inception, or inceptive action. In the same way, (6.19)-(6.25) may tell us something about how Bima speakers segment events; they too may perceive the inception of an event as a natural event boundary. What Bima Impossible Verb SVCs apparently do not describe are two independent actions of equal weight, even where it might be expected that these actions would go together naturally and so be perceived as ‘recognizable event types’.3 Across languages, SVC non-substitutes for clauses headed by single verbs could thus be telling us diﬀerent things about how speakers conceptualize actions and events; some SVCs may indicate which action combinations naturally constitute recognizable event types, while others may indicate where speakers perceive event boundaries. A diﬀerent case again is Kalam, in which SVCs like (6.27) include all of the necessary elements of a well-formed event report listed in (6.28):

(6.27) B ak am mon p-wk d ap ay-a-k. man that go wood hit-break get come put-3sg-past ‘The man fetched some ﬁrewood.’ (literally: ‘The man went and chopped and took some ﬁrewood and brought it back.’) (Kalam; Pawley 1993, 95)

(6.28) 1: Whether or not the actor had to move from his previous location (scene 1) to the scene of the action(s) (call it scene 2). 2: What he did at scene 2. If his action was aimed at obtaining something, whether or not he obtained it. 3: Whether or not he moved from scene 2 to another location (scene 3) and whether or not he carried the object obtained in scene 2. 4: What he did with the object obtained at scene 2 (or if transported, at scene 3).

3Bima cannot describe the naturally cooccurring actions of cooking then eating in SVC form, for example, although the one SVC with ‘equally weighted’ verbs I did encounter was the following: Reho sangcicu mab’u.na Reho trip fall.3 ‘Reho tripped and fell.’ This SVC may be acceptable for the reason that the actions of tripping and falling are so very tightly connected (temporally and causally). CHAPTER 6. CONCLUSION 310

To not include all of the components listed in (6.28) would render a sentence ungrammatical, and so Kalam, in addition to placing event boundaries further from the ‘main’ event, also conventionalizes this kind of event segmentation grammatically. Where there is data on what the limitations on SVCs in certain languages are, this can tell us some rather interesting things. But where there is no data, or where the data is indicative of there being no limits on SVC formation, this opens up some new questions about serialization in the languages concerned. Edo has already been singled out as a language which seems to allow every imaginable combination of verbs in its SVCs, or at very least Stewart (2001) gives no indication of any limit on the formation of these SVCs, which describe such seemingly unrecognizable event types as buying and reading LGB or selling then cooking yams:

` (6.29) Ozó dé. LGB tié. Ozo buy LGB read ‘OzoboughtLGBandreadit.’ (Edo;Stewart2001,60)

` (6.30) Ozó khié. n `ıyán lé. Ozo sell yam cook ‘Ozo sold the yams and cooked them.’ (Edo; Stewart 2001, 49)

The following, very contentful Saramaccan example has also been held up as beggaring belief about what could constitute a recognizable event type for the speakers of this language:

(6.31) A k´ısi d´ı fou náki k´ıi limbó bói njan. 3sg catch det bird hit kill clean cook eat ‘He caught the bird, struck it dead, cleaned, cooked and ate it.’ (Saramaccan; Veenstra 1996, 103)

The questions that these examples open up are, ﬁrstly, might there be some SVCs which in fact tell us nothing about event conceptualization or segmentation in the languages concerned? This seems the most likely interpretation – that the use of SVC structure in (6.29)-(6.31) signiﬁes nothing more than the iconic reading that the actions described occur in a tighter temporal connection than would otherwise be CHAPTER 6. CONCLUSION 311

