Event Structures in Linguistic Form and Interpretation 9783110925449, 9783110190663

Table of contents :
Contents
Introduction
Section I Event Structure and Syntactic Construction
Patients in Igbo and Mandarin
Event decomposition and the syntax and semantics of durative phrases in Chinese

Citation preview

Event Structures in Linguistic Form and Interpretation

Language, Context, and Cognition Edited by Anita Steube

Volume 5

w DE

G Walter de Gruyter · Berlin · New York

Event Structures in Linguistic Form and Interpretation Edited by Johannes Dölling Tatjana Heyde-Zybatow Martin Schäfer

w DE

G Walter de Gruyter · Berlin · New York

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication

Data

Event structures in linguistic form and interpretation / edited by Johannes Dölling, Tatjana Heyde-Zybatow, Martin Schäfer, p. cm. — (Language, context, and cognition) Includes bibliographical references and index. ISBN 978-3-11-019066-3 (alk. paper) 1. Grammar, Comparative and general — Syntax. 2. Grammar, Comparative and general — Verb phrase. 3. Semantics. I. Dölling, Johannes. II. Heyde-Zybatow, Tatjana, 1973— III. Schäfer, Martin, 1975P295.E96 2007 415—dc22 2007005990

ISBN 978-3-11-019066-3 Bibliographic information published by the Deutsche

Nationalbibliothek

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

© Copyright 2008 by Walter de Gruyter G m b H & Co. KG, D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording or any information storage and retrieval system, without permission in writing from the publisher. Printed in Germany Cover design: Christopher Schneider, Berlin Printing and binding: Hubert & Co. G m b H & Co. KG, Göttingen

Contents Introduction

IX

Section I: Event Structure and Syntactic Construction Patients in Igbo and Mandarin Alexander Williams

3

Event decomposition and the syntax and semantics of durative phrases in Chinese Jo-wang Lin

31

Syntactic decomposition of events in Korean and Standard Indonesian

55

Minjeong Son and Peter Cole

Section II: Event Structure and Modification Unifying illegally Kyle Rawlins

81

Adverbial modification of adjectives: Evaluatives and a little beyond Marcin Morzycki

103

The structure of criterion predicates Kjell Johan Sceb0

127

Reference to embedded eventualities

149

Markus Egg

Section III: Event Structure and Situation Aspect Two puzzles for a theory of lexical aspect: Semelfactives and degree achievements Susan Rothstein The notion of 'path' in aspectual composition: Evidence from Japanese Eri Tanaka

175

199

VI

Contents

Reflexive intransitives in Spanish and event semantics Eric McCready and Chiyo Nishida

223

Scalar complexity and the structure of events

245

John Beavers

Section IV: Event Structure and Plurality On the plurality of verbs Angelika Kratzer

269

Event quantification and distributivity Kimiko Nakanishi

301

The event structure of irreducibly symmetric reciprocals Alexis Dimitriadis

327

Existential readings for bare plurals in object position

355

Sheila Glasbey

Section V: Event Structure and Temporal Location Tense and adverbial quantification Cornelia Endriss and Stefan Hinterwimmer

389

Phase structure and quantification Marko Malink

413

Cohesion in temporal context: The role of aspectual adverbs Alice G.B. ter Meulen

435

Mandarin sentential -le, perfect and English already

447

Hooi Ling Soh and Meijia Gao

Section VI: Event Structure and Natural Language Ontology The lower part of event ontology Regine Eckardt

477

Verbs of creation Christopher Piñón

493

Contents

VII

Portraits of the Authors

523

Index

527

Introduction This volume comprises a selection of papers presented at the workshop "Event Structures in Linguistic Form and Interpretation", which took place at the University of Leipzig in March 17-19, 2004. The workshop was hosted by the research project "Event Structures: Grammatical and Conceptual Components of Utterance Interpretation" at the Department of Linguistics of the University of Leipzig. The central topic to be addressed was how conceptual information on event structure is encoded in linguistic expressions and how such information can be reconstructed from utterances. Answers to these questions essentially contribute to a better understanding of the relationship between lexical semantics, syntactic structure, pragmatic inference, and world knowledge in a broader cognitive perspective. Among the many collections on event-based semantics and syntax appearing over the last ten years (e.g. Rothstein, 1998; Tenny and Pustejovsky, 2000; Higginbotham, Pianesi and Varzi, 2000; Lang, Maienborn and FabriciusHansen, 2003; Austin, Engelberg and Rauh, 2004; Maienborn and Wöllstein, 2005; and Verkuyl, de Swart and van Hout, 2005), this volume adopts a decidedly applied attitude in that the existence of Davidsonian event arguments is taken as given and that problems of the fundamental methodology are of minor concern. Instead, it demonstrates how the idea of event structure can be successfully applied to a wide range of empirical problems in an increasing number of languages. Thus, the topic is discussed not only on the basis of English and German but, among others, of Mandarin Chinese, Japanese, Korean, Indonesian, and Igbo as well. The contributed papers fall into several broad classes - accounting for event structure in connection with syntactic construction, modification, situation aspect, plurality, temporal location, and natural language ontology. Accordingly, the volume is organized into six sections.

Section I: Event Structure and Syntactic Construction The assumption of an event-related framework in research on verb meaning raises important questions for the analysis of the syntax/semantics interface. While it is uncontroversial that event structure is to some degree reflected in syntactic structure, there is an ongoing discussion about the sort of verbal information that should play a role in the syntactic derivation. Two different strategies of meaning decomposition, which can be traced back to the Genera-

χ

Introduction

tive Semantics tradition, dominate. One line (cf. e.g. Dowty, 1979; Jackendoff, 1990; Pustejovsky, 1991; Bierwisch, 1997; Rappaport and Levin, 1998) assumes that verbs have to be decomposed in the lexicon, independently of syntax, into more primitive predicates representing the structure of events designated. The fundamental assumption behind such lexicalist accounts is that the lexical-semantic entry of a verb determines its syntactic behaviour. The second, more recent line (cf. e.g. Hale and Keyser, 1993; von Stechow, 1996; Ritter and Rosen, 1998; Travis, 2000; Ramchand, 2003; Borer, 2005) follows the idea that event structure is explicitly encoded in syntax. Accordingly, the meaning of verbs is viewed as being compositionally constructed from primitives linked immediately to functional heads and abstract verbal roots. In this connection, one essential problem is whether thematic role information is projected from the verb's lexical representation, or imposed by the structural context in which the verb occurs. In particular, Kratzer (1996, 2003) has given strong evidence for the claim that external arguments are attached in the syntactic derivation through the mediation of Davidsonian event arguments. Building on the proposal by Kratzer, Alexander Williams additionally questions the common assumption that patients have to be treated as internal arguments of the verb. He lays out data from resultative constructions (RCs) in Igbo and Mandarin to show that even if a verb in a simple clause requires a direct object - interpreted as patient - the same verb has no such requirement if it serves as a means predicate in a RC. These data can be explained if the patient relation is introduced by VP structure, and not by the verb. Williams argues that languages differ with respect to the number of lexically encoded arguments within the range of possible participants of the described event. This makes it possible that the lexical entries of verbs in English contain the patient argument position but the entries of verbs in Igbo and Mandarin do not. A further important general consequence of his analysis is that one has to recognize patient as a basic thematic predicate. The paper by Jo-wang Lin is concerned with the syntactic distribution and the semantics of durative phrases in Mandarin. He argues that they can be adjoined to every maximal projection, provided that we can interpret them there without violating the homogeneity requirement. According to Lin, the event structure of verbs should be decomposed in overt syntactic structures, similar to von Stechow's treatment of the scope ambiguity of wieder ('again'). According to the author, this decompositional analysis explains why result-related durative phrases are syntactically more restricted - they can only occur after the direct object of the sentence - than process-related durative phrases. Finally, Lin discusses the implication of his structural account for data containing durative phrases in conjunction with incremental theme verbs, arguing that they have to be analyzed as inherently telic. Minjeong Son and Peter Cole also present empirical support for the view that the decomposition of verb meaning is explicitly reflected in (morpho-)

Introduction

XI

syntax. According to them, such an approach provides satisfactory explanation not only for the scope ambiguity of tasi ('again') associated with morphological as well as lexical causatives in Korean, but also for the apparent lexical ambiguity of the verbal suffix -kan in Standard Indonesian. For Korean, the authors assume that an abstract CAUSE can be overtly realized by the suffix i-, while the Indonesian suffix -kan is argued to be the overt instantiation of a RESULT predicate. Furthermore, Son and Cole suggest that their account unifies the different uses of the suffix in causative and benefactive constructions. A broader question raised by this analysis is in what extent languages differ in the way conceptual components of event structure are mapped into the syntax.

Section II: Event Structure and Modification Sentences containing adverbial modification were one of the main reasons for Davidson (1967) to introduce event variables into semantic representations. According to him, modifiers of verbal expressions can be analyzed as first order predicates that add information about the events introduced by the respective expressions. One advantage of this approach is that it allows one to draw inferences that relate to adverbial modifiers by virtue of conjunction reduction. While Davidson's account was restricted to instrumental, local and temporal adverbials, Parsons (1990) argued that the event variable can be used in the formalization of all VP modifiers that is, including, e.g., manner adverbs such as slowly, gently or quietly. In the wake of Parsons, the event-based framework has been applied to a far broader range of modification phenomena and has yielded many fruitful results. However, a number of areas have shown themselves to be somehow intricate. Thus, it is well-known that many adverbials can receive different interpretations depending on their syntactic position (cf. e.g. Ernst, 2002). Besides a few proposals covering specific types of modification (cf. e.g. Maienborn, 2001), till now there is no general approach to the systematic effects that syntactic variation bears on the resulting event structure. Moreover, an open question is to what extent adverbial modification can involve reinterpretations which are triggered by semantic mismatches and executed by shifting of ontological event type (cf. e.g. de Swart, 1998; Geuder, 2000; Dolling, 2003, 2005). The paper by Kyle Rawlins starts from the afore-mentioned observation that the meaning of modifying adverbs may vary depending on their syntactic position. In particular, the author is concerned with several modifier uses of adverbs like illegally: (a) a clausal use (= high position), (b) a manner use (= low position) and (c) two pre-adjectival uses. Rawlins argues that the meaning differences induced by placing such an adverb in different positions result purely from scope, especially from the relative scope with respect to the existential quantification over events and with respect to tense. After introducing a semantics for all four uses of illegally separately, he proposes one lexical entry

XII

Introduction

that expresses the core meaning of this adverb, which is identical with its interpretation in the high position. To enable adverbs to compose in a variety of positions, Rawlins assumes a family of type-shift operators that coerce sentence modifiers into modifiers of the respective types. Adverbs in a pre-adjectival position stand in the centre of the paper by Marcin Morzycki. It examines adverbs like remarkably and surprisingly which modify adjectival phrases and give rise to judgments about having a property to a particular degree, although they are not degree words. On the approach proposed, the evaluative adverbs have the effect of domain widening, similar to effects observed for embedded exclamatives, and are interpreted as arguments of an unrealized degree morpheme in much the same way as nominal measure phrases have been proposed to be. At the same time, Morzyski shows that they themselves have the same denotation as the corresponding adjectives and their adverbial counterparts in the clause-modifying position. Finally, the author suggests how the analysis can be extended to ad-adjectival subject-oriented adverbs. Kjell Johan Sseb0 is interested in the question how the modification of abstract predicates by instrumental fey-phrases can be accounted for in an eventbased framework. Criterion predicates like obey doctor's order and do me a favour form one type of these predicates and have remained ill-understood until now. Also, there is no consensus on the proper analysis of the second major type of abstract predicates, the manner-neutral causative predicates like create a fiction and ruin my reputation. As is argued by the author, criterion predicates as well as manner-neutral causatives are characterized by a certain degree of indeterminacy. Whereas the former do not specify the physical criteria which an action must meet, the latter are unspecific about the way in which the change of state is brought about. Saeb0 proposes an analysis where both types of predicates involve an indeterminate event predicate and where the function of the instrumental fry-adjunct is to fill it with content via unification. Another aspect of modification of event expressions is highlighted by Markus Egg. He offers a unified approach to cases in which modifiers or affixes refer to embedded eventuality arguments in the semantics of the modified expression or base, respectively. Addressing modification of deverbal nouns by an adjective like beautiful in beautiful dancer, restitutive readings of againsentences, and the effect of prefixation in German verb nominalisations like Losgerenne, the author argues that all three cases can be analyzed as interface phenomena in which a modifier or affix may semantically apply to only a part of the semantic contribution of its modified expression or base. Within this part, the eventualities that are bound in the semantics of the modified expression or base as a whole emerge as open arguments. The analysis is modelled in a framework based on underspecification and makes use of potential scope ambiguities in the semantics of the expressions discussed.

Introduction

XIII

Section III: Event Structure and Situation Aspect Situation aspect (cf. Smith, 1991), also called lexical aspect or Aktionsart, which is concerned with the internal temporal constitution of events and its linguistic reflection, is perhaps the most prominent semantic field where the Davidsonian approach has achieved remarkable advances. Although the literature on this phenomenon is vast, there remain many questions that are not convincingly solved yet. Since Vendler's (1967) classification of verbal predicates into four classes - activities, states, accomplishments and achievements - researchers aim at the clarification of the determining factors of these classes. Features like dynamicity, durativity, telicity and gradability play a central role in the discussion. Do such features characterize the underlying events or are they part of the verbal meaning or both of them? If they are considered to be features of the events, then one has to distinguish between several ontological types (or sorts) of events, i.e. between events of change, processes, and states (cf. e.g. Bach, 1986a,b; Parsons, 1990; Piñón, 1995; Dölling, 2005). Another possibility that has proven itself to be of great benefit is the approach proposed by Krifka (1989, 1992, 1998), which identifies quantization (telicity) and cumulativity (atelicity) as properties of verbal predicates. Here, the tight interaction between the referential information of a verb and that of its arguments (or adjuncts) is crucial for the explanation of aspectual composition (cf. e.g. Filip, 1999; Engelberg, 2000; Rothstein, 2004). Susan Rothstein argues that the semantics behind the Vendlerian verb classes is best captured with the help of two sets of features: whether or not the event in its denotation is durative, and whether or not it denotes an event of change. But the verbal class of semelfactives (e.g. knock) and that of degree achievements (e.g. cool) pose notorious problems for the Vendlerian classification. Rothstein proposes an operation of S-summing in order to explain why degree achievements, although verbs of change, also have atelic interpretations. This operation forms singular events out of sums of temporally adjacent events. While S-summing normally does not apply to verbs of change, because two events of change cannot be immediately adjacent to each other, it can apply to degree predicates, because they are characterized as changes in values on a scale. Semelfactives differ from activities in that they come with natural beginnings and endpoints and therefore can be perceived as atoms. If that is the case, the interpretation of a semelfactive verb is telic. Eri Tanaka's paper aims at an explanation for why Japanese lacks socalled strong adjectival resultative constructions. Its starting point is the assumption that one should differentiate between incremental theme verbs and motion verbs on the one hand, and change of location/state verbs on the other hand. Only the latter should be represented by the BECOME-operator that serves as a source for their telic interpretation. Verbs of the first group are atelic in Japanese and are interpreted as referring to a path or scale. Telic inter-

XIV

Introduction

pretations with verbs of this group are only possible if the path/scale is bounded by the object-NP or a postposition (e.g. -made 'up to'). Tanaka takes the complementary distribution of the two postpositions -ni ('in/at/to') and made as further evidence for the distinction between the two verb groups. She suggests that weak resultatives are formed on the basis of change of location/state verbs - hence BECOME-verbs - and that the adjectives modify the result state. Strong adjectival resultatives are based on verbs of the first group, but Japanese adjectives lack the possibility to bind/to limit the path/scale and therefore Japanese lacks this sort of construction. The paper by Eric McCready and Chiyo Nishida is concerned with Spanish reflexive intransitives (Ris), i.e. constructions containing the reflexive clitic se in conjunction with a non-transitive verb. After establishing the differences between RIs and their transitive counterparts, the authors note that (dynamic) RIs have three distinctive properties. They require a quantized subject NP, they license a dative argument which stands in some relation to the event described by the sentence, and they coerce the verbs they appear with to achievements denoting a transition, or the onset or the final end of a process or a state. McCready and Nishida model these facts by introducing a presupposition of telicity on the VP associated with reflexive se, allowing RIs to be associated with an additional, possibly implicit, argument, and, finally, placing meaning postulates on verb classes associating the temporal interval of the event with the initial or final point of a path. John Beavers focuses primarily on the factors governing durativity in dynamic predicates to build a broader picture of the aspectual behaviour of descriptions of events of change. His main objective is to show that there is a general correlation between the durativity of an event and the gradability of the scale of change of a participant. He argues that Rrifka's homomorphism model designed originally to explain the nature of telicity of incremental theme verbs can be generalized to cover a wide range of dynamic predicates following a scalar approach to telicity and, in consequence, also explains the durativity/gradability correlation. Essential for the analyses by Beavers is a distinction between two types of mereological complexity: structures with two sub-parts (begin, end) and structures with at least three sub-parts (begin, middle and end). In addition, the paper outlines relevant lexical, pragmatic and contextual constraints on durativity and gradability and discusses their possible origins.

Section IV: Event Structure and Plurality Quite a number of recent event-based studies deal with the question of how distributive, collective and cumulative readings arise. Traditional analyses attribute this difference between the addressed readings to the quantificational make-up of the NPs occurring in the respective sentences. But since the work of Schein (1993) and Lasersohn (1995), it has been recognized that event

Introduction

XV

structure must play a major role in accounts of plurality, including distributive, collective and cumulative interpretations (cf. e.g. Landman, 2000; Kratzer, 2003). In this connection, there is an ongoing discussion about the question of whether verbal predicates are lexically already marked for plurality. Moreover, the plurality question points towards a broader problem: although there is a huge amount of literature on quantification, the formal approaches to these phenomena rarely make use of events. With the growing acceptance of event semantics it became apparent that a combination of event-based theories and theories of quantification is necessary. In this context, the influence of the event type on the existential force of the subjects and objects of verbs presents another challenge. Angelika Kratzer pursues some of the consequences of the idea that there are at least two sources for distributive/cumulative alternation of readings in English. One source is lexical pluralization. Following a proposal by Krifka and Landman, she assumes that all verb stems are born as plurals independent of the particular language and the particular nature of its NPs. The second source of cumulative/distributive interpretation in English is directly provided by plural NPs. Kratzer proposes that such phrases with plural agreement features are able to pluralize adjacent verbal projections. The difference between the two possibilities lies in the fact that distributive/cumulative interpretations on the basis of lexical pluralization allow the occurrence of singular NPs within the sentences, which is impossible with the second type. She shows that the phenomena discussed in her paper all pose conceptual problems for analyses not based on events, but can be given elegant accounts within a version of the Davidsonian event semantics. The objective of the contribution by Kimiko Nakanishi is closely connected to the last one: It examines empirical data on collective and distributive interpretations of constructions in Japanese and German, where a quantificational expression appears separated from its host NP. The author assumes that such split quantifiers measure the events in the verbal domain, but only if the relevant VP denotes a part-whole structure, i.e. a lattice of events. Furthermore, she assumes that, with the help of the mapping based on the homomorphism from events to objects, the measure function applies indirectly to the events by measuring the objects denoted by the host NP. According to her, this mechanism of event measurement satisfactorily accounts for why split quantifier constructions usually disallow collective readings. In addition, Nakanishi discusses some examples where, by contrast, a collective reading is available in such a construction and argues that the proposed analysis is capable of handling these cases, too. A specific type of multiple participants in a single event is focussed by Alexis Dimitriadis' paper. The author presents a wide range of data on reciprocals, especially on irreducibly symmetric and discontinuous reciprocals, found in numerous languages. It is essential for an irreducibly symmetric

XVI

Introduction

predicate like meet that it expresses a relationship between two arguments where, in contrast to ordinary reciprocals, both arguments necessarily have thematically identical participation in any event described by means of it. As it is shown by Dimitriadis, irreducible symmetry plays a prominent role in many discontinuous constructions, in which the logical subject of a reciprocal verb appears to be split between the syntactic subject and a comitative vv/i/i-phrasc. Taking into account the well-known problems with assigning the same thematic role to different participants, he proposes an analysis where the symmetric events are decomposed into sub-minimal events specifying the distinct relations of each participant to the complex event. Sheila Glasbey investigates which verbal predicates disallow existential readings for their bare plural objects and, in addition, which adjectival predicates disallow existential readings for their bare plural subjects. One of her observations is that, contrary to the usual picture, many verbs and adjectives which might well be classified as individual level predicates give not only generic but also existential interpretations. Using a situation-theoretic framework, the author argues that, generally, such a reading is made possible by the presence of a localising situation, which may be provided by the event argument of the verb or by an appropriate context. According to her, psych-verbs with experiencer subjects like hate as well as adjectival predicates lack event arguments and, therefore, may allow an existential reading for their bare plurals only with the help of a specific context. The proposal is that, in both cases, the reading comes from a so-called existential inference which is licensed by the respective situation.

Section V: Event Structure and Temporal Location The significance of temporal location for the determination of event structure should be obvious (cf. e.g. Parsons, 1990; Kamp and Reyle, 1993). According to common understanding, the tense of the verb serves primarily to localize the respective event in time - notably before, around, or after the time of utterance (or speech). A more recent approach takes another stance (cf. Klein, 1994): Tense does not directly specify the time of the event; rather, it locates the time interval about which the utterance asserts something with respect to the utterance time. Viewpoint (or grammatical) aspect (cf. Smith, 1991), i.e. the device for making distinctions between, e.g., perfective and imperfective, on the other hand, concerns the relationship of event time to tense time. Thus, not only tense but also viewpoint aspect is defined in terms of temporal relations. While there is a huge amount of research on the two grammatical categories and on the interaction between them, there is much less work on how they interact with quantificational NPs or with quantificational adverbials in terms of scopes, as well as with particles like already or still. In addition, specific questions arise from languages which lack at least one of the categories to express

Introduction

XVII

the temporal location of events. For example, Mandarin has no tense marker and, hence, no grammaticalized means to impose a constraint on the time about which the assertion is made. An important problem which has to be investigated is how the temporal location system fills such a 'gap'. In their paper, Cornelia Endriss and Stefan Hinterwimmer present novel evidence for an analysis of Quantificational Variability Effects (QVEs) as byproducts of a quantification over events/situations. They compare adverbially quantified sentences containing indefinites modified by relative clauses with sentences containing corresponding quantificational NPs modified by relative clauses, showing that the former have to obey a constraint in order for QVEs to arise that does not hold for the latter: the tense of the relative clause verb has to agree with the tense of the matrix verb. While this is completely unexpected under the assumption that QVEs come about via direct quantification over individuals, the authors show that a natural explanation is possible under the assumption that the events quantified over have to be located in a salient time interval, and that this interval is determined on the basis of a pragmatic strategy dubbed Interval Resolution Strategy that favours local information. Marko Malink deals with the interplay of negation, quantifiers and phase particles schon ('already'), noch ('still'), noch nicht ('not yet') and nicht mehr ('not anymore'). In particular, he provides a scope analysis of German sentences such as Einige sind noch nicht da ('Some people are not there yet') or Niemand ist mehr da ('Nobody is there anymore'). In order to correctly account for the intuitive meaning of such sentences, he proposes to split negative phase particles such as noch nicht and negative quantifiers such as niemand into a negation and a purely positive part. He then shows that the negation has a different scope position than the positive part of the phase particle or quantifier to which it belongs. He goes on to specify a compositional formal account of quantified phase structures within a generalized quantifier framework. Malink argues that in the sentences under consideration, phase particles have a bridging function connecting the VP and the subject-quantifier. A different view on properties of phase particles or, in other words, aspectual adverbs is taken by Alice G. B. ter Meulen. With the overall aim of a better understanding of temporal coherence of information states, she examines the relation between the objective content of English aspectual adverbs and the subjective information which can be also conveyed by them. According to the author, the four basic aspectual adverbs - not yet, already, still and not anymore - constitute a logical polarity square in the temporal domain of events, showing the fundamental relationship between current, past and future reference times. In addition, they have usages associated with marked high pitch prosody, where, besides the factual information, the speaker's attitude regarding the flow of events or its perceived speed is expressed. To capture them, ter Meulen introduces a modal operator which quantifies over alterna-

XVIII

Introduction

tives to the current course of events, dependent upon the speaker's epistemic state. Hooi Ling Soh and Meijia Gao's paper investigates the meaning of Mandarin sentential -le with specific focus on its relation to English perfect tense and to the particle already. They propose that the transition marker -le encodes the assertive meaning that the situation in question is realized prior to a reference time (in a Reichenbachian sense), along with the presupposition that a situation opposite to the one described by the sentence exists immediately before the point of realization. The reference time is either the utterance time, or, when the particle jiu is used, some specified time in the past or the future. The authors discuss the similarities that have been previously noted between sentential -le and English perfect in terms of the presence of a result state and a continuative reading and argue that the relevant readings are not entailed by le. On their analysis, the transition marker shares its assertive meaning with perfect tense and its presupposition with already.

Section VI: Event Structure and Natural Language Ontology There is perhaps no topic in semantics in which questions of meaning are so intimately interwoven as with ontological questions such as the concept of event. In order to analyse what information on event structure is encoded in linguistic expressions one needs an account of the ontology underlying natural language. Such a theory of the world, i.e. of what basic sorts of entities there are, what fundamental properties the entities have and how they are related, does not necessarily correspond to the categorial commitments of current natural science, especially of micro physics. Instead, natural language ontology (cf. Bach, 1986a) has to accommodate every kind of entity which can be the object of ordinary talking and thinking. More specifically, the ontological analysis is not primarily concerned with the way the world really is but rather with the way we conceptualize it for the purposes of every-day life. Thus, it has to do with entities which result from projecting our cognitive framework onto environmental input and, for this reason, are also dependant on the concrete point of view we take. Over the past years, an increasing number of domains of such entities have been distinguished in investigating the common sense world. Following the crucial innovation by Link (1983) to assume specific lattice structures on the domains of pluralities and of quantities of matter, Bach (1986b) and, in particular, Krifka (1989, 1992, 1998) have extended this algebraic approach to events. Many others, of which only some could be referred to in this introduction, have contributed to this far-reaching field of research. As a result, semantics is provided now with a system of assumptions that precisely characterize the structure of the domain of events and its relationship to the domain of objects.

