Event Arguments: Foundations and Applications 9783110913798, 9783484305014

Table of contents :
Introduction
Section I: Events - states - causation
The event structure of CAUSE and BECOME
Stativity, supervenience, and sentential subjects
Do states have Davidsonian arguments? Some empirical considerations
Ser and estar: The syntax of stage level and individual level predicates in Spanish
Sentence connection as quantificational structure
Section II: Event nominals
Gerund types, the present participle and patterns of derivation
Referential arguments of nouns and verbs
Section III: Events in composition
Building resultatives
Reconciling “possessor” datives and “beneficiary” datives - Towards a unified voice account of dative binding in German
Event arguments and modal verbs
Section IV: Measuring events
Types of degrees and types of event structures
Too poor to mention: Subminimal events and negative polarity items
Semantic properties of split topicalization in German
Author index
Subject index

Citation preview

Linguistische Arbeiten

501

Herausgegeben von Hans Altmann, Peter Blumenthal, Klaus von Heusinger, Ingo Plag, Beatrice Primus und Richard Wiese

Event Arguments: Foundations and Applications Edited by Claudia Maienborn and Angelika Wöllstein

Max Niemeyer Verlag Tübingen 2005

Bibliografische Information der Deutschen Bibliothek Die Deutsche Bibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.ddb.de abrufbar. ISBN 3-484-30501 -0

ISSN 0344-6727

© Max Niemeyer Verlag, Tübingen 2005 Ein Unternehmen der K. G. Saur Verlag GmbH, München http://www. niemeyer. de Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Printed in Germany. Gedruckt auf alterungsbeständigem Papier. Druck: Laupp & Göbel GmbH, Nehren Einband: Industriebuchbinderei Nadele, Nehren

Contents

Claudia Maienborn and Angelika Introduction

Wöllstein 1

Section I: Events - states - causation Manfred

Bierwisch

The event structure of

CAUSE

and

BECOME

Stefan Engelberg Stativity, supervenience, and sentential subjects Anita

11

45

Mittwoch

Do states have Davidsonian arguments? Some empirical considerations

69

Kay-Eduardo Gonzalez- Vilbazo and Eva-Maria Remberger Ser and estar: The syntax of stage level and individual level predicates in Spanish

89

Horst Lohnstein Sentence connection as quantificational structure

113

Section II: Event nominals Artemis Alexiadou Gerund types, the present participle and patterns of derivation Ingrid

139

Kaufmann

Referential arguments of nouns and verbs

153

Section III: Events in composition Angelika Kratzer Building resultatives

177

Daniel Hole Reconciling "possessor" datives and "beneficiary" datives - Towards a unified voice account of dative binding in German

213

Werner Abraham Event arguments and modal verbs

243

vi

Contents

Section IV: Measuring events Patrick Caudal and David Nicolas Types of degrees and types of event structures Regine

Eckardt

Too poor to mention: Subminimal events and negative polarity items Kimiko

277

301

Nakanishi

Semantic properties of split topicalization in German

331

Author index

357

Subject index

361

Claudia Maienborn

and Angelika

Wöllstein

Introduction

Since entering the linguistic stage in the late sixties, Davidsonian event arguments have taken on an important role in linguistic theorizing. The central claim of Donald Davidson's seminal (1967) work "The logical form of action sentences" is that events are spatiotemporal things, i.e., concrete particulars with a location in space and time. This enrichment of the underlying ontology has proven to be of great benefit in explaining numerous combinatorial and inferential properties of natural language expressions. Among the many remarkable advances achieved within the Davidsonian paradigm since then figure most prominently the progress made in the theoretical description of verb semantics, including tense and aspect, and the break through in analyzing adverbial modification. Numerous monographs and collections attest to the extraordinary fertility of the Davidsonian program; see, e.g., Rothstein (1998), Tenny & Pustejovsky (2000), Higginbotham, Pianesi & Varzi (2000), Lang, Maienborn & Fabricius-Hansen (2003), Austin, Engelberg & Rauh (2004) to mention just a few more recent collections. In the course of the evolution of the Davidsonian paradigm, two moves have turned out to be particularly influential in terms of expanding and giving new direction to this overall approach. These are, first, the "Neo-Davidsonian turn" introduced by Higginbotham (1985, 2000) and Parsons (1990, 2000), and, secondly, Kratzer's (1995) merger of event semantics with the stage-level/individual-level distinction. The neo-Davidsonian approach has lately developed into kind of a standard for event semantics. It is basically characterized by two largely independent assumptions. The first assumption concerns the arity of verbal predicates. While Davidson introduced event arguments as an additional argument of (some) verbs, neo-Davidsonian accounts take the event argument of a verbal predicate to be its only argument. The relation between events and their participants is accounted for by the use of thematic roles. The second neoDavidsonian assumption concerns the distribution of event arguments. They are considered to be much more widespread than originally envisaged by Davidson. Hence, neoDavidsonian approaches typically assume that it is not only (action) verbs that introduce Davidsonian event arguments, but also adjectives, nouns, and prepositions. Thus, nowadays event arguments are widely seen as a trademark for predicates in general.1 The second milestone in the development of the Davidsonian program is Kratzer's (1995) event semantic treatment of the so-called stage-level/individual-level distinction, which goes back to Carlson (1977) and, as a precursor, Milsark (1974, 1977). Stage-level predicates (SLPs) express - roughly speaking - temporary or accidental properties, whereas

A note on terminology: Bach (1986) coined the term "eventuality" for the broader notion of events, which includes, besides events proper, i.e., accomplishments and achievements in Vendler's (1967) terms, also processes and states. Other labels for event arguments in the broad sense are, e.g., "spatiotemporal location" (Kratzer 1995), "Davidsonian argument" (Chierchia 1995), or "E-position" (Higginbotham 1985).

2

Claudia Maienborn and Angelika Wöllstein

individual-level predicates (ILPs) express (more or less) permanent or inherent properties. 2 On Kratzer's (1995) account, the SLP/ILP-distinction basically boils down to the presence or absence of an extra event argument. Stage-level predicates are taken to have an additional event argument, while individual-level predicates lack such an extra argument. This difference in argument structure is then exploited syntactically by the assumption of different subject positions for SLPs and ILPs; see Diesing (1992). Since then interest has been directed towards the role of event arguments at the syntax/semantics interface and the impact they have on syntax proper in terms of, e.g., event phrases. All in all, Davidsonian event arguments have become a very familiar "all-purpose" linguistic instrument over the past decades, and recent years have seen a continual extension of possible applications far beyond the initial focus on verb semantics and adverbials. These developments are accompanied by a newly found interest in the linguistic and ontological foundation of events. To the extent that more attention is paid to less typical events than the classical 'Jones buttering a toast' or 'Brutus stabbing Caesar', which always come to the Davidsonian semanticist's mind first, there is a growing awareness of the vagueness and incongruities lurking behind the notion of events and its use in linguistic theorizing. A particularly controversial case in point is the status of states. The question of whether state expressions can be given a Davidsonian treatment analogous to process and event expressions (in the narrow sense) is still open for debate; see Maienborn (2005) and the commentaries to this target article for some of the pros and cons. The present volume grew out of a workshop "Event arguments in syntax, semantics and discourse" that the editors organized in February 26-28, 2003, in Munich (as part of the annual meeting of the German association for linguistics, DGfS), and in which we invited contributions geared towards drawing an interim balance of the use of and motivation for event arguments in linguistic theory. The articles presented here offer proposals towards this end from different empirical and theoretical perspectives. The leading question shared by the majority of the articles could be phrased in the following way. How do lexical semantics, syntax, and pragmatics conspire to project event

structure?

Discussing a wide range of linguistic phenomena (mostly pertaining to English, German and Romance) the articles (a) supply fresh evidence for the virtually ubiquitous presence of event arguments in linguistic structure; (b) they provide new, event-based, solutions as superior alternatives to already existing analyses; and/or (c) they shed new light on the nature of event arguments and the way these are handled by the linguistic machinery.

2

See, e.g., Higginbotham & Ramchand (1997), Jäger (2001) for overviews of the linguistic phenomena that have been associated with the stage-level/individual-level distinction.