the case if multi-clausal structure were instead used. But the other question worth asking is whether (6.29)-(6.31) are really SVCs. Edo and Saramaccan ‘SVCs’ of this type appear suspicious for the reason that they apparently are so very productive, expressing event structural combinations that go above and beyond the upper limits on SVC formation in other languages. Another suspicious feature is the cooccurrence of covert coordination structures in these languages. Could (6.29)-(6.31) actually be covert coordinations and not SVCs? The difficulty is that (6.29)-(6.31) fulfill all the properties held to define SVCs set out in the conclusion to Chapter 2, including the Theta Structure Property. (6.29)-(6.31) do display the surface configurations of arguments found in clauses headed by a single verbs in Edo and Saramaccan. If (6.29)-(6.31) were to be excluded from the class of SVCs, then some new SVC-defining property would need to be formulated that excludes them. But what would it look like, and how far below the surface of all the overt properties characterizing monoclausal behavior would it have to delve? The other possibility is that (6.29)-(6.31) may fail to fulfill language-specific properties of what it means to be monoclausal, meaning that no additional SVC-defining property would be necessary. I will conclude this dissertation here, on this question of whether or not sentences like (6.29)-(6.31) really are SVCs. For while it may well be the case that all serial verb form may signify in these cases is the tighter temporal connection of the events described, at the same time I want to emphasize that these apparently event structurally unlimited SVCs do stand in contrast to the SVCs of all the other languages considered in this dissertation. The SVCs of all the other languages do display some kind of event structural limit, where this applies also to those languages whose SVCs show the same syntactic characteristics (such as non-contiguity) as Edo and Saramaccan. Fongbe SVCs, for example, are syntactically similar to those of Edo, yet Lefebvre and Brousseau (2002) observe that the former are far more limited in terms of what they can describe than Stewart’s (2001) Edo SVCs. What is striking is that the majority of languages, whose SVCs do display event structural limits, give such interesting and contentful answers to the question why are these Impossible Verb SVCs acting as if they contain only one verb? – answers that can actually tell us something about how speakers perceive and conceptualize events. In this light it becomes hard to accept CHAPTER 6. CONCLUSION 312

that there may be some languages whose SVCs tell us nothing very much. On the one hand, it may just be that this is a reality that must be accepted; this is probably the more sensible position. But on the other hand, event structurally seemingly unlimited SVCs like (6.29)-(6.31), and their cooccurrences with covert coordinations, may open up new avenues in the study of non-contiguous SVCs. For if one of the ﬁrst major achievements in the study of SVCs was to distinguish SVCs from covert coordinations, the next step could be to identify covert coordinations which disguise themselves rather more cunningly as SVCs. Bibliography 313

Bibliography

Abraham, R. C. 1958. Dictionary of modern Yoruba. London: University of London Press.

Aikhenvald, A. 2006. Serial verb constructions in typological perspective. In A. Aikhenvald and Dixon (Eds.), Serial Verb Constructions, 1–68. Oxford: Oxford University Press.

Alsina, A. 1996. The Role of Argument Structure in Grammar. Stanford, CA: CSLI Publications.

Amoida, A., F. Amoida, and T. Amoida. 1978. Wanl¨oKontu fu Anasi. Paramribo: Instituut voor Taalwetenschap.

Amoida, F. 1974a. Bunu ku Hogi 2. Paramaribo: Instituut voor Taalwetenschap.

Amoida, F. 1974b. Fufuu, huku, ku taanga. Paramaribo: Instituut voor Taalweten- schap.

Amoida, F. 1974c. Wakama Oto. Paramaribo: Instituut voor Taalwetenschap.

Amoida, F. 1974d. Wookoma Oto. Paramaribo: Instituut voor Taalwetenschap.

Amoida, S., and C. Kuse. 1974. Wanl¨oFesiten Oto. Suriname: S.I.L.

Ansre, G. 1966. The verbid – a caveat to ‘serial verbs’. Journal of West African Languages 3(1):29–32.

Awobuluyi, O. 1973. The modifying serial construction: a critique. Studies in African Linguistics 4(1):87–111.

Aw´oyal´e, Y. 1988. Complex predicates and verb serialization. Lexicon Project Work- ing Papers 28.

Baker, M. 1989. Object sharing and projection in serial verb constructions. Linguistic Inquiry 20(4):513–553. Bibliography 314

Bamgbos.e, A. 1973. The modifying serial construction: a reply. Studies in African Linguistics 4(2):207–217.

Bamgbos.e, A. 1974. On serial verbs and verbal status. Journal of West African Languages 9:17–48.

Baxter, A. N. 1988. A grammar of Kristang. Canberra: Paciﬁc Linguistics.

Beejën, L. 1982. MötëBuku. Paramaribo: Instituut voor Taalwetenschap.

Bierwisch, M. 1983. Semantische und konzeptuelle Repr¨asentationen lexikalischer Einheiten. In R. Ruzicka and W. Motsch (Eds.), Untersuchungen zur Semantik, 61–99. Berlin: Akademie-Verlag.

Bodomo, A. B., O. S. C. Lam, and L. S. S. Yu. 2003. Double object and serial verb benefactive constructions in Cantonese. In M. Butt and T. Holloway King (Eds.), Proceedings of the LFG03 Conference, University at Albany, State University of New York. Stanford, CA, CSLI Publications.

Bohnemeyer, J. 1999. Event representation: some primordial soup for the evolution of a research project on event representation in language and cognition. Manuscript.