Introduction

XIX

Regine Eckardt addresses an apparent conflict between two applications of event ontology for semantics. Thus, scholars aiming at modelling the difference between telic and atelic predicates commonly assume that certain properties - the properties of homogenous reference - of events are inherited by all their parts, no matter how small they are. However, proposals modelling negative polarity items are forced to assume a level where parts of events are so small that they can no longer inherit properties denoted by a natural language predicate. After summarising the two positions, Eckardt proposes three possible solutions for this dilemma, and discusses two of them, the construction of infinitesimal events and a solution capitalizing on the Sorites paradox. Both of the treatments appear to capture some of the essence of how in every-day reasoning very small events are thought about. She concludes that till now there are no conclusive arguments in favour of one or the other of these options. Another subject which is immediately connected with basic problems of ontology is investigated by Christopher Piñón. His paper gives an account of verbs of creation (e.g., build, compile, draw, write) which are known to take an internal argument denoting a physical object that is effected or brought into being as a result of the event named by the verb. The author argues that such verbs are actually ambiguous with respect to the sortal character of their internal argument. In particular, the internal argument of a verb of creation may also denote what he calls a template, i.e. an abstract individual that is physically instantiated in the course of the event described by the verb. This idea allows to analyze performance verbs of creation (as sing in sing a song) and those denoting the creation of templates (as compose in compose a symphony). Moreover, Piñón can also explain data such as Sarah built the house that Rebecca designed where the house that Rebecca designed designates a house template (or design) which Sarah builds an instantiation of. To this end, he proposes a set of sort-shifters that serve to capture systematic ambiguity among verbs of creation like build. We like to thank the reviewers of the workshop for their great help and patience: Markus Egg, Veronika Ehrich, Stefan Engelberg, Hana Filip, Fritz Hamm, Angelika Kratzer, Manfred Krifka, Claudia Maienborn, Susan Rothstein, Arnim von Stechow, and Henriette de Swart. Furthermore, we want to thank the authors of the contributed papers for their cooperation all the way through. The conference received financial support from the Centre for Higher Studies at the University of Leipzig, which is gratefully acknowledged. The research project "Event Structures: Grammatical and Conceptual Components of Utterance Interpretation" at the Department of Linguistics of the University of Leipzig as well as the preparation of manuscript for publication were funded by the German Research Foundation (DFG).

XX

Introduction

Sebastian Hellmanii and Stefan Keine deserve special thanks for their accurate formatting of the manuscript. Thanks also to David Dichelle and Ryan Young for checking the English of non-native speakers involved in the volume. Leipzig, January 2007 Johannes Dölling, Tatjana Heyde-Zybatow and Martin Schäfer

References Austin, Jennifer, Stefan Engelberg and Gisa Rauh (eds.) (2004): Adverbials. The Interplay between Meaning, Context, and Syntactic Structure. Amsterdam, Philadelphia: Benjamin. Bach, Emmon (1986a): 'Natural Language Metaphysics'. In: R. Barcan Marcus, G. Dorn and P. Weingartner (eds.): Logic, Methodology and Philosophy of Science. Vol. VII., Amsterdam: Elsevier, 573-595. Bach, Emmon (1986b): 'The Algebra of Events'. Linguistics and Philosophy 9: 5-16. Bierwisch, Manfred (1997): Lexical Information from a Minimalist Point of View. In: Ch. Wilder, H.-M. Gärtner and M. Bierwisch (eds.): The Role of Economy Principles in Linguistic Theory. Berlin: Akademie Verlag, 227-266. Borer, Hagit (2005): Structuring Sense. An Exo-Skeletal Trilogy. Vols. 1 and 2. Oxford: Oxford University Press. Davidson, Donald (1967): 'The Logical Form of Action Sentences.' In: N. Resher (ed.): The Logic of Decision and Action. 81-95. Pittsburgh: University of Pittsburgh Press. Dölling, Johannes (2003): 'Flexibility in Adverbal Modification: Reinterpretation as Contextual Enrichment'. In: E. Lang and C. Maienborn and C. Fabricius-Hansen (eds.): Modifying Adjuncts. Berlin, New York: Mouton de Gruyter, 511-552. Dölling, Johannes (2005): Semantische Form und pragmatische Anreicherung: Situationsausdrücke in der Äußerungsinterpretation. Zeitschrift für Sprachwissenschaft 24 : 159-225. Dowty, David (1979): Word Meaning and Montague Grammar. Dordrecht: Kluwer. Engelberg, Stefan (2000): Verben, Ereignisse und das Lexikon. Tübingen: Niemeyer. Ernst, Thomas (2002): The Syntax of Adjuncts. Cambridge: Cambridge University Press. Filip, Hana (1999): Aspect, Eventuality Types and Nominal Reference. New York: Garland Publishing. Geuder, Wilhelm (2000): Oriented Adverbs. Issues in the Lexical Semantics of Event Adverbs. Dissertation, Universität Tübingen.

Introduction

XXI

Hale, Kenneth and Keyser, Jay (1993): 'On argument structure and the lexical expression of syntactic relations'. In: K. Hale and J. Keyser (eds.): The view from building 20: A festschrift for Sylvain Bromberger. Cambridge, MA: MIT Press, 53-108. Higginbotham, James and Pianesi, Fabio and Varzi, Achille C. (eds.) (2000): Speaking of Events. Oxford: Oxford University Press. Jackendoff, Ray (1990): Semantic Structures. Cambridge, MA: MIT Press. Kamp, Hans and Reyle, Uwe (1993): From Discourse to Logic. Introduction to Modeltheoretic Semantic of Natural Language. Dordrecht: Kluwer. Kratzer, Angelika (1996): 'Severing the external argument from its verb'. In: J. Rooryck and L. Zaring (eds.): Phrase Structure and the Lexicon. Dordrecht: Kluwer, 109-137. Kratzer, Angelika (2003): The Event Argument and the Semantics of Verbs. Ms. University of Massachusetts, Amherst. Krifka, Manfred (1989): Nominalreferenz und Zeitkonstitution. Zur Semantik von Massentermen, Pluraltermen und Aspektklassen. München: Fink. Krifka, Manfred (1992): 'Thematic relations as links between nominal reference and temporal constitution'. In: I. A. Sag and A, Szabolcsi (eds.): Lexical Matters. Stanford, CA: CSLI Publications, 29-53. Krifka, Manfred (1998): 'The origins of telicity'. In: S. Rothstein (ed.): Events and Grammar. Dordrecht: Kluwer, 197-235. Klein, Wolfgang (1994): Time in Language. London and New York: Routledge. Landman, Fred (2000): Events and Plurality. The Jerusalem Lectures. Dordrecht: Kluwer. Lang, Ewald and Maienborn, Claudia and Fabricius-Hansen, Claudia (eds.) (2003): Modifying Adjuncts. Berlin, New York: Mouton de Gruyter. Lasersohn, Peter (1995): Plurality, Conjunction and Events. Dordrecht: Kluwer. Link, Godehard (1983): 'The Logical Analysis of Plurals and Mass Terms: A LatticeTheoretical Approach'. In: R. Bäuerle,Ch. Schwarze and A. von Stechow (eds.): Meaning, Use, and Interpretation of Language. Berlin, New York: de Gruyter, 303323. Maienborn, Claudia (2001): 'On the Position and Interpretation of Locative Modifiers'. Natural Language Semantics 9: 191-240. Maienborn, Claudia and Wöllstein, Angelika (eds.) (2005): Event Arguments: Foundations and Applications. Tübingen: Niemeyer. Parsons, Terence (1990): Events in the Semantics of English. A Study in Subatomic Semantics. Cambridge, MA: MIT Press. Piñón, Christopher (1995): An Ontology for Event Semantics. PhD Thesis. Stanford University. Pustejovsky, James (1991): 'The syntax of event structure'. Cognition 41: 47-81.

XXII

Introduction

Swart, Henriette de (1998): 'Aspect shift and coercion'. Natural Language and Linguistic Theory 16: 347-385. Ramchand, Gillian (2003): First Phase Syntax. Ms. Oxford University. Rappaport Hovav, Malka and Levin, Beth (1998): 'Building Verb Meanings'. In: M. Butt and W. Geuder (eds.): The Projection of Arguments: Lexical and Compositional Factors. Stanford: CSLI Publications, 97-134. Ritter, Elizabeth and Rosen, Sara Thomas (1998): 'Delimiting Events in Syntax'. In: M. Butt and W. Geuder (eds.): The Projection of Arguments: Lexical and Compositional Factors. Stanford, CA: CSLI Publications, 135-164. Rothstein, Susan (ed.) (1998): Events and Grammar. Dordrecht: Kluwer Academic Publishers. Rothstein, Susan (2004): Structuring Events. A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Schein, Barry (1993): Plurals and Events. Cambridge, MA: MIT Press. Smith, Carlotta (1991): The Parameter of Aspect. Dordrecht: Kluwer. Stechow, Arnim von (1996): 'The different readings of wieder "again": A structural account'. Journal of Semantics 13: 87-138. Tenny, Carol and James Pustejovsky (eds.) (2000): Events as Grammatical Objects. Stanford, CA: CSLI Publications. Travis, Lisa (2000): 'Event structure in syntax'. In: C. Tenny and J. Pustejovsky (eds.): Events as Grammatical Objects. Stanford, CA: CSLI Publications,, 145-185. Vendler, Zeno (1967): Linguistics in Philosophy. Ithaca, NY: Cornell University Press. Verkuyl, Henk J. and de Swart, Henriette and van Hout, Angeliek (eds.) (2005): Perspectives on Aspect. Dordrecht: Springer.

Section I

Event Structure and Syntactic Construction

Alexander Williams (Maryland)

Patients in Igbo and Mandarin 1 Introduction The development of event semantics has facilitated discussion of a basic question in verbal grammar. For a given thematic relation between (the meanings of) a verb and a phrase local to it, is that relation introduced by the lexical representation of the verb, or by the structural context in which the verb occurs? More briefly, is the phrase an argument of the verb or not? By supplying a parameter of which thematic relations can be conjunctively predicated - the event variable - event-based representations have allowed questions and answers on this topic to be formulated in a clear and useful way (e.g. Carlson 1984, Dowty 1989, Kratzer 1996, Marantz 1997, Rothstein 2001). Yet the two ways of modeling a thematic relation, as projecting from the verb or as imposed by its context, are often hard to distinguish empirically. In this paper I discuss a case where the facts seem decisive: resultative constructions in Mandami and Igbo, like (1) and (2) respectively.1 (1)

(2)

tati duàii -le nàtiào miibän. 3s kick snap -PFV that plank 'He made that plank snap by kicking.' O ku wa -ra çba ahu. 3sS strike split - F A C T gourd that 'He made that gourd split by striking.' (ex. Hale, Ihionu, and Manfredi 1995, trans. AW)

The grammar of these constructions demonstrates that, characteristically in Igbo and Mandarin, neither agents nor patients are arguments of the verb. Basic the1

Mandarin is a Sinitic language and the national language of China. In glosses of Mandarin, PFV means 'perfective,' and PRT means 'sentence final particle'. Igbo ([ißo]) is a Benue-Congo (or Eastern Kwa) language spoken mainly in Nigeria (see Swift et al. 1962, Green and Igwe 1963, Emenanjo 1978, and Igwe 1999). Glosses of Igbo use the following abbreviations. FACT means 'factative'; roughly, a predicate in the factative has past time reference when eventive and nonpast time reference when stative. Β VC means 'bound verb cognate' ( see Nwachukwu 1987 and Emenanjo 1978). The BVC is a nominalization of the verb group; in all the data presented here, it is used solely to satisfy the requirement that a verb group in the factative has not be clause-final (Nwachukwu 1987: 19-21). PROG means 'progressive,' SBRD means 'subordinate verb prefix,' and Ρ means 'all-purpose preposition.'

4

Alexander Williams

matic relations are instead introduced by the structure in which the verb occurs. I have defended this conclusion for agents elsewhere (Williams 2004, 2005). In this paper I make the case for patients, a more surprising hypothesis. Almost always, patients are treated as arguments of the verb, and strong conceptual support has been given for this decision, notably in Kratzer (2003). Here I counter with a grammatical argument. For Igbo and Mandarin, the relative distribution of verbs and patients is best explained if patients are introduced by verb pirrase structure, and not by the verb. Based on this I abstract a broader conclusion. If the event of a verb necessarily has a patient as a participant, we still cannot conclude that the verb has a patient as an argument. Lexical meaning is not lexical valence. The program is as follows. I begin in section 2 by sketching what it means for an argument to be introduced by the verb or by its context. Sections 3 and 4 then define what resultative constructions are, and how they can be used to test for the valence of verbs within them. The lessons of English resultatives are discussed in section 5, before I lay out the target data from Igbo and Mandarin in section 6. Section 7 is the core of the paper, setting out the claim that the facts of section 6 are best explained if patients are not arguments of the verb. Sections 8 and 9 consider and dismiss various alternatives. I then mention a semantic objection to the theory in section 10 before concluding.

2 Projectionist and nonprojectionist models Suppose that the direct object in a simple clause names the patient of the verb's event, as in (3). There are two ways to model this, differing in how the relation is introduced in deriving the semantic representation of the clause (see Carlson 1984, Dowty 1989, Kratzer 1996, Borer 2003). (3)

\A1 pounded the cutlet\ = 3e.pound{e) Λ PAT(e) = c Λ AG(e) = a

We might say that the patient relation is introduced by the lexical representation of the verb, as in (4) for example. In this case the patient relation projects from the verb, and the patient is a lexical argument of the verb. (4)

¡poimdj = Xy(Xx)Xe.pound(e)

Λ PAT(e) = y (Λ AG(e) = χ)

Or we might say that the patient relation is not introduced by the verb, but by the structure in which it occurs, perhaps as in (5).2 Then the patient, while it is identified by an argument category in the clause, is not a lexical argument of the verb.

2

In writing "PAT(e ) = [DP]]," I am presuming as a convenience that D P here denotes in the set of individuals.

5

Patients in Igbo and Mandarin

(5)

(a)

\pound\ = \e.pound {e)

(b)

[ [VP V D P ] ] = Ae. [ V ] (e) Λ P A T ( e ) =

[DP]

I will call models of the first type projectionist, and models of the second type nonprojectionist. The two models differ in how many lexical arguments they assign the verb. But they need not differ in how many participants they assign the event-type that the verb describes. If a verb has an argument to which it assigns a certain thematic relation, then it describes an event involving a participant who bears that relation. But the converse is not necessarily true. Compare (4) and (5a) for example. Given existential closure of y (and x) in the fonner, each defines a predicate of events. These two predicates do not necessarily differ in extension. Any event that verifies the predicate from (4) must have a patient; this is stated explicitly in the formula. But the same might be true of (5a), albeit implicitly, if the metalanguage predicate pound is defined to have (6) as a consequence (see Dowty 1989: 85). In that case (4) and (5a) will describe exactly the same set of events. (6)

Ve [pound (e) —s· 3j/.PAT(e) = y]

Thus the choice between projectionist and nonprojectionist models is primarily not semantic, in the strict sense, but grammatical. Given two theories of either sort which assign the verb the same type of event, which one yields the simpler and more explanatory grammar? It is sometimes suggested that, pursuant to some universal principle, a verb has a lexical argument for each participant in its event (see e.g. Lidz, Gleitman and Gleitman 2003, Kratzer 2003: Chapter 1, page 18). But this is an empirical hypothesis, not an a priori principle. And if my understanding of Igbo and Mandarin is correct, it is a hypothesis challenged by the facts of these languages (see also Davis and Demirdache 2000, Bhatt and Embick 2004).

3 Resultative constructions Resultative constructions, henceforth RCs, are single clause constructions comprising two predicates, a means predicate (M) and a result predicate (R). Neither M nor R is introduced by a conjunction, adposition, or complementizer. (7) is an English example, where M is pound and R is flat. (7)

Al pounded the cutlet flat.

Semantically, RCs express a relation of causation between the eventualities described by M and R, without this relation being indicated by any overt morpheme (Dowty 1979, a.o.): (7) says that pounding caused flatness. One aspect of this

6

Alexander Williams

meaning is that some object changes state, entering the result condition defined by R. The pirrase that names this object, I will say, controls R. In (7) the cutlet controls flat, since (7) entails that the cutlet became flat. (1) and (2), repeated here, are RCs from Mandarin and Igbo. In (1) M is ti 'kick,' R is duàn 'snap,' and R is controlled by nàtiào mìiban 'that plank.' The sentence says that kicking caused snapping, and what wound up snapped was the plank. (2) says that striking caused splitting, and what wound up split was the gourd. M here is ku 'strike' and R is wa 'split.' (1)

(2)

tati duàn-le nàtiào mùbàn. 3s kick snap -PFV that plank ' H e made that plank snap by kicking.' Ο ku wa -ra çba ahu. 3sS strike split -FACT gourd that ' H e made that gourd split by striking.'

My glosses will follow a fixed fonnat: 'subject made object R by M'ing.' I will discuss the syntax of Mandarin and Igbo RCs briefly in section 6. What will filterest me primarily are the understood thematic relations of subject and object to the event of M, the means event. In (7) Al names the agent of pounding and the cutlet names its patient. In (1) and (2) as well, the subject is the agent of the M event, and the object is the patient. But we will see in section 6 that Mandarin and Igbo differ from English in not requiring this particular pattern of relations. And this will form the basis of my central conclusion, that verbs are typically without arguments in Igbo and Mandarin. One last distinction sfiould be made, between what I will call transitive and intransitive RCs. In English the distinction is readily made in tenns of surface syntax. Transitive RCs have a subject and an object, (8), while intransitive have only a (surface) subject, (9). (8) (9)

(a) (b) (a) (b)

Al pounded the cutlet flat, Al yelled Iiis throat hoarse. The lake froze solid, The door swung shut.

But the criterion of the distinction, as I intend it, is semantic. Transitive but not intransitive RCs include an argument that is construed as the agent of causation3 (i.e. the causer); as it happens, this is always the subject. Concomitantly there is a difference in control of R. Control is by the object in transitives and the (surface) subject in intransitives (Simpson 1983, Y. Li 1995). Classing RCs along these lines assists in cross-linguistic comparison, allowing generalizations 3

I assume a very broad understanding of the agent relation, similar to what Van Valin and Wilkins 1996 have for their "Effector" relation. I do not assume that agents must be animate or volitional actors.

Patients in Igbo and Mandarin

7

to emerge which are otherwise obscured by independent differences in syntax (Williams 2005). In this paper I rely on it only to limit the scope of discussion: I will discuss only transitive RCs.

4 The relevance of resultative constructions RCs are sometimes analyzed as complex predicates (e.g. Dowty 1979, Larson 1991). It is assumed, that is, that M contains no argument positions. Instead the means verb combines directly with R to the exclusion of the object, (10). 4 (10) [ Object [ Vmeans R ] ]

(linear order irrelevant)

Insofar as this analysis is correct for a given RC, the construction will provide a diagnostic environment. By putting a verb in M, we stand to learn something about its lexical argument structure. Let us see why. Suppose we are choosing between two denotations for pound, (11) and (12). The two options make different predictions when pound occurs in M, if the RC is a complex predicate. (11) [pound ] = XyXe.pound(e) Λ PAT(e) = y (12) [pound J = \e.pound{e) Given (11), the verb, when it occurs in M, will have an argument that is not immediately sahnated, since there the verb's sister will be R, an expression that cannot provide a patient. We consequently expect that the complex predicate will inherit this unsaturated argument from M. We expect, for example, that the denotation ofpoundflat will have the outlines in ( 13 ). (13) [poundflat ] = \y . . . pound(e) Λ PAT(e) = y . .. So by assigning the argument to the verb lexically, we encode an expectation that the verb will be subject to the same requirement in an RC as in a simple clause. In both contexts it will cooccur with a phrase understood as its patient. Any deviations from this expectation will count as special cases, in need of explanation. For example, we might need to posit a covert operator that binds the verb's unsaturated argument. We have no such expectation, however, if the patient is not an argument of the verb, (12). Then there will be no argument left unsaturated when M and R combine, and no argument to pass along to the complex predicate. So there will be no assumption, based solely on the verb's lexical representation, that it will 4

Some of the points I make about the complex predicate analysis can also be made about the Small Clause analysis (Kayne 1985, Hoekstra 1988), according to which RCs have the structure: [ Vmeans [ Object R ] ]. Yet I will not discuss the Small Clause analysis here, as I find it unattractive for the languages under consideration (cf. Sybesma 1999).

8

Alexander Williams

enter the same thematic relation in RCs as it does in simple clauses. It will not come as a surprise if no noun pirrase in an RC is interpreted as the patient of the means event. If they are complex predicates, therefore, RCs can provide evidence for whether or not a given thematic relation projects from the verb. If the relation obtains wherever the verb occurs, equally in RCs and simple clauses, then it is likely introduced by the verb itself, lexically. But if it should matter where the verb occurs - with the relation required in simple clauses, but not in RCs - then perhaps it is introduced not by the verb, but by its context.

5 English resultative constructions and verbal valence Before heading into Igbo and Mandarin, it will be useful to consider English as a counterpoint. We will see that, in English, the grammar of RCs can be used to argue that (at least) patients are arguments of the verb. In English, a verb is typically subject to the same requirements in an RC as m a simple clause. A verb will require a patient (or theme) when in M, for example, to the same extent, and under the same conditions, that it requires one m a simple clause (Dowty 1979: 222, Carrier and Randall 1992: 187, Levin and Rappaport Hovav 1995: 39, but cf. Boas 2003: 113). The verb yell, for example, does not require identification of its theme (i.e. that which is yelled) in simple clauses, (14), and the same is true in RCs, (15). (14) Al yelled. (15) Al yelled his throat hoarse.

The verb hammer generally does occur with an object naming the patient of hammering. But sometimes, particularly when the hammering is repetitive, the patient may go unexpressed, (16). Again, this is true in RCs as in simple clauses; (17) does not tells us what was hammered. (16) Al hammered ? (nails). (17)

7Al hammered his wrist sore.

Finally, verbs like cut and cany do not tolerate drop of their patients in simple clauses (18), and the same intolerance is shown in RCs (19). Carrier and Randall (1992: 187) illustrate the same point for the verb frighten, (20). (18) (a) (b)

Al cut *(the frozen meat). Al carried *(the luggage).

(19) (a) (b) (20) (a)

*Al cut the knife dull. *Al carried his neck sore. The bears frightened *(the campers).

(b)

*The bears frightened the campground

empty.

Patients in Igbo and Mandarin

9

Thus in each case the behavior of the verb in RCs corresponds to its behavior in simple clauses.5 The same pattern governs grammatical relations. A verb in M will find its thematic relata bearing the same grammatical relations in the RC clause that they would have in a simple clause. In the simple clauses (21a) and (22a), yell and pound find their agent in the subject and their (theme or) patient in the object; the opposite arrangement is impossible, (21b), (22b). (21) (a) (b) (22) (a) (b)

A! yelled slogans. *The slogans yelled Al. Rocky's fists pounded the frozen meat. * The frozen meat pounded Rocky 's fists.

Just so, neither verb can occur in a RC where the object names its agent and the subject names its theme, (23), (24). Notice that the intended meanings here are entirely plausible. (23) *The slogans yelled Al hoarse. Intended: 'The slogans made Al hoarse by Iiis yelling them.' (24) * The frozen meat pounded Rocky 's fists bloody. Intended: 'The meat made the fists bloody by their pounding it.'

When a verb is subject to the same argument requirements in both simple clauses and RCs, I will say that it shows uniform projection. And when it is characteristic of a language that its verbs show uniform projection, I will say that the language has the uniform projection property, or UPP. Thus English has the UPP. Knowledge of this is revealed by our reaction to a quote attributed to Mormon pioneer Brigham Young, (25). (25) '"God almighty will give the United States a pill that will puke them to death,' Young said during tensions in the late 1850's." (T. Egan, New York Times, 3 February 2002)

From this unusual sentence we deduce immediately that Young's grammar must have allowed sentences like (26). Were the UPP not a characteristic of English, the strength of this inference would be surprising. (26) This bitter pill will surely puke you.

Uniform projection is explained if argument requirements are stated as lexical properties of the verb, since they will then be expressed wherever the verb occurs. English RCs have therefore been take to support a projectionist model 5

Essentially the same pattern is found with respect to agent arguments. A verb will require or refuse an agent to the same degree in M as in simple clauses.

10

Alexander Williams

of argument relations, most emphatically in Levin and Rappaport Hovav 1995 (chapter 2).6

6 Arguments in Igbo and Mandarin In this section we will see that Igbo and Mandarin do not have the UPP. Systematically, verbs that must cooccur with a patient in simple clause have no such requirement in RCs. First let us get a clearer idea of the basic structure of RCs in these languages, which differs somewhat from that of the RC in English. In English, M is a verb but R never is. In Igbo and Mandarin, however, M and R are both verbs, roots that can serve as the sole predicate of a clause without auxiliary support. Thus the R predicates in (27) and (29) can head clauses on their own, (28) and (30). Notice that the R verb is moreover not constrained to be stative; in both these examples, it is eventive. (27) tä ti duàii -le nàtiào mùbàii. 3s kick snap -PFV that plank 'He made that plank snap by kicking it.' (28) nàtiâo niùbân dnàn -le. that plank snap -PFV 'That plank snapped.' (29) O ku wa -ra oba ahu. 3sS strike split -FACT gourd that

'He made that gourd split by striking it.' (30) Oba ahu wa -ra awa. gourd that split -FACT BVC

'That gourd split.'

In English, R is phrasal, in that it may contain modifiers in addition to its head, (31). But the head of R cannot be modified in Igbo and Mandarin, as shown for Mandami in (32). Thus R is a verbal head simply, and not a pirrase. (31) AI pounded the cutlet very flat. (32) tä zá (*hën) ping -le nàknài ròu. 3s pound (*very) flat -PFV that meat 'He pounded that meat (*very) flat.'

Tense and aspect suffixes follow both verbs in Igbo and Mandami, and do not attach to M independently. The direct object likewise follows both M and R, and 6

We should appreciate that Levin and Rappaport Hovav's conclusion is persuasive only if English RCs are complex predicates. If instead the means verb combines immediately with the object NP, as argued by Carrier and Randall ( 1992), then the local syntactic context of the verb will be the equivalent in RCs and in simple clauses. A n d in that case we would expect the U P P pattern whether the arguments project f r o m the verb or not.

Patients in Igbo and Mandarin

11

cannot occur between them, (33), (34). (33)

*tä ti (-le) nàtiào rniibän diiàn (-le). 3s kick ( - P F V ) that plank snap ( - P F V ) Intended: 'He made that plank snap by kicking.' (34) *0 ku (-ru) oba ahu wa (-ra). 3sS strike ( - F A C T ) gourd that split ( - F A C T ) Intended: 'He made that gourd split by striking.'