Introduction

3

The volume is organized into four sections: Events - states - causation; Event nominals; Events in composition; Measuring events. Section I: Events - states - causation addresses mainly foundational issues concerning the nature of events and states, how they relate to causation, and how they show up in the linguistic structure. Manfred Bierwisch discusses the anchoring and accessibility of event arguments in semantic structure. He compares the different ways in which event arguments are structurally anchored in Davidsonian, neo-Davidsonian, and Reichenbachian approaches and presents arguments in favor of the latter variant. Bierwisch then goes on to argue that, no matter how complex a verb's internal event structure might be, only the highest event argument is made accessible for reference, quantification, modification, etc. This means, in particular, that inchoative and causative verbs will never project a target state into their argument structure. Apparent counterevidence as provided by durational adverbials, which obviously serve to specify the duration of an inchoative's target state, is accounted for by assuming that the operator BECOME is of an elusive nature. That is, target state modification of inchoatives relies on the improper absence of BECOME. Stefan Engelberg draws attention to one of the classes of verbs that do not fit easily into the Davidsonian picture, namely dispositional verbs such as German helfen (help), gefährden (endanger), erleichtern (facilitate). These verbs may have an eventive or a Stative reading depending on whether the subject is nominal or sentential. Trying to account for their readings within the Davidsonian program turns out to be challenging in several respects and provides new insights into the different nature of events and states. Engelberg advocates the philosophical concept of supervenience as a useful device to account for the evaluative rather than causal dependency of the effect state expressed by these verbs. The proper analysis of state expressions is taken up again by Anita Mittwoch. She examines the arguments raised by Katz (2000, 2003) and Maienborn against extending the Davidsonian approach to (all) state expressions and rejects most of them, thereby corroborating the general neo-Davidsonian approach. On this view, states, rather than being different things, are merely somewhat poor examples of event(ualitie)s. Engelberg's and Mittwoch 's considerations concerning the ontological nature of states are supplemented by an article on the syntax of copular state expressions. Kay-Eduardo Gonzälez-Vilbazo and Eva-Maria Remberger present a minimalist account of the Spanish copula forms ser and estar, which figure as lexical exponents of the stagelevel/individual-level distinction. Ser and estar are analyzed as syntactic default strategies (last resort) that are introduced into the derivation at different functional layers: tense (T°) and predication (Pr°). Motivation for this comes from current semantic analyses of the ser/estar alternation for which the authors strive for a more transparent syntactic correlation. Finally, causality is taken up again by Horst Lohnstein, who proposes a uniform account of the semantics of clause-connectives (while, i f , when, because etc.) in terms of an invariant quantificational structure whose components are subject to parametrization. Lohnstein shows how different interpretive effects as, e.g., the temporal vs. adversative reading of German während (while/whereas) can be derived in this framework.

4

Claudia Maienborn and Angelika Wöllstein

Section II: Event nominals presents a syntactic and a lexicalist approach towards an analysis of the argument structure of deverbal nominalizations. Artemis Alexiadou discusses nominal and verbal gerunds in English within the framework of Distributed Morphology suggesting that the different properties associated with these forms follow from different attachment sites of a nominal -ing affix. Whereas nominal gerunds result from attaching -ing directly to the verbal root, verbal gerunds result from combining -ing with AspectP. On Alexiadou's perspective, argument structure is derived syntactically via an event structure which in turn is introduced by a special type of functional layer in the syntax. Ingrid Kaufmann, instead, pursues a lexicalist approach according to which argument structure is basically determined at the level of lexical-semantic structure. Kaufmann's analysis is based on a corpus study of German nominalized infinitives showing that nominalized infinitives display two different patterns of argument realization whose distribution is determined by genuine semantic and pragmatic conditions. In order to account for these findings Kaufmann proposes an "ontological" solution according to which the two different patterns of nominalized infinitives differ in the way how the verb's event argument is referentially anchored. Section III: Events in composition focuses on the role of event arguments at the syntax/semantics interface. The studies aim at uncovering the combinatorial mechanisms that lead to the formation of complex event descriptions. Angelika Kratzer develops a novel analysis of German and English adjectival resultatives along the lines of serial verb constructions. In expressions like to drink my teapot dry the adjective is taken to combine with an empty CAUSE-affix. The resulting causing event is identified with the event expressed by the verb via the combinatorial operation of Event Identification. Kratzer succeeds in showing (a) how several syntactic and semantic properties of resultative constructions can be derived from her analysis and (b) that the direct object in a resultative construction is not a true argument of the verb but always starts out from within the adjectival phrase. Working within Kratzer's framework, Daniel Hole proposes an analysis of possessor and beneficiary datives in German that extends Kratzer's Event Identification into a more general combinatorial operation, called Variable Identification. This mechanism serves to augment an event description by an additional thematic argument that will be bound by an already existing argument. Thus, operations like Event Identification and Hole's dativeinduced Variable Identification can be seen as a specific implementation of the neoDavidsonian program of building up complex event descriptions from a maximally coherent conjunction of a set of smaller predications. Werner Abraham is concerned with the deontic and epistemic readings of modal verbs in the Germanic languages. Putting special emphasis on their Aktionsart-sensitivity, Abraham suggests to account for the polyfunctionality of modal verbs by assuming a control analysis for the deontic reading and a raising analysis for the epistemic reading. This syntactic analysis is correlated with a semantic analysis according to which epistemic modal verbs inherit both the theta properties and the event characteristics of the embedded full verbs, whereas deontic modal verbs project event and thematic arguments of their own.

Introduction

5

Finally, Section IV: Measuring events provides a particularly clear picture of the many ways in which event arguments can be involved in measuring expressions. Patrick Caudal and David Nicolas explore the relationship between degree structure and event structure by an analysis of various degree adverbials. Differences in distribution and interpretation are accounted for by assuming different types of degree scales. Degree modifiers like partially, completely act as modifiers on quantity scales, whereas extremely, perfectly and the like act as modifiers on intensity scales. The proposal rests on the assumption that most verbal predicates, including Stative predicates, can receive a degree argument, either for inherent lexical reasons, or by virtue of their structural context. On this basis, Caudal and Nicolas introduce a new - and broader - characterization of (a)telicity in terms of a mapping between degrees and events. Regine Eckardt draws attention to negative polarity items such as bat an eyelash, lift a finger, which serve to single out events of a particularly insignificant size. Eckardt develops an event-based variant of the pragmatic approach to NPI licensing proposed by Krifka (1995), showing that her event-based variant has several advantages compared to Krifka's event-free original account. On Eckardt's analysis, the respective NPIs turn out to be a special kind of adverbial modifier denoting functions from event predicates to event predicates. Weak NPIs map event predicates to the minimal events in their extension whereas strong NPIs yield so-called subminimal events, i.e., events that are even below the extension of an event predicate. Besides accounting for the different licensing contexts for weak and strong negative polarity items, Eckardt's approach also offers new insights into the ontology of events in terms of mereological structure. Finally, Kimiko Nakanishi examines measure phrases that are separated from their host NP in German split topicalizations as opposed to measure phrases that are adjacent to their host NP. Nakanishi proposes to account for their different semantic properties in terms of different domains of measurement. Whereas the non-split case involves the measurement of individuals in the nominal domain, measure phrases in split topicalizations are analyzed as a means of measuring events in the verbal domain. Several semantic restrictions on split measure phrases such as the incompatibility with single-occurrence events, the incompatibility with individual-level predicates, and the unavailability of collective readings follow from monotonicity constraints applying to the verbal domain. In their entirety, the articles collected here offer a representative overview of the questions, assumptions and strategies that are presently being pursued in the further development of the Davidsonian program. Our aim is that they will offer further impulses to work in this area. We wish to thank all the authors for their enthusiasm and cooperation during all stages traversed along the way from the DGfS conference to the publication of this volume. We are particularly grateful to those who agreed to review one or more of the submitted papers: Artemis Alexiadou, Manfred Bierwisch, Miriam Butt, Patrick Caudal, Regine Eckardt, Stefan Engelberg, Werner Frey, Kay-Eduardo Gonzalez-Vilbazo, Daniel Hole, Gerhard Jäger, Graham Katz, Ingrid Kaufmann, Manfred Krifka, Ewald Lang, Jürgen Lenerz, Jörg Meibauer, Anita Mittwoch, Kimiko Nakanishi, David Nicolas, Susan Olsen, Luis Paris, Christopher Pinon, Beatrice Primus, Irene Rapp, Eva-Maria Remberger, Barbara Stiebels, Thomas Ede Zimmermann.

6

Claudia Maienborn and Angelika

Wöllstein

Fabienne Fritzsche deserves special thanks for her competent help in formatting the manuscript. Finally, we would like to thank the team at Niemeyer, especially Brigitta Zeller and Wolfgang Herbst, for their helpfulness and expertise that have contributed to the successful completion of this volume.