Bohnemeyer, J. 2000. Constraints on motion event coding: Vectors or path shapes? Manuscript.

Bohnemeyer, J. 2001. Linguistic principles of motion event segmentation. Paper presented at the ALT IV, Santa Barbara.

Bohnemeyer, J. 2003. The unique vector constraint: The impact of direction changes on the linguistic segmentation of motion events. In E. van der Zee and J. Slack (Eds.), Representing direction in language and space, 86–110. Oxford: Oxford University Press.

Bohnemeyer, J., N. J. Enﬁeld, J. Essegbey, I. Ibarretxe-Antunano, S. Kita, F. L¨upke, and F. K. Ameka. 2007. Principles of event representation in language: The case of motion events. Language 83(3):495–532. Bibliography 315

Bohnemeyer, J., and C. Stolz. 2006. Spatial reference in Yukatek Maya: a survey. In S. Levinson and D. Wilkins (Eds.), Grammars of space : explorations in cognitive diversity, 273–310. Cambridge, UK: Cambridge University Press.

Bowden, J. 2001. Taba: a description of a South Halmahera language. Canberra: Research School of Paciﬁc Studies.

Bradshaw, J. 1993. Subject relationships within serial verb constructions in Numbami and Jabˆem. Oceanic Linguistics 32(1):133–161.

Bresnan, J. 1982. The passive in lexical theory. In J. Bresnan (Ed.), The Mental Representation of Grammatical Relations. Cambridge, MA: The MIT Press.

Bril, I., and Ozanne-Rivierre (Eds.). 2004. Complex Predicates in Oceanic Languages. Berlin: Mouton de Gruyter.

Bril, I. 2004. Complex verbs and dependency strategies in Nˆelˆemwa (New Caledonia). In I. Bril and Ozanne-Rivierre (Eds.), Complex Predicates in Oceanic Languages, 167–198. Berlin: Mouton de Gruyter.

Brousseau, A.-M. 1998. R´ealisations argumentales et classes de verbs en Fongb`e. Paris: Peeters.

Bruce, L. 1984. The Alamblak Language of Papua New Guinea (East Sepik). Can- berra: Paciﬁc Linguistics.

Bruce, L. 1986. Serialisation: the interface of syntax and lexcion. Papers in New Guinea Linguistics (24):21–37.

Brunson, B. 1992. Thematic discontinuity. PhD thesis, University of Toronto.

Butt, M. 1995. The Structure of Complex Predicates. Stanford, CA: CSLI Publica- tions.

Butt, M. 2003. The light verb jungle. Workshop on Multi-verb Constructions. At: www.ling.hf.ntnu.no/tross/Butt.pdf. Bibliography 316

Byrne, F. 1985. Verb Serialization and Predicate Complementation in Saramaccan. PhD thesis, University of Arizona.

Byrne, F. 1987. Grammatical Relations in a Radical Creole: Verb Complementation in Saramaccan. Amsterdam: John Benjamins.

Campbell, R. 1996. Serial verbs and shared arguments. The Linguistic Review 13:83–118.

Carstens, V. 1988. Serial verbs in yoruba. Paper presented at the 2nd workshop in Niger-Congo syntax and semantics. MIT, Cambridge, MA.

Carstens, V. 2002. Antisymmetry and word order in serial verb constructions. Lan- guage 78(1):3–50.

Choi, S. 2005. Multiple verb constructions in Korean. PhD thesis, University of Sussex.

Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton.

Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris.

Christaller, Rev, J. G. 1875. A Grammar of the Asante and Fante Language called Tshi. Basel: Gregg, Ridgewood.

Clark, E., and H. Clark. 1979. When nouns surface as verbs. Language 55:767–811.

Cohen, N. 1976. Some interclausal relations in Jeh. The Mon-Khmer Studies Journal 5:153–164.

Collins, C. 1993. Topics in Ewe syntax. PhD thesis, M.I.T.

Collins, C. 1997. Argument sharing in serial verb constructions. Linguistic Inquiry 28:461–497.

Collins, C. 2002. Multiple verb movement in }Hoan. Linguistic Inquiry 33:1–29. Bibliography 317

Comrie, B. 1995. Serial verbs in Haruai (Papua New Guinea) and their theoretical implications. In J. Bouscaren, J.-J. Franckel, and S. Robert (Eds.), Langues et langage: Problèmes et raisonnement en linguistique, mélanges offerts àAntoine Culioli, 25–37. Paris: Presses Universitaires de France.

Croft, W. 1991. Syntactic Categories and Grammatical Relations. Chicago: University of Chicago Press.