Mainly because M and R are in this way inseparable, it is widely agreed that Mandami and Igbo RCs are complex predicates. The means verb combines directly with R, and does not combine first with any noun phrase arguments (see e.g. Thompson 1973, Y. Li 1990, and Huang 1992 for Mandami; Lord 1975 and Hale, Ihiono, and Manfredi 1995 for Igbo). I will assmne that, more specifically, the first node dominating both M and R is a as in (35a), with an intermediate head introducing the relation of causation between events.7 Details of the semantic derivation will be proposed in section 7; but in outline it will proceed as in (35b). (35) (a)

α

CAUSE (b)

[Q'] =

V R

[CAUSE]([V

R

])([VMJ)

6.1 Unrealized patients In both Mandami and Igbo, a verb that must cooccur with a patient in simple clauses need not do so when serving as M in an RC. For Mandarin this observation is commonplace (L.Li 1980, Lii 1986, Ma 1987, Tan 1991, among others). Take the verb qië 'cut,' for example. In simple clauses, (36), it requires an object naming the patient of cutting. Thus sentences like (36b) or (36c) can only be analyzed as including a silent object pronomi, referring to some individual salient in the discourse. They cannot mean simply that there was an event of Lao Wei cutting something, or that there is such an event ongoing. (36) (a)

7

Läo Wèi qië -le zhúsún. L.W. cut -PFV bamboo shoot 'Lao Wei cut bamboo shoots.'

I have no strong objection to an alternative structure where the node a contains just the two verbs, and the relation of causation is introduced by semantic rule. But there is some slight morphological evidence for the presence of the head in Mandarin.

12

Alexander Williams

(b)

(c)

*Läo Wèi qië -le. L.W. cut - P F V Intended: Ί cut' (Can mean: 'He cut it.') *Läo Wèi zài qië. L.W.

PROG cut

Intended: 'Lao Wei is cutting.' (Can mean: 'Lao Wei is cutting it.')

Wlien qië 'cut' is the means verb of an RC, however, no such requirement holds. The RC in (37), for example, can mean just that the subject made the knife dull by cutting something. No nomi phrase names what is cut. (37) tä hái qië dim -le nïde càidâo. 3s also cut dull -LE your food.knife 'He also made your cleaver dull by cutting.' (Adapted from Ma 1987: 428)

(37) does not contain a silent object pronomi, referring to the patient. Syntactically the sentence has no space for a second object, (38). (38)

*ta hái qië dim -le (zhúsún) nïde càidâo (zhúsün). 3s also cut dull -LE (bamboo) your food.knife (bamboo) Intended: 'He also made your cleaver dull by cutting bamboo.'

Pragmatically, moreover, (37) is not constrained to occur only in a context that would license silent pronominal reference to the patient of cutting. The context of (39a), for instance, does not license pronominal reference to anything but the cleaver, yet (39b) is felicitous nonetheless. (39) (a)

(b)

càidâo zënme húishi a? cleaver how happened Ρ RT 'What happened with the cleaver?' Lào Wèi qië dim -le [pro]. L.W. cut dull -PFV it 'Lao Wei made it dull by cutting.'

Should the speaker of (37) want to identify what was cut, this can be done (among other ways) by adjoining an adverbial verb phrase, as in (40). Yet regardless of whether this addition is required by the conversation, it is not required by the syntax.8 (40) Lào Wèi qië zhúsün, qië dim -le càidâo. L.W. cut bamboo shoots, cut dull -PFV cleaver 'Cutting bamboo shoots, Wei made the cleaver dull by cutting.'

Patients in Igbo and Mandarin

13

Filially we cannot say that the direct object in (37), càidâo 'cleaver,' is itself an argument of the means verb. The cleaver is indeed the instrument of the means event; but in simple clauses qië 'cut' cannot take an instrument as its direct object, (41). (41)

*tâ qië -le ni de càidâo. 3s eut -LE your cleaver Intended: 'He cut with your cleaver.'

This pattern is systematic. With few exceptions, any verb in M can occur without the patient argument required in simple clauses. (42)-(44) give further examples. (42) wö cä zâng -le liângkuài móbù. Is wipe dirty-PFV two towels Ί made two towels dirty by wiping.' (ex. Wang 1995: 148, trans. AW) (43) tä pai téng -le shöu. 3s smack hurt -PFV hand Can mean: 'He made Iiis hand hurt by smacking [something else].' (Adapted from L. Li 1980: 98, trans. AW) (44) tä mäi -köng -le qíanbao. 3s buy -empty -PFV wallet 'He bought (so much that) his wallet (got) empty.' (ex. and trans. Tan 1991: 100)

It can be shown, just as it was for (37), that none of these RCs includes a nomi phrase naming the patient (or theme) of the means event; yet in each case M is a verb that must cooccur with a patient (or theme) argument hi simple clauses, and cannot take an instrument as its direct object. Igbo displays the same pattern as Mandami, just as systematically. A verb required to cooccur with a patient in simple clauses is subject to no such requirement when in M. Take the Igbo verbs bi 'cut' and gwu 'dig out', for example. 9 In simple clauses like (45) and (46), these verbs must cooccur with an argument nomi phrase that identifies what was cut or what was dug out. (45) O bi -ri osisi. 3sS cut -FACT wood 'He cut wood.'

8

9

Such VPs are considered adjuncts not only because they can be dropped, but also because they cannot include aspectual suffixes or modal verbs. The Igbo data I present here come mainly from primary research I conducted with native speakers from Nigeria, now living in the Philadelphia area. More information on Igbo RCs can be found in Lord 1975; Nwachukwu 1987; Uwalaka 1988; Déchaîne 1993; Hale, Ihionu, and Manfredi 1995; and Igwe 1999.

14

Alexander Williams

(46) O

gwu -ru

3sS dig

ji.

-FACT y a m

'He dug up yams.'

Unlike Mandarin, Igbo has no silent object pronouns; so (47) and (48) have no grammatical analysis at all. (47) (a)

*0 bi -ri

(ebi).

3 s S c u t -FACT ( B V C )

(b)

Intended: 'He cut [stuff].' *0 na ebi (ebi). 3 s S PROG SBRD- c u t ( B V C )

(48) (a)

Intended: 'He is cutting [stuff]' *0 gwu -ru (egwu). 3sS dig

(b)

-FACT ( B V C )

Intended: 'He dug up [stuff].' *0 na egwu (egwu). 3 s S PROG S B R D - d i g

(BVC)

Intended: 'He is digging up [stuff].'

Yet when bi 'cut' and gwu 'dig out' appear in M, there is no need for a patient. (49) and (50) are perfectly natural, despite the absence of any nomi pirrase identifying what was cut or what was dug out. (49) O 3sS 'He (50) O 3sS 'He

bi -kpu -ru mma. cut -blunt -FACT knife made his knife blunt by cutting [stuff].' gwu -ji -ri ogu. dig.up -snap -FACT hoe made the hoe snap by digging up [stuff].'

Again, these are not cases of silent anaphora, since Igbo has no silent object pronouns. Nor do they express alternative argument structures for bi 'cut' and gwu 'dig up', alternatives which select an instrument rather than a patient as object. In simple clauses an instrumental object is impossible, (51), (52). (51)

*0 bi -ri mma (n' osisi). 3sS cut -FACT knife (P wood) Intended: 'He cut with a knife (at wood).' (52) *0 gwu -ru ogu (na ji). 3sS dig.up -rV hoe (P yam) Intended: 'He dug with Iiis hoe (at yams).'

We can only conclude that the requirement associated with these verbs hi simple clauses is absent hi RCs. Should the speaker want to identify the patient of the means event, this can be done by means of an adjunct PP, as in (53) and (54).

Patients in Igbo and Mandarin

(53) O 3sS 'He (54) O 3sS 'He

15

bi -kpu -ru mma n' osisi. cut -blunt -FACT knife Ρ wood made his knife blunt cutting wood.' gwu -ji -ri ogunaji. dig.up -snap -FACT hoe Ρ yarn made the hoe snap digging up yams.'

But the addition of this information is not syntactically required. The behavior of bi 'cut' and gwu 'dig out' is in no way exceptional. Verbs that require a patient in simple clauses do not when in M, quite generally. Three more examples are given in (55)-(57). (55) O so -ja -ra osisi. 3sS poke -splayed -FACT wood 'He splayed the stick by poking [with it].' (Can also mean: 'He splayed the stick by poking it.') (56) O de -ji -ri pensili. 3sS write -snap -FACT pencil 'She made the pencil (nib) snap by writing.' (57) O bu -lio -ro olu (η' ibu). 3sS carry -sore -FACT neck (Ρ load) 'She made her neck sore by carrying (a load).'

In none of these RCs is there a noun pirrase identifying the patient of the means event. Yet so 'poke', de 'write', and bu 'carry (on the head)' are all verbs that require a patient in simple clauses, and cannot take an instrument as object. 6.2 Patients in unexpected places The suspension of simple-clause requirements is also evident in the correspondence between thematic and grammatical relations. A verb constrained to find its patient in the direct object of a simple clause may seem to find a patient in the subject of an RC. For Mandarin this has been observed in L. Li (1980), Lli (1986), Ma (1987), Tan (1991), and elsewhere; the most widely known discussions are in Y. Li ( 1990 and 1995). Consider (58) and (59) for example. (58) (a)

(b)

jiëjië xï -le ylfú. elder sister wash -PFV clothes 'Big sister washed (the) clothes.' *ylfu xï -le jiëjië. clothes wash -PFV elder.sister Intended: 'Big sister washed the clothes.'

16

Alexander Williams

(59) ylfu χι lèi -le jiëjië. clothes wash tired -PFV elder.sister 'The clothes made big sister tired from [her] washing [them].' (ex. Ren 2001: 326, trans. AW)

In simple clauses, (58), the verb xi 'wash' is constrained to find its patient in the object and its agent in the subject. Yet in the RC (59), the subject is understood as naming the patient of the means event, and the object, its agent: big sister washes the clothes. So constraints on the correspondence between grammatical and thematic relations in simple clauses are apparently voided when the verb is in M. Tan ( 1991 ) suggests that sentences like these reflect the possibility of the verb in M occurring intransitively and nonagentively, as in (60).111 (60)

ylfu

xi

-le.

c l o t h e s w a s h -PFV

'The clothes are washed.'

She then proposes that, in RCs like (59), the means verb occurs in its intransitive guise, and consequently assigns its patient role to the subject. But this cannot be correct. Construed as nonagentive intransitives, sentences like (60) have a result-state interpretation. (60) means that the clothes are in the state that results from washing, for example. Yet this meaning is no part of (59). (59) does not mean: 'The clothes being in a washed state made big sister tired.' It means rather that washing the clothes made her tired. The contribution of the means verb here is eventive, and not (result-) stative. Thus we should assmne that the verb in (59) is the eventive transitive of (58) and not the result-state intransitive of (60). Now let us turn to Igbo. My interviews with Igbo speakers have hinted that sentences like Mandami (59) are possible in Igbo as well: transitive RCs where the subject is the patient of the means event while also being the agent of causation. Of the four speakers I consulted with most regularly, two accepted (61) and two rejected it. (61) % Ji alili gwu -ji -ri oguya. yam that dig.out -snap -FACT hoe 3sPOSS 'That yam snapped Iiis hoe by digging [it] out'

For those who accept this sentence, the subject, ji ahu 'that yam,' is understood as the patient of the means event: the yam is what was dug out. In simple clauses, however, gwu 'dig out' must find its patient in the object, (62). 11 10

Tan demonstrates that sentences like (60) do indeed have an intransitive analysis, under which there is no silent pronoun referring to an agent, and the patient NP is the (surface) subject. This conclusion accords with the consensus in the Chine se-language literature (e.g. Gong 1980), and with the perspective in Li and Thompson ( 1994); but see LaPolla ( 1988) for disagreement.

Patients in Igbo and Mandarin

(62) (a)

O

gwu

-rii

17

ji.

3 s S d i g . o u t -FACT y a m

(b)

'He dug out yams.' *Ji aliu gwu -ru

ya.

y a m t h a t d i g . o u t -FACT 3s

Intended: 'He dug out that yam.'

So for some speakers of Igbo, constraints on the correspondence between thematic and grammatical relations are relaxed when a verb appears in an RC. I do not know what to make of the disagreement among speakers. But it is interesting that (61) was sensible to any speakers at all. Contrast the English caique in (63), which provokes only bafflement. (63) *Thatyam

dug his hoe apart.

6.3 Summary of the data The thematic relations an Igbo or Mandarin verb must enter in simple clauses, it need not enter when in the means predicate of an RC.12 Correspondingly, while the interpretation of subject and object is fixed with respect to the verb in a simple clause, it is largely free with respect the means verb in an RC. At the same time, two correlated aspects of interpretation remain fixed. The subject names the agent of the event of causation, and the object names the thing caused to enter the result state defined by R. Unlike thematic relations to the means event, these relations to the event of causation are never reversed (Y. Li 1995). The understood causer, for example, is never named by the object. And while it may happen that the sentence has no noun phrase naming the patient

11

One of the speakers who accepted (61 ) also accepted (i). This sentence could not be tested with m y other consultants, however, as their dialects do not include the verb no 'tired, sore' (Green and Igwe 1963: 232, Igwe 1999: 559). (i)

Ibu bu -no -ro ya olu. load carry -sore -FACT 3s neck 'The load made his neck sore f r o m carrying.' Here M is bu 'to carry on the head', and the subject names what is carried. But this is impossible when bu is on its own, ( ii ). ( ii )

*Ibu bu -ru ya. load carry -FACT 3s Intended: 'He carried the load.' The speaker w h o accepted ( ii )grew up in the Isu-ikwu-ato region of an area now known as A m biya, formerly a part of Imo State. The dialect studied in Green and Igwe ( 1963) was spoken "near Umuahia by the people known as O h u h u " ( 1963: xiii). 12

It is also true that certain HOH-semantic requirements shown in simple clauses m a y be suspended when a verb is in M. Verbs that are required to cooccur with a certain semantically vacuous object noun phrase in simple clauses are under no such obligation when in M. This is a major topic in the two seminal papers f r o m which the present work derives, Thompson ( 1973 ) and Lord ( 1975 ). Yet I lack the space to discuss it here.

18

Alexander Williams

of the means event, the 'causee' in the event of causation is always identified, namely by the phrase that controls R. A theory of Igbo and Mandarin must therefore answer three questions. Why does the observed degree of freedom in interpretation obtain only in RCs? W i y is interpretation in RCs free only with respect to the means event? And how are Igbo and Mandami are different from English? I believe the only explanatory answers to these questions are provided by the theory I will now describe, the No Argument Theoiy for Igbo and Mandami, or NAT.

7 The No Argument Theoiy The facts of section 6 follow directly if we assmne that patients, as well as agents, are not arguments of the verb in Igbo and Mandami. 13 The typical transitive verb in these languages characteristically has no arguments lexically; it simply denotes a sortal on events, as illustrated in (64) for verbs meaning 'cut.' (64)

Mandarin 'cut': Igbo 'cut':

\qie\ [6¡]

= =

Xe.cut(e) λ e. cut. (e)

Correspondingly, thematic relations are introduced by the environment the verb occurs in. Kratzer's (1996) proposal for introducing agents structurally is familiar and I will adopt it here. (65) sketches the proposal by pairing syntactic nodes with their interpretation. Α υ head denoting the agent relation combines with VP, and the resulting υ is interpreted by a rule known as "event identification." (65) À * . 3 e [ [ V P J ( e ) Λ Η ( χ ) (e)]

VP

VAG

ÀyÀei.AG(ei) = y

I add that, in Igbo and Mandarin, patient relations are also introduced structurally. This is done, I will assmne, by means of a semantic rule that applies at VP, as in (66). 14 (66)

VP Àe.[V](e)APAT(e) = [DP] DP

13 14

V

Lin (2001 ) arrives at similar conclusions for Mandarin, but by a very different route. In (66) I have the direct object preceding the verb, on the assumption that verb raising will derive the correct surface order. As nothing here depends on this, however, the reader is free to reject my assumption, and to presume instead that V precedes DP underlyingly.

19

Patients in Igbo and Mandarin

Others might prefer to posit a head that denotes the patient relation, combining this with the verb by event identification, just as in the case of agents. For a simple clause whose verb is Mandarin qië 'cut' or Igbo bi 'cut,' (66) yields (67a) as the denotation for VP. Plugging this into (65) then yields (67b) for υ . (67) (a)

[ [VP D P qië ] ] = [ [ V P D P bi ] ] = λ e. cut (e)

(b)

Λ PAT(e) =

[ [„' *>AG [ V '

\x\e.cut{e)

D P qië]]

[DP] ] = [ [„, I'AG [ V ' D P bi ]] ] =

Λ PAT(e) = [DP] Λ AG(e)

=x

(67b) states directly that the object DP is the patient of pounding, and the subject, when it comes along, will be its agent. The grammar thus predicts correctly that the interpretation of subject and object in simple clauses will be fixed. But crucially, the same grammar yields a vague interpretation for subject and object in a RC, given two ordinary assumptions: the M verb and R constitute a complex predicate, and this predicate has the distribution of a simple verb. If the M verb fonns a complex predicate with R, it does not combine first with an object. In the present context, this means it does not first enter any structure that introduces a patient. Moreover, the minimal assumption about the semantics of combining M and R is that it introduces no content beyond the relation of causation. The smallest constituent containing both M and R, then, has the interpretation in (68). 15 I assiune, recall, that the CAUSE' relation here is introduced by a silent head, CAUSE, located between the two verbs; this head can be taken to denote as in (69). (68)

[ M R ] = Ae.3ei3e2[CAUSE'(e,ei,e2) Λ [M](ei) Λ [R](e2)]

(69)

[CAUSE] = ATCA_MAe.3ei3e2[CAUSE/(e,ei,e2)

Λ M(eι)

Λ

TZ(e2)]

Given the lexical denotations in (70a), therefore, the complex predicates qië dim 'cut dull' and bi kpu 'cut dull' will denote as in (70b).

(70) (a) (b)

[ dim ] = [ kpu ] = Ae.dull(e) [ qië dim ] = [ bi kpu ] = A e . 3 e i 3 e 2 [ C A U S E ' ( e , e i , e 2 )

A cut(e 1) Λ dull fa)]

15

(68) is similar to Rothstein's rule of "resultative conjunction," (i). (i)

Resultative conjunction (Rothstein 2001: 158)

A + RB = A ' j / A e . 3 e i 3 e 2 [ ( e = e i U e 2 ) Λ ( C U L (e χ ) Π. e 2 ) Λ A f a , Y) Λ B f a , y ) \ (i) presupposes an analysis of what (68) has as "CAUSE'" into a sequence of two relations, namely the first two conjuncts in the body of the formula. But we are free to import this analysis into ( 68 ). The only real difference between ( 68 ) and ( i ) is that ( i ) identifies the presumed internal arguments of M and R by lambda-abstraction, while (68) includes no such operation, since it combines verbs that have no internal arguments. If this difference is factored out, (68) and (i) can be seen as equivalent.

20

Alexander Williams

The RC predicate thus denotes a predicate true of events e wherein one event ei causes another e 2 - but it specifies no thematic relations to the means or result events individually. Now let us assmne that the minimal RC predicate has the same syntactic distribution as a simple verb. This assumption is common in the literature, where Igbo and Mandami RCs are often described as compound verbs. Here it means that complex predicates like qië dim 'cut dull' and bi kpu 'cut dull' will occur in the V slot of the VP structure in (66), yielding (71). Plugging this into the υ structure of (65) yields (72) in turn. (71) [ [VP DP [V qië dim ]] ] = [ [Vp DP [ v bi kpu ]] ] = Ae.3ei3e2[CAUSE'(e, ei, e 2 ) Λ cut (e ι) Λ dull fa) Λ PAT(e) = [DP] ] (72) [ [ T/ Υ [VP DP [V qië dim ]]]] = [ [„, Υ [VP DP [ v bi kpu ]]] ] = AiAe.CAUSE'(e, ei, e 2 ) Λ cut (e ι) Λ dull fa) Λ PAT(e) = [DP] Λ AG(e) = χ The VP and υ structures introduce thematic relations. But as a matter of locality, these relations predicate of the mam event of causation, and not of its subevents of means and result. The semantics thus tells us that the subject is the agent of causation and the object is its patient, but says nothing explicit about their relations to either the means or the result events. Interpretation with respect to these events is consequently free - except insofar as it is constrained, semantically and pragmatically, by being the agent and patient of a certain event of causation. This predicted degree of vagueness is exactly what the Mandarin and Igbo data show, I suggest. 16 The subject and object of a RC may be construed as bearing any plausible thematic relation to the means event, or no relation at all, because the semantic representation insists on none in particular. Construal with respect to the result event, on the other hand, is limited by the one semantic constraint that seems natural. Any definition of the basic predicates PAT and CAUSE' should have (73) as a theorem. (73) If χ is the patient of e m causing e r , then χ is the patient of e r . 17 So if a plank is the patient of kicking causing snapping, then the plank is the patient of snapping, and hence winds up snapped. This is simply what it means to be the patient of an event of causation. Parsons makes essentially the same claim for his "Themes" of "BECOME" events - which, after all, can be regarded 16

17

Sybesma ( 1999) has similarly suggested that vagueness is what is behind the facts of Mandarin, though his analysis of the RC is otherwise different. In case e r is a state, rather than an event of state change, we will have to consider the patient of a state as its holder. If this is unacceptable, we can simply restate ( 73 ) less gracefully as (i ). (i)

If χ is the patient of e m causing e r , then χ is the holder of the result state in e r .

21

Patients in Igbo and Mandarin

as events of causation with no means event or agent specified (Dowty 1979, Parsons 1990): "The Theme of [BECOME's] event is the same as the Theme of its Target state: BECOME(e,s) -*· [Theme(e,z) Ξ Theme(s,x)]" (Parsons 1990: 119). Given this semantics, it follows definitionally that the direct object in an RC, because it necessarily names the patient of the CAUSE' event, also controls the result predicate R. Take (74), for example. (74) (a)

(b)

Láo Wáng ti duàii -le nàtiào mùbàii. L.W. kick snap -PFV that plank 'Lao Wang made that plank snap by kicking.' [(74a)] = À x À e . C A U S E ( e , e i , e 2 ) Λ kick (e ι) Λ Λ PAT(e) = plank Λ AG(e) = laowang

snap(e2)

Here the object controls R, but not because the denotation in (74b) states any relation between the plank and the snapping. Rather, it establishes a patient relation between the plank and the event of kicking causing snapping. The relation to the snapping event, in virtue of which we say that the direct object controls R, is a definitional consequence. Evidently the meaning of CAUSE' does not entail identity between the agent or patient of causation and any particular participant in the means event. But there do seem to be default inferences; strongest among them, the inference that the agent of causation is in general the agent of the means event. Apparent variation in the strength of this inference cross-linguistically is discussed in Williams (2005).

8 Attractions and alternatives Two aspects of the NAT are attractive. First, it relies on no special valencereducing operations, posited ad hoc in the RC context, without morphological motivation. The account derives just from defining the lexical primitives, and observing that RCs are complex predicates, at least in Mandami and Igbo. Second, it implies a natural point of cross-linguistic difference. We can assume that Igbo and Mandarin differ from English just in the lexical valence of verbs which describe the same sort of event, (75). In English the patient is an argument of the verb, and consequently English shows uniform projection (see sections 4 and 5). (75) (a)

(b) (c)

Mandarin'cut': \_qie\ =

Xe.cut.(e)

Igbo 'cut': [fa'] = λ e. cut (e) English 'cut': [c*rf] = Xx . . . Xe.[cut(e) Λ PAT(e) = χ . . .}

That verbs with similar meaning may differ in apparent valence is a familiar observation. Discuss and argue describe very similar activities, but only discuss

22

Alexander Williams

requires a direct object to identify the topic of conversation. Wliat the NAT asks us to assume is just that languages may exhibit characteristic differences in how many arguments they assign to a verb lexically, within the range allowed by the number of participants in its event. 18 This seems a plausible assumption. Now let us consider alternative accounts. How might one model the Mandarin and Igbo data while assuming, contra the NAT, that (at least) patients are arguments of the verb? I see three clear possibilities, but I think they all fail as explanations. First, we could say that each verb has multiple lexical argument structures, but most are permitted only in the M context. Perhaps xf 'wash' has several lexical entries, for example, corresponding to the several denotations in (76), but only the entry with denotation (a) occurs freely. The others are constrained to occur only in M. (76) [ * i ] = (a)

XyXxXe.wash{e) A PAT(e) = y A AG(e) = χ

(b) (c)

\x\e.wash(e) \e.wash(e)

(d)

\x\y\e,wash(e)

A AG(e) = χ

A PAT(e) = y A AG(e) = χ

Second, we might keep lexical verbs unambiguous, granting them only those argument structures that are manifested in simple clauses, and locate ambiguity in the complex predicate instead. The same pair of unambiguous verbs in M and R, that is, might yield a complex predicate with several distinct argument structures, (77). These differ in the thematic relations they establish between the means event and the subject or object referents. (77) [ xïlèi ] = (a)

Aj/AiAe.CAUSE'(e,ei,e 2 ) Λ (wash(ei)

A PAT(ei) = y

A AG(ei) = x) A (tired{e 2 ) A PAT(e 2 ) = y) (b)

A j / A i A e . C A U S E ' ( e , e i , e 2 ) Λ (wash(ei)

A AG(ei) =

x)

A (tired(e2) A PAT(e 2 ) = y) (c)

Aj/AiAe.CAUSE'(e,ei,e 2 ) Λ (wash(ei)

A PAT(ei) = χ

A AG(ei) = y) A (tired(e2) A PAT(e 2 ) = y) The operation of resultative predicate formation would then not define a function. More specifically, it would have the effect of arbitrarily permuting or deleting the lexical arguments of the means verb. A version of this solution was developed by Y.Li (1990, 1995). Finally, formation of an RC predicate might suppress the lexical arguments of the verb in M, through deletion or existential binding. The scheme for interpret18

The NAT itself says nothing about whether languages differ in how many participants in the verb's event must be identified in a simple clause. It says only that languages may differ in how many arguments in a simple clause are lexical arguments of the verb.