References

Austin, Jennifer R., Stefan Engelberg & Gisa Rauh (eds.) (2004): Adverbials. The interplay between meaning, context, and syntactic structure. Amsterdam, Philadelphia: Benjamins. Bach, Emmon (1986): "The Algebra of Events." - In: Linguistics and Philosophy 9, 1-16. Carlson, Gregory (1977).· Reference to kinds in English. - Doctoral dissertation, University of California. Chierchia, Gennaro (1995): "Individual predicates as inherent generics." - In: Gregory N. Carlson & Francis J. Pelletier (eds.): The generic book, 176-223. Chicago: University of Chicago 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. Diesing, Molly (1992): Indefinites. - Cambridge Mass.: MIT Press. Higginbotham, James (1985): "On Semantics." - In: Linguistic Inquiry 16, 547-593. (2000): "On Events in Linguistic Semantics." - In: J. Higginbotham, F. Pianesi & A. Varzi (eds.): Speaking of Events, 49-79. New York, Oxford: Oxford University Press. Higginbotham, James & Gillian Ramchand (1997): "The Stage-Level/Individual-Level Distinction and the Mapping Hypothesis." - In: Oxford University Working Papers in Linguistics, Philology & Phonetics 2, 53-83. Higginbotham, James, Fabio Pianesi & Achille C. Varzi (eds.) (2000): Speaking of Events. New York, Oxford: Oxford University Press. Jäger, Gerhard (2001): "Topic-Comment Structure and the Contrast between Stage Level and Individual Level Predicates." - In: Journal of Semantics 18, 83-126. Katz, Graham (2000): "Anti Neo-Davidsonianism: Against a Davidsonian Semantics for State Sentences." - In: C. Tenny & J. Pustejovsky (eds.): Events as Grammatical Objects, 393-416. Stanford, CA: CSLI Publications. (2003): "Event arguments, adverb selection, and the Stative Adverb Gap." - In: E. Lang, C. Maienborn & C. Fabricius-Hansen (eds.): Modifying Adjuncts, 455-474. Berlin, New York: Mouton de Gruyter. Kratzer, Angelika (1995): "Stage-level and individual-level predicates as inherent generics." - In: G. N. Carlson & F. J. Pelletier (eds.): The Generic Book, 125-175. Chicago: Chicago University Press. Krifka, Manfred (1995): "The semantics and pragmatics of polarity items." - In: Linguistic Analysis 25, 209-257. Lang, Ewald, Claudia Maienbom, & Cathrine Fabricius-Hansen (eds.) (2003): Modifying Adjuncts. Berlin, New York: Mouton de Gruyter. Maienbom, Claudia (2005): "On the limits of the Davidsonian approach. The case of copula sentences." - To appear in: Theoretical Linguistics 31/3. Milsark, Gary L. (1974): Existential Sentences in English. - Doctoral dissertation, MIT. - (1977): "Toward an Explanation of Certain Peculiarities of the Existential Construction in English." - Linguistic Analysis 3, 1-29.

Introduction

7

Parsons, Terence (1990): Events in the Semantics of English. A Study in Subatomic Semantics. Cambridge/Mass.: MIT Press. - (2000): "Underlying States and Time Travel." In: J. Higginbotham, F. Pianesi & A.Varzi (eds.): Speaking of Events, 81-93. New York, Oxford: Oxford University Press. Rothstein, Susan (ed.) (1998): Events and Grammar. Dordrecht: Kluwer. Tenny, Carol & James Pustejovsky (eds.) (2000): Events as Grammatical Objects. Stanford, CA: CSLI Publications. Vendler, Zeno (1967): Linguistics in Philosophy. - Ithaca, New York: Cornell University Press.

Section I:

Events - states - causation

Manfred Bierwisch The event structure of

CAUSE

and

BECOME

Adopting the widely held view that verbs are predicates of events or states and refer to situations or eventualities, the present paper explores consequences of this assumption for the structure of inchoative and causative verbs. Inchoatives like aufaachen (wake up) are supposed to be semantically characterized by the operator BECOME, causatives like wecken (wake up) by the operator CAUSE. While BECOME specifies the transition from a source state to a target state, CAUSE identifies the causal connection between a cause and its effect, the latter often specified by an event of the inchoative sort. The paper argues that, with respect to its syntactically based compositional interpretation, a verb refers to one and only one state or event, irrespective of the complex structure of causatives and inchoatives involving causation, cause, effect, transition, source-, and target-state. Technically, the event-reference of a verb is based on the highest position in its argument structure. This position absorbs (or unifies with) the qualification represented by (extensional) modifiers, as in I woke him up at nine by a phone call. Comparing the status of the fact variable proposed by Reichenbach with the event variable introduced by Davidson, Reichenbach's referential operator is argued to provide the more appropriate analysis for compositionally complex verbs. Further problems created by the "Neo-Davidsonian" variant of argument structure are argued to provide additional motivation for the view adopted here. The analysis proposed for verbs carries over to event nouns, as shown by the parallel structure of the change of the schedule last week and the schedule changed last week. Apparent counterexamples are related to durational adverbials like for two weeks, which are usually supposed to be incompatible with proper events. But they are compatible with inchoatives in cases like the schedule changed for two weeks. In these cases, however, the adverbial clearly specifies the duration of the target state, rather than the duration of the event. The paper shows how this effect derives from the proposed analysis, if independently motivated assumptions about the status of BECOME are added.

1.

The problem

The fairly trivial observation that natural language expressions are about situations, or more specifically states, events, and processes, raises non-trivial questions concerning the reference to situations and its proper analysis. Current proposals for dealing with these questions derive in one w a y or the other from Reichenbach (1947) or Davidson (1967). Both accounts are based on the assumption that a proposition ρ is to be enriched by an additional variable e which establishes the reference to a situation which is characterized by p. The technical details o f the two proposals are different, but for a wide range o f problems their consequences are the same and their representations can be translated into each other, as w e will see. Both approaches are primarily concerned with the logical form and semantic interpretation o f linguistic expressions, paying only marginal attention to the question o f how the relevant representations are built up syntactically, and which role in particular the situation

12

Manfred Bierwisch

or event variables play within the morpho-syntactic structure of linguistic expressions. Against this background, I will be concerned in this paper with the following questions: A Β

Which of the semantic event variables are syntactically accessible, and how? Which effects of event variables can be assigned to their syntactic and semantic selection restrictions?

Question A presupposes that semantic variables are accessible for syntactic specification, assuming that this is in fact the function of theta roles a head assigns to its syntactic complements, and it queries which event variables may realize a function of that sort in which way. Question Β presupposes that it is by means of theta roles that a head realizes its semantic restrictions and morpho-syntactic or categorial requirements, called s-selection and c-selection, respectively, and it raises the non-trivial question of whether event variables, which are not normally specified by syntactic complements, can be associated with selection restrictions, and what their effects may be. These relevant issues will be pursued with respect to the semantic predicates CAUSE and BECOME and their combinations appearing in lexical items like close, kill, change etc. I will adopt the basic assumptions about these elements developed in Dowty (1979), extended by proposals discussed a.o. in Bierwisch (2002, 2003). The problems to be faced are illustrated by cases like these: (1)

a. b. c.

He woke up for a while, but then he slept quietly for hours. Mach am Abend bitte ein paar Minuten das Fenster auf. In the evening, please open the window for a few minutes. Yesterday, he came quite a while to my office.

According to standard and in fact well motivated assumptions, events denoted by verbs like open, come, wake up etc. can be temporally located by adverbials such as yesterday, then, in the evening, etc, but they cannot be modified by durational adverbials like for hours, (for) quite a wile, a few minutes, which combine freely with processes and states like sleep, rest, or wait. In (la) however, durational adverbials combine with the event wake up as well as the state sleep. In (lb), moreover, the same event denoted by aufmachen (open) seems to be modified by the temporal adverbial in the evening and the durative adverbial a few minutes. In a similar vein, come is modified by both yesterday and quite a while in (lc). On closer inspection, one has to note, however, that the temporal adverbs locate the event in question, while the durational adverbs qualify the resulting state, rather than the event. Similarly, the durative adverbial for a while modifying wake up in (la) specifies only the state of being awake. This illustrates the problem to be pursued here, viz. the question of how events and states and their properties are to be systematically accounted for. Some remarks about background assumptions needed to deal with these problems seem to be in point.

13

The event structure of CAUSE and BECOME

2.

Background assumptions

Linguistic expressions relate a Phonetic Form PF to the representation of its meaning, the linguistic aspect of which is called Logical Form or Semantic Form SF.1 Hence a linguistic expression is a pair , where PF determines its pronunciation and SF its conceptual interpretation. Each expression is furthermore categorized by syntactic and morphological features Cat, classifying e.g. dreams as alternatively verb, present, and third person singular, or as noun and plural. Finally, a linguistic expression is characterized by its Argument Structure AS, which consists of a hierarchically structured sequence of argument positions or theta roles, determining the properties of constituents the expression may or must combine with. In particular, the theta roles in AS impose conditions called s-selection and cselection, determining the semantic and morpho-syntactic properties of expressions saturating the theta role in question. 2 Thus the theta role to be saturated by the subject of the Verb dreams requires semantically a human (or personal) entity and syntactically a nominative singular DP. These two types of constraints are naturally determined by the semantic aspect of the argument position and by morpho-syntactic features associated with it. More technically, a theta role Θ, is a pair < λχ,, F,>, where λχ, abstracts over a variable in SF, and F, is a (possibly empty) set of features to be matched by the features in Cat of a constituent saturating 0j. The s-selection of 0 j can now be understood as an effect of the predicates applying to Xj in SF, while the c-selection is due to features that follow from grammatical rules or principles, or are specified by lexical idiosyncrasy. A preliminary illustration of the assumptions sketched so far is given in (2), the lexical entry for the German temporal preposition nach (after), which differs grammatically from the directional preposition nach (towards) by the categorization [ - Directional] : (2)

/ nach /

PF

[-V,-N,-Dir]

λχ λy [+Obl]

Cat

AS

[Τy>Τχ ]

SF

GF As indicated in (2), Cat and AS constitute jointly what one might call the Grammatical Form GF of a linguistic expression, because Cat and AS together determine essentially its grammatically controlled combinatorial potential. On the other hand, AS and SF together

1

I need not go here into details related to the terminological decision. While LF, as used in Chomsky (1981) and subsequent work, is primarily concerned with syntactically determined aspects of meaning, SF is concerned also with the (grammatically relevant) internal structure of lexical items determining their contribution to the meaning (or conceptual interpretation) of linguistic expressions. For some discussion of these matters and the overlap in orientation between LF and SF see Bierwisch (1997). As I am concerned here with issues that clearly relate to word-internal conditions of semantic representation, I will take SF to provide the relevant representational format.