Croft, W. 1998. Event structure in argument linking. In M. Butt and W. Geuder (Eds.), The projection of arguments, 21–63. Stanford, CA: CSLI Publications.

Crowley, T. 2002. Serial Verbs in Oceanic. Oxford: Oxford University Press.

Davidson, D. 1969. Truth and meaning. In J. Davis, D. Hockney, and W. Wilson (Eds.), Philosophical Logic, 1–20. Dordrecht: Reidel.

D´echaine, R.-M. 1993. Predicates Across Categories. PhD thesis, Amherst.

Dillon, C. 2004. Verbs of motion in nigerian pidgin english: Manner, direction, and serial verb constructions. At: https://www.indiana.edu/ iulcwp/pdfs/03- dillon04.pdf.

Dowty, D. 1979. Word meaning and Montague Grammar. Dordrecht: Reidel.

Durie, M. 1997. Grammatical structures in verb serialization. In A. Alsina, J. Bresnan, and P. Sells (Eds.), Complex Predicates, 289–354. Stanford, CA: CSLI Publica- tions.

Early, R. 1993. Nuclear layer serialization in Lewo. Oceanic Linguistics 32(1):65–93.

Enﬁeld, N. J. 1998. Event typicality and culture-related constraints on serial verb combinations. ALI Ethnosyntax workshop handout, Brisbane.

Enfield, N. J. 2002. Functions of ‘give’ and ‘take’ in Lao complex predicates. In R. S. Bauer (Ed.), Collected Papers on Southeast Asian and Pacific Languages, 13–36. Canberra: Pacific Linguistics. Bibliography 318

Engelberg, S. 1994. Valency and aspectuality. In D. Halwachs and I. St¨utz (Eds.), Sprache, Sprechen, Handeln I, 53–59. T¨ubingen: Niemeyer.

Engelberg, S. 2000. Verb meaning and event structure. Lacus Forum 26:257–268.

Fillmore, C. 1968. The case for case. In E. Bach and R. Harms (Eds.), Universals in Linguistic Theory, 1–87. New York: Holt, Rinehart and Winston.

Fillmore, C. 1982. Frame semantics. In TLS of Korea (Ed.), Linguistics in the Morning Calm, 111–137. Seoul: Hanshin Publishing Co.

Foley, W. A. 2006. Events and serial verb constructions. Handout from talk at School of Oriental and African Studies.

Foley, W. A., and M. Olson. 1985. Clausehood and verb serialization. In J. Nichols and A. C. Woodbury (Eds.), Grammar Inside and Outside the Clause, 17–60. Cambridge, UK: Cambridge University Press.

Foley, W. A., and V. Valin. 1984. Functional Syntax and Universal Grammar. Cam- bridge, UK: Cambridge University Press.

Francois, A. 2004. Constraints and creativity in the macro-verb strategies of Mwotlap. In Bril and Ozanne-Rivierre (Eds.), Complex Predicates in Oceanic Languages, 107–143. Berlin: Mouton de Gruyter.

George, I. 1976. Verb serialization and lexical decomposition. Studies in African Linguistics 6:63–72.

George, I. 1982. The Na ... Na construction in Nupe. Jolan: Journal of the Linguistic Association of Nigeria 1:35–45.

George, I. 1985. Complex verbs in Nupe and Yoruba. Studies in African Linguistics 16:295–321.

Giv´on, T. 1991. Some substantive issues concerning verb serialization: Grammat- ical vs. cognitive packaging. In C. Lefebvre (Ed.), Serial Verbs: Grammatical, Comparative and Cognitive Approaches, 137–184. Amsterdam: John Benjamins. Bibliography 319

Good, J. 2003. Strong Linearity: Three Case Studies Towards a Theory of Mor- phosyntactic Templatic Constructions. PhD thesis, U.C. Berkeley.

Grimshaw, J. 1990. Argument Structure. Cambridge, MA: MIT Press.

Grimshaw, J. 1994. Semantic content and semantic structure. Ms, Rutgers University.

Grimshaw, J., and S. Vikner. 1993. Obligatory adjuncts and the structure of events. In E. Reuland and W. Abraham (Eds.), Knowledge and Language II, 143–155. Kluwer.

Hale, K. 1991. Misumalpan verb sequencing constructions. In C. Lefebvre (Ed.), Serial Verbs: Grammatical, Comparative and Cognitive Approaches, 199–216. Amster- dam: John Benjamins.

Hamel, P. 1993. Serial verbs in Loniu and an evolving preposition. Oceanic Linguistics 32(1):111–132.