Patients in Igbo and Mandarin

23

ing RC predicates might be as in (78), for example, where 3z binds a presumed lexical argument of M. (78)

[MR] =

Aj/Ae.3;3ei3e2[CAUSE'(e,ei,e2) Λ [M](¿)(ei) Λ [R](j/)(e2) ]

We would then be free to assume that xi 'wash' does have a patient argument lexically, since suppression of this argument under complex predicate formation would ensure that it is assigned to no pirrase in the RC clause. Any understood thematic relation to the means event would be regarded as the result of inference, just as proposed within the NAT (see also Sybesma 1999). This last alternative is the most attractive. It neither multiplies dubious lexical entries nor introduces a nonfunctional operation into the grammar. But it shares with the other alternatives one basic problem. Each proposes that the M context is somehow special. It licenses argument structures not otherwise licit; it allows the verb's lexical arguments to be permuted; or it supresses them altogether. But why should the M context have these effects? More pressingly, why should it have these effects in Igbo and Mandarin but not English? If M's arguments are existentially bound in Mandarin and Igbo, for example, why shouldn't the same be true in English? Unless these questions find a good answer, the descriptive postulates of all three alternatives will seem ad hoc. I believe there is no good answer, no independent feature of the M context, just m Mandarin and Igbo, that should have any special effect on the argument structure of its occupant. Sometimes changes in valence are linked to changes m aspectuality (e.g. eventive versus stative), or in what event the verb describes (e.g. a spontaneous change versus one wrought by an agent). But it is clear that no such change affects the means verb in Igbo and Mandarin. Or changes m valence accompany changes in lexical category; Dowty (1989), for example, suggests that verbs lose their arguments under lexical nominalization. But there is no evidence that the lexical category of a root is different in M than in simple clauses. And finally, there is no fonnal indication that passive or antipassive operations apply to the means verb in Mandarin or Igbo. Only one aspect of RCs m these languages has any allure as an explanatory factor: they, unlike the RCs of English, involve a compound of two verbal heads. Yet I will show in section 9 that this prospect too is a dead end. The alternatives are therefore empirical failures. So long as we presmne that, m Igbo and Mandami as in English, patients are arguments of the verb, the RC data cannnot be explained. Yet once this presumption is removed, an explanation follows, just from the agreed fact that RCs are complex predicates with the distribution of simple verbs.

24

Alexander Williams

9 Size does not matter English allows R to be phrasal, but Igbo and Mandarin do not. For this reason Igbo and Mandarin RCs are often described as compounds, and sometimes as compounds fonned 'in the lexicon.' If we needed to claim that the M context of Mandarin and Igbo has special effects on argument structure, this difference in syntax might seem to promise an explanation of why. But in fact it could provide no explanation, for three reasons. 19 First, there is no a priori reason that combining two lexical heads, whether m the lexicon or in the syntax, should cause the argument structures of either one to be modified or suppressed. Any such effect would have to be stipulated specially, and on no clear basis. Second, the required stipulation would conflict with many studies of compounds, which have found it useful to assiune that, if a verb has argument structure, it is preserved under compounding. The interpretation of compounds like English god-fearing, for instance, has often been explained by assuming that the root verb (here,fear) maintains its argument structure, and assigns the noun its internal thematic role (see Grimshaw 1990). Third and most importantly, there is direct evidence from Mandarin that the size of R is not what matters. What accounts for the lack of uniform projection is the formation of a complex predicate, regardless of whether its secondary predicate is a head or a phrase. The evidence comes from another complex predicate type in Mandarin, called the V-de construction. The V-de construction consists of a verb, transitive or intransitive, bearing the enclitic -de, followed possibly by a noun phrase (NP 0 ) and necessarily by a verb phrase (VP2), (79). V P 2 is controlled by N P 0 when present, and otherwise by the nearest NP outside V P i . (79)

V-ife c o n s t r u c t i o n : [VPlV-de

(NP0) VP2 ]

There are two or three subtypes of V-de construction, differing in what semantic relation holds between the meanings o f V and VP 2 (Huang 1988, Lamarre 2001, Yue 2001). In one the meaning is roughly causative, and here glossing -de as 'such that' yields an appropriate paraphrase. (80)

tä h ä n

-dé w o m e n d ö u l ù o x i à -le

3s s c r e a m -DE w e

all

fall

yânlèi.

-PFV t e a r

' H e s c r e a m e d s u c h t h a t w e all s h e d tears.' (ex. L . Li 1963: 4 0 5 , trans. A W )

Two major studies of the V-de construction are L. Li (1963) and Huang (1992). 19

Most likely the size of R does explain a regular difference in word order. Among SVO languages with RCs, those that constrain R to be a head place the direct object after both M and R ( Sbj M R Obj), while the others have the object interceding (Sbj M Obj R); see Williams (2005).

25

Patients in Igbo and Mandarin

Both conclude that V and VP 2 form a complex predicate to the exclusion N P 0 ; V combines first with VP2 and then the result combines with NP 0 , 2 " Underlyingly, therefore, the V-de construction is isomorphic to the verb-verb RC. Both are complex predicates; they differ just in the size of their secondary predicates. The NAT therefore predicts that, in V-de constructions as in an RCs, a verb will not be subject to the cooccurrence requirements it is subject to in simple clauses. And this is correct. Compare (81) and (82). In the simple clause (81), kiiä 'praise' must cooccur with a patient, but in the V-de construction (82), it need not. (81)

*wö kiiä -le. Is praise -PFV Intended: Ί praised.' (Can mean: Ί praised him/her.') (82) wö päi Läo Wèi -de mäpi, knä -dé lián ta tàitài yë Is smack Lao Wei's

bùhâoyisï

horse-rump, praise -DE even his wife also embarrassed

le. PRT

'Flattering Lao Wei, I praised [him] such that even his wife got embarrassed.'

(82) cannot be analyzed as containing a silent pronomi, serving as the object of kiiä and referring to the understood recipient of praise. Syntactically there is no space for such a pronomi, either before the verb or after, (83). (83)

*wö päi Läo Wèi -de mäpi, (tä) kuä (tä) -dé (tä) lián tä tàitài Is smack L.W.'s horse rump, (him) praise (him) -DE (him) even his wife yë bùhàoyìsi le. also embarrassed PRT Intended: Same meaning as (82).

We also find that, again, notional thematic relata may be found hi unusual syntactic positions, as hi (84). (84) wandòu chi -dé rén tùi fa ruän. peas eat -DE people legs go soft 'Peas make people go weak in the knees from eating them.' (L.Li 1963:405, quoting Liu Ke)

Here the understood patient of eating, wandòu 'peas,' is the subject of the clause, and the understood agent is the object, rén 'people'. This arrangement is not possible hi simple clauses, (85).

20

For Huang (1992) the surface discontinuity of the predicate is an effect of verb-raising, which here applies to V alone. Huang also regards what I label VP2 as a clause whose subject is a silent anaphor, controlled by the nearest noun phrase.

26 (85)

Alexander Williams

*wandòu chi rén. peas eat people Intended: 'People eat peas.'

One can plausibly object that wanddu 'peas' in (84) is a topic, whose thematic relation to chT 'eat' is only inferred and not assigned grammatically. But in that case chTenters no patient relation in (84), and this is itself significant, since in simple clauses the patient relation is required, (86). (86)

*Läo Wèi chi -le. L.W. eat -PFV Intended: 'Lao Wei ate.' (Can mean: 'Lao Wei ate it.')

The fact that Mandami verbs seem to lose their arguments in verb-verb RCs is thus part of a larger pattern. Their arguments seem to get lost in any complex predicate, whether its secondary predicate is a single verb or a pirrase. One cannot use the size of R to explain the lack of uniform projection in Mandarin, therefore, without missing a major generalization. Given this, I will assmne that the size of R cannot explain why Igbo verbs do not show uniform projection either, or why English verbs do. This seems to be the null hypothesis.

10 Might patients be a vice? There is much doubt about the semantic legitimacy of a generalized patient (or theme) relation. More so than agents, the presmned patients of distinct event-types share few distinguishing properties (Parsons 1990, Dowty 1991). As Kratzer (2003) observes, we generally cannot recognize the patient of an event except under a particular description of the event imposed by the verb. Thus it is unclear whether a highly general patient predicate could be given truth conditions of any substance. Correspondingly, if there is such a predicate, it will surely require that events are individuated to a very fine grain - fine enough that each event will, so to speak, wear its patient on its sleeve (cp. Parsons 1990 and Landman 2000 on agents). Yet such finegrainedness tends to undermine a motivating ambition in semantic theory, that of relating language to a denotational domain of significantly independent structure. For this reason, the idea of a basic predicate that denotes a generalized patient relation is often considered suspect. With this paper I mean only to cast an opposing doubt, based on the distributional facts of Igbo and Mandarin. In these languages patient arguments seem to be introduced syntactically, and not only in a small group of special cases. If introducing an argument syntactically means introducing a thematic predicate mto the semantic derivation (as is commonly but not always assumed 21 ) then the

Patients in Igbo and Mandarin

27

grammars of Igbo and Mandarin do include patient as a basic predicate. And if this is correct, then perhaps our semantic ambition should be moderated.22

11 Conclusion The grammar of resultatives in Igbo and Mandami is explained directly if patients in these languages are not lexical arguments of the verb. The explanation is attractive because it does not require any special operations on argument structure that apply in Mandami and Igbo only. There is no independent indication that such operations do apply, and if they were to be postulated, it would be hard to say why they don't apply in English resultatives as well. By adopting the No Argument Theory, therefore, we afford ourselves an account of resultative structure that is cross-linguistically more uniform. The source of the observed variation is relocated to the lexicon; or, more precisely, to differences in whether a certain argument type is introduced by the verb or by the structure it occurs in. These conclusions imply two claims of general relevance. First, we need to distinguish between what sort of event a verb describes, and what combinatory requirements are associated with the verb lexically. A verb need not have as many lexical arguments as its event has thematic participants. The idea that it should has guided much research, both grammatical and psycholinguistic. But if I am right, there is empirical evidence against it. Second, we need to include patient in the inventory of basic thematic predicates, despite semantic arguments to the contrary.

References Bhatt, Rajesh, and David Embick (2004): Causative derivations in Hindi. Unpublished manuscript. University of Texas at Austin and University of Pennsylvania. Boas, Hans (2003): A Constructional Approach to Resultatives. Stanford: CSLI. Borer, Hagit (2003): Exo-skeletal and endo-skeletal explanation: Syntactic projections and the lexicon. In The NaUire of Explanation in Linguistic Theory, John Moore and Maria Polinsky (eds.). Chapter 3. Stanford: CSLI. Carlson, Greg (1984): Thematic roles and their role in semantic interpretation. Linguistics 22: 259-279. Carrier, Jill, and Janet H. Randall (1992): The argument structure and syntactic structure of resultatives. Linguistic Inquiry 23: 173-233. 21

22

For Rothstein (2001 ), syntactic introduction of an argument does »of mean introducing a thematic relation. It just means abstracting over a designated variable in a structured denotation. If this view were preferred - and if the NAT could be implemented in its terms without formal difficulty (which is not perfectly clear) - then it might be possible to state the NAT without making any reference to a patient relation. I lack the space to discuss this possibility here. With Landman (2000), we might accept that a denotational domain of fine-grained events is linguistically necessary, but proceed to relate this domain to one of coarser-grained "situations," by construing events as properties of situations.

28

Alexander Williams

Davis, Henry, and Haniida Demirdache (2000): On lexical meanings: Evidence from Salish. In Events as Grammatical Objects, Carol Tenny and James Pustejovsky (eds.), 97-142. Stanford: CSLI. Déchaîne, Rose-Marie (1993): Predicates across Categories: towards a Category Neutral Syntax. Amherst [Mass.]. Doctoral diss. Dowty, David (1979): Word Meaning and Montague Grammar. Dordrecht: Reidel. Dowty, David (1989): On the semantic content of the notion of 'thematic role.' In Properties, types and meanings, II, Gennaro Chierchia, Barbara H. Partee, and Raymond Turner (eds.), 69-130. Dordrecht: Kluwer. Dowty, David (1991): Thematic proto-roles and argument selection. Language 67 (3): 547-619. Emenanjo, Nolue (1978): Elements of M o d e m Igbo Grammar. Ibadan: University Press Limited. Goldberg, Adele (1995): Constructions. Chicago: University of Chicago Press. Gong, Qianyan (1990): Xiandai Hanyuli-de shoushi zhuyu j u [Patient-subject clauses in M o d e m Chinese]. Zhongguo Yuwen 1980 (5): 335-344. Green, M. M., a n d G . E. Igwe (1963): A Descriptive Grammar of Igbo. Berlin: Akademie Verlag. Grimshaw, Jane (1990): Argument Structure. Cambridge, MA: The MIT Press. Hale, Ken, U. Peter Ihionu, and Victor Manfredi (1995): Igbo bipositional verbs i n a syntactic theory of argument structure. In Theoretical Approaches to African Linguistics, Akinbiyi Akinlabi (ed.), 83-107. Trenton: Africa World Press. Hoekstra, Temi (1988): Small Clause results. Lingua 74: 101-139. Huang, C.-T. James. (1988): Wo pao de kuai and Chinese phrase structure. Language 64: 274-311. Huang, C.-T. James. (1992): Complex predicates in Control In Control and Grammar, Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham (eds.), 109-147. Dordrecht: Kluwer. Igwe, G. Egemba (1999): Igbo-English Dictionary. Ibadan: University Press PLC. Kayne, Richard (1985): Principles of particle constructions. In Grammatical Representation, Jean Mark Gawron, Hans G. Obenauer, and Jean-Yves Pollock (eds.) 101-140. Dordrecht: Foris. Kratzer, Angelika (1996): Severing the external argument from its verb. In Phrase Structure and the Lexicon, Johan Rooryck and Laurie Zaring (eds.), 109-137. Dordrecht: Kluwer. Kratzer, Angelika (2003): The Event Argument. Manuscript from www.semanticsarcliive.net. Lamarre, Christine (2001): Verb complément constructions in Chinese dialects: Types and markers. In Sinitic Grammar, Hillary Chappell (ed.), 85-120. Oxford: Oxford University Press. Landman, Fred (2000): Events and Plurality. Dordrecht: Kluwer. LaPolla, Randy J. (1988): Topicalization and the question of the lexical passive in Chinese. In Proceedings of the Third Annual Conference on Chinese Linguistics, Mar-

Patients in Igbo and Mandarin

29

jorie Chan and Thomas Emst (eds.), 170-188. Bloomington: Indiana University Linguistics Club. Larson, Richard (1991): Some issues in verb serialization. In Serial Verbs, Clair Lefebvre (ed.), 185-210. Philadelphia: John Benjamins. Levin, Beth, and Malka Rappaport Hovav (1995): Unaccusativity. Cambridge, MA: The MIT Press. Li, Charles N., and Sandra A. Thompson (1994): On 'middle voice' verbs in Mandarin. In Voice: Forni and Function, Barbara Fox and Paul J. Hopper (eds.), 231-246. Philadelphia: John Benjamins. Li, Linding (1963): Dai de zi de buyu ju [Sentences with complements including the morpheme de]. Zhongguo Yuwen 1963 (5): 396-410. Li, Linding (1980): Dongbu ge jushi [Sentences with the verb-complement construction]. Zhongguo Yuwen 1980 (2): 93-103. Li, Yafei (1990): On V-V compounds in Chinese. Natural Language and Linguistic Theory 8: 177-207. Li, Yafei (1995): The thematic hierarchy and causativity. Natural Language and Linguistic Theory 13: 255-282. Lidz, Jeffrey, Henry Gleitman, and Lila Gleitman (2003): Understanding how input matters: The footprint of universal grammar on verb learning. Cognition 87: 151-178. Lin, Tzong-hong (2001): Light verb syntax and the theory of phrase structure. Ph.D. diss., Department of Linguistics, University of California at Irvine. Lord, Carol (1975): Igbo verb compounds and the lexicon. Studies in African Linguistics 6 (1): 23^18. Lii, Shuxiang (1986): Hanyu jufa de linghuoxing [The flexibility of Chinese sentence grammar]. Zhongguo Yuwen 1986 (1): 1-9. Ma, Xiwen (1987): Yudonjieshi donci youguan de mouxie juzhi [Some sentence patterns relevant to verbs in the verb-result construction]. Zhongguo Yuwen 1987 (6): 424441. Marantz, Alec (1997): No escape from syntax: Don't try morphological analysis in the privacy of your own lexicon. U. Penn Working Papers in Linguistics 4 (2): 201-226. Nwachkwu, P. Akujuobi (1987): The argument structure of Igbo verbs. Technical Report 18, MIT Lexicon Project Working Papers. Parsons, Terence (1990): Events in the Semantics of English. Cambridge, MA: The MIT Press. Ren, Ying (2001): Zhubin kehuanwei dongjieshi shuyujiegou fenxi [An analysis of verbresult complement constructions where the subject and object can switch places]. Zhonguo Yuwen 2001 (4): 320-328. Rothstein, Susan (2001): Predicates and Their Subjects. Dordrecht: Kluwer. Simpson, Jane (1983): Resultatives. In Papers in Lexical Functional Grammar, Malka Rappaport, Lori Levin, and Annie Zaenen (eds.), 143-157. Bloomington: Indiana University Linguistics Club. Swift, L.B., A. Ahaghotu, and E. Ugorji (1962): Igbo Basic Course. Washington, D.C.: Foreign Service Institute. Sybesma, Rint (1999): The Mandarin VP. Dordrecht: Kluwer.

30

Alexander Williams

Tail, Fu (1991): Notion of subject in Chinese. Ph.D. diss.. Department of Linguistics, Stanford University. Thompson, Sandra A. (1973): Resultative verb compounds in Mandarin Chinese: A case for lexical rules. Language 42 (2): 361-344. Uwalaka, Mary Angela A.N. (1988): The Igbo Verb: A Semantico-Syntactic Analysis. (Beitrage zur Afrikanistik, Band 35.) Vienna: Institute für Afrikanistik mid Aegyptologie der Universität Wien. Van Valili, Robert D., and David P. Wilkins (1996): The case for 'Effector': Case roles, agents and agency revisited. In Grammatical Constructions: Their Forni and Meaning, Masayoshi Shibatani and Sandra A. Thompson (eds.), 289-322. Oxford: Clarendon. Wang, Hongqi (1995): Dongjieshi shubu jiegou peijia yanjiu [Studies on the valence of resultative complement constructions]. In Xiandai Hanyu Peiji Yufa Yanjiu, Yang Shen and Ding'ou Zheng (eds.), 144-167. Beijing: Beijing University Press. Williams, Alexander (2004): Intransitive resultatives and Igbo. Paper presented at the 74th Annual Meeting of the Linguistic Society of America. Williams, Alexander (2005): Complex causatives and verbal valence. Ph.D. diss., Department of Linguistics, University of Pennsylvania. Yue, Anne (2001): The verb complement construction in historical perspective with special reference to Cantonese. In Sinitic Grammar, Hillary Chappell (ed.), 232-265. Oxford: Oxford University Press.

Jo-wang Lin (Taiwan)

Event decomposition and the syntax and semantics of durative phrases in Chinese 1

Introduction

This paper deals mainly with the syntactic distribution and semantics of durative phrases in Chinese. It has been pointed out that durative phrases in Chinese may be interpreted differently depending upon the situation type of the sentence that they modify. According to Ernst (1987) and Li (1987), one interpretation of durative phrases is that they measure the duration of an event such as (1). This interpretation obtains when the situation described involves no change of state or non-completion of the event. By contrast, when the situation described involves a change of state or the completion of an event, they claim that durative phrases are interpreted as describing the time elapsed since completion of the event as the translation in (2) indicates. (1)

Women we

zou-le

liang-ge

walk-Asp two-CL

xiaoshi hour

'We walked for two hours.' (2)

Ta yijing jiehun

san

nian le

he already get-married three year PAR 'It has been three years since he got married.'

With regard to the interpretations of durative phrases in Chinese, two remarks are in order here. First, unlike English durative /or-phrases which are generally incompatible with change of state verbs unless the sentence is coerced into a repetitive or iterative interpretation, it is very natural for Chinese durative phrases to occur with such verbs without meaning coercion. Second, although

*

An earlier version of this paper was presented at the second Workshop on Formal Syntax and Semantics in Academia Sinica in Taipei on September 27-28, 2003 and at the Workshop on Event Structures in Linguistic Form and Interpretation at the University of Leipzig on March 17-19, 2004. I would like to thank the participants for their questions and comments. I am also grateful to Waltraud Paul for her written comments, though I am not able to incorporate most of them into the text due to constraint of space. This work was supported by National Science Council of Taiwan, grant No. 93-241 l-H-009-011.

32

Jo-wang Lin

Emst (1987) and Li (1987) have described the second reading of durative phrases as "the duration since completion of an event" (SCE reading), it is more correct to say that this reading involves measuring the duration of the consequent state resulting from an event. 1 In fact, one of the goals of this article is to clarify the interpretation of durative phrases which occur with change of state verbs. I will refer to durative phrases associated with a change of state verb as R-related (result-related) duratives, pending my arguments in a later section. In turn, I will refer to the first reading of durative phrases as the P-related (process-related) interpretation. P-related and R-related durative phrases raise many interesting semantic and syntactic questions. For example, semantically speaking, do the two interpretations represent lexical ambiguity of durative phrases, as Ernst's (1987) and Li's (1987) description of the two readings as the 'since completion' and 'event duration' readings might imply? It is the goal of this paper to show that durative phrases in Chinese are not lexically ambiguous. Syntactically, R-related duratives differ from P-related duratives in that the latter may precede or follow the direct object, whereas the former can only follow the direct object. This is illustrated by the examples in (3) and (4).

(3)

(a)

Wo

(yijing) kai

jichengche ershi

nian

le

I

already

taxi

year

PAR

drive

twenty

Ί have already driven a taxi for twenty years.' (b)

Wo

yijing

kai

ershi

I

already

drive

twenty year

nian jichengche taxi

le PAR

Ί have driven a taxi for twenty years.' (4)

(a)

Women yijing

dida

zhongdian

shi

fengzhong

le

we

reach

destination

ten

minute

PAR

already

'It's already been ten minutes since we reached the destination.' (b)

*Women we

yijing

dida

shi

fengzhong

zhongdian

le

already

reach

ten

minute

destination

PAR

'It's already been ten minutes since we reached the destination.'

Another goal of this article is to explain why R-related duratives behave differently from P-related duratives with respect to the word order problem. In addition to the above syntactic and semantic issues, durative phrases are worth studying because of their interaction with Incremental Theme verbs. It is well-known that the combination of an Incremental Theme verb with a quantized object NP produces a quantized, i.e., telic, predicate (Krifka 1989, 1992,

Although I do not agree with Ernst (1987) and Li (1987) about the "since completion" reading, I will still use the smce-construction to translate the Chinese sentences in many cases, as this is the most idiomatic way to translate it. However, this has no implication for the analysis of these sentences. Also see Paul (1988) for a different description of the two interpretations.

Event decomposition

33

1998; Filip 1999, among many others). This then predicts that durative phrases in an Incremental Theme sentence should elicit an R-related reading, and this prediction is correct. To our surprise, however, such sentences also allow durative phrases to receive the P-related reading, as is evidenced by the ambiguity of

(5). (5)

Zhe-dong

fangzi wo

gai-le

san

nian

le

this-CL

house

build

three

year

PAR

I

(i)

Ί have been building this house for three years'

(ii)

'It's three years since I built the house./I built this house and the result state has existed for three years.'

The final goal of this article is to explore how the ambiguity of the durative phrase in (5) arises and discuss the implications of such constructions for the semantic representation of Incremental Theme verbs.

2 Previous analyses of durative phrases Several syntactic analyses of Chinese durative phrases have been proposed in the literature (Teng 1975, Huang 1982, 1991, 1997, Li 1987, Ernst 1987, Tang 1990,1994, Sybesma 1999). However, none of them adequately accounts for all of the relevant data discussed in the last section. I will discuss these analyses in turn. To begin with, let us consider Tang's (1990, 1994) analysis. Following Bowers (1988) and Larson (1988), she has assumed that direct objects are projected in the specifier position of V P and the verb is moved to a higher functional head. Moreover, she proposes that durative (and frequency) phrases can be adjoined to VP, as is shown in (6). (6)

[FP F

4

[vp Duration [ w

Object

[ν· V

Duration]]]

I

In (6), when the verb is raised to the functional head F, the durative-object order is derived. To derive the object-durative order, on the other hand, she proposes, following Larson's (1988) idea that oblique expressions may be base-generated as the complement of V, that durative phrases in Chinese can also be projected under the minimal V ' as the complement of V. Sybesma (1999) proposes that durative phrases function as massifiers (mass-classifiers) just as a numeral-classifier does in a nominal structure such as san ping jiu 'three bottle wine'. Therefore, the durative phrase may precede the object NP. As for the object-durative order, he has assimilated the structure to

34

Jo-wang Lin

locative resultatives with zai and dative structures. The exact details of Sybesma's proposal are beyond the scope of this article. The major empirical problem with Tang's and Sybesma's solutions to the distribution of duratives is that both analyses fail to distinguish P-related duratives from R-related duratives. As noted, while P-related duratives may precede or follow the direct object, R-related duratives only follow the object. As far as I can see, their solutions would apply blindly to any type of verb, thus incorrectly predicting that R-related duratives should be able to precede the object NP just like P-related duratives. Huang (1991, 1997) does not directly discuss the word order problem of durative phrases. He does, however, make a very specific proposal about their syntactic position. According to Huang, a sentence like Ta kan-le san tian shu 'he read-Asp three days books' is analyzed as having a gerundive IP embedded to a light verb DO. The durative phrase is adjoined to I' or is the specifier of the IP. After the verb moves to DO, the durative-object word order is obtained. The problem with this approach is that it is not clear how the object-durative word order is derived. Nor is it clear how his analysis prevents R-related duratives from appearing before the object NP. Teng (1975), Ernst (1987) and Li (1987) do not discuss word order variation in detail but they do provide explanations for why the R-related duratives do not appear before the object NP. According to Teng, an R-related durative must occur after the object NP, because it is not a constituent within the VP but the main predicate of the sentence. 2 Li (1987) does not argue against Teng's main predicate hypothesis of R-related durative phrases. She does, however, convincingly demonstrate that in addition to functioning as the main predicate of a sentence, R-related duratives (and P-related duratives) can also be VP constituents. This manifests itself in the light of the distribution of adverbs such as yijing 'already', which may occur either before the verb or before the durative phrase, and the scope interaction between negation markers and durative phrases. Since R-related duratives can remain within the projections of a verb, the question still needs to be answered as to why they cannot occur before the object NP. Li's answer to this question is that, when a durative phrase occurs before the object NP, it forms a constituent with the object NP and is the specifier of the NP. She argues that this analysis is supported by the fact that de, a general modification marker, can be inserted between the durative phrase and 2

That a durative phrase can be (part of) the main predicate is supported by the fact that the verb you 'have' can be placed before the durative as is shown in (i) (Teng 1975, Ernst 1987). (i) Ta lai meiguo you liang nian le he come America have two year PAR 'It has been two years since he came to America.' In this article I will not discuss the status of you. It can be analyzed as part of the main predicate, as in Teng (1975) and Ernst (1987). The major concern in this article will be the occurrences of the duratives that are not the main predicate of the sentence.