2

The terminology - semantic or s-selection and categorial or c-selection - is due to Chomsky (1986), where selectional restrictions were not formally associated with theta roles, however.

14

Manfred

Bierwisch

can be considered as the Extended Semantic Form ESF, according to which e.g. a preposition like nach is a two-place relation. 3 This will turn out to be crucial for the semantic combinations an expression may enter into. Within the SF of (2), Τ is a functor that assigns a time interval to its argument, and the two place predicate > represents an ordering-relation over the set of time intervals. Hence χ and y are variables over individuals susceptible to ordering in time. Hence the s-selection associated with both argument positions of nach requires entities to which a time interval can be assigned. The c-selection determined by the Object Position is expressed by the feature [+Oblique] requiring a Dative-DP. 4 The other argument of nach - and of prepositions in general - , sometimes called the external or designated argument, does not specify features of c-selection, a point to which we will return. The conditions of s- and c-selection just mentioned must be met e.g. by the object of nach in a phrase like nach der Wahl (after the election). Suppose for the sake of illustration that something like (3) abbreviates the representation of the object-DP in question, where [DEF e [ ELECTION e ] ] identifies a definite individual:5 (3)

/der Wahl/

[+N,-V,+Obl ]

[ DEFe[ELECTIONe] ]

Merging (2) and (3) yields a PP with the representation indicated in (4), where the SF of (3) replaces the variable χ in (2) as an effect of lambda-conversion, triggered by the combination of (2) and (3) through functional application: (4)

/ nach der Wahl /

[-V,-Ν,-Dir]

λγ[ Τ y > Τ [ DEF e [

ELECTION

e ]] ]

(4) illustrates in a rather simplified form the result of combining a head with its complement. In addition to this type of combination called complementation, we need an account of the operation that merges a head with an adjunct, as e.g. in Besuch nach der Wahl (visit after the election), where nach der Wahl is a modifier, not a complement of the head Besuch. Abbreviating the representation of Besuch by (5), we get something like (6) as the result of merging a head with an adjunct: (5)

/Besuch/

[+N,-V,...]

λζ [VISITζ]

3

Technically, ESF is an expression in a so-called lambda-categorial language. Assuming that for principled reasons the SF o f major syntactic constituents is to be construed as an expression o f type t, i.e. as a proposition, ESF b e c o m e s an η-place predicate with A S defining its arity, i.e. the number and type o f its arguments. See Bierwisch (1997, 2003) for further discussion.

4

The feature [+Obl] is in fact predictable, being the default case for objects o f prepositions in German. Hence it would not have to be specified in the entry (2). The principles and conditions controlling such regularities will largely be ignored in the present context, except where event positions are involved.

5

This is, o f course, an oversimplification in various respects. First, e must be construed as referring to a definite eventuality o f the sort discussed in Bach (1986), a point to which w e will return. Second, the definiteness operator DEF is actually a short-hand for a number o f assumptions that cannot be discussed here. It must, however, provide a referential binding for the argument position o f the Noun Wahl, turning it into a definite description, as will be discussed shortly.

The event structure of CAUSE and (6)

15

BECOME

/ Besuch nach der Wahl /

[ +N, - V,... ] λ ζ [ [ VISIT Ζ ] & [ Τ Ζ >

Τ [ DEF e [ ELECTION e ] ] ] ]

What (6) is supposed to account for is the observation that head and modifier are semantically combined by logical conjunction, and more specifically that the condition abbreviated by VISIT specifies an event that is additionally subject to the temporal location expressed by nach der Wahl. To this effect, the argument position λy of the adjunct (4) is absorbed by (or unified with) the argument position λζ of the head (5). As both operators abstract over the same sort of variables, the absorbing theta role does not violate the s-selection of the absorbed Role, which furthermore does not impose c-selectional constraints that could be violated. Absorption of a theta role must furthermore be assumed to have two consequences: First, the SF of the adjunct is added to that of the head by logical conjunction &. 6 Second, the variable bound by the absorbed operator is substituted by the variable of the absorbing operator. In the present case, λy is absorbed by λζ and y is substituted by z. This account of (extensional) modification follows essentially the proposal made in Higginbotham (1985). It must be generalized in non-trivial ways if e.g. intensional modification as in der angebliche Besuch (the alleged visit) is to be included, since an alleged visit is not something that is a visit and an alleged event. In Bierwisch (2003), I have argued that in head-adjunct-combinations the head characteristically absorbs a theta role, as opposed to head-complement-combinations, where the head discharges a theta role. We will return to these matters below. To sum up the framework sketched so far, we have lexical entries as sketched in (2), the argument structure of which specifies their s- and c-selectional properties as illustrated above. On the basis of these entries, syntactically complex expressions are created by the operation Merge as illustrated in (4) and (6) for complementation and adjunction, respectively. Merge combines two (basic or complex) expressions X and Y into a complex expression Z. One o f Z ' s constituents is its head, determining its categorization. A provisional formulation of the properties of Merge is given in (7), presupposing that complex expressions have the same basic organization as lexical items. (7)

Merge (X, Y) ==> Z, where a. PF of Ζ is the linear combination of PF of X and Y,7 b. Cat of Ζ is projected from Cat of X iff X is the head of Z, and either c. X discharges the lowest (i.e. next available) position of its AS to Y by functional application with subsequent lambda-conversion within SF, or d. Y discharges the lowest position in its AS to X by lambda-absorption, followed by logical conjunction of the SF of X and Y.

6

For principled reasons, discussed e.g. in Wunderlich (2000) and Bierwisch (2002), the conjunction & is asymmetrical, at least in the sense that one conjunct is closer to the functor than the other. Whether and which semantic consequences are connected to this asymmetry need not concern us at the moment.

7

T w o qualifications are to be made at this point. First, I will ignore here morphological processes with non-linear aspects of combination. Second, the linear ordering of head and complement or adjunct is subject to complex conditions of various sorts that must be left aside here.

16

Manfred Bierwisch

(7c) and (7d) determine the argument structure and the Semantic Form of Ζ (i.e. the ESF as noted above) under complementation and adjunction, respectively, where (7d) covers only extensional modification and is thus in need of further elaboration. As a consequence, conditions of s- and c-selection are imposed either according to (7c) by the head on the complement, or according to (7d) by the adjunct on the head. 8

3.

S o m e aspects of event arguments

Within this framework, the status of argument positions providing event reference is to be made explicit in two steps. First, as noted initially, an event variable, originally proposed as "fact variable" in Reichenbach (1947) and reinvented, in a somewhat different guise, in Davidson (1967), is assumed to explicitly represent states and events as entities in SF. The formal ontology of the values to be assigned to this variable is developed in Bach (1986). 9 The second step takes up the notion of a referential theta role, proposed in Williams (1981) and elaborated in Higginbotham (1985), and others. Originally, Williams considered this type of role as characteristic for nouns, creating the basis for reference and quantification as in this man, every book, some problems etc. 10 In Higginbotham (1985) and Bierwisch (1988) it was also taken as the basis for extensional modification, as sketched in (6) and (7d). With these prerequisites, it is a natural move to assume that verbs refer to events in roughly the same way in which nouns are assumed to refer to individuals to be assumed for nouns. The point is illustrated by the parallel between (8) and (9) compared to (10): (8)

a. b. c.

8

9

10

Sie ändern den Fahrplan. They change the schedule. Sie ändern häufig/oft den Fahrplan. They often change the schedule. Sie ändern am Montag den Fahrplan. On Monday, they change the schedule.

It might be added that Merge is deliberately based on the operation Merge as introduced e.g. in Chomsky (1995), with the following amendments: (i) Merge as defined in (7) does not project the full set of features of the head, but only those in Cat, (ii) it does not only merge the phonetic and syntactic information but also the information in ESF, thereby realizing the selection restrictions. For the time being, I will ignore the much debated difference between events and states, both covered by what Bach called "eventualities". It must be noted that the notion of Referential Role is crucially different from that of agent, theme, goal etc., although Williams (1981) is not quite clear in this respect. While agent, theme, etc. are supposed to relate somehow to the conceptual content of an argument position, referentiality has to do exclusively with the way in which variables relate to the domain of interpretation. In fact, agent, theme, patient, etc. all can become referential roles, as e.g. in murderer, proposal, employee, which are referential by means of the agent, theme, and patient role, respectively.