Hellan, L., D. Beermann, and E. S. Andenes. 2003. Towards a typology of serial verb constructions in akan. In Proceedings of the Workshop on Multi-verb Construc- tions, Trondheim. At: http://edvarda.hf.ntnu.no/ling/tross/hellanBeermann.pdf.

Hyslop, C. 2001. The Lolovoli Dialect of the North-East Ambae Language, Vanuatu. Canberra: Paciﬁc Linguistics.

Iwasaki, S., and P. Inkaphiromu. 2005. A Reference Grammar of Thai. Cambridge, UK: Cambridge University Press.

Jackendoﬀ, R. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press.

Jackendoﬀ, R. 1983. Semantics and Cognition. Cambridge, MA: MIT Press.

Jackendoﬀ, R. 1990. Semantic Structures. Cambridge, MA: MIT Press.

Jagacinski, N. 1992. The ?au usages in Thai. In C. J. Compton and J. F. Hartmann (Eds.), Papers on Tai Languages, Linguistics, and Literatures. Northen Illinois University. Bibliography 320

Jansen, B., H. Koopman, and P. Muysken. 1978. Serial verbs in the creole languages. In Amsterdam Creole Studies II. University of Amsterdam Instituut voor Algemene Taalwetenschap.

Jarkey, N. 1991. Serial Verbs in White Hmong: a Functional Approach. PhD thesis, University of Sydney.

Jauncey, D. G. 1997. A Grammar of Tamambo, the Language of Western Malo, Vanuatu. PhD thesis, Australian National University, Canberra.

Kaufmann, I. 1995. What is an (im)possible verb? Restrictions on semantic form and their consequences for argument structure. Folia Linguistica 29:67–103.

Kaufmann, I., and D. Wunderlich. 1998. Cross-linguistic patterns of resultatives. In Working Papers SFB 282 Theory of the Lexicon, 109. University of D¨usseldorf.

Kayne, R. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press.

Kiparsky, P. 2001. Structural case in Finnish. Lingua 111:315–376.

Klamer, M. 1998. A grammar of Kambera. Berlin/New York: Mouton de Gruyter.

Koenig, J.-P., G. Mauner, B. Bienvenue, and K. Conklin. 2007. What with? The anatomy of a (proto)-role. Journal of Semantics 25:175–220.

Kroeger, P. 2004. Analyzing Syntax. Cambridge, UK: Cambridge University Press.

Lakoﬀ, G. 1968. Instrumental adverbs and the concept of deep structure. Foundations of Language 4:4–29.

Lane, J. 1991. Kalam serial verb constructions. MA thesis, Dept. of Anthropology, University of Auckland.

Larson, R. 1991. Some issues in verb serialization. In C. Lefebvre (Ed.), Serial Verbs: Grammatical, Comparative and Cognitive Approaches, 185–210. Amsterdam: John Benjamins. Bibliography 321

Law, P. 1996. A note on the serial verb construction in Chinese. Cahiers de Linguistique-Asie Orientale 25:199–233.

Law, P., and T. Veenstra. 1992. On empty operators in serial verb constructions. In MIT Working Papers in Linguistics 17: Proceedings of the Kwa Comparative Syntax Workshop, 183–203.

Lefebvre, C. 1991. Take serial verb constructions in Fon. In C. Lefebvre (Ed.), Serial Verbs: Grammatical, Comparative and Cognitive Approaches, 37–78. Amsterdam: John Benjamins.

Lefebvre, C., and A.-M. Brousseau. 2002. A Grammar of Fongbe. Berlin: Mouton de Gruyter.

Levin, B. 1993. English Verb Classes and Alternations: A Preliminary Investigation. Chicago, IL: University of Chicago Press.

Levin, B., and M. Rappaport Hovav. 1999. Two structures for compositionally derived events. In Proceedings of SALT, Vol. 9, 199–223.

Levin, B., and M. Rappaport Hovav. 2005. Argument Realization. Cambridge, UK: Cambridge University Press.

Levin, B., and M. Rappaport Hovav. 2011. Lexicalized meaning and manner/result complementarity. In B. Arsenijevi´c, B. Gehrke, and R. Mar´ın (Eds.), Subatomic Semantics of Event Predicates. Dordrecht: Springer. (22 pages, revised June 2011).

Levinson, S. 2003. Space in Language and Cognition: Explorations in Cognitive Di- versity. Cambridge, UK: Cambridge University Press.