Event decomposition

35

the object NP. Moreover, she proposes that in such a structure the durative phrase is interpreted as quantifying over the verb, i.e., "the quantity or extent of certain activity". According to Li, this is why an achievement sentence like *Ta lai san-ge yue (de) meiguo 'It's been three months since he came to America' is ill-formed. The problem with Li's approach is that it is not possible that a durative phrase is always the specifier of an NP. Notice that ¿/¿-insertion is possible only when the object NP is a bare noun. If the object NP is a full NP such as zhe-ben shu 'this-Cl book', de-insertion is impossible (cf. * san-ge xiaoshi de zhe-ben shu 'three hours of this book'). Li's analysis seems to have nothing to say about this. Her explanation also raises the question of why a [duration + NP] phrase quantifies over events rather than result states. For constructions involving an activity verb and a bare noun object, this constraint is understandable, because atelic activity situations have no result states. For telic situations, on the other hand, it is not clear why the restriction holds. Suppose that there is a non-instantaneous change of state verb which denotes a property of the result state. Then there is no a priori reason why a result state cannot be quantified over. Therefore, unless there is a better explanation of why the structure of [V + duration + NP] is restricted to quantification over the extent of an event rather than the result state of that event, Li's analysis is not so much an explanation as a stipulation. Ernst's (1987) account for why R-related duratives may not precede object NPs is inadequate as well. According to Ernst, this is so because "the semantic rule interpreting this marked reading restricts it to VP-final position". This answer, however, does not help us better understand the nature of the R-related duratives, if no concrete proposal is offered concerning what the semantic rule is and why this rule has the property it has. In view of the above problems of the previous analyses of R-related durative phrases, I would like to pursue a different analysis in the following section that may accommodate the relevant data in a more enlightening way. I will propose a semantics-based structural account of the distribution of durative phrases inspired by von Stechow's (1995, 1996) analysis of the German word wieder 'again'.

3 An alternative account Before proposing my analysis of durative phrases in Chinese, it is helpful to first discuss the German word wieder 'again', which displays the following interpretive contrast (von Stechow 1995, 1996).3 (7)

(a)

Ali Baba

Sesam

wieder öffnete

Subj

Obj

again

(restitutive/repetive)

opened

(7a) and (7b) should be understood as subordinate clauses.

36

Jo-wang Lin

(b)

AH Baba Subj

wieder Sesam öffnete (only repetitive) again Obj opened

(7a) has both a restitutive reading in which wieder modifies the result state and a repetitive reading in which wieder modifies the whole event, whereas (7b) provides only the latter reading. To account for the above data, von Stechow (1995, 1996) argues that the ambiguity of wieder is a reflection of syntactic scope. 4 He has assumed a rather abstract syntax which decomposes telic verbs into their subcomponents (CAUSE + BECOME + STATE). Following Kratzer (1996), he suggests that the external argument of a verb is introduced by a functional head called Voice, which identifies the CAUSE component. Above the VoiceP is AgroP, where accusative Case is checked. Below the VoiceP is VP, whose head is BECOME, which in turn takes a result state projection. More concretely, his proposal can be represented as follows:

FP

(8) REPETITIVE

Àe[VoiceP(e)]

MODIFIER again NP

λχ:Υοκ:ε'

SUBJECT

(&)

voice

ACTIVE V

agent(x)(e)

BECOME RESTITUTIVE MODIFIER again

(9)

(a) (b)

NP

X

OBJECT

RESULT

3e[again(Ae[agent(Ali Baba)(e) & BECOME (opened(fcam))(e)])] 3e[agent(Ali Baba)(e) & BECOMEfagain (opened(5eram))(e)]]

In the above structure, the restitutive again is adjoined to the result XP, hence only modifying the result state, whereas the repetitive again is adjoined to a position higher than VP, hence modifying the whole event. Both the subject and the object have to be raised out of VP to check their Case in [SPEC, AgrsP] and [SPEC, AgroP], respectively. In this analysis, (7a), in which again follows both the subject and the object, is ambiguous because again can be adjoined to either the result XP or a recursive VP. In contrast, in (7b), in which again follows the 4

See also Dowty's (1979) treatment of again and/or-phrases in English.

Event decomposition

37

subject and precedes the object, it can only be adjoined to a position higher than AgroP. Hence, only the repetitive reading is possible. A very important aspect of von Stechow's analysis is its overt syntactical expression of the sub-eventuality of a bigger event, hence making that sub-eventuality accessible to syntactic modification and semantic composition. In a similar vein, Foli (2002) and Ramchand (2003) have recently argued that there are three sub-event projections for a lexical expression that entails a result state. These three projections are vP, VP and RP. According to Ramchand, "vP introduces the causation event and licenses different types of external argument", "VP specifies the nature of the change or process and licenses the entity undergoing change or process" and "RP gives the 'telos' or 'result state' of the event and licenses the entity that comes to hold the result state". In this paper, I share the above authors' view of the syntax of event structure and will propose an account for Chinese durative phrases based on such event decomposition in overt syntax. Following Ramchand (2003), I assume that each verb carries some semantic features that need to be checked by a head in the syntax. In particular, I assume that a result predicate in the representation of a verb meaning has to check the result feature of the head R of RP, the event predicate has to check the process feature of the head V of VP, and the Agent predicate has to check the agent feature of VoiceP. The last point about my syntactic assumption of the phrase structure is the position of AgroP. As noted, von Stechow (1996) has assumed that, in German, AgroP is located above VP. In Kratzer's (1996) original proposal, however, accusative Case is actually assigned below VoiceP in the specifier position of VP via government. Kratzer's analysis indicates that AgroP can, in principle, be located below VoiceP if her framework is translated into Chomsky's minimalist program. Indeed, in this paper, I will assume that, in Chinese, AgroP is located below VoiceP and above VP. Similar assumptions for this kind of structure can be found in Basilico (1998) and Sanz (2000). The different positions of AgroP can be thought of as choices of different parameters. In addition to the above syntactic assumptions, there is a semantic requirement of durative phrases that needs clarifying before I am able to explain their distribution. It is well-known that expressions like for two days/in two days are used to distinguish between accomplishments and achievements, and activities (processes) and states. While expressions like for two days occur more naturally with processes and states, expressions like in two days generally occur with accomplishments and achievements. This suggests that durative phrases impose an aspectual homogeneity requirement on the constituent that they modify (Moltmann 1991, Dini and Bertinetto 1995). I assume that durative phrases in Chinese have this homogeneity requirement as part of their selectional restriction. Now let us return to the distribution of durative phrases. As we saw earlier in (3), when a durative phrase occurs with a process/activity, the durative phrase

38

Jo-wang Lin

may precede or follow the direct object. The permutation of word order can be explained by adjoining the durative phrase to AgroP or VP as shown below. ( 1 0 ) UgrsP

W0 k [voiteP t k [voice'kaij [ Ag roP e r s h i niai) [AgroP j i c h e n g c h e i [ A g r o' AgrOj [ v p e r s h i n i a n [ v p t ¡ [ v V¡]]]]]]11]

rive

i

twenty year

i ^

ta|i

twenty year

Τ

In (10), just as the subject NP has to move to the specifier position of AgrsP to check its Nominative Case, the object NP jichengche 'taxi' must move to the specifier position of AgroP to check its Accusative Case. The motivation of such movement has been discussed thoroughly in the syntax literature. As for the movement of the verb, we can assume Ramchand's (2003) feature checking system, in which the verb has to check all relevant eventuality features. When the durative phrase is adjoined to VP, the object-durative order is derived; when the durative phrase is adjoined to AgroP, the durative-object order is derived. It is important to note here that in (10) the durative phrase can be adjoined to VP, AgroP or VoiceP, because they are all process projections which meet the homogeneity requirement. However, durative phrases are rarely adjoined to VoiceP or higher projections unless they are contrastive as is shown in (11) below. (11)

Ta

san

he

three day

tian

han

baba

with

father sleep

shui,

tian

han

mama

shui

four day

si

with

mother

sleep

'He sleeps with his father for three days and sleeps with mother for four days.'

In (11), the durative phrases can also occur after the verb shui 'sleep' without affecting the meaning. Although I do not know exactly why P-related durative phrases have to be contrastive in order to occur in a preverbal position, I would like to speculate the following: Indefinite NPs (in Chinese and in many other languages) occurring in a preverbal position must receive a specific interpretation. Since durative phrases are mainly used to assert cardinality rather than existence of referents, special contexts are required to license their presence in a preverbal position. A contrastive context is one context that may remove the specificity requirement. Whether or not this explanation is correct, the fact that P-related duratives may appear in a preverbal position is predicted by the homogeneity requirement of durative phrases. Similarly, the homogeneity account predicts that durative phrases may appear in the sentence-initial position, but this prediction is not borne out. Again, an account like what the above suggested may apply to this case. The sentence-initial position is the topic position, which requires an NP to be definite or specific. Because durative phrases are not definite or specific NPs, they are excluded from the sentence-initial position. The situation here is similar to other

39

Event decomposition

indefinite object NPs as the ungrammatically of the sentence *Liang-ben shu, wo kan-le 'Two books I read-Asp' shows. Next, let us consider cases when the main verb is inherently telic. In such cases, the phrase structure contains a result state projection. Take (4a) for instance. It is represented as follows: 5 (12)

[AgrsP

women k

[Volccp

I

tk

[voice 1

didaj [ AgroP *shi fenzhong [ AgroP zhongdian, [Agr„· Agro; reach

ten minute

destination

[vp *shi fenzhong [ w t, [ v · Vj [RP shi fenzhongfap ek [R. t¡ t¡] ]]]]]]]]]]]] ten minute

ten minute

In (12), among the five projections, RP, VP, AgroP, VoiceP and AgrsP, only the most deeply embedded projection RP represents a homogeneous eventuality. Neither a process plus a result state nor a causing event plus a process plus a result state can be homogeneous. Therefore, it follows from the homogeneity requirement of durative phrases that they can only attach to the result projection RP in (12). Since the verb has to be raised to the Voice head and the object NP must move to AgroP, it follows that the durative NP must follow the direct object. This not only explains why a durative phrase always has an R-related interpretation when it occurs with a verb that lexically entails a result state, but also accounts for why R-related duratives may not appear before the direct object NP.

4 The semantics of durative phrases In this section, I turn to the semantics of durative phrases. I will show that the very same meaning of durative phrases may apply to both P-related and R-related durative phrases, thus arguing against an ambiguity approach to the interpretations of Chinese durative phrases. I define the meaning of a durative NP as in (13): (13)

[[Durative NP]] =: À R Xe/s[R (e/s) & MU(x(e/s)) = Num Λ Hom(e/s)]

where e is a variable ranging over events, s is a variable ranging over states, τ is a function which, when applied to an eventuality, gives its temporal trace, MU is a measure unit function which yields a number when applied to a temporal trace and Horn is a predicate representing homogeneity, 'e/s' means either a variable of e or variable s.6 As for the notation 'R ', it is intended to stand for a variable with a flexible semantic type involving a sequence of one or more eventuality

'e¡¡' in (12) represents an empty category that is co-referential with the subject NP. We can use a single eventuality variable here. However, in order to make the meaning clearer, I distinguish e from j so that no confusion will occur.

40

Jo-wang Lin

arguments. That is, ' R ' can be of type , , < s c , s < s , t » > , etc. (See Piñón 1999 for the use of such a notation.) The motivation for this notation will be explained later. (13) says that a durative phrase takes an expression of type R as an argument, and that the duration of the first eventuality argument of R in terms of the unit M U is the number Num as specified by the durative NP, and finally that this eventuality is homogeneous. Now let us see how this meaning of durative phrases apply to the examples discussed in the last section. Consider the representation in (10) first, in which the durative phrase elicits a P-related interpretation. Assume that when a verb is raised, it leaves a variable of the same semantic type, and that object NPs leave a trace of type e. The step by step computation of (10) is as follows: (14)

When the durative phrase is adjoined to VP [[Vj]] = R < s , t » and 6 E xtÌs of type . CAUSE is defined as 'for all eventualities e, e\ CAUSE (e, e')=l iff e ' is a caused eventuality of e.5 This analysis is in line with the Distributed Morphology assumption that the current analysis is based on, which hypothesizes that morphemes are bundles of features and that it is these feature bundles that occur in the terminal nodes of syntax (Halle and Marantz 1993, among others). Therefore (11) maps two interpretable features into a single syntactic head and yields the structure and the interpretation as (12) and (13), respectively. (12)

CAUSE-Voice Ρ J o / m ^ ^ C A U S E . Voice' [CAUSE, θ Ε χτ]

(13)

VP

CAUSE-Voice Ρ Mary

STEP2: (ΘΕΧΤ (CAUSE VP)): λχλε. 3e' [ΘΕΧΤ (e, χ) & melt(e') & Th(e', ice) & CAUSE (e, ε')] STEP 1: (CAUSE (VP)): λε. 3ε' [melt(e') & Th (e', ice) & CAUSE (e, e')] [CAUSE,

ΘΗΧΤ]

VP (melticc)

Adopting Pylkkänen's (2002) proposal concerning the interpretation of the CAUSE-Voice head in (12), we assume that the contents of a CAUSE-Voice head in which two features are bundled together is semantically complex, as shown in (13). In other words, CAUSE and θ Ε χτ are a unit only syntactically, but not semantically. The interpretation of the CAUSE-Voice head thus is 4

Pylkkänen (2002) s8parat8s CAUSE from an 8xternal-argument-introducing Voice based on a cross-linguistic observation that in some languages a causative meaning is possible without an external argument, e.g., unaccusative causatives (e.g., Finnish desiderative causatives). She further proposes that this s8paration may be universal and that CAUSE itself would never introduce an external argument. Since we observe no empirical evidence for the 8xist8nce of unaccusative causatives in Korean, we assume that two features responsible for the causative meaning and the introduction of an external argument, CAUSE and θΕΧτ, form a syntactic unit that receives an overt realization as -/-.

5

The term 'eventuality' is a neutral term for events, proposed by Bach (1981) in event semantics, given that sometimes the term event excludes states (8.g., Kratzer 1996; Harley 1995).

Minjeong Son and Peter Cole

62

carried out in two steps, in which CAUSE applies before ΘΕΧΤ (see Pylkkänen 2002 for justification of this approach). We assume that causation associated with causative predicates in Korean is expressed by the same syntactic head that introduces an external argument (henceforth, CAUSE-Voice). The semantic and syntactic representation of CAUSE-Voice presented in (13) is assumed to be the same in the underlying functional architecture of all types of causatives in Korean. 2.2.3 Semantic and syntactic decomposition of causative predicates Based on the assumptions regarding the position of an external argument and the representation of CAUSE outlined above, let us now consider how morphological causatives are represented syntactically and semantically in Korean. Example (5b) is repeated as (14). (14)

Chelswu-ka

sakwacup-ul

nok-i-ess-ta.

Chelswu-NOM

apple juice-ACC

melt-CAUSE-PST-DEC

'Chelswu melted the apple juice.'

(14) receives a semantic and syntactic representation as (15).6 (15)

CAUSE-VoiceP λε. 3e'[Ag(e, Chelswu) & melt(e') & Th(e', apple juice) & CAUSE (e, e')] Chelswu

STEP2: λγλε. 3e'[Ag(e, y) & melt(e') & Th(e', apple juice) & CAUSE (e, e')] STEP1: λε. 3e'[melt(e') & Th(e', apple juice) & CAUSE(e, e')] VP

apple juice

CAUSE-Voice

Xf λβ. 3e'[f (e') & CAUSE (e, ε')]

Vmelt λχλε. πιείτε) & Th(e, χ)

The interpretation based on the semantic composition in (15) can be paraphrased as 'there exists e ' such that the apple juice melts in e ' and e ' is a result state of e such that Chelswu is an agent in e.' By having a VP as a complement of CAUSE under the decompositional approach, an explanation for the scope ambiguity associated with tasi 'again'

6

For the und8rlying representation of the inchoative counterpart nok- melt (intran.)', we a s s u r e that VP merges with BECOME-Voice, which does not introduce an external argument in its specifier position. This assumption is in line with Harley (1995) and Folli and Harley (2002), who argue that eventiveness of predicates is determined by the presence of an Event Phrase (equivalent to VoiceP in our analysis) and that there are different kinds of Voice heads (e.g., CAUSE-Voice, BECOME-Voice, etc.). See Son (2004) for a discussion on how the argument structure of predicates in Korean is determined based on this assumption.

Syntactic decomposition of events in Korean and Standard Indonesian

63

falls out naturally; tasi can be adjoined either to the VP, where it takes scope over the caused event, or to the CAUSE-VoiceP where it takes scope over CAUSE (i.e., a causing event). Different syntactic positions of tasi, therefore, trigger two different interpretations, the restitutive reading, as in (16b), and the repetitive reading, as in (16a). (16)

(a)

CAUSE-VoiceP again ([λε. 3e'[Ag(e, Q & melt(e') & Th(e\ apple juice) & CAUSE (e, e')]](e") tasi

CAUSE-VoiceP Xe.3e'[Ag(e, Q & melt(e') & Th(e\ apple juice)

repetative^^ Chelswu

'

& CAUSE (e, e')] STEP2: λyλe. 3e'[Ag(e, y) & melt(e') & Th(e', apple juice) & Cause(e, e')] STEP1: λβ. 3e'[melt(e') & Th(e', apple juice) & CAUSE(e, e')]

(b)

VP tasi

CAUSE-Voice λ^β,Ρ· λβ. 3e'[f (e') & CAUSE(e, e')] VP

restitutive^-^' apple juice

again ([λβ'. melt(e') & Th(e', apple juice)])(e")

_ Vmelt λχλβ. melt(e) & Th(e, x)

For the meaning of 'again,' we adopt the semantics proposed by von Stechow (1996); 'again' is of type « s , t > < s , t » , a function taking two event arguments. Its meaning is defined as (17) and (18).7 (17)

(18)

Let Ρ be a property of eventualities and let e be an eventuality. again(P)(e) is defined only if 3e' [ [MAX] (P)(e') = 1 & e' < e ] Where defined, [again] (P)(e) = 1 iff P(e) = 1 MAX is a symbol of type « s , t > , < s , t » . [MAX] (P)(e) = 1 iff P(e) and there is no e' such that e is a proper part of e' and P(e') = 1

The semantics of again presupposes that the same kind of event had occurred previously. Therefore, the denotation of the higher VP in (16b), the restitutive reading, can be read as 'e ' is an event such that the apple juice melted in e ' and there has been a maximal event of the same kind before'. The repetitive reading is derived by adjoining tasi to the CAUSE-VoiceP. The denotation of the higher CAUSE-VoiceP in (16a), therefore, can be read as 'there exists e' such that the apple juice melted in e ' and e ' is a caused event of e in which Chelswu is an agent in e, and there has been a maximal event of the same kind as e before'.

The definition is read as 'P is a property of eventualities, and '' : w £ Ds A w' £ f~] b(w)] (w' £ p) (16) Singly-relative possibility·. A proposition ρ is a possibility in a world w in view of a conversational background b iff - ι ρ is not a necessity in w in view of b. (17) [illegallyg]™' c is defined iff c picks out exactly one conversational background that is legal. 5 In this case, [illegally H ]™ c = λ ρ £-D(st) · ^

Λ

'-*{p is a possibility in w in view of bc)

)

where bc is the legal conversational background picked out by c.

(17) straightforwardly casts the adverb as a negated (due to "il-") possibility operator, with two modifications. Unlike a normal modal, illegallyu is "factive". While it rules out the truth of a proposition in deontically ideal worlds, it still asserts the truth of the proposition in our world. I have cast this as straightforward entailment, but it might be considered to be a presupposition.6 This also results in the entailment that the proposition is illegal hi our world. Secondly, the conversational background is lexically restricted hi a stronger fashion than is typical for a modal. That is, modals impose only large-scale restrictions hi the kind of c.b. they use, such as deontic vs. epistemic. To the best of my knowledge, no true modals force the background to be something as specific as sets of rules or laws. Now for an example. I will show the computation of (10) to demonstrate how the high use works. Later I'll give a denotation of this sentence with a more complex event semantics, but for this example I'll assmne that move denotes a predicate of individuals: (18) (19)

[move]™' c = \x £ De . χ moved in w [illegallyg, white moved] " ' 1 iff [illegallya] ™ c ( [white moved]

= 1

iff [illegallyJJ] (Aw/ £ Ds . white moved i W ) = 1 White moved in w \

( Λ

-.((λw' £ Ds . White moved in w') is a possibility in w in view of bc)

J

The sentence asserts two things: (i) White moved in the evaluation world, and (ii) In all legally ideal worlds (w.r.t. the body of law picked out, which is the laws of chess) it's false that White moved. 5

6

I will not be specific about how a context picks out a c.b.; there are several ways to go about it. One simple way would be to assign the adverb an index and treat backgrounds as references in the assignment function. Wyner (1997) has proposed that adverbs like "wisely" in high position are predicates of facts. Under this kind of analysis, the use of such an adverb might presuppose the existence of a fact verifying the main sentence, effectively presupposing that the main sentence is true. Here I will continue to treat the apparent factivity as entailment, for reasons that will be apparent later.

Unifying illegally

87

3 Low position The predominant trend in analyzing low/manner adverbs in recent work has been to take them to be predicates of events, following Davidson (1967)'s treatment of other adverbials. I follow this trend, and in particular I assume the neo-Davidsonian analysis (Parsons 1990, Wyner 1994, Eckardt 1998, Landman 2001, among others)7. Consider the sequence of claims, again about a chess game, in (20)-(23). (20) (21 ) (22) (23)

White moved illegally. White moved a pawn illegally. White moved a pawn diagonally illegally. White moved a pawn backwards illegally.

A sentence like (20) is true if some aspect of White's move constituted a violation of the law. We are not told which aspect, and there may have been other aspects which did not violate the law. I take this to be the most important intuition about the low position sentences. Unlike the parallel high-position sentence (10), (20) says nothing about moves that White did not make. The sentence declares the particular move White made to have been illegal, and nothing more. All the other sentences are the same, except that they provide more information about the move. In fact, they could all be true in the same context, a property which the high-position sentences in ( 10)-( 13) do not have. I will make two basic assumptions about eventualities. First, I will take them to be particulars, in their own domain Dv. Second I will assume Kratzer (1996)'s "severing" of the agent argument from the verb. On this view, the verb does not lexically have an external argument, but this argument is added in by a Voice head in the Infi range. (24)

[Voice]™' c = XP e D{vt)

. λχ £ De . \e € Dv . P ( e ) Λ Agent (e, χ)

In most neo-Davidsonian systems, the event argument is saturated by an existential quantifier at some point in the derivation. It is not clear where this quantifier is introduced compositionally. Kratzer (1998) makes it part of the meaning of Aspect morphemes, but other authors (e.g. Landman 2001, Chung and Ladusaw 2004) assmne a type-shift or compositional operation called Existential Closure (EC), which gets applied when we have an unsaturated predicate and need a saturated one. I will assmne that EC is a type-shift applying as a last resort, and that composition of a sentence is not complete unless the top node is a truth value. (25) Existential Closure If a constituent α denotes a function of type (vX) for any type X, then E C ( [ a < r - c ) = 3e £ Dv . [ « Γ " ·

7

The analysis here should be broadly compatible with non-Davidsonian theories of adverbial modification. See Wyner ( 1989), Landman (2001 ) §3.4.2. It is not compatible with McConnell-Ginet ( 1982 )'s analysis of manner modification.

88

Kyle Rawlins

Verb denotations take the following fonn: (26)

[ m o v e ] ™ ' c = \e

E Dv

. e is a m o v i n g i n w

A verb binds the event to the evaluation index, as well as picking out its type. I am informally assuming a theory of counterpart relations that applies to the domain of eventualities as well as individuals (Lewis 1968,1986, among others). An event that has no counterpart in w would make a verbal predicate false at w - this is the main effect of the "in w". Given these assumptions, we are now in a position to define illegally^ As with the high use, I will use deontic modality, but the immediate application of it is slightly more complicated than for the high case. If an event is illegal, it has some aspect that is not allowed in deontically ideal worlds. We can model this by ruling out occurrences of the actual event (that are sufficiently similar to it) in the ideal worlds. I will assmne that the ordering semantics or theory of inertia worlds will cause us to consider only worlds that are close enough so that if an event has an occurrence (i.e. a counterpart), that occurrence shares all properties of the actual event. No event will have a counterpart in an ideal world that is not identical to it. Thus, a predicate like [move]"'' 0 will be true in w of an event that occurs in w' only if that event has an identical counterpart in w. 8 By abstracting over the evaluation world and applying the event argument, we can fonn a proposition of the right kind. (27)

[ i l l e g a l l y L ] ™'c is d e f i n e d iff c p i c k s out e x a c t l y o n e c o n v e r s a t i o n a l b a c k g r o u n d that is legal. I n this case, [ i l l e g a l l y L ] w'c = \P

( l

£ ß ^ t »

· Ae £ Dv

.

P{e) Λ

-i((Aii)'

\ £ Ds

. P(w')

( e ) ) is a p o s s i b i l i t y i n w i n v i e w of bc)

J

w h e r e bc is t h e legal c o n v e r s a t i o n a l b a c k g r o u n d p i c k e d out b y c.

This denotation has two mam components. Using the letter V to stand hi for the constituent illegallyL modifies, a V-ing illegallyl is also a V-ing in the evaluation world. This is the low position correlate of factivity. Additionally, in legally ideal worlds, it is not possible for such an event to take place. By ruling out identical counterparts of the event, we entail that some property of the event 8

Pete Alrenga (p.c.) has suggested to me that this is effectively the same assumption we make about sentences like "John might fail the exam" when we hold the facts about John, the exam, and what constitutes failing constant in ideal worlds (even if a listener is not necessarily aware of all of them). In Kratzer's ordering semantics, this would be modeled by using a circumstantial modal base that fixes the relevant facts, and a deontic ordering source that picks out the closest ideal worlds satisfying the circumstances. Note also, as pointed out to me by Donka Farkas (p.c.) that this is not a problem about counterpart theory, as the same fixing of the facts is necessary in other theories of transworld identity (e.g. Kripke ( 1980)). One alternative would be to quantify explicitly over properties of events, effectively moving the work into the semantics proper. This would be similar to Wyner ( 1994 )'s treatment, which quantifies over subparts of the event.