The event structure of CAUSE and BECOME

(9)

a. b. c.

(10) a. b. c.

17

Der Fahrplan ändert sich. The schedule changes. Der Fahrplan ändert sich häufig/oft. The schedule changes often/frequently. Der Fahrplan ändert sich am Montag. On Monday, the schedule changes. Die Änderung des Fahrplans The change of the schedule Die häufige/*oft Änderung des Fahrplans The frequent change of the schedule Die Änderung des Fahrplans am Montag The change of the schedule on Monday

Besides the specification of reference by means of tense and complementizer or determiner, the event-reference is parallel for the causative verb, the inchoative verb, and the noun, both in German ändern, sich ändern, and Änderung and in English verb and noun change. Also, frequency and temporal modifiers apply to verbal and nominal heads in the same way. The fact that oft and often are restricted to verbal heads, while häufig can modify verbs as well as nouns, is due to c-selection by the adjunct, with oft imposing something like [ +V ]." Furthermore, the event reference of verbs can enter standard anaphoric relations and may be picked up by appropriate pronouns, as shown by the italicized elements in (11): (11) a. b. c.

Sie ändern häufig den Fahrplan. Das macht viel Ärger. Sie ändern häufig den Fahrplan, was viel Ärger macht. They change the schedule frequently, that/which is very irritating.

Again, this is essentially parallel to the referential character of nouns, with the anaphoric relations based on their referential argument, as shown in (12): (12) a. b. c.

Er kritisiert die häufige Änderung des Fahrplans. Sie macht viel Ärger. Er kritisiert die häufige Änderung des Fahrplans, die viel Ärger macht. He criticizes the frequent change of the schedule, which/it is disappointing.

As already noted, c-selection imposed by morpho-syntactic features of adjuncts can restrict them to verbal heads - as in oft (often), heute (today), jetzt (now) - or to nominal heads - as in häufig (frequently), heutig (today's), or jetzig (present). The familiar semantic restrictions, on the other hand, based on s-selection and depending on the content of SF, carry over from verbs to nouns, preventing (proper) events from durational modifiers like for

'1 A different, but comparable condition on c-selection restricts heute (today), gestern (yesterday), damals (then), bald (soon) and others to verbal heads, as opposed to heutig, gestrig, damalig, baldig modifying nominal heads. A closely related distinction is realized more systematically (but not without exceptions) by the English suffix -ly . For further discussion of this point see Bierwisch (2003).

18

Manfred

Bierwisch

hours, and states from delimitations like quickly, or within a few minutes, as indicated in (13) and (14). 12 (13) a. b.

(14) a. b.

Das The Die The

Haus wurde {wiederholt/*stundenlang/ziemlich rasch} zerstört. house was {repeatedly/*for hours/rather quickly} destroyed, {wiederholte/*stundenlange/ziemlich rasche} Zerstörung des Hauses. {repeated/rather quick} destruction of the house {*for hours}.

Das The Die The

Haus wurde {gestern/stundenlang/*ziemlich rasch} beobachtet. house was observed {yesterday/for hours/*rather quickly}. {gestrige/stundenlange/*ziemlich rasche} Beobachtung des Hauses. {*rather quick} observation of the house {yesterday/for hours}.

Event variables cannot only enter anaphoric relations, they are also subject to quantification, with frequency adverbials like always, often, occasionally, seldom, etc. acting as quantifiers over eventualities. Thus the Logical or Semantic Form of (15a) should be something like (15b), or slightly more formally (15c): (15) a. b. c.

The schedule changes frequently. There are many e such that e is a change of schedule. For many e [ the schedule changes (e) ]

Quantification applies not only to events but just as well to states, if instances are separable, as in (16), where states are individuated by relevant occasions: (16) a. b.

Die Leitung ist immer besetzt. The line is always busy, Peter wiegt selten zu viel. Peter seldom weighs too much.

Participating in quantification, event variables can furthermore be involved in scope relations. Thus the preferred reading of (17) assures that mail delivery occurrs regularly on Monday, not on other days. In other words, on Monday qualifies the regular delivery, i.e. it has scope over regularly. The preferred reading of (18), on the other hand, claims that on Monday the delivery of mail is regular, i.e. regular qualifies the delivery on Monday, and has, in this sense, scope over on Monday. (17)

12

Die Post wird regelmäßig am Montag zugestellt. Mail is delivered regularly on Monday.

It should be noted that acceptability judgements can be obscured by a coerced, event-like interpretation o f observe, such that e.g. they will observe the house in three hours is construed as they will start the observation of the house in three hours. But coercion o f this sort confirms, rather than spoils the tenet that s-selection is based on semantic conditions. I will return to these matters in detail in section 7 and 8.

The event structure

(18)

of CAUSE and BECOME

19

Die Post wird am Montag regelmäßig zugestellt. On Monday, mail is delivered regularly.

Whether and under which conditions the preferred interpretation can be replaced by other options need not concern us here. In part it is a matter of stress and focus-assignment, which must be left aside. The point to be made is merely that event variables cannot only be quantified - as in (15) and (16) - , but participate in standard relations of regular variables. 13

4.

Implementing event reference

As already noted, event reference has been introduced into standard semantic (or logical) representations in different ways. The most direct proposal is due to Davidson (1967), who suggested that the main predicate of an action sentence is to be extended by an additional argument, which refers to the event characterized by the sentence in question. More technically, a proposition of the general form (19) should in fact be analyzed as (20), where P' is an n+1 place predicate that relates Ρ and its arguments to the event e. (19)

P(x,,...,xn)

(20)

3e[P'(e,x,,...,xn)]

This proposal is illustrated in (21b), where the transitive verb butter of (21a) is analyzed as a three-place relation between e and the arguments of the verb. Past tense, provisionally indicated by T(e) < T(u), ordering e temporally before the utterance time T(u), and adverbials like in the kitchen are now treated as predications of e, conjoined to the main proposilion.14 (21) a. b.

Fred buttered the toast in the kitchen 3e [ buttering (e, Fred, the toast) & T(e) < T(u) & in the kitchen (e) ]

Twenty years earlier, Reichenbach (1947) had already proposed a more general way to introduce event variables. Instead of adding an argument to the major predicate, Reichenbach defined an event function [ ρ ]* which turns a proposition ρ into a property of events. Substituting in this function the proposition (19) for p, one gets (22), which corresponds to (19) very much like Davidson's (20) corresponds to the initial (19). 13

Scope variation of the sort illustrated in (17) and (18) does not carry over to nominalization, as shown by (i) as opposed to (ii). This is due to conditions of DP-syntax that are not to be pursued here. (i)

the regular delivery o f mail on Monday

(ii) * the delivery of mail on Monday regular(ly) 14

The treatment o f tense as a conjunct on a par with adjuncts must be modified for reasons to which we return. For the time being it simply indicates the specification imposed on e.

20

Manfred Bierwisch

(22)

3e[P(xls...,xn)]*(e)

In these terms, the analysis of (21a) comes out as (23), with tense and adverbial modification represented again by conjoined propositions: (23)

3e [ [ buttering (Fred, the toast) ]*(e) & T(e) < T(u) & in the kitchen (e) ]

Reichenbach's proposal is more general than Davidson's, as it introduces an event variable by a general event function 15 , rather than by extending the arity of particular (classes of) predicates. 16 It could thus apply to any proposition, including those specifying e.g. locative or temporal properties. This requires an empirically restricted occurrence of the eventfunction, ultimately converging with the specification needed for event arguments of the Davidsonian style. With this proviso, and ignoring certain consequences of the different theoretical contexts of the two proposals, Davidson's and Reichenbach's event variables are intended to account for roughly the same range of phenomena. 17 In particular, both Reichenbach and Davidson represent adverbial modification by conjoined predications of the event argument, such that e.g. Fred buttered the toast follows from (21a) by the rules of standard logic. A rather different way to treat the event variable has been proposed a. o. by Parsons (1990). This so-called neo-Davidsonian theory replaces (19) by (24), turning Ρ into a oneplace predicate P" of events to which the arguments of Ρ are then related by thematic relations Rj!, (24)

3e [ P"(e) & R,(e, x,) & ... & R n (e, x n ) ]

Thematic relations are taken from the usual set of theta roles like agent, theme, source, goal, etc. Under this proposal, the analysis of (21a) would come out as something like (25): (25)

3e [ buttering (e) & Agent(e, Fred) & Patient (e, the toast) & T(e) < T(u) & Location (e, the kitchen) ]

The move from (19) to (24) is - in spite of the deceptive terminology - a radical defection from Davidson's original intention. Separating the core predicate from its original arguments has far-reaching and fatal consequences. I will briefly sketch three of them.