Levinson, S., and D. Wilkins. 2006. Towards a semantic typology of spatial description. In S. Levinson and D. Wilkins (Eds.), Grammars of Space : Explorations in Cognitive Diversity, 512–552. Cambridge, UK; New York: Cambridge University Press. Bibliography 322

Li, C. N., and S. A. Thompson. 1973. Serial verb constructions in Mandarin Chinese. In You Take the High Node and I’ll Take the Low Node, 96–103. Chicago: Chicago Linguistic Society.

Lord, C. 1993. Historical Change in Serial Verb Constructions. Amsterdam: John Benjamins.

Manning, C. 1996. Ergativity: Argument Structure and Grammatical Relations. Stan- ford, CA: CSLI Publications.

Marchand, H. 1969. The Categories and Types of Present-Day English Word- Formation: A Synchronic-Diachronic Approach. Munich: Verlag C. H. Beck.

Margetts, A. 1999. Valence and Transitivity in Saliba: an Oceanic Language of Papua New Guinea. Nijmegen: Max Planck Institute.

Margetts, A. 2004. Core-layer junctures in Saliba. In Bril and Ozanne-Rivierre (Eds.), Complex Predicates in Oceanic Languages, 65–88. Berlin: Mouton de Gruyter.

Matisoﬀ, J. 1973. The Grammar of Lahu. Berkeley/Los Angeles: University of California Publications in Linguistics.

Matsumoto, Y. 1996. Complex Predicates in Japanese. Stanford, CA: CSLI Publica- tions.

McClure, W. 1994. Syntactic Projections of the Semantics of Aspect. PhD thesis, Cornell University.

McWhorter, J. 1992. Substratal inﬂuence in Saramaccan serial verb constructions. Journal of Pidgin and Creole Languages 7(1):1–53.

McWhorter, J. 2004. Saramaccan and Haitian as young grammars: The pitfalls of syntactocentrism in Creole genesis research. Journal of Pidgin and Creole Lan- guages 19:77–137. Bibliography 323

Mithun, M. 1986. Evidential diachrony in Northern Iroquoian. In W. Chafe and J. Nichols (Eds.), Evidentiality: the Linguistic Encoding of Epistemology. Nor- wood, NJ: Abex.

Moyse-Faurie, C. 2004. Complex predicate constructions in East Uvean. In Bril and Ozanne-Rivierre (Eds.), Complex Predicates in Oceanic Languages, 199–223. Berlin: Mouton de Gruyter.

Muansuwan, N. 2002. Verb Complexes in Thai. PhD thesis, University at Buﬀalo, SUNY.

Muysken, P. 1988. Parameters for serial verbs. In V. Manfredi (Ed.), Niger-Congo Syntax and Semantics, 65–75. Boston: Boston University African Studies Center.

Muysken, P., and T. Veenstra. 1995. Serial verbs. In J. Arends et al. (Ed.), Pidgins and Creoles, 289–301. Amsterdam: John Benjamins.

Naess, A. 2004. Serial verbs and complex constructions in Pileni. In Bril and Ozanne- Rivierre (Eds.), Complex Predicates in Oceanic Languages, 225–250. Berlin: Mou- ton de Gruyter.

Needleman, R. 1973. Tai Verbal Structures and some Implications for Current Lin- guistic Theory. PhD thesis, UCLA.

N’Guessan, J. K. 2000. Les séries verbales en Baoulé: questions de morphosyntaxe et de sémantique. Studies in African Linguistics 29(1):75–90.

Nishiyama, K. 1998. V-V compounds as serialization. Journal of East Asian Liguistics 7:175–217.

Noonan, M. 1985. Complementation. In T. Shopen (Ed.), Language Typology and Syntactic Description, Volume II, 42–140. Cambridge, UK: Cambridge University Press.

Noonan, M. 1992. A Grammar of Lango. Berlin: Mouton de Gruyter. Bibliography 324

O’Keefe, J. 1996. The spatial preposition in English, Vector Grammar, and the Cognitive Map Theory. In P. Bloom, M. Peterson, L. Nadel, and M. Garrett (Eds.), Language and Space, 227–316. Cambridge, MA: MIT press.

Oleyaran, O. 1982. On the scope of serial verb constructions in Yoruba. Studies in African Linguistics 13(2):447–468.

Osam, E. K. 1994. From serial verbs to prepositions and the road between. Language Typology and Universals 47:16–36.

Pawley, A. 1993. A language which deﬁes description by ordinary means. In W. Foley (Ed.), The Role of Theory in Language Description, 87–129. Berlin: Mouton de Gruyter.

Pesetsky, D. 1995. Zero syntax: Experiencers and cascades. Cambridge, MA: MIT Press.

Pustejovsky, J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press.