89

Unifying illegally

has caused it to be disallowed from occurring in worlds that are legally ideal. Let us consider the example computation of (20), assuming the structure in (20)' and that White is a proper name, (20') [ip White [Voice [ y p moved illegally] ] ]

In the following formulas, bc always refers to the unique legal background picked out by the context. I will start from the bottom and work upwards. IllegallyL and moved need to combine via Intensional Function Application, as the denotation of illegallyL takes the intension of a predicate of events for its first argument. Combining the two (and assuming the context will provide a suitable conversational background) gives us: • [illegallyL] ™ ' c ( [moved] w ' c φ) = • Ae £ Dv . e is a moving in w

( Λ

\

-i((Au>' £ Ds . e is a moving in w') is a possibility in w in view of bc)

J

Next, the Voice head introduces an argument place in the composition for the external argument. • [Voice] ™ ' c ( [illegallyL] ™'c ( [moved] ™ ) ) =

• Xx £ De . Xe £ Dv . / Λ

Agent (e, χ ) e is a moving in w

Λ

-i((Aii)' £ Da . e is a moving in w')

\

\

is a possibility in w in view of bc)

/

Next, the external argument composes straightforwardly by Function Application, and Existential Closure saturates the event argument to create a truth-value. Substituting the definition of possibility, we get the final result: (28)

[White Voice moved i l l e g a l l y L ] p c = 1 iff [3e : e e Dv] / Agent (e, White) \ Λ e is a moving in w

Λ \

[Vu/ : w' £ Da Aw' £ ΠΜ™)1( -i(e is a moving in w'))

)

The previous example demonstrating high use did not make use of an event semantics. For the purposes of comparison, it is useful to compute the denotation of that example (10) with an event semantics. The result is:

90

Kyle Rawlins

(19') [illegallyH, white moved]" '

/

r_

A \

„ , / Be : e É DJ L V a [Vw' : w' £ Ds Aw'

1 iff

e is a moving in w λ „7, . , A, ent e , whlte ë (' ) J £ f]bc(w)}(

-i([3e : e £ Dv] ( \ Λ

\

Agent(e, White) λ e is a moving in w J /

Comparing (19') and (28) reveals that the difference between the two sentences derives from the relative scope of the universal quantifier over worlds (i.e. the possibility operator due to the adverb) and the existential quantifier introducing the event variable. In the low use, the existential quantifier over events scopes above the universal quantifier over worlds. The high use reverses this, and the universal takes scope over the existential.

4 Pre-adjectival uses The picture becomes more complex when we consider adverbial modification of adjectives. Examples of this are most natural when the adjective is directly deverbal, but are not limited to this case. Here are some naturally occurring examples involving a variety of adverbs, and some constructed examples involving illegally. (29) ...with zoomy homage to the age of the camcorder and a clumsily realistic spontaneity among its performers. (Google) (30) It's called The Score, an appropriately generic title for a droning, high-toned little heist picture with no dash and no raison d'être. (Google) (31) ...the consequences of operating an illegally uninsured business could bring significant criminal and civil consequences. (Google) (32) When they find an illegally colored house, they'll kick down the door and drag the homeowner off to jail. (Google) (33) Alfonso noticed an illegally red house. (34) Alfonso noticed an illegally built house.

Here I will not try to account for the conditions under which illegally can modify an adjective, but will provide an analysis that gives the correct truth-conditions for sentences like (31)-(34). Sentences involving illegally in this use seem to have two readings. In (33), for example, it could be that zoning laws in the town forbid red houses entirely. In this case, the sentence says that it was illegal for the house to be red at all. This meaning, like the high use, gives rise to the entailment that no other shade of red would have been acceptable; the law bans any red houses. I will call this the whole-predicate use, and write this version of illegally as illegallyW P . This use is directly analogous to the high use.

Unifying illegally

91

The second reading can be paraphrased by saying that the way/manner in which the house is red is not allowed. The law might allow some shades of red, but ban others. It is only the particular shade of red that matches the house that is called illegal. I'll call this second use the sub-predicate use, and write this version of illegally as illegally sp. This use is directly analogous to the low use. With a deverbal case like (34), this second reading comes out more clearly, although the first is still present. It may be that it was illegal to build the house in the first place (e.g. if the land is zoned for factories), or that some part of the house violates building codes (e.g. the foundation does not meet earthquake code, and the house is in California). 4.1 The whole-predicate use Intuitively, we want illegallywp to declare that any instantiations of the adjectival predicate combined with its nomi are declared illegal. If a house is illegally WP red, we want no legal ideals to include houses that are red at the evaluation index. This can be accomplished in a similar way to the low and high cases, where we ruled out eventualities satisfying the verbal predicate, and propositions satisfying the sentence, respectively. Here we mie out individuals satisfying the adjectival predicate. The following denotation works only for attributive adjectives that normally combine with a nommai predicate via Predicate Modification. 9

where bc is the legal conversational background picked out by c.

This denotation is quite similar to the denotation for the low use, except that the predicates involved are predicates of individuals, not events.1" The correlate of factivity in this domain is to assert that the property holds of the entity argument at the evaluation world. Here is the denotation of "illegally WP red house" assuming standard denotations for the other items.

9

10

We could also create a higher-typed version of illegallywp ι if some adjectives that allow adverbs need to have higher types such as ((s(et))(et)). This denotation is also a candidate for the meaning of illegal, since the types involved are the same. However, with the adjective, we run into issues of stage vs. individual level illegality that there is not space to discuss here. Note also that this coincides with Geuder ( 2000 )'s suggestion that some adjectives are not more basic than their adverbial counterpart.

92

Kyle Rawlins

[illegally red house] w * c = (IFA, PM)

/ Xx £ De

Λ

χ is red in w χ is a house in w

Λ

-i((Aii)' £ Ds . χ is red in w')

\

\ is a possibility in w in view of bc) / where bc is the legal conversational backgroimd picked out by c.

In English, it is not possible for the house to have counterparts in ideal worlds and still be red. The house in the evaluation world, however, is red. 4.2 The Sub-predicate use I will analyze this reading by assuming that illegallysp modifies a predicate of states, just as illegallyL modifies an event-predicate. This requires assuming that at least some adjectives have a state argument that is saturated by existential closure. This has been independently proposed by Parsons (1990) §10.4. We also must assume that there is such a thing as a particular state. An illegally red house on this reading is a house where there is a state of redness, the house is in that state, and the particular state of redness that the house is in does not occur m legally ideal worlds. If this analysis is right, the denotation of illegally SP would be identical to the denotation for the low use, assuming that Dv contains states as well as eventualities of other kinds.

5 Unification In the previous sections, I have accounted independently for the high (clausal) and low (maimer) uses of illegally, as well as two pre-adjectival meanings. This section unifies these uses, extracting a single core lexical meaning, and a family of type-shifts which apply to allow the adverb to compose in a variety of positions. For the sake of comparison, here are the denotations of each use laid out next to each other. Each denotation, as before, is defined only when the context picks out exactly one legal backgroimd, and bc in each denotation refers to this backgroimd. (17)

[illegallyHr-c= λρ £ Ds, fl „])

(not final)

This merely predicates remarkable-ness of the proposition expressed by Clyde is tall, yielding a meaning that might be paraphrased 'it is remarkable that Clyde

114

Marcin Morzycki

is tall' (which is an inadequate paraphrase for reasons discussed in section 3.1). To introduce the effect of domain widening, we might modify (29b) by existentially quantifying over a domain larger than the contextually-supplied domain provided by the resource domain variable C: (30)

[ Clyde is remarkably

talle ]

(not final)

= remarkable(A3d3C'[C/DC Λ deC'A tall(Clyde)(d) Λ d>s,fl„]) This amounts to loosening the requirement that a degree of Clyde's talhiess be among the contextually salient degrees, permitting it instead to be either among these degrees or in some larger domain C that includes these degrees. Still, this is not yet quite adequate, because remarkably adverbs, like exclamatives, contribute domain widening in a particular sense that (30) does not reflect. Unlike the variety of widening that Kadmon and Landman (1993) argue any involves, exclamatives and remarkably adverbs impose the further requirement that the degree quantified over not be in the imwidened portion of the domain. For Clyde to be remarkably tall, it is not sufficient that he be tall to a degree that is among the contextually salient ones. Rather, Clyde has to be tall to some degree that is not among the degrees already contextually salient - he must be tall to a degree that has been added to the domain by widening, as (31 ) reflects: (31)

[ Clyde is remarkably

talle ]

= remarkable(A3d3C'[C/DC Λ deC'-C Λ tall(Clyde)(d) Λ d>s,fl„]) This requires that there be a degree to which Clyde is tall which exceeds the standard and that it is in the portion of the widened domain C that excludes the original domain C. This denotation seems to be an adequate representation of the meaning of Clyde is remarkably tall. It reflects that this sentence involves a claim that something is remarkable, and that what is remarkable is not merely that Clyde is tall or even that there is some particular degree such that it is remarkable that he is tall to that degree. Rather, what is claimed to be remarkable is that Clyde's height is so great that it exceeds all the heights one would otherwise have entertained. In this way, this denotation reflects the same sort of domain-widening that an embedded exclamative would contribute, thereby explaining the semantic correlation with the embedded exclamative paraphrase. The factivity entailment that is also characteristic of both remarkably adverbs and exclamatives is predicted here, too, because this denotation requires that there be a degree to which Clyde is tall that exceeds the standard for talhiess. Maintaining this requirement of exceeding the standard is cmcial to capturing the factivity entailment - the requirement of widening the domain on its own would not suffice, since it would not rule out the possibility that Clyde is tall to a degree smaller than any in the

Abverbial modification of adjectives

115

domain, and that what is remarkable about his height is how small it is.6 Other remarkably adverb sentences can be given interpretations analogous to this one.

4 Assembling the pieces The previous section arrived at a model of the interpretation of remarkably adverb sentences, but nothing has so far been said about how this interpretation is assembled compositionality. It will emerge in this section that the familiar means of semantically combining an adverb and an expression it modifies are not adequate for the task that needs to be perfonned here - and that a further examination of the syntax suggests another path to take. 4.1 The trouble with the usual options The most basic means of interpreting a modifier is intersectively, by a rule like Heim and Kratzer (1998)'s Predicate Modification. There is no straightforward way of doing this for remarkably adverbs. The principal difficulty is that for two expressions to be interpreted intersectively they must be of the same semantic type. In order to implement an intersective interpretation for remarkably adverbs and the adjectival projections they modify, it will thus be necessary to find a single type for the denotations of both the remarkably adverb and its sister. But what could this type be? One possibility that seems initially appealing is that both the remarkably adverb and its sister denote properties of degrees. This, though, is problematic, and at a minimum requires complicating the ontology of degrees significantly. To begin with, it would be necessary to find a way to construe the remarkably adverb itself as a property of degrees. Given denotations like the one arrived at above, it is at best highly unclear how this might be done. Of course, one might conclude from this that there is something severely wrong with these denotations. It could in principle be that remarkably adverbs are interpreted simply by predicating them directly of degrees. This has the appeal of simplicity, but, among other difficulties, such an approach would have to be spelled out far more before it could be made sense of. Certainly, if a degree is simply an interval on a scale as assumed here (following Kennedy 1997 and Schwarzschild and Wilkinson 2002), predicating of this interval that it is remarkable or surprising or disappointing or strange would at a minimum fail to make obvious predictions, and at worst might be as irredeemably incoherent as a claim like '12 is remarkable'. Another, perhaps less serious but non-trivial difficulty is what one might do with the type that would result when a remarkably adverb and its sister are interpreted - if this type is itself a property of degrees, as would result from an intersective interpretation, an account would have to be provided of how this 6

This presupposes that the standard will always be in the domain of quantification - an assumption natural at least, and perhaps unavoidable.

116

Marcin Morzycki

can ultimately be predicated of individuals. Certainly, there are ways in which this can be done, both by altering syntactic assumptions or semantic ones. One interesting semantic approach toward this problem may be available if degrees are formalized, as Faller (2000) proposes, as vectors in a Vector Space Semantics (Zwarts 1997, Zwarts and Winter 2000, Winter 2001). In this sort of framework, there are independently necessary type shifts that map properties of vectors ( qua degrees) to properties of individuals. Any approach in which a remarkably adverb is predicated directly of a degree also faces the problem of explaining what the relationship is between predicating a remarkably adverb of a degree and predicating its adverbial or adjectival cognates of individuals and propositions (and perhaps eventualities). One can certainly claim that remarkably simply denotes a property of remarkable degrees, and remarkable a property of remarkable individuals - but this merely conceals the problem behind the metalanguage predicate 'remarkable'. When an intersective denotation for a modifier is not possible, one usually simply adopts a higher, predicate-modifier type denotation - construing it as a function that applies directly to the modified expression. But for remarkably adverbs, this road too has dangerous pitfalls. If remarkably adverbs were predicate modifiers, they would presumably denote functions from AP denotations to AP denotations - given the assumptions here, expressions of type {{e, dt), (e, dt)). This would certainly help with the problems noted in the previous section, since the remarkably adverb could now 'have access' to the adjectival denotation in a way that would make it possible to build up a denotation like the one arrived at in section 3. But this is inconsistent with the syntactic behavior of these expressions. As we have already seen (in ( 5 )-( 6 ) ), remarkably adverbs project further structure: (32) (a) (b) (c)

Clyde is [[quite remarkably] tall]. Floyd is [[rather surprisingly] ugly]. Many voters are [[pretty horribly] conservative].

In light of this, it is not the remarkably adverb itself but rather the extended AdvP in which it occurs which must have the higher-type denotation. But to achieve this, barring some complicated, previously unattested type shift, it would be necessary to assume that other elements of the adverbial extended projection - including comparative morphology, very, and all other Degs - are systematically ambiguous between their regular denotations and ones that yield this very high AP-modifying type. This would be an exceptionally implausible and costly assumption at best. 4.2 Building up more syntax: analogy to measure phrases If, as the previous section argued, remarkably adverbs cannot be interpreted intersectively or as predicate modifiers, how should they be interpreted? A closer examination of the syntax suggests an answer.

Abverbial modification of adjectives

117

One especially clear aspect of the syntax of these expressions is that they resemble nominal measure phrases - they occur in the same linear position, and they are in complementary distribution with overt degree words modifying the APs in which they occur: (33) (a) (b)

Floyd is {six feet/remarkably} tall {*six feet/*remarkably}. Floyd is {*six feet/*remarkably} very tall.

It seems reasonable, then, to pursue a parallel syntactic analysis. I will assume that APs with absolute adjectives and measure phrases have a structure like the one in (34), in which the measure phrase occupies the specifier position of a Deg(ree) head (Abney 1987, Corver 1990, Grimshaw 1991, Kennedy 1997): (34)

DegP DP sixfeet

Deg' Deg

AP

[ABS]

tall

Under other circumstances, the Deg head can be spelled out overtly as a comparative morpheme (or other degree morpheme) or as a degree word. With absolute adjectives, it cannot be overtly spelled out; in these cases, Kennedy (1997) suggests the Deg head is instead occupied by a null degree morpheme [ABS]. In light of the similarities, it is natural to assign remarkably adverbs a similar structure, in which their phrasal projections likewise occupy the specifier position of DegP: 7 (35)

DegP

Proposals of roughly this form for degree adverbs in general - by which is typically meant any true adverbs in AP - have been made before. Abney (1987) suggests a structure similar to (35), with adverbs in a specifier position, and the structures Jackendoff (1977) has in mind would have ones like (35) among their contemporary analogues. 7

I use DegAdvP here to distinguish the degree projection of the adjective and that of the adverb.

118

Marcin Morzycki

Kennedy's [ABS] has in (35) been replaced with a similar feature [R]. Although a stronger reason to distinguish these will emerge shortly, there are at least two other, purely syntactic reasons this distinction may be useful. One of these is that [ABS] licenses a DP in its specifier, so it is Case-licensing. Remarkably adverbs, on the other hand, have no need to check Case. Another consideration here is a slight difference in distribution - measure pirrases, unlike remarkably adverbs, are possible in comparatives: (36) Clyde is {two feet/*remarkably/*siirprisingly} taller than Floyd.

It will be necessary, then, to distinguish the ability to license measure pirrases and remarkably adverbs in order to reflect that certain Degs may license one but not the other. In light of this independent necessity, there would not be any advantage to uniting the ability to license measure pirrases and remarkably adverbs in one Deg, [ABS].

This sort of structure has a number of syntactic advantages. It can account for the complementary distribution of measure pirrases and remarkably adverbs, since these both occupy the same structural position. It can account for why remarkably adverbs are obligatorily left of the adjective. And it can account for why they are in complementary distribution with overt Degs, since they require a particular (null) Deg to license them. 8 4.3 Putting the syntax and semantics together With these syntactic structures in place, it is now possible to look on the semantic compositionality puzzle with a fresh eye. For measure-phrase structures like (34), Kennedy suggests that the semantics is assembled as in (37): (37)

[ Clyde is six feet [ABS] tall ] = [ [ABS] ] ( [ tall] )([six

feet ] )([ Clyde

])

The Deg [ABS] yields a property of individuals as the denotation of the DegP. It does the semantic work of relating the AP and the measure pirrase. Given the parallels, it is natural to suppose that semantic composition works similarly in (35). The [R] feature can be taken to be interpretable, and paralleling [ABS], to be what relates the AP and the remarkably adverb semantically: (38)

[Clyde

is remarkably

[R] tall J =

[ [R] ] ( [ tall ] )( [ remarkably

8

] )( [ Clyde ] )

This structure also predicts that it should not be possible to stack remarkably adverbs, but that it should be possible to introduce them recursively. That is, while exactly one remarkably adverb phrase can occur for each AP, a remarkably adverb phrase can itself contain a remarkably adverb (e.g. [[surprisingly [terrifyingly]] ugly]).

Abverbial modification of adjectives

119

This means of putting the pieces together, via the mediation of [R], will be the key to solving the compositionality problem and arriving at the desired interpretation. It is now possible to suppose that the denotation of remarkably is identical to that of the adjective remarkable. To illustrate this, though, it will be useful to make two simplifying assumptions purely for exposition. First, I will omit the degree argument in the denotation of both remarkably adverb and their corresponding adjectives. Second, I will for the moment suppose that these corresponding adjectives denote properties of propositions rather than, say, ordinary individuals. Both remarkably and remarkable can thus be taken to have the denotation in (39): (39) [remarkable J = [remarkably ] = λρ . remarkable(p)

This is a very simple denotation, and reflects only the barest, most minimal lexical core of the meaning of these expressions. It is a long way from the making the semantic contribution that was attributed to remarkably adverbs in section 3. But the challenge of getting from one to the other can now be met straightforwardly - the additional semantic work that needs to be done can be attributed not to the adverb itself, but rather to the [ R] feature in Deg that licenses it. Just as the adverb itself can now have as its denotation only the irreducible essence of its lexical semantics, so too the [R] can now have as its denotation only those aspects of meaning that characterize the class of remarkably adverbs generally, independent of the particular choice of adverb: (40) [ [R] ] = \A{e,{dj))

XR{SIJ) λχ . R(A3d3C'[C'DC Λ deC'-C Λ A(x)(d) Λ d>s A ])

This denotation reflects exactly the semantic properties identified in section 3 as characteristic of remarkably adverbs- among the more prominent ones, domain widening. It also serves as a kind of semantic glue, helping hold together typetheoretically the adjective and adverb denotations. These pieces fit together in a way that yields the desired result: (41)

[Clyde is remarkably [R] tall J = ['[R] j([tallj)([remarkablyj)([Clydej)

= remarkable(A3d3C'[C/DC Λ deC'-C Λ tall(Clyde)(d) Λ d>s,fl„]) This is exactly the denotation ultimately arrived at in section 3 in (31 ). 4.4 Problems averted This division of labor avoids the problems raised by the alternative approaches to introducing remarkably adverbs into semantic composition. The problems associated with an intersective interpretation do not arise here because this ap-

120

Marcin Morzycki

proach does not impose the requirement that the adverb and its sister be of the same type. Consequently, we are not forced into any uncomfortable further assumptions to sustain these types. In particular, there is no analytical pressine on this view to treat remarkably adverbs as properties of degrees. Rather, the denotation of a remarkably adverb is ultimately predicated of a proposition, as seems most natural. The problems associated with a predicate modifier denotation are avoided as well. On the current account, the type of the remarkably adverb and its projections remains very simple, and more important, the same as the corresponding adjective. So it is no surprise - and indeed, expected - that remarkably adverbs should support their own degree words and project the full adverbial extended projection. It will not be necessary to assmne either massive systematic ambiguity of Degs or any novel otherwise unmotivated type-shifts, because the types of all elements of the adverbial projection will be exactly the same as they would otherwise be.

5 Relation to clausal counterparts What has now been introduced is a kind of decomposition - the apparent meaning of remarkably adverbs has been split into two parts, one associated with the remarkably adverb itself and one associated with its position. Among the chief advantages of having done things this way is that it provides a simple theory not only of the relation to the corresponding adjectives, but also of the relation to clause-modifying uses. Given the same denotation for remarkably proposed in (39), the right interpretation for its clause-modifying use in (42) will follow: (42) (a) (b)

[ Clyde is talle ] = 3d[df=C Λ tall(Clyde)(d) Λ d>s,fl„] \Remarkably, Clyde is talle ] = remarkable(A3d[d£C Λ tall(Clyde)(d) Λ d>s,flH] )

The denotation in (42) requires only that it be remarkable that Clyde is tall, which seems to reflect what the clause-modifying use of remarkably means.

6 Approaching ad-adjectival subject-oriented adverbs This approach is certainly not a comprehensive theory of ad-adjectival adverbial modification. It addresses only one class of adverbs that occur in the extended AP - but there are others. Among them is a class, to which I now fimi, that bears some resemblance to subject-oriented adverbs in VP and hence takes us one step closer to the most basic broader questions about adverbial modification. These AP-modifying adverbs include some of the canonical examples of subjectoriented adverbs, which seem to contribute roughly their usual interpretation: 9 9

'Subject-oriented' is an especially unfortunate term in this context, but I stick with it for its familiarity. Many of the examples in this section have their roots in a collection of naturally-occurring

Abverbial modification of adjectives

121

(43) Clyde seemed {intentionally/deliberately/accidentally/willingly} reliant on Herman.

That these do in fact have the agentive semantics that is a signature of subjectorientation in VP is clear from the coimterpragmatic inferences they create in an environment like (44): (44) #When he was served to his hungry Martian overlords, Clyde seemed {defiantly/imapologetically/rudely} raw on the inside.

This sentence leads us to suppose Clyde had some control over his being raw on the inside, in a way that we would not if the adverb were absent. Another distinguishing feature of (VP-)subject-oriented adverbs is focus-sensitivity, a characteristic ordinary maimer adverbs do not have (Wyner 1994, Geuder 2000). These adverbs pattern with subject-oriented VP adverbs in this respect, too, as the non-synonymy of (45a) and (45b) reflects: (45) (a) (b)

Greta seemed rudely reliant on HERMAN to clean up after her. Greta seemed rudely reliant on Herman to CLEAN UP after her.

But despite this evidence for thinking these adverbs are in a meaningful sense subject-oriented, there is a fundamental problem here. Adjectives by then very natine are stative; subject-orientation by its very natine requires agentive or voluntary eventualities, which states cannot in principle be. There is another difficulty as well, a compositional one similar to the one remarkably adverb gave rise to: If these adverbs are essentially subject-oriented, how can they compose with an (extended) AP denotation? As before, a simple intersective interpretation does not lead very far here. It is very unclear at best how an adequate denotation for the adverb could be framed in the appropriate way. Subject-oriented adverbs may denote properties of events, but - even setting aside the sortal difficulty about states versus events - there is no obvious place to plug such a thing into the structure of an AP. Moreover, an intersective interpretation is inherently symmetrical, which makes the prediction that (46a) should feel redundant, which it does not, and that (46b) should be a contradiction, which it is not: (46) (a) (b)

Clyde seems both rudely vocal and vocally rude. Clyde didn't seem rudely vocal; rather, he seemed vocally rude.

Perhaps it might be possible to swat this sort of observation away by appeal to pragmatics. This does not seem implausible, but it is definitely swimming against the empirical current.

examples gathered by Tom Ernst.

122

Marcin Morzycki

Naturally, the alternative of simply assigning these adverbs predicate modifier denotations remains. This is, in fact, how Wyner (1998) treats subject-oriented adverbs (in VP), proposing that they denote functions from properties of events to properties of events. This sort of approach, though, presents the same problem with respect to further AdvP structure encountered above with remarkably adverbs. If these adverbs denoted predicate modifiers, they would not be compatible with degree word denotations. But adverbs of this sort occur with degree words quite readily: (47) Clyde appeared {rather rudely/quite thoughtlessly/very cleverly} indifferent to others. The puzzle this leaves us with is an echo of the one remarkably adverb presented. It seems only natural, then, to consider applying the same tools here. The model of the interpretation of remarkably adverbs developed above, in which the adverb can enter semantic composition through the mediation of a Deg that stitches things together type-theoretically and makes its own particular semantic contribution, can help address both of these problems. To illustrate this fact about the semantic combinatorics, it will be necessary to sidestep the extremely important and obviously relevant but still rather murky issue of what exactly the semantics of subject-orientation is. I thus will adopt the toy semantics - certainly inadequate - in (48), in which deliberately and deliberate both simply denote properties of events:1" (48) [deliberately^ = Ae . deliberate(e) Accepting this, it seems plausible that a sentence such as (49a) might receive an interpretation like (49): (49) (a) (b)

Clyde is deliberately reliant on Herman. [ Clyde is deliberately reliant on Herman ] = 3s3e3e'[reliant-on-Herman(Clyde)(s) Λ deliberate(e) Λ agent(e)(Clyde) Λ cause(e)(e') Λ become(e')(Clyde)(reliant-on-Hennan)]

What (49) means, then, is that Clyde was the agent of a deliberate event that caused an event of Clyde becoming reliant on Herman. The claim that underlies this is that a subject-oriented adverb in the extended AP is interpreted with respect to an event that stands in a particular (causal) relation to the state associated with the adjective. Because these adverbs and remarkably adverbs have apparently the same distribution inside AP, the syntax from which this sort of denotation will be built can mirror the one proposed for remarkably adverbs above: 10

Attempts to address the semantics of subject-orientation in earnest include McConnell-Ginet ( 1982), Wyner ( 1994, 1998), Geuder (2000), Ernst (2002) and Rawlins (2003).