15

16

17

Reichenbach explicitly uses fact function and event function synonymously. The distinction between facts on the one hand and events and states on the other made in Vendler (1967) and subsequent discussions corresponds more (but not exactly) to the distinction Reichenbach makes between objective or situational fact functions and propositional fact functions. Davidson originally assumed event arguments for verbs of change and action. Later on, various extensions have been discussed by various authors, including e-arguments not only for state verbs but also for certain types of adjectives and heads of locative PPs. I will return to this matter below. A hint to different notational variants appearing in the literature might be useful. In essentially the sense of Reichenbach's event function, Kamp & Reyle (1993) use the colon to associate a proposition ρ with an event e, Wunderlich (2000) uses curly brackets, and Bierwisch (1988) an operator inst. Thus [p]*(e), e:p, {p}e, and e iNSTp all specify an event e instantiating a proposition p.

The event structure

of CAUSE and BECOME

21

First, as shown by (25), arguments and adjuncts of a verb cannot differ with respect to their semantic effect: Both are conjuncts added in the same way to the event predication. 18 Now, a major point in Davidson's treatment of events and adverbials was to provide a systematic account for the inference from e.g. (26a) to (26b) by standard conjunction reduction: (26) a. Fred met Eve in Paris. b. Fred met Eve. c. * Fred met in Paris. d. * Fred met. According to the neo-Davidsonian analysis, (26c) and even (26d) should be derivable by conjunction reduction in the same way, obviously a wrong conclusion without any empirical justification. Notice that the deviance of (26c) and (26d) is not merely a matter of the syntactic surface, violating conditions of c-selection, but indicates rather a semantic defi19

ciency. The second point, directly related to this problem, concerns the fact that the number and type of arguments a predicate requires belong to its essential, intrinsic properties. The event expressed by the verb give, for instance, requires necessarily what is usually called an agent, a theme, and a recipient; it cannot get along with, say, an experiencer and a goal or just a theme. Similarly think needs an experiencer and a theme, while sleep requires an experiencer, but excludes a theme. This is not a matter of arbitrary incidences, but systematically determined by the respective event predicates. That requires highly intricate sets of postulates, determining not only the required, but also the excluded thematic relations. Such postulates, which have never been considered by neo-Davidsonians in an even remotely adequate way, do nothing but supply information that has artificially been stripped away from the core predicates - an arbitrariness that becomes particularly obvious if one takes into account the internal structure of complex predicates of the sort to be looked at below. Third, representations of the neo-Davidsonian style are in conflict with requirements of standard logic in a much wider range of the cases than those illustrated in (26). They yield inappropriate results also in lots of other cases, notably with respect to negation. Thus according to (24), the representation of (27a) - ignoring tense - would be (27b), which is equivalent to (27c). 18

The basic distinction between arguments and adjuncts is not obviated by the fact that there are similarities or even borderline cases. Thus in Paris is a locative argument selected by stay in (i), it is a free locative adjunct in (ii), and something in between, often called argument-adjunct, optionally selected by the noun stay in (iii). (i)

He stayed in Paris

(ii)

He visited me in Paris

(iii) his stay in Paris For some discussion o f these similarities and borderline cases see Bierwisch (1988, 2003). The present problem is in no way affected by these phenomena. 19

It might be noted that these considerations apply also to the condition T(e) < T(u) indicating past tense in (25). This is one o f the reasons requiring a different treatment of tense, as mentioned in fn.13.

22

Manfred

(27) a. b. c.

Bierwisch

He doesn't sell it. - 3 e [ sell (e) & Agent (e, he) & Theme (e, it) ] Ve [ -i sell (e) ν - , Agent (e, he) ν - , Theme (e, it)]

According to normal understanding, (27a) is true if and only if there is no instance of his selling it, whatever he and it are apt to refer to. The three options by which (27a) could be falsified according to (27b) or (27c) are simply besides the point: There is no way to understand what it would mean that someone referred to by he is not the agent of the selling event e or something referred to by it is not the theme of that event, even if one ignores problems related to quantification over events.20 Notice that this is different for something like he doesn Ί sell it today, where the negation - in line with the original Davidsonian approach - can appropriately apply to the event of his selling it and to the temporal location of that event. These and a number of further points concerning the controversial nature of separated thematic roles strongly argue against the neo-Davidsonian approach. This leaves us with two possibilities to incorporate event reference into the notational system sketched in section 2. (28a) illustrates the Reichenbach-version of the verb sleep (using Kamp's ":" rather than Reichenbach's "[ ]*"), while (28b) follows Davidson's proposal, extending the property SLEEP into a relation between an individual and an event: (28) a. b.

/ sleep / [ +V, -Ν ] λχ λε [ e : [ SLEEP Χ ] ] / sleep / [ +V, -Ν ] λχ λε [ [ SLEEP' χ ] e ]

Assuming that eventualities like individuals are elements of type e, the one-place predicate SLEEP in (28a) is of type (e, t), taking χ to build up a proposition of type t, and the colon : is formally an operator of type (t,{e, t)), turning a proposition into a predicate of events. The two-place predicate SLEEP' in (28b), on the other hand, is of type (e,(e, t)), turning two individuals into a proposition. In both versions, both variables are bound by argument positions, providing the subject position and the event reference, respectively. The difference between (28a) and (28b) is in one respect more than merely a notational variant, however. As the domain of eventualities includes events, processes, and states, the variable e is subject to a sortal choice, depending in one way or the other on the predicate that takes e as its argument. In (28b) this choice is directly determined by the predicate SLEEP', while in (28a), it must somehow be proliferated from the property SLEEP to the argument of the event operator ":". I will return to this issue below. It is worth noting that corresponding to the verb sleep, we have the entry (29) for the event noun sleep, which differs merely by its categorization: (29)

20

/ sleep / [ -V, +N ] λχ λβ [ e : [ SLEEP χ ] ]

It must be emphasized that what is at issue are the propositions that he and it are agent and theme of e, not the identity of the individuals referred to. The identity of the individuals could be focused and negated, as in HE doesn't sell it (but his BROTHER). - This is barely possible for the object N P in (27), however, for independent reasons, preventing contrastive stress on it.

The event structure of

CAUSE

and

23

BECOME

This difference has important consequences beyond the categorization as such, because argument positions differ systematically for nouns and verbs with respect to their syntactic properties. Thus, argument positions of nouns are generally optional, except for the referential role, which happens to be the event position in (29). Hence λχ must be saturated by the subject in (28), but can be left unspecified in (29), as in the sleep last night, etc. In (29), as in lexical entries in general, the c-selectional properties associated with the argument positions (such as case requirements) are largely predictable by the categorization. Thus λχ requires nominative for the verb in (28), but (possessive) genitive for the noun in (29). See e.g. Bierwisch (1997) and Wunderlich (2000) for further discussion. Given the assumptions about SF sketched in section 2, the entry (28a) would now support representations like (30b) and (31b), again with obvious simplifications in various respects: (30) a. b.

(31) a. b.

Fred slept in the kitchen 3 e [ [ T e < T u ] [ e : [ SLEEP F R E D ] &

[ e IN [ D E F y [ KITCHEN y ] ] ] ] ]

Fred slept for an hour in the kitchen 3e

[ [ T e < T u ] [ e : [ SLEEP F R E D ]

&

[ EXTENT e •

1 HOUR ]

&

[ LOC e IN [ D E F y [ K I T C H E N y ] ] ] ] ]

It must be added that (31b) is at best a first approximation, as it does not account for the relative scope of temporal and locative adverbials in relation to their syntactic position.

5.

T h e e v e n t structure o f inchoativity a n d c a u s a t i v i t y

The verb sleep refers to a particular sort of state. The same type of state provides the source of the change referred to by the verb awake (and wake up) and also the target state of the inverse change referred to by fall asleep. The transitive variant of wake up furthermore exemplifies the possibility to add an agent identifying the source of the change denoted by the intransitive verb, a pattern characteristic for so-called ergative verbs like break, close, change, etc. Following familiar assumptions, deriving from McCawley (1973), Dowty (1979) and related work, grammatically relevant semantic relations within and between causative and inchoative verbs based on the state characterized by SLEEP can be represented as follows, with [ ACT y ] specifying the event which brings about the relevant change of state: 21

21

Actually, the proposition [ ACT y ] used here and in the sequel is a shorthand in various respects. First, ACT must be construed as a predicate subsuming all sorts of appropriate activities by which the effect in question can be brought about. Thus ACT comes close to a variable ranging over activity predicates. For a more detailed discussion of cause, effect, and causation see section 7 below. Second, ACT is treated as a predicate applying to an individual that provides the argument position o f the agent, as exemplified in (i):

24

Manfred Bierwisch

(32) a. b. c.