Rappaport Hovav, M., and B. Levin. 1998. Building verb meanings. In M. Butt and W. Geuder (Eds.), The projection of arguments, 97–134. Stanford, CA: CSLI Publications.

Rappaport Hovav, M., and B. Levin. 2001. An event structure account of English resultatives. Language 77:766–797.

Rappaport Hovav, M., and B. Levin. 2008. The English dative alternation. Journal of Linguistics 44:129–167.

Rappaport Hovav, M., and B. Levin. 2010. Reﬂections on manner/result complementarity. In E. Doron, M. Rappaport Hovav, and I. Sichel (Eds.), Syntax, Lexical Semantics, and Event Structure, 21–38. Oxford, UK: Oxford University Press.

Reinig, J. 2004. Serial and complex verb constructions in Teop. In Bril and Ozanne- Rivierre (Eds.), Complex Predicates in Oceanic Languages, 89–106. Berlin: Mou- ton de Gruyter. Bibliography 325

Savon, S. 1974a. Taku Hatima Oto. Paramaribo: Instituut voor Taalwetenschap.

Savon, S. 1974b. Wanl¨oBakaa Kontu 1. Paramaribo: Instituut voor Taalwetenschap.

Schachter, P. 1974. A non-transformational account of serial verbs. Studies in African Linguistics Supplement 5:278–282.

Schlesinger, I. 1979. Cognitive structures and semantic deep structures: the case of the instrumental. Journal of Linguistics 15:307–324.

Sebba, M. 1987. The Syntax of Serial Verbs. Amsterdam: John Benjamins.

Senft, G. 2004. Serial verb constructions in Austronesian and Papuan languages. In Bril and Ozanne-Rivierre (Eds.), Complex Predicates in Oceanic Languages, 49–64. Berlin: Mouton de Gruyter.

Senft, G. (Ed.). 2008. Serial Verb Constructions in Austronesian and Papuan Lan- guages. Canberra: Paciﬁc Linguistics.

Seuren, P. 1990. Serial verb constructions. In B. Joseph and A. Zwicky (Eds.), When Verbs Collide: Papers from the 1990 Ohio State University Mini-Conference on Serial Verbs, 14–33.

Slobin, D. I. 2004. The many ways to search for a frog: linguistic typology and the expression of motion events. In S. Str¨omquist and L. Verhoeven (Eds.), Relating Events in Narrative: Vol. 2 Typological and Contextual Perspectives, 219–257. Mahwah, NJ: Lawrence Erlbaum.

Slobin, D. I. 2006. What makes manner of motion salient?: Explorations in linguistic typology, discourse, and cognition. In M. Hickmann and S. Robert (Eds.), Space in Languages, 59–81. Philadelphia, PA: John Benjamins.

Sperlich, W. 1993. Serial verb constructions in Namakir of Central Vanuatu. Oceanic Linguistics 32(1):95–110.

Stahlke, H. 1970. Serial verbs. Studies in African Linguistics 1(1):60–99. Bibliography 326

Steever, S. B. 1988. The serial verb construction formation in the Dravidian languages. Delhi: Motilal Banarsidass.

Stewart, J. 1963. Some restrictions on objects in Twi. Journal of African Languages 2(2):145–149.

Stewart, O. 2001. The Serial Verb Construction Parameter. New York: Garland.

Sudmuk, C. 2005. The Syntax and Semantics of Serial Verb Constructions in Thai. PhD thesis, University of Texas.

Talmy, L. 1972. Semantic structures in English and Atsugewi. PhD thesis, U.C. Berkeley.

Talmy, L. 1976. Semantic causative types. In Shibatani (Ed.), Syntax and semantics (vol 6) : The grammar of causative constructions, 43–116. New York: Academic Press.

Talmy, L. 1988. Force dynamics in language and cognition. Cognitive Science 12:49– 100.

Talmy, L. 2000. Toward a Cognitive Semantics. Cambridge, MA: MIT Press.

Thepkanjana, K. 1986. Serial Verb Constructions in Thai. PhD thesis, University of Minnesota.

Thepkanjana, K. 2010. Ditransitive constructions in Thai. In A. Malchukov, M. Haspelmath, and B. Comrie (Eds.), Studies in Ditransitive Constructions. Berlin: Mouton de Gruyter.

Uwalaka, M. 1988. The Igbo Verb: a Semantico-syntactic Analysis. Wien: AFRO- Pub.