Abverbial modification of adjectives

(50)

123

DegP

deliberately

Deg

I [AGT]

reliant on Herman

The price to be paid here is that in order to build (49), a distinct Deg [AGT] will have to be posited. With that done, though, the composition is relatively straightforward, and proceeds along the same lines as Kennedy's for measure phrases and the one for remarkably adverbs above: (51)

[ Clyde is deliberately [AGT] reliant on Herman ] = [ [AGT] ] ( [ reliant on Herman ] )( [ deliberately ] )( [ Clyde ] )

The additional agentive meaning, and the additional events that underlie it, will b e i n t r o d u c e d b y [AGT]: (52)

[ [AGT] ] = AAASAxAs . 3e3e'[A(x)(s) Λ S(e) Λ agent(e)(x) Λ cause(e)(e') Λ become(e')(x)(A)] (53) [ [AGT] ] ( [ reliant on Herman ] )([ deliberately ])([ Clyde ]) = As3e3e'[reliant-on-Herman(Clyde)(s) Λ deliberate(e) Λ agent(e)(Clyde) Λ cause(e)(e') Λ become(e')(Clyde)(reliant-on-Herman)]

So the same theoretical architecture that provided an account of remarkably adverbs above seems to provide solutions to both of the problems this section began with. The compositional issue is solved exactly as before - the adverb is interpreted as an argument of degree morphology, which does the essential compositional work. And the problem of relating subject-orientation, which is a notion bound up with events, and the semantics of the extended AP, which is stative, is solved as well: the degree head contributes a semantics that makes available to the adverb a causing event of which it can be predicated. As before, the adverb can have a simple, first-order denotation that (plausibly) remains constant across its uses in various positions. Though no substantive proposal will be offered here of how ad-adjectival uses of subject-oriented adverbs relate to subject-oriented uses in VP, it is worth noting that elements of the account suggested here bear a surprising resemblance to the model of subject-orientation that Wyner (1998) constructs. He argues that a verbal functional head, a 'volitional' fonn of the passive auxiliary be, is crucial in explaining why the interpretation of subject-oriented adverbs can be affected by passivization. Both this element and [AGT] occupy functional heads in the

124

Marcin Morzycki

extended projection of the modified expression, and both contribute an agentivity inference to the interpretation of subject-oriented adverbs.

7 A final word The principal argument here has been that the syntactic and semantic architecture of at least one and perhaps two classes of AP-modifying adverbs involves precisely the same semantics for the adverb itself as in other positions, with additional, specifically ad-adjectival meaning arising through its interaction with a degree morpheme that introduces it. For remarkably adverbs, this additional semantics seems to involve widening the domain of degrees; for subject-oriented ad-adjectival adverbs, it seems to involve some notion of agentivity. That this 'factoring-out' approach proved useful in both cases may suggest that it may be fruitfully applied more widely, perhaps in some f o n n to more prototypical adverbiale as well. But one way or another, these AP-modifying adverbs offer a novel perspective on familiar larger questions about adverbial modification.

References Abney, S. (1987): The English Nomi Phrase in Its Sentential Aspect. Doctoral dissertation, MIT. Bartsch, R. (1976): The Grammar of Adverbials. North-Holland, Amsterdam. Bierwisch, M. (1989): 'The semantics of gradation'. In Bierwisch, Manfred and Ewald Lang, eds., Dimensional Adjectives. Springer-Verlag, Berlin. Cinque, G. (1999): Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford University Press, New York. Corver, N. (1990): The Syntax of Left Branch Extractions. Doctoral dissertation, Tilburg University. Cresswell, M. J. (1976): 'The semantics of degree'. In Partee, Barbara H., ed., Montague Grammar. Academic Press, New York. Elliott, D. E. (1974): 'Toward a grammar of exclamations'. Foundations of Language 11:231. Emst, T. (2002): The Syntax of Adjuncts. Cambridge University Press, Cambridge. Faller, M. (2000): 'Dimensional adjectives and measure phrases in vector space semantics'. In Faller, M., S. Kaufmann, and M. Pauly, eds., Formalizing the Dynamics of Information. CSLI Publications, Stanford. Geuder, W. (2000): Oriented Adverbs: Issues in the Lexical Semantics of Event Adverbs. Doctoral dissertation, Universität Tübingen. Grimshaw, J. (1979): 'Complement selection and the lexicon'. Linguistic Inquiry 19(2):279. Grimshaw, J. (1991): 'Extended projection'. In Lexical Specification and Lexical Insertion. Erlbaum, Hillsdale, NJ. Groenendijk, J. and M. Stokhof (1984): 'On the semantics of questions and the pragmatics of answers'. In Landman, Fred and Frank Veltman, eds., Varieties of Formal Semantics. Foris, Dordrecht.

Abverbial modification of adjectives

125

Hamblin, C. (1973): 'Questions in Montague English'. Foundations of Language 10(1):41. Heim, I. and A. Kratzer (1998): Semantics in Generative Grammar. Blackwell Publishers, Oxford. Jackendoff, R. (1972): Semantic Interpretation in Generative Grammar. MIT Press, Cambridge, Massachusetts. Jackendoff, R. (1977): X-Bar Syntax: A Study of Phrase Structure. No. 2 in Linguistic Inquiry Monographs. MIT Press, Cambridge, Mass. Kadmon, N. and F. Landman (1993): 'Any'. Linguistics and Philosophy 16(4):353. Karttimen, L. (1977): 'Questions revisited'. Unpublished manuscript, The Rand Corporation. Kennedy, C. (1997): Projecting the Adjective: The Syntax and Semantics of Gradability and Comparison. Doctoral dissertation, University of California, Santa Cruz. Published in 1999 by Garland, New York. Kennedy, C. and L. McNally (2004): 'Scale structure and the semantic typology of gradable predicates'. Under review for Language. Klein, E. (1991): 'Comparatives'. In von Stechow, Arnim and Dieter Wunderlich, eds., Semantik: Ein internationales handbuch der zeitgenössischen forschimg. Walter de Gruyter, Berlin. McCawley, N. (1973): 'Boy, is syntax easy!' In Coram, C., T. Cedric Smith-Stark, and Ann Weiser, eds., Papers from the Ninth Regional Meeting of the Chicago Linguistic Society. CLS. McConnell-Ginet, S. (1982): 'Adverbs and logical forni: A linguistically realistic theory'. Language 58:144. Michealis, L. and K. Lambrecht (1996): Conceptual Structure, Discourse, and Language. CSLI, Stanford. Moltmann, F. (1997): Parts and Wholes in Semantics. Oxford University Press, New York. Rawlins, K. (2003): Ά study in some adverb denotations'. B.A. honors thesis, University of Massachusetts Amherst. Rullman, H. (1995): Maximality in the Semantics of lf7?-Constractions. Doctoral dissertation, University of Massachusetts. Russell, B. (1905): 'On denoting'. Mind 14:479. Schwarzschild, R. and K. Wilkinson (2002): 'Quantifiers in comparatives: A semantics of degree based on intervals'. Natural Language Semantics 10(1 ): 1. Seuren, P. A.M. (1973): 'The comparative'. In Generative Grammar in Europe. Reidel, Dordrecht. von Fintel, Κ. (1994): Restrictions on Quantifier Domains. Doctoral dissertation, University of Massachusetts Amherst. von Stechow, A. (1984): 'Comparing semantic theories of comparison'. Journal of Semantics 3:1. Westerstâhl, D. (1985): 'Determiners and context sets'. In van Bentham, Johan and Alice terMeulen, eds., Generalized Quantifiers in Natural Language. Foris, Dordrecht. Winter, Y. (2001): 'Measure phrase modification in vector space semantics'. In Megerdoomian, Karine and Leora A. Bar-el, eds., Proceedings of WCCFL XX. Cascadilla Publications, Somerville, Mass. Wyner, A. (1994): Boolean Event Lattices and Thematic Roles in the Syntax and Seman-

126

Marcin Morzycki

tics of Adverbial Modification. Doctoral dissertation, Cornell University. Wyner, A. (1998): 'Subject-oriented adverbs are thematically dependent'. In Rothstein, Susan, ed., Events in Grammar. Dordrecht: Kluwer. Zanuttini, R. and P. Portner (2003): 'Exclamative clauses: At the syntax-semantics interface'. Language 79(1 ):39. Zwarts, J. (1997): 'Vectors as relative positions: A compositional semantics of modified PPs'. Journal of Semantics 14:57. Zwarts, J. and Y. Winter (2000): 'Vector space semantics: A modeltheoretic analysis of locative prepositions'. Journal of Logic, Language and Information 9:171.

Kjell Johan Sasbo (Oslo)

The structure of criterion predicates 1 Introduction A number of verbal predicates have an intuitively rather abstract meaning; they may specify some higher-order, typically modal, property of an action but they remain tacit on what is actually going on. Here are some examples: (1 )

(a)

(b)

obey doctor's orders, do me a favour, transgress Holy Law, give way, respond, start the nest, badger the bureaucracy, take revenge on the remote father create a fiction, ruin my reputation, surprise the nation, waste fuel, help the campaign of Senator John Kerry, undermine the war on terrorism

The items in (lb) are causative predicates, more precisely such that do not specify the way in which the change of state is brought about; Kearns (2003) calls them causative upshot predicates. I call them manner-neutral causatives. The predicates in (la) are not causative. Ryle (1949: 125-147) classifies them as achievements; Kearns (2003: 599) refers to them as criterion predicates: The key notion here is that there is some conventional criterion an action must meet in order to qualify as an event of the criterionmatching kind. While criterion predicates specify conventional (nonnative) or intentional criteria, they are imspecific about the physical criteria an action must meet. Usually, there is a need for more information on how the action is perfonned. If you ask me to do you a favour, I will want to know what it is. If you tell me that you are obeying doctor's orders, you are probably alluding to a familiar action. Very often, the context will, in various ways, fill in the picture. Much the same is true of manner-neutral causatives. One way of specifying more concrete criteria is to modify the VP with an adjunct; a clause or a PP. In English, the natural choice is a by adjunct with a present participle complement, as in (2)-(5). (2) (3)

The City retaliated by electing its own mayor. Mowgli kept a promise by killing Shere Khan.

128

Kjell Johan Sœbo

(4)

By sending rain, Yahweh had usurped the function of Baal.

(5)

It tries to escape by moving as fast as possible away from the predator.

This extends to manner-neutral causatives, as in (6)-(9). (6) (7) (8) (9)

By calling and dancing, he entices a female to Iiis bower. Yahweh saved the Israelites by opening the Sea of Reeds. Yahweh made Adam by scooping up some clay and breathing on it. In Germany they portrayed the Plague as a maid travelling through the air like a blue flame, killing her victims by raising an arm.

(I will subsume criterion predicates and manner-neutral causatives under the tenn abstract predicate.) There is a strong intuition that in each case, the merge of the by plnase and the plnase it modifies denotes one set of events, and that somehow, the by plnase predicate fills a slot in the abstract predicate. My intent is to account for these intuitions through a fonnal analysis of the abstract predicate, the instrumental adjunct, and the way they are composed. The rest of the paper is structured as follows. In Section 2,1 review the recent work of Kearns (2003) on abstract predicates and argue that it is incomplete, both as it stands and as a basis for an analysis of the by locution. Section 3 provides a critical survey of work on the by locution (Bennett 1994) and on by adjuncts in connection with causative predicates (Dowty 1979). In Section 4, I develop my own analysis based on unification in recent DRT. In Section 5, I discuss the limits to the by locution, arguing that they can be stretched through causative or criterial interpretations of predicates that are not strictly causative or criterion predicates. Section 6 offers conclusions.

2 Kearns 2003 and the Anscombe thesis According to Kearns (2003), criterion predicates do not refer to basic actions or events but to actions or events that depend on basic actions or events, to parasite events depending on host events (p. 600). What is ... parasitic about criterion predicates,... , is that the eventualities described cannot simply come about, but must be realized in the occurrence of some event which is describable in different tenns. (Kearns 2003: 627) In the acfiial analysis, there is only one event involved, but there is a (usually rather vacuous) host and a parasite description of that event. The latter, the criterial component of the predicate, is an "individual-level predicate on events", as indicated in paraplnases like those in (10) (p. 628): (10) (a) (b)

Jones broke the law. 'Jones did something, and what he did was illegal'

The structure of criterion predicates

(c)

129

3 e (do(j, e) & illegal(e))

This analysis is questionable for two reasons. First, it is hardly reasonable to ascribe properties like legality to events. Generally, it would seem that what must meet conventional criteria are event types - predicates - and not event tokens.1 Note that cases like (11a) and (12a) turn out to be trivial - contradictory or tautologous - on an analogous analysis, as demonstrated in (11c) and (12c). Obviously, one and the same event e can only occur once. (11) (a) (b) (c) (12) (a) (b) (c)

Joan made a common mistake. 'Joan did something, and what she did was a common mistake' 3 e (do( j, e) & mistake(e) & common(e)) Joan did something noone had ever done before. 'Joan did something, and what she did noone had ever done' 3 e (do( j, e) & noone had ever done e before)

In fact, there is another possible formalisation in the style of the (c) formulae, equally in accordance with the (b) paraphrases, avoiding these problems: (10) (d) (11) (d) (12) (d)

3 e 3 Ρ (P( j, e) & illegal(P)) 3 e 3 Ρ (P(j, e) & mistake(P) & common(P)) 3 e 3 Ρ (P(j, e) & noone had ever done Ρ before)

This is close to what I will propose in Section 4. Second, this analysis is questionable because it fails to fonn a sound basis for an analysis of the by locution. Kearns does not offer an explicit analysis of the by locution. In fact, she assmnes that criterion predicates, as opposed to causative predicates, occur more naturally with in -ing adjuncts (p. 629). It may be that in adjuncts of the type illustrated in (13) and (14) preferably modify criterion predicates. But cases of such predicates with by adjuncts abound in corpora, and Kearns herself discusses (p. 599) a case in which "a soldier obeys an order by fixing his bayonet". (13) In naming him, Putin ended a guessing game that had begun to overshadow a predictable presidential election two weeks from now that is seen as a sure tiling for Putin. (14) In naming him, Putin avoided creating an alternative center of power or a rival for the political spotlight.

In any case, it is clear that Kearns considers in or by adjuncts to offer host descriptions, potentially specifying the profonna "do" predicate in formulae like (10c) above. One can thus conjecture that (15a) may analyse as (15c).

1

There is an interesting parallel to the notion of felicity: Krifka ( 1998) argues, contra, e.g., Rothstein ( 2004 ), that telicity cannot be a property of event tokens but must be a property of event types.

130

Kjell Johan Sœbo

(15) (a) (b) (c)

Jones broke the law by limiting. 'Jones limited, and Iiis limiting was illegal' 3 e (hunt( j, e) & illegal(e))

This corresponds to what has been referred to in the philosophical literature as the "Anscombe thesis" in its very simplest fonn. The Anscombe thesis (according to Bennett 1994) If someone 0s by 7ring, and F is the act which makes it the case that she 0s, and Ρ is the act which makes it the case that she 7rs, then F is P. In other words, the modified predicate and the by adjunct describe one event in two ways. To be sure, there is a strong intuitive basis for this assumption. However, as spelt out in (15c) or any formula where the host and the parasite are parallel predicates and where the latter is a first-order predicate on events, it causes two (closely related) problems. First, it predicts that the construction is closed under weakening, which it is arguably not: (16) (a)

(b)

He broke the Jungle Law by limiting at the pool in a drought He broke the Jungle Law by limiting at the pool He broke the Jungle Law by limiting 3 e (hmit(j, e) & illegal(e) & atthepool(e) & inadrought(e)) 3 e (hmit(j, e) & illegal(e) & atthepool(e)) =>· 3 e (hmit(j, e) & illegal(e))

Second, as pointed out, i.a., by Bennett (1994), the Anscombe thesis is liable to predict a symmetry between the by adjunct and the modified predicate: (16) (c) (d)

He broke the Jungle Law by limiting. ? He limited by breaking the Jungle Law.

3 Second-order predicates decomposed In this section, I discuss two approaches to the semantics of the by adjunct which avoid the two problems noted above by treating what the by adjunct adjoins to as a composite expression containing an existential quantification over things such as those expressed in the by adjunct. This is a significant step forward. What these approaches do not provide is a compositional analysis. 3.1 Bennett 1994 and the "namely" analysis According to Bennett (1994), the asymmetry of the by construction falsifies the Anscombe thesis that we have two descriptions of one and the same event. His analysis differs from the one sketched in the last section in two ways:

The structure of criterion predicates

131

1. At the relevant level of analysis, the by complement does not denote a set of events; in fact, it denotes a (true) proposition (a fact). 2. At the relevant level of analysis, the pirrase modified by the by pirrase denotes a second-order entity; in fact, a set of true propositions (facts). (I will argue that 1. is inessential while 2. is essential.) Bennett paraphrases (17a) as (17b). A formalisation in the style of the formula in (10c) or (15c) (ignoring tense) could yield (17c): (17) (a) (b) (c)

Jones broke a promise - by - coming home late. Some fact about his behavior conflicted with a promise he had made earlier namely the fact that - he came home late. comehomelate(he) & promise( hiot(comehomelate(he)))(he)

Without violating the spirit of this analysis, one could reintroduce events to represent the verb pirrase break a promise by being late as (17d) (simplified): (17) (d)

λ χ λ e ( la.te(i)(e) Λ 3 ei ( promise(

3 ei (late(x)(e2)))(x)(ei)))

As a representation of the modified VP, this is intuitively not far off the mark, and it seems to avoid the two problems noted above: The construction is not predicted to be closed under weakening or to be symmetric. The reason is that the "parasite", the criterial component, is not at the same level as the "host"; it is one level up and has an argument place for the host. In principle, this pattern generalises to manner-neutral causative predicates, but this is yet to be a semantic analysis - it is not clear how the meaning of the modified VP comes from the meaning of its two daughters through the "namely" operation. Indeed, it is not easy to develop a compositional analysis along these lines. 3.2 Dowty 1979 and the by postulate Dowty (1979: 227-229) treats by adjuncts as modifying causative VPs. He first considers ascribing a causative element to by, but rejects this because it seems to result in a double causation. 2 It may be added that such a move is also problematic in connection with criterion predicates, for which the result does not seem to involve any element of causation. What Dowty proposes, instead of a translation of the preposition, is a meaning postulate: VpVPVQ Vx • [ b y f ( P ) ( y [Q{y} CAUSE > ] ) ( x ) - - [P{x} CAUSE > ] ] This ensures that if John awakened Mary by shaking her, then his shaking her awakened her - a welcome result. Dowty did not use events, but in principle, 2

In fact, one may be tempted to such a move by considering predicates, activities or achievements, that are neither clear causative nor clear criterion predicates; cf. Section 5.

132

Kjell Johan Sœbo

the by postulate could be reformulated in terms of events. And in principle, it generalises to criterion predicates (on an appropriate decomposition). But of course, the postulate does not amount to a compositional analysis. It does not specify the meaning of the by pirrase, and in particular, it does not say what, if any, predicates the by pirrase cannot meaningfully modify. If the by pirrase combines with a predicate not of the fonn (y [Q{y} CAUSE p]), the meaning postulate does not apply, so it is unable to predict negative facts like those m (18). tied his necktie combed his hair b u t t o n e d his shirt polished his nails p u t on his t o p h a t

Í

> by . . . J

The scope of such negative facts may be debatable; Dowty himself (p. 229) mentions the case in which John "hammers the metal flat by pounding it with a pipe wrench". The boundary between abstract and concrete predicates is fuzzy and flexible, but the cases in (18) are evidence that there are predicates that are definitely too "concrete" to be modified by by pirrases. I will retimi to this topic in Section 5. One way to build a compositional analysis is to give the abstract predicate a separate argument place for a by phrase predicate (simplified): "AwakenMary": APAe 3ei[Cause(Become(awake(m))(ei))(P(e))] "Keep a promise": λ Ρ A e [ P(e) Λ Promise(P)( Agent(e)) ] But this is hardly plausible considering the cases where the abstract predicate occurs "on its own", without being modified by anything more specific. It would seem, therefore, that one must look farther afield for a compositional analysis preserving the ideas of Bennett and Dowty.

4 The analysis The discussion hi the last two sections has suggested that abstract predicates should not be described (only) as predicates of events but (also) as predicates of predicates of events, that is, as second-order predicates of events, and that when modified by a by phrase, they are predicated of the by phrase predicate. More precisely, there is reason to assume the following hypothesis: Hypothesis If someone 0s by 7ring, then φ says that she does a φ such that ... (for instance, φ is something promised, or her doing -φ causes something), and -φ is π.

The structure of criterion predicates

133

To develop this hypothesis into a viable analysis, I will first show how one can make fonnal sense of it in a version of Discourse Representation Theory. Next, I will show how the problem of composing the representations can be overcome through the notion of unification used in recent DRT. I illustrate various combinations of predicates and show how negative facts can follow from a failure of unification. On the resulting analysis, • Bennett's and Dowty's ideas are rendered in a compositional version • the Anscombe thesis is vindicated: There are two descriptions of one event • the symmetry problem is solved: There is symmetry at event token level but asymmetry at event type level • a prepositional notion of causation is vindicated. 4.1 The desideratum I will assmne that the result of by adjunction denotes a set of events: Ae [...], and that the by phrase predicate is predicated of those events; if the by phrase is by reversing, we have Ae [... reverse(e)...}. Considering a sentence like (19a), I will assmne that the by phrase adjoins at the level of the VP, cf.(19b), and that the Agent relation comes into play at a later stage (cf. Kratzer 1996). (19) (a) (b)

(c)

Neither would give way by reversing. give way by reversing

Ae [. .. reverse(e). . . ]

On top of this VP, three functional heads round off the sentence: • Voice: (e.g.) XxXe [Agent(x)(e)] (a function from objects to sets of events) • Aspect: (e.g.) \P\t3e [P(e) Λ Perfective(e)(í)] (a function from sets of events to sets of times) • Tense: (e.g.) Past(í 0 )(íi) (a time) These three functions will be disregarded in the following. What remains in the skeletal representation of the result of by adjunction Ae [... P(e)... ], where Ρ is the by phrase predicate, is a representation of the modified abstract predicate that involves P , and this requires decomposition. Wien, as in (19a-c), the modified abstract predicate is a criterion predicate, decomposition is especially difficult. Let us begin with a causative predicate, for which we have some experience with decomposition. Consider (20a-c):

134 (20) (a) (b) (c)

Kjell Johan Sœbo She maddened me by dancing. madden me by dancing \e [. . . dance(e) . . . ]

According to an event-based notion of causation which has become customary over the last years (cf. e.g. Pylkkänen 2002), one would expect (20c) to take the more specific fonn of (20d): (20) (d)

\e 3 ei [Become('mad(?'))(ei) Λ dance (e) A Ca.use(ei)(e) ]

But this is a representation of the causative predicate (madden ) which does not involve dance, the by pirrase predicate, and it is difficult to see how the symmetry problem can be overcome on such an analysis. There is, however, an alternative decomposition, more in line with Dowty's work (1976, 1979), where causation is not a relation between events but between propositions (although intensions are notationally disregarded below): (20) (e)

Àe[3ei [Become('mad(î))(ei) Λ dance(e) Λ Cause(Become(mad(¿))(ei))(dance(e)) ]

Here, it is clear that the abstract predicate involves the by pirrase predicate dance occurs twice in the representation. To be sure, it is debatable whether this is the best formulation of causative verb causation, but in any case, a coimterfactual analysis of causation (Lewis 1973) is more natural on the basis of a formulation where, as here, the causing event type enters into the causation relation than on the basis of one where only the causing event token enters into it. Turning to criterion predicates with by adjuncts, a similar pattern emerges: To the extent that a decomposition is feasible, it will involve the by adjunct. Take the predicate give way from (19a). This seems to entail doing something another party insists upon, although it does not follow from objective norms. Let us assume that a decomposition along these lines is theoretically possible (although rather indeterminate). Then give way by reversing seems to be the same plus the condition that reversing is that something. On the assumption that something similar holds of all criterion predicates, we can focus on one case where a decomposition is not merely possible but relatively practical, keep a promise (still, (21c) is a simplification): (21) (a) (b) (c)

She kept a promise by dancing. keep a promise by dancing \e 3 ei [Pro'mjse(dance(pro))(Agent(e))(ei) Λ dance(e)]

In order to prepare the ground for a compositional analysis - how to arrive at representations like (20e) and (21c) - it is useful to cast the representations in a Discourse Representation Structure fonnat ((20b)=(22a), (21b)=(23a)):

The structure of criterion predicates

(22) (a)

135

madden me by dancing

ei (b)

(23) (a)

Ae

dance (e) Become(mad(?'))(ei) Cause(Become ( m a d (?') ) (ei ) ) {dance (e) )

keep a promise by dancing

ei Q (b)

Ae

dance (e) dance C Q Promise (Q (pro) ) (Agent (e) ) (ei )

So far, so good - but the problem is that it is far from obvious how to derive these structures in a compositional maimer. We can go some way towards identifying the contribution of the abstract predicate and that of the adjunct through formulations corresponding one-to-one to the above hypothesis; "if someone by 7ring, then φ says that she does a -φ such that... and -φ is π": ei (22) (c)

Ae

Ρ

P(e) Become(mad(?'))(ei) Cause(Become(mad(?'))(ei))(P(e)) ! Ρ = dance !

ei Ρ Q (23) (c)

Ae

P(e) PÇQ Promise

(Q (pro) ) (Agent (e) ) (ei ) ' Ρ = dance '

This serves to isolate the problem: The sole contribution of the by adjunct seems to consist in the condition Ρ = dance', but if the contribution of the modified predicate is everything but that condition, it is difficult to see how the by phrase can have access to the Ρ event type discourse referent - as long as we maintain traditional wavs of comnosition.

136

Kjell Johan Sœbo

4.2 Composition by unification: Stores and binding conditions There are a variety of ways of composing, . . .

Duke Ellington Recent work in DRT (e.g. Bende-Farkas and Kamp 2001, Kamp 2001) uses unification rather than functional application as a method of composition. So far, this method has mainly been used for the representation of semantic incorporation (see below); I will argue that abstract predicate modification represents another case for which it can make a positive difference. First, it is necessary to describe the novel features in general tenns. A preliminary representation of a node consists of a store and a content. A store consists of triples: A variable, constraints, and a binding condition. Here I will assmne just pairs: A variable and a binding condition, ( , }. A content is a DRS:

When two nodes meet, the unification of store variables of the same type is driven by the binding conditions, and the two content DRSs are then merged. The binding conditions that a store variable may be subject to include: definite (BCdef), indefinite (BC in d e f), and quantificational (BCq). A quantificational store variable must find an indefinite store variable to bind. Bende-Farkas and Kamp (2001) use this to account for Definiteness Effects in semantic incorporation (cf. Bende-Farkas 1999 and Farkas and de Swart 2003); for instance, there be in English comes with a quantificational variable, and if the matching variable from the sister NP is quantificational or definite, unification will fail and the merge will be incoherent. Indefinite store variables, on the other hand, do not need to be bound, although they easily are; if they are not, they eventually enter the content DRS as normal (indefinite) discourse referents. I will use three binding conditions, • Λ for 'classical' abstraction, • indefinite, and • constant as a subsort of BCQ,

The structure of criterion predicates

137

and I will assume that the by phrase introduces a constant predicate variable while the abstract predicate introduces another, indefinite predicate variable. When the two phrases meet, the fonner will bind the latter. If the by pirrase meets a "concrete" predicate not introducing an indefinite predicate variable, the unification fails and the composition terminates. If the abstract predicate does not meet a by pirrase (or a similar modifier), the indefinite predicate variable enters the content DRS as an ordinary discourse referent.