/ sleep / [ +V ] λχ λ ε [ e : [ SLEEP χ ] ] / awake / [ +V ] λχ λβ [ e : [ BECOME - , [ SLEEP χ ] ] ] / awake / [ +V ] λχ λ γ λε [ e : [ [ A C T y] [ C A U S E [ B E C O M E

[ SLEEP

χ

] ] ] ] ]

The transitive verb awake and its intransitive variant should, of course, be based on (roughly) the same lexical item. (32a) and (32b) can in fact be collapsed into one entry as shown in (33), where heavy parentheses include optional parts, the subscripts indicating that they must simultaneously be present or absent: (33)

/ awake / [ +V ] λ χ ( α λ γ ) λε [ e : ( Α [

[ ACT

y]

[CAUSE) [ BECOME - . [ SLEEP

χ

] ] ] ] ]

Thus, according to (33), if an agent is present, it is realized as the grammatical subject, otherwise the argument of SLEEP becomes the subject. In German, the items integrated in (33) require separate entries shown in (34c) and (34d), which cannot be collapsed, even though they are etymologically related. German furthermore provides a lexical entry denoting the inverse event of (33), as shown in (34b). 22 (34) a. b. c. d.

/ / / /

schlaf- / ein + schlaf- / auf + wach- / (auf+) weck- /

[ +V [ +V [ +V [ +V

] ] ] ]

λχ λχ λχ λχ

λε λε λε Xy

[ε: [ S L E E P Χ ] ] [ε: [ BECOME [ SLEEP χ ]]] [e: [ B E C O M E - , [ S L E E P χ ] ] ] λβ [e: [ [ ACT y] [ CAUSE [ BECOME - , [ SLEEP x]]]]]

It might be added, that 8rgative verbs like (33) are represented in German by cases like brechen(break), heilen(hea\), schmelzen(melt) and others. However, the dominating pattern relating inchoative and causativc constructions of the same verb in German is reflexivization of the type (sich) öffnen (open), (sich) drehen (turn), (sich) biegen (bend), (sich) ändern (change) etc. (Cf. sie ändern den Fahrplan vs. der Fahrplan ändert sich in (8) and (9) above). A lexical entry of ändern (change) that would account for this aspect is sketched in (35), where heavy parentheses again indicate optionality: 23

22

23

(i) Paul woke me up (ii) A sudden noise woke me up As shown by (ii), however, the subject position of a causative verb can also be assigned to an expression referring to an event rather than the relevant actor. Now, mutual substitution of actor and event is a rather general phenomenon. It therefore needs a systematic account, which cannot be pursued here any further. The differences between (33) and (34) are in fact typical phenomena of lexicalization, exploiting general principles of lexical representation in idiosyncratic ways. This includes the incidental "overload" by the almost synonymous entries awake and wake up. A similar overload appears in German with aufwachen and erwachen being largely synonymous. It is worth noting on this background that the lack of a straight causative counterpart for einschlafen (fall asleep) is not a mere idiosyncrasy: the verb einschläfern (lull asleep), which would fill this position morphologically, has the highly specialized interpretation of narcotize, obviously due to the fact that falling asleep is internally triggered and cannot directly be caused by an external agent. As a side-remark it might be mentioned that the predicate D I F F E R E N T is an abbreviation, to be defined provisionally as follows, where { ρ } indicates that ρ is presupposed, as discussed below:

The event structure of CAUSE and BECOME

(35)

25

/ änder- / [ +V ] λχ λγ λε [ e : [ ( [ ACT y] [ CAUSE )[ BECOME [ DIFFERENT χ ] ] ] ] ]

In contrast to ergative verbs like (33), de-causativization is simpler in (35): No position from the argument structure is deleted, optionality applies only to the causative component in SF, leaving a spurious position Xy, as a consequence of which the object position is realized by a reflexive pronoun. 24 Thus although the reflexive construction looks superficially more complex than the un-ergative use of a causative verb, the lexical information it requires is surprisingly simple. Turning now to the event structure based on lexical items with the internal make up illustrated in (32) - (35), we notice first that inchoative verbs referring to an event e, involve at least three eventualities: a source state Sj and a target state sk, such that the event e, is to be defined as the transition from Sj to s k . Thus for einschlafen, Sj and sk are the states of being awake and being asleep, respectively. Generally, the properties of the source state are defined by those of the target state simply by negation. Hence if c, is just the transition from Sj to s k , the information needed for the SF of an inchoative verb is only the specification of BECOME and the proposition ρ specifying the target state.25 Second, causative verbs referring to an eventuality en involve at least a cause em and an effect ei, such that en consists in the causation of ei by e m . The effect ei might be a process, as in the truck moved the trailer steadily, or a state as in the squad kept the rope straight, but in the majority of cases it is an event, as in Max opened the bottle, Eve woke the kids up, Macy's changed the schedule, etc. In case of an event causation, the event's source and target state are to be distinguished. Hence causation of an event involves (at least) five eventualities: (36) a. b. c. d. e. (i)

the the the the the

causing event em effect ei, identified as the change e, causation en of ei by the cause em source state Sj of e; target state sk of e;

[DIFFERENTX] =def 5 (P, s) [ { SI [ P x ] } -, [ Ρ X ] ]

In other words, for something to be different with respect to some property P, a state meeting this condition is presupposed. It is only this presupposed condition with respect to which a difference can be identified in the first place. 24

This analysis implies (i) specific assumptions about improper positions in AS, i.e. operators that do not bind a variable in SF, and (ii) a natural, but non-trivial assumption about reflexive anaphors, according to which the antecedent of a reflexive pronoun provides the value for its argument position. See Bierwisch (1997) for some discussion of both assumptions.

25

For inchoatives like close, open, wake up, redden, etc. this fact manifests itself even in their morphological make-up. But also inchoatives like come and receive, where the target state is not marked morphologically, derive the initial from the final state by negation. There are, however, at least two types of lexical amendments that can be added to this basic pattern. First, for very few cases the source state may impose additional conditions. A case in point is melt, which requires its theme to start out as solid, rather than merely not liquid. Second, in cases like ersticken (choke), ertrinken (drown), erfrieren (freeze to death), all with the target state not alive, the transition is to be qualified by the mode of dying. Again the morphological make-up is relevant in many cases, an issue that must be left aside here.

26

Manfred

Bierwisch

Corresponding to the target state, by which inchoatives are determined, the effect of the causation tends to be characteristic for causatives. This is in fact the essence of the pattern illustrated in (33), which captures the crucial property of so-called ergative verbs. It furthermore turns in many cases the resulting state into the defining condition of the causation as a whole. Obvious examples are the causative variants of open, close, wake up, dry, clean or German schwärzen (blacken), kühlen (cool), töten (kill) etc.26 Differing from the source state of inchoatives, the cause of causatives can be and often is lexically specified: erschlagen (slay) erstechen (stab (to death)), erschießen (shoot), erdolchen (stab (with a dagger)) differ from töten (kill) by specifying the action, left open in kill. As already mentioned, ACT in (33) is a kind of dummy, in causatives like stab, shoot, or hang it is replaced by a lexical specification of the pertinent action.27 Besides these differences in lexical specification, the eventualities listed in (36) differ with respect to their temporal structure, their logical status, and their referential accessibility, as shown in sections 6 and 7.

6.

The eventualities involved in BECOME

Taking BECOME as the core component of inchoatives, I will characterize its basic temporal structure by means of conditions proposed e.g. in Dowty (1979), taking BECOME as an operator of type (t, t) with the properties indicated in (37), where ρ specifies the target-state, and I, J, Κ are time intervals as schematized in (38): (37) [ BECOME ρ ] is true at I if and only if (i) there is an interval J containing the initial bound of I such that —.p is true at J, and (ii) there is an interval Κ containing the final bound of I such that ρ is true at K. (38)

J

Κ 1

[' 1

"

:

>

I Two problems must be clarified here. First, as it stands, the interval I can extend over arbitrary parts of the source as well as the target state, such that Fred woke up would hold for a situation that includes arbitrary parts of Fred's sleep and of his being awake. Second, if one relies on strictly two-valued logic, no interval I' between J and Κ is possible, as at any time 26

27

Again, the defining target state may, but need not be morphologically realized. Thus while German töten is related to tot (dead), the resulting state has no overt reflex in kill. Similarly give, show, or convince are characterized by the resulting state, viz. have, see, and believe, respectively, without morphological relationship. The basic causative pattern can be enriched by further conditions, as in assassinate, murder, donate, etc. Amendments of this sort don't change the event reference and can thus be ignored here.

The event structure

of CAUSE and BECOME

27

either ρ or —.p must hold, with no transition. Hence any change would have to be strictly momentary. Dowty suggests to avoid these problems first by means of some sort of Gricean maxim, which picks out the shortest non-empty interval appropriate under conditions of encyclopedic or common sense knowledge, and second by acknowledging intervening time-intervals with undecided (or not two-valued) truth conditions. This ambivalent time structure of the event ej, which overlaps with both Sj and s k , is reflected by the fact that normally ej is not available for durational adverbials, as shown in (39a), but might still be qualified for extension in time in (39b): (39) a. * The cat died for three hours, b. The cat died very slowly. The next point to be noted is the different status to be assigned to the event, its initial and its final state. According to standard criteria, they instantiate what is usually called assertion, presupposition and implication, respectively. Consider (40) for illustration: (40) a. b. c. d.