Valin, R. V. 2005. Exploring the Syntax-Semantics Interface. Cambridge, UK: Cam- bridge University Press. Bibliography 327

Valin, R. V., and R. L. Polla. 1997. Syntax: Structure, Meaning, and Function. Cambridge, UK: Cambridge University Press. van Hout, A. 1996. Event Semantics of Verb Frame Alternations. PhD thesis, Tilburg University, Tilburg, The Netherlands. van Hout, A. 2000a. Event semantics in the lexicon-syntax interface: Verb frame alternations in dutch and their acquisition. In C. Tenny and J. Pustejovsky (Eds.), Events as Grammatical Objects, 239–281. Stanford, CA: CSLI Publications. van Hout, A. 2000b. Projection based on event structure. In P. Coopmans, M. Ever- aert, and J. Grimshaw (Eds.), Lexical Specification and Insertion, 403–427. Ams- terdam: John Benjamins. van Klinken, C. 1999. A grammar of the Fehan dialect of Tetun, an Austronesian language of West Timor. Canberra: Pacific Linguistics. van Staden, M., and B. Narasimhan. 2004. Granularity in the cross-linguistic encoding of motion and location. At: http://www.mpi.nl/world/persons/private/bhuvana/granularity.doc. van Staden, M., and G. Reesink. 2008. Serial verb constructions in a linguistic area. In G. Senft (Ed.), Serial Verb Constructions in Austronesian and Papuan Languages. Canberra: Pacific Linguistics.

Veenstra, T. 1996. Serial Verbs in Saramaccan. Dordrecht: HIL Dissertations.

Volker, C. 1998. The Nalik language of New Ireland, Papua New Guinea. In Berkeley Models of Grammars, Vol. 4. New York: Peter Lang.

Voorhoeve, J. 1975. Serial verbs in creole. Paper presented at Hawaiian Pidgin and Creole Conference.

W¨alchli, B. 2005. Co-compounds and Natural Coordination. Oxford: Oxford Univer- sity Press. Bibliography 328

Wechsler, S. 2003. Serial verbs and serial motion. In D. Beermann and L. Hel- lan (Eds.), Proceedings of the Workshop on Multi-Verb Constructions. Trondheim Summer School, Norwegian University of Science and Technology, Trondheim.

Westermann, D. 1930. A Study of the Ewe Language. London: Oxford University Press.

Williams, W. 1976. Linguistic Change in the Syntax and Semantics of Sierra Leone Krio. PhD thesis, Indiana University.

Williams-van Klinken, C., J. Hajek, and R. Nordlinger. 2002. Serial verbs in Tetun- Dili: A preliminary account. Based on chapter 9 of ‘Tetun Dili: A Grammar of East Timorese Language’ Canberra: Paciﬁc Linguistics.

Williamson, K. 1963. The syntax of verbs of motion in Ijo. Journal of African Languages 2:150–154.

Williamson, K. 1965. A Grammar of the Kolokuma Dialect of Ijo. Cambridge, UK: Cambridge University Press.

Winford, D. 1992. Directional serial verb constructions in Caribbean English Cre- oles. In F. Byrne and J. Holm (Eds.), Atlantic Meets Paciﬁc: a global view of pidginization and creolization. Amsterdam: John Benjamins.

Wingerd, J. 1977. Serial verbs in Haitian Creole. In Language and Linguistics Prob- lems in Africa: Proceedings of the VIIth Conference on African Linguistics, 452– 466. Hornbeam Press.

Wunderlich, D. 1991. How do prepositional phrases ﬁt into compositional syntax and semantics? Linguistics 29:591–621.

Wunderlich, D. 1997a. Argument extension by lexical adjunction. Journal of Seman- tics 14:95–142.

Wunderlich, D. 1997b. Cause and the structure of verbs. Linguistic Inquiry 28:27–68. Bibliography 329

Wunderlich, D. 2000. Predicate composition and argument extension as general options. In Stiebels and Wunderlich (Eds.), Lexicon in Focus, 247–270. Berlin: Akademie Verlag.

Wunderlich, D., and R. Lak¨amper. 1999. On the interaction of structural and semantic case. In Working papers of the SFB Theory of the Lexicon 112. Dusseldorf.

Zlatev, J., and P. Yangklang. 2004. A third way to travel: the place of Thai in motion- event typology. In L. V. S. Stromqvist and R. Berman (Eds.), Relating Events in Narrative: Cross-linguistic and Cross-contextual Perspectives. Mahwath, NJ: Erlbaum.

Zwicky, A. 1990. What are we talking about when we talk about serial verbs? In B. Joseph and A. Zwicky (Eds.), When Verbs Collide: Papers from the 1990 Ohio State University Mini-Conference on Serial Verbs, 1–13.