4.3 Examples Let us first look at preliminary representations of two abstract predicates, one causative and one criterial. (Note that these representations abstract away from intensions, and that they represent oversimplifications in other respects as well. Recall that I follow Kratzer 1996 in assuming the Agent relation to come into play at a later stage; more on agentivity below.)

(24) (a)

(b)

(25) (a)

(b)

madden me

λ), ( Ρ, indefinite )

keep a promise

λ), ( Ρ, indefinite }

Next, let us see what a representation of a simple by adjunct might look like. I will assmne that the function of the preposition is purely identificational: Essentially, it takes a predicate and returns a store-content pair introducing a constant predicate variable with the content identifying this as the predicate: (26) (a) (b)

by dancing ( Π , constant. ) J>

138

Kjell Johan Sœbo

When a by phrase like this modifies an abstract predicate such as (24a) or (25a), the variable Π from the fonner binds the variable Ρ from the latter, the latter being substituted for the fonner and entering the universe of the merged content DRS. Below are some illustrations of this, as well as illustrations of cases in which there is no Ρ store variable or no Π store variable is provided.

4.3.1 Simplex cases Wien a manner-neutral causative predicate is modified by a by adjunct, (24c) depicts how the by pirrase predicate identifies the maimer by unification:

(24) c. make me cry ; e , A), I Ρ, indefinite )

unification

by calling {Π, constant

make me ay by calling ei Ρ

{ (e,A>

}

Pie) Bec(cry(?'))(ei) Cause(Bec(cry(?;))(ei))(P(e)) Ρ = Xe

collie)

ei ie, A

call{e) Bec(cry(?'))(ei) Cause(Bec(cry(-¿))(ei))(caZZ(e))

The structure of criterion predicates

139

Note that although the by adjunct is only ascribed an identifieational meaning (and in particular, not a causative meaning), it is fully possible to represent what seems to be the negation of this meaning, as in (27a); unification occurs but the bound variable is claimed to be different from the constant predicate.

(27) (a)

sadden me (but) not by dancing (but... by singing) ei Ρ

P(e) (b)

Bec(sad(?'))(ei) Cause(Bec(sad(i))(ei))(P(e))

Ρ φ \e

dance (e)

Note, also, that when an agent is eventually connected to the modified VP, via the relation Agent(x) (e) (Kratzer 1996), it is the causing event, described by the modifier, that is assigned agentivity; the caused event may well be unintentional. This is as it should be. It is an interesting question whether the empty grammatical subject of the by phrase is always an external argument, essentially an agent; as it stands, the analysis presupposes that it is. As the by phrase is not represented with a PRO subject, a theme trace variable cannot be bound by anything. This predicts, in particular, that there should be no passives in by adjuncts, and passives are indeed very rare; when they cannot be interpreted as covert actives, along the lines of (28b), they seem rather marginal, cf. (29): (28) (a) By being defeated, you have ruined everything. (b) By letting yourselves be defeated, you have ruined everything. (29) ? The mullah lost his honour by being lifted off the floor.

On the other hand, if desired, it would be possible, only more complicated, to supply the representation of the by phrase with a PRO subject, controlled by an agent DP or by a raised theme argument binding a theme trace variable in the modified VP; hi the latter case, PRO could bind a theme trace variable hi the by phrase. Whether this is deshable is primarily an empirical question. As long as subjects seem to be agents, I will assmne that they are. The composition of criterion predicates like (25a) and by adjuncts like (26a) will parallel the composition of causative predicates like (24a) and by adjuncts as shown in (24c) above. So will, hi principle, the composition of criterion predicates and more complex by adjuncts, as shown hi (30) below.

Kjell Johan Sœbo 4.3.2 Complex cases

keep a

promise / Q

; e , A),

P(e) PÇQ Promise(Q(j>To)) (Agent (e)) (/)

I Ρ, indefinite )

unification

by killing Shere

Khan

Q ei ' Π, constant

keep a promise

Π = Ae

by killing Shere /

p

Khan

Qi (pro)) (Agent (e)) (/)

Promise{Qi

(e,A>

}•

PQQi P(e) Q e2 Ρ = Aei

Q

e2

/

tW(ei) Cause(t(s)(e1))(Q(e))

Q(e i ) tW(e2) Cause(t(s)(e2))(Q(e1))

Qi

Pí'Oí7M'se(Qi(p]-o))(Agent(e))(/) Cause(t(s)(e2))(Q(e)) tW(e2) Ae

/ Aei

\

Q e2 Q(ei) tW(e2) Cause(t(s)(e2))(Q(e1))

Ç Qi

/

141

The structure of criterion predicates

(30) illustrates the modification of a criterion predicate through a by adjunct whose predicate is itself complex, here a manner-neutral causative predicate (whose indefinite predicate store variable is entered into the content DRS). The bottom structure says that to keep a promise by killing Shere Khan is to do something causing the death of Shere Khan such that doing something causing the death of Shere Khan entails something that has been promised. To be sine, the exact fonn of the representations is open to modifications. This is particularly true of the decomposition of the concept keep a promise, where modal and temporal parameters, while relevant, are not made explicit. The event of keeping a promise must succeed the event of making a promise, and the indefinite predicate referent Ρ may stand for a predicate in intension. The essential thing is the unification of that referent, made accessible as a store variable, and the predicate Π from the representation of the by phrase. Note that when a manner-neutral causative predicate by phrase modifies another manner-neutral causative predicate, there is an asymmetry between the two: With reference to (7), repeated here as (31a), to save the Israelites by opening the Sea is not to do something causing both the Sea to become open and the Israelites to become safe but to do something causing the Sea to become open such that that doing something causing the Sea to become open causes the Israelites to become safe, - as illustrated in (3 lb): (31) (a) (b) Q

Yahweh saved the Israelites by opening the Sea of Reeds, save the Israelites by opening the Sea:

ei e2

Q(e) Bec(ja/è(î))(ei) Cause(Bec(o/7en(s))(e2))(Q(e))

/

Ae

Q

Bec(open(s))(e2^

e2

Q(e)

Cause(Bec(ja/è(î))(ei)) V

Bee (open (s))(e2) Cause (Bec(op>en(s))(e2))(Q(e))

/

Note, finally, that the modification-by-unification mechanism is recursive; there is no difficulty in representing the appropriate meaning of, say, (32): 3 (32) Mowgli kept a promise by killing Shere Khan by stampeding buffalo through a ravine.

4.3.3 Concrete parasites If composition by unification is to succeed when a predicate is merged with a by adjunct, the predicate must provide an indefinite predicate store variable. Not all 3

That is, there is no theoretical problem; in practice, however, such a representation will easily become very complex.

142

Kjell Johan Sœbo

predicates do. I will argue later, in Section 5, that this is not a sharp, absolute distinction and that predicates can be quite flexible in this regard; but some are simply too concrete or manner-specific to be interpreted as providing an indefinite predicate variable playing a part in their interpretation. These predicates supply the negative facts about by pirrase modification.

(18) ?? Fred ^

tied his necktie combed his hair b u t t o n e d his shirt polished his nails p u t on his t o p h a t

} by . . .

(33) illustrates the failure of composition by unification for the event type spew all over a man and a woman and the by pirrase by getting blind dnmk on seven gins and umpteen pints (inspired by Saturday night and Sunday morning by Alan Sillitoe), a combination which would not be implausible were the by pirrase to convey a causal relation on its own: (33)

spew ail over a man and a woman ({

(e, Λ) I ,

spew...

(e)

^

by getting blind dnmk on seven gins and umpteen pints ', Π, constant )

fail. Unification fails because the constant binding condition for Π, a subsort of Q for Quantifieational, necessitates the binding of a variable with air indefinite binding condition in the store of the sister. Here there is none to be found, or even accommodated. Note the parallel to presupposition failure as failure of anaphora binding; store elements with binding conditions of the Quantificational sort can be viewed as intrasentential presuppositions-as-anaphora. 4.3.4 Lone parasites The by adjunct requires a predicate providing an indefinite predicate variable, but not vice versa: A causative or a criterion predicate can very well occur on its own, without any sort of modifier, because the indefinite predicate store variable is transferred to the content DRS as a normal discourse referent if nothing

The structure of criterion predicates

143

happens. It does not need to be bound; if it is not, it stays indefinite, as in these examples: (34) (a) (35) (a) (36) (a)

He did me a favor. You have done a great deed. The boy insulted me in your bar.

This is not to say that it stays indefinite in a broader context. In isolation, the sentence may be represented with an indefinite predicate discourse referent: e t

eC t

Ρ t < n

P(e)

Agent(e)(yoz/)

great(P) But this Ρ can serve as a source or a target for intersentential unification, so that the final representation of the discourse includes conditions of the fonn Ρ = Xe ... and e = . . . , or equivalently, much as if there were a by adjunct. This seems particularly common with criterion predicates, as shown below: (34) (b)

(35) (b) (36) (b)

"You want to tell me what this is about?" "He did me a favor. I want to say thanks. That's all it is." "It must have been quite a favor" I said. "Do you mind if I ask what he did?" "He showed me a kindness when I was down on my luck." You have saved the world from the evil witch. [ . . . ] You have done a great deed. The boy insulted me in your bar. He told me to shut up.

The establishment of such binding relations is based on pragmatic reasoning and accompanied by discourse relations. In (35b), Ρ succeeds its unifier, and we may speak of abstraction, while in (36b), we may speak of elaboration (Asher and Lascarides 2003: 204-207, Behrens and Fabricius-Hansen 2002).

5 The scope of abstractness Sometimes, a by phrase adjoins to a predicate which does not seem to be an abstract predicate, either a criterion or a manner-neutral causative predicate; typically, then, the by phrase seems to convey a causal relation on its own. Such cases are, of course, a threat to the analysis proposed in the last section. My defense will be to argue that on closer inspection, predicates which do not appear to be abstract really are, at least under the given circumstances; that is, predicates that may not be intrinsically causative or criterial can, under the influence of certain factors, be interpreted as causative or criterial, one of these factors being a merge with a by phrase. There is independent support for this, and at a

144

Kjell Johan Sœbo

general level, there is reason to embrace the idea that criteriality and causativity are not fixed and lexical but flexible and contextual categories. Let us first consider some activity predicates modified by by pirrases. (37) (38) (39) (40)

Snakes move by throwing their bodies into backward-moving waves. They feed by filtering food particles from the water. It swims by flexing its body from side to side. The majority of their people live by fanning.

It is not unreasonable to assume that the predicates move, feed, swim, live are used here in a slightly derived, abstract sense: • to move or swim in the relevant sense is to propel oneself (through water) • to feed in the relevant sense is to obtain food m the genetically encoded way • to live m the relevant sense is to sustain life; to satisfy one's "basic needs" Thus interpreted, the apparently intransitive activity (or even stative) verbs are m actual fact transitive and causative accomplishments (though temporally, they remain atelic due to iterativity or to the fact that a change of state is not brought about but prevented (cf. Dowty 1979: 124)). As such, they introduce indefinite predicate variables for the causing activity. The verb feed features an additional criterion that the indefinite predicate must satisfy, as suggested by the formulation "in the genetically encoded way". Let us next consider some achievement predicates modified by by phrases. Much the same story can be told about them: (41 ) They find prey by detecting minute vibrations from a distance away. (42) . . . . a project to reach India not by following the coastline of Africa ... but rather by plunging boldly into the unknown Western ocean. (43) He was forced to forfeit the medal he had won by cheating. (44) He claimed that he had escaped by crossing the Congo.

In fact, a relevant story has already been told about such cases: To account for "progressive achievements", Rothstein (2004: 45-50, 136-139) proposes that achievement predicates have a double nature: They can be coerced, or shifted, to activities culminating in achievements, that is, to accomplishments: S H I F T ( V P p u n c t u a i ) : Ae.(BECOME)(e) Ae.3eieo [e = s (eiUeo)A(DO(a)(ei)A(BECOME(VP))(eo)ACul(e) = e2]

The structure of criterion predicates

145

The dummy predicate DO corresponds to the indefinite predicate variable Ρ m the representations of abstract predicates in the last section. To find prey, to escape, to reach India, or to win the medal in the broader, accomplishment sense is to do something culminating in finding prey, reaching India, winning the medal, or escaping in the narrower, achievement sense. If, following Dowty (1979: 183), we take the presence of a causal event to be the most salient distinction between achievements and accomplishments, we can represent, e.g., escape in the shifted sense, escape+, as: ei ,

Ì

( Ρ, indefinite } J

P(e) ay«7/?e(Agent(e))(ei) C ause (escape ( Agent ( e ))( e ι ))( Ρ ( e ) )

Rothstein's shifting operation is supposed to be triggered by progressive aspect; by adjunction now emerges as another factor triggering accomplishment readings of apparent achievements. It should be accentuated that assuming criteriality and causativity to be elastic notions is in no way a costly concession. On the contrary, it is what we should expect. It would be surprising if the class of abstract predicates were closed and solely lexically determined. Elasticity is welcome because it reflects the basically relative (functional, pragmatic) nature of abstractness. In this light, it is not surprising that the limits to the by locution are fuzzy. They are, we may say, just as fuzzy as they ought to be.

6 Conclusions It seems, then, that the key to a better understanding of the by locution is a better understanding of the things it modifies, namely, abstract predicates, and vice versa. The need to overcome the "symmetry problem" forces a reassessment of criterion predicates and manner-neutral causatives as predications not merely over events but over sets of events. Conversely, once it is appreciated that predicates with a by adjunct involve a second, indetenninate predicate, it becomes clearer what the contribution of the by adjunct should consist in: The determination of that second predicate. This does not proceed on its own, however. A lexical decomposition where the indetenninate "second predicate" is visible remains useless as long as this second predicate is inaccessible for determination through the by adjunct. Some innovative method of composition is called for, and in fact available: Recent work in DRT supplements (or supplants) β reduction by unification. Constituent representations are bipartite, and discourse referents figuring in the content section are entered as variables in the store section along with (constraints and) so-called Binding Conditions that drive the unification. The by phrase can thus

146

Kjell Johan Sœbo

be translated as a structure where the embedded predicate is represented by a store variable with a condition ensuring its unification with the store variable for the "secondpredicate" 4 This can be carried out in recent DRT; it can presmnably be modelled in another framework as well. However, by pirrase or other intrasentential modification of abstract predicates is part of a larger picture encompassing intersentential fonns of unification between (in)determinate predicate referents. Here, DRT will make a positive difference, inasmuch as even the representations of full root sentences are in this framework only preliminary, open to linkings and bindings driven by more or less "pragmatic" presuppositions. Ahead lies a better understanding of a discourse relation like elaboration. There are negative facts about by phrase modification, and they can be accounted for; but the limits to the locution are not that sharp. This reflects the vagueness and context sensitivity of the boundary between the abstract and the concrete. A predicate appearing concrete in one perspective may appear abstract in another. It may be assmned that an instrumental adjunct can itself effect such a shift in perspective. The Anscombe thesis is vindicated: Just one action is indeed perfonned if one signals by waving one's arm. This thesis seemed to rim afoul of the symmetry problem as long as predicates like signai were not taken apart; once there is asymmetry at event type level, however, symmetry at event token level ceases to be a problem. An appealing intuition is thus proven viable.

References Anscombe, Gertrude (1957): Intention. Oxford: Blackwell. Asher, Nicholas and Alex Lascarides (2003): Logics of Conversation. Cambridge: Cambridge University Press. Behrens, Bergljot and Cathrine Fabricius-Hansen (2002): Connectives in contrast: A discourse semantic sUidy of Elaboration. In Information StracUire in a Cross-Linguistic Perspective, Hilde Hasselgârd et al. (eds.), 45-61. Amsterdam: Rodopi. Bende-Farkas, Agnes (1999): Incorporation as Unification. Proceedings of the 12th Amsterdam Colloquium, Paul Dekker (ed.). Amsterdam: ILLC. Bende-Farkas, Agnes and Hans Kamp (2001): Indefinites and Binding: From Specificity to Incorporation. ESSLLI, Helsinki. (http://www.helsinki.fi/esslli/courses/readers/K29.pdf) Bennett, Jonathan (1994): The "Namely" Analysis of the by Locution. Linguistics and Philosophy 17: 29-51.

4

It may be debatable in how strict a sense this scheme adheres to compositionality; if it is ultimately judged to transcend one's preferred compositionality notion, it is at least in good company with recent work on incorporation and related matters arguing the need for moderately innovative methods of composition; cf. e.g. Farkas and de Swart 2003.

The structure of criterion predicates

147

Dowty, David (1976): Montague Grammar and the Lexical Decomposition of Causative Verbs. In Montague Grammar, Barbara Partee (ed.), 201-245. New York: Academic Press. Dowty, David (1979): Word Meaning and Montague Grammar. Dordrecht: Reidel. Farkas, Donka and Henriette de Swart (2003): The Semantics of Incorporation: From Argument Structure to Discourse Transparency. Stanford: CSLI Publications. Kamp, Hans (2001 ): The Importance of Presupposition. In Linguistic Forni and its Computation, Rohrer, Christian, Antje Rossdeutscher and Hans Kamp (eds.), 207-254. Stanford: CSLI Publications. Kearns, Kate (2003): Durative Achievements and Individual-Level Predicates on Events. Linguistics and Philosophy 26: 595-635. Kratzer, Angelika (1996): Severing the External Argument from its Verb. In Phrase Structure and the Lexicon, Rooryck, Johan and Laurie Zaring (eds.), 109-137. Dordrecht: Kluwer. Krifka, Manfred (1998): The Origins of Telicity. In Events and Grammar, Rothstein, Susan (ed.), 197-235. Dordrecht: Kluwer. Lewis, David (1973): Causation. Journal of Philosophy 70: 556-567. Pylkkänen, Liina (2002): Introducing Arguments. MIT Dissertation. Rothstein, Susan (2004): Structuring Events: A Study in the Semantics of Aspect. Oxford: Blackwell. Ryle, Gilbert (1949): The Concept of Mind. Harmondsworth: Peregrine Books.

Markus Egg (Groningen)

Reference to embedded eventualities 1 Introduction In the interfaces between morphology or syntax and semantics, semantic contributions of the involved syntactic or morphological constituents are opaque, i.e., processes of semantic construction may only handle them as a whole, they are blind to the inner structure of these constituents. This principle is violated in many cases where the semantic contribution Cs of a syntactic or morphological constituent C includes embedded eventualities, i.e., eventualities that do not show up as arguments of Cs. (The tenn 'eventuality' refers to states of affairs of all kinds; following Davidson (1967), verbs and their projections have an additional eventuality argument in their semantics.) Reference to such embedded eventualities by a syntactic or morphological sister constituent of C should be impossible - but it is not. Consider e.g. the preferred reading of (1), 'person who usually dances beautifully'. Here the adjective qualifies an eventuality of dancing. This eventuality is contributed in the semantics of the verb stem, however, it is embedded in the semantics of dancer (a set of individuals) and thus does not surface as an eventuality argument in the semantic contribution of (1) as a whole (Larson 1998): (1)

beautiful dancer

(1) has an additional reading 'beautiful person who usually dances'. Here the adjective pertains semantically to the individual argument of the semantics of dancer, hence, semantic construction for this reading is trivial. Another well-known challenge for semantic construction, viz., the so-called institutive readings of aga/n-sentences can likewise be subsumed under this phenomenon. E.g., the restitutive reading of (2) can be paraphrased as 'Max manipulated the window, and, as a result, the window was open, and it had been open before': (2)

Max opened the window again.

Here the idea is that change-of-state verbs like open lexically specify the aftermath or result state of eventualities in their extension. The transitive verb open,

150

Markus Egg

for instance, specifies the result state as a state of the object NP referent (in (2), the window) being open (Dowty 1979). However, this result state is not an argument of the semantics of the verb as a whole. Again presupposes that a specific kind of eventuality occurred before. In the restitutive reading, it must be of the same kind as the result state. I.e., the restitutive reading of (2) follows immediately if we assmne that the modifier again may refer to the embedded result state of the verb. In addition, there is another, repetitive reading of (2), in which the adverbial refers to the mam eventuality argument of the verb. Consequently, the preceding eventuality which is presupposed in this reading is one of Max manipulating the window, which results in the window being open (i.e., the whole causation is a repetition of a previous one). In analogy to the restitutive reading of (2), (3) can be interpreted as 'Max manipulated the window and, as a result, it was open for two hours': (3)

Max opened the window for two hours.

A literal interpretation is barred here due to the aspectual incompatibility of the durative adverbial for two hours (which selects for unbounded expressions) and the expression it modifies (roughly, the rest of the sentence, which is bounded). 1 In contrast, the aftermath predicate is unbounded, which allows the application of the adverbial semantics. Finally, reference to embedded eventualities may also show up at the morphology-semantics interface, in particular, for the 'bracketing paradox' noted by Liideling (2001) for nominalisations of German separable prefix verbs like (4) and analysed by Müller (2003). (4)

Los- gerenn -e start iter_nonii rim iter_noni2 'iteration of events of starting to rim'

In the morphology, the order of affixation is circmnfix Ge... e before prefix los-. In the opposite order, the first part of the circmnfix could not be placed correctly between prefix and stem (Müller 2003). In its semantics, the circmnfix contributes an iterative operator and maps the semantics of its base onto the property of being an iteration of a kind of eventuality as determined by the base. I.e., Gerenne refers to an iteration of running eventualities. If we now assmne that the prefix contributes an inchoative operator that expresses the beginning of an eventuality, prefixation of ios- to Gerenne should return an expression denoting the beginning of an iteration of running eventualities. However, the 1

Properties of eventualities are bounded (formally, in the extension of a property of properties of eventualities B D ) iff they do not apply to proper parts of elements in their own extension ( Krifka 1992): (i) VP.BD(P) Oe^^P(e')

Reference to embedded eventualities

151

prefix refers not to the main eventuality of the iteration, it refers to the embedded eventualities (that make up the iteration), which returns an expression that refers to repeated beginnings of running eventualities. The examples (l)-(4) show that reference to embedded eventualities leads to ambiguity if it is optional. Thus, (1) and (2) have additional readings in which the modifier refers to the main eventuality argument of the modified expression. In contrast, the modifier in (3) and the prefix los- in (4) cannot refer to the main eventuality argument of its modified expression or its base, respectively, i.e., there is no ambiguity. The structure of the paper is the following. After rephrasing the phenomenon m a fonnal framework (the λ-cal cuius) to highlight the common ground between the presented examples in section 2 , 1 will devote section 3 to a discussion of previous analyses to the phenomena presented above. Then section 4 provides a brief introduction into imderspecification formalisms, on which my analysis is based. The proposed analysis is sketched in section 5, finally, I conclude with a brief outlook in section 6.

2 Formalisation of the data In this section, I will offer a fonnal account of the phenomenon. The goal of this section is also to bring out the common ground between the presented data. 2.1 Agentive nouns Reconsider the first example (1). Its preferred reading (approximately, 'person who usually dances beautifully' 2 ) can be derived in two steps. The first step consists in dividing the semantics of agentive nouns like dancer into the stem and the affix meaning, where the stem semantics emerges as an argument of the fimctor which is the semantic contribution of the affix: (5)

'person who usually' affix meaning

...

'dance s stem meaning

The crucial point in this division is that the embedded eventuality argument from the semantics of dancer is no longer embedded in the stem semantics. The second step then is to pertain the adjective semantically to the verb stem only. As a consequence, it can refer to the relevant eventuality argument in a straightforward fashion, because this argument is open in the verb stem semantics, hence, accessible to a modifier like beautiful. Since the affix meaning

2

I do not attempt to reconstruct the semantics of these agentive nomináis fully, since for the line of argumentation in the present paper the exact spellout of the affix semantics is not relevant. All that matters is that it comprises an operator that has the verb stem semantics in its scope.

152

Markus Egg

'person who usually X-es' is applied to the meaning of the stem only after modification by the adjective, the adjective ends up in the scope of the affix -er. The following fonnal spellout of this basic intuition in the λ-calculus is based on the reconstruction of the semantics of the agentive affix -er in (6a) as a function from the stem semantics Ρ (a relation between η [η G Ν] individual arguments and one eventuality argument) to a set of individuals. (This is considerably simplified in that issues of argument structure and binding are ignored for the purposes of this paper.) Elements of this set of individuals are identical to an individual χ such that when χ participates in an eventuality e (this is expressed by the relation in), then e is usually a P-eventuality where χ is the agent. Here ' j ' is shorthand for a sequence of zero or more individual arguments of the verb. The definition of the generic quantifier GEN in (6b) adopts one of the versions of the quantifier that are discussed in Krifka et al. (1995): (6)

(a) (b)

XPXz.GEN[e,x](xineAz = x,3_v.P(x,v)(e)) GEN[e,x](Ä(x)(e),C(x)(e) ) iffÁ(x)(e) usually entails C{x){e)

According to (6a), the semantics of dancer is reconstructed in (7a) as the set of people such that when they are participating in an eventuality, it is usually an eventuality of them dancing. Here the semantic contribution of the verb stem is underlined, in this expression, there is a free eventuality argument. 3 If the semantics of the adjective pertains to only this underlined part, the result is the representation (7b) for the preferred reading of ( 1 ). Here the adjective semantics refers to the embedded eventualities of dancing, thus, the expression refers to people who are usually dancing beautifully. Its other reading is represented by (7c), which refers to beautiful people who are usually dancing: (7)

(a)

Xy.GEN[e,x](x in e Ay = x, dance'(x)(e) )

(b)

Xy.GEN[e,x](x in eAy = χ, dance'(χ)(e) Λbeautiful'(e)

(c)

Xy.GEN[e,x](x in e Λ y = χ, dance' (χ) {e ) ) Λ beautiful' (y)

)

2.2 Change-of-state verbs For change-of-state verbs, I follow the decomposition approach as introduced in the Generative Semantics tradition and worked out in detail in Dowty (1979). I.e., the meaning of complex verbs is broken down into simpler meanings that are linked by suitable operators. Consider e.g. the semantic decomposition for the transitive verb open: 3

Here and in the following, I will use V\-equality (equivalence of Xx.P(.x) a n d P , if .v does not occur free in the function P ) to keep semantic representations readable. E.g., the underlined expression dance'(.Y) in (7a) is equivalent to Xf.dance'(.r)(f ), which highlights the open eventuality argument. In addition, α-equality will be used (representations are semantically equivalent if they are only distinguished by consistent renaming of bound variables ).

Reference to embedded eventualities

(8)

153

λνλτλ?. CAUSE (x, BECOME(Xe'.be-open'(y ) (