The cat died. The cat didn't die. Did the cat die? Didn't the cat die?

Asserting and denying the cat's dying equally requires the initial state of the event, viz. the cat's being alive, to hold before. It must also hold for both types of question (40c) and (40d) to be appropriate. The negation of the initial state, i.e. the cat's being dead before the event, is compatible with the negation (40b) only as a correction of the presupposition. The target state on the other hand, viz. the cat being dead afterwards, follows from the truth of the assertion (40a), while its negation, that the cat is still alive, follows from the denial (40b). These observations are expressed more formally in (41) for the target state and in (42) for the source state, where t o t ' represents (temporal) overlap of t and t', and { φ } ψ indicates that φ is presupposed by ψ : 28 (41)

Ve [ 5s' [ e: [ BECOME ρ ] implies s ' : [ ρ ] ] ] where Τ e = t, Τ s' = t', t' ο t , t' ο t " , t < t" .

(42)

Ve [ 3s [ e : [ BECOME [ ρ ] ] ] => [ { s: ρ ] } [ e : [ BECOME [ ρ ] ] ] ] ] where Τ s z>c Τ e (i.e. s precedes e immediately)

(41) requires the target state s' to share its time in part with e and in part with the subsequent interval t". Because of (41), the source state s can overlap only with the initial part of the event e. One might construe (42) as an operation that expands the expression to the left of the arrow into that to the right of it, supplying automatically the presupposed source state

28

This notation is adopted from Kamp (2001), where properties o f presuppositions are explored more generally.

28

Manfred Bierwisch

of an inchoative event. In any case, (41) and (42) spell out the properties of BECOME and the predictable aspects of inchoatives based on it. I will now turn to the intriguing question to what extent the eventualities involved in a change of state are accessible for reference and modification. The analysis proposed for wake up, einschlafen, or aufwachen (in (32b) and (34)) suggests that it is just the main event, which the referential position Xc makes available to this effect. This seems to be born out by cases like (43), where apparently tense, temporal and modal adverbials all apply to the main event: (43)

Dann schlief sie innerhalb von drei Minuten ganz sanft ein. Then she fell asleep very softly within three minutes

Similarly, adverbial quantification by frequently, usually, mostly, occasionally etc. as in (44a) must rely on the same variable, given that adverbial quantifiers range over events, as proposed e.g. by von Fintel (1994). Simplifying with respect to irrelevant details, (44a) is thus to be analyzed as (44b), where [ MOST e ] must be construed as a quantifier with the restrictor given by the SF of Eva schläft ein and the nucleus in zehn Minuten·. (44) a. b.

Eva schläft meistens in zehn Minuten ein. Eva usually falls asleep within ten minutes MOST e

[ e : [ BECOME

[ SLEEP EVA ] ] ] [ Τ e c

1 0 MINUTES ] ]

As noted right in the beginning, this is not the whole story, though. A crucial problem, already illustrated in (1) above, is shown by the minimal pair in (45): (45) a. b.

Er ist in kurzer Zeit eingeschlafen. (He fell asleep within a moment) Er ist für kurze Zeit eingeschlafen. (He fell asleep for a moment)

The temporal delimitation within a moment in (45a) characterizes the change, while the durational adverbial for a moment in (45b) can only concern its resulting state. In other words, different aspects of the complex eventuality must be available for modification. One way to account for this observation has been proposed by McCawley (1973) within the framework of Generative Semantics. According to this proposal, the system of prelexical syntax provides two syntactic positions for an adverbial in cases like (45): within a moment commands [ BECOME [ SLEEP χ ] ], while for a moment commands just the predication [ SLEEP χ ]. 29 The pros and cons of pre-lexical syntax need not be repeated here, as the observation illustrated by (1) and (45) has various ramifications not naturally accounted for on the basis of pre-lexical syntax. Notice first, that the alternative interpretation illustrated in (45) carries over to adnominal modification, as shown by the parallel properties of the (a)- and (b)-cases in (46) and (47): 29

This approach has been pursued in a number of ways, especially with respect to elements like almost and again e.g. in von Stechow (1996). Alternative accounts of these facts, which do not rely on pre-lexical syntax, are discussed e.g. in Kamp & Roßdeutscher (1994).

The event structure of CA USE and BECOME

(46) a. b.

(47) a. b.

29

Er kehrte nach kurzer Zeit/für kurze Zeit heim. He returned home after/for a short time Seine Heimkehr nach so kurzer Zeit/für so kurze Zeit His return home after/for such a short time Das Wetter änderte sich plötzlich/dauerhaft. The weather changed suddenly/permanently Die plötzliche/dauerhafte Änderung des Wetters The sudden/permanent change of the weather

The alternative furthermore persists under adverbial quantification as illustrated in (44). Thus, a generic sentence like (48a) clearly requires quantification and modification to apply to the event as a whole, while in (48b) only the temporally restricted sleeping period is quantified over: (48) a. b.

Ein normaler Patient schläft meistens in ungefähr einer Stunde ein. An average patient usually falls asleep within roughly one hour Ein normaler Patient schläft meistens fur ungefähr eine Stunde ein. An average patient usually falls asleep for roughly one hour

In principle, this type of interpretation again carries over from adverbial cases like (49) to the adnominal modification in (50), which might be considered as clumsy, but neither as ungrammatical nor unclear in interpretation: (49) a. b.

(50) a. b.

Gelegentlich ändert sich das Wetter in wenigen Minuten. Occasionally the weather changes within a few minutes Gelegentlich ändert sich das Wetter fur mehrere Wochen. Occasionally the weather changes for several weeks Gelegentliche Änderungen des Wetters in wenigen Minuten waren absehbar Occasional changes of the weather for within a few minutes were to be expected Gelegentliche Änderungen des Wetters für mehrere Wochen waren absehbar Occasional changes of the weather for several weeks were to be expected

With respect to the background assumptions sketched in section 2, the question arises whether and how the state s', instantiating according to (41) the result of the change, should be available for reference and modification in the same way as the event e. Formally, two options can be adumbrated, if we assume that both the event e and its target state s' are actually available in SF for abstraction by argument positions, an assumption that requires the SF of an inchoative verb like einschlafen as illustrated in (34b) to be modified as shown in (51), with the obvious modification in (41) and (42). The two options to be considered can then be illustrated by (52). (51)

[ e : [ BECOME [ s ' : [SLEEP χ ] ] ] ]

30

Manfred

(52) a. b.

Bierwisch

/ ein + schlaf- / [ +V ] λχ Xs' Xe [ e : [ BECOME [ s': [ SLEEP χ ] ] ] ] / ein + schlaf- / [ +V ] λχ X(e,s') [ e : [ BECOME [ s': [ SLEEP χ ] ] ] ]

Both of these possibilities raise non-trivial problems. In (52a) an additional, presumably optional, in any case rather specific position would have to be introduced into AS requiring various conditions determining its properties and behavior. Even if s-selection would guarantee that only an appropriate adverbial can get its argument position absorbed by either Xs' or Xe, there are still a fair number of unsolved problems raised by the additional, improper referential position. These problems would not arise in (52b), where no additional position is introduced, but merely λε, the regular event reference, is replaced by the position supporting the complex variable (e,s') instead of the original e. This would require, however, an intricate and completely ad hoc regime of lambda abstraction, dealing with complex variables and their effects. Hence instead of exploring artificial ways to adapt one of the solutions hinted at in (52), it seems reasonable to stick to already available means as far as possible, getting along without an additional eventuality-variable squeezed into AS. Two observations seem to with respect to event or state time, the companionship, and characterize the target state, it (53) a. b.

Peter Peter Peter Peter

be relevant in this respect. First, adverbials that are neutral apply by default to the overall event. Thus even though the the localization of Peter's change of place could just as well is interpreted as a specification of the event:

kam gestern abend. came last night kam unerwartet nach Hause. came home unexpectedly

Even though (53a) would be compatible with the truth of Peter was here last night, it clearly does not semantically represent that proposition. Corresponding comments apply to (53b). Second, tense and time adverbials seem to be forced to apply to the same eventuality. Thus, the durational modification in (54a) concerns the target state, which the past tense locates before the utterance time, while the manner adverbial in (54b) modifies the change, which the present tense locates (preferably) at utterance time. (54) a. b.

Das The Das The

Tor öffnete sich fur fünf Minuten. door opened for five minutes Tor öffnet sich langsam. door opened slowly

These observations suggest that there is only one event reference available, which must support both the change or - under appropriate conditions - its result. This would be the natural effect if inchoative verbs with the operator BECOME have the representation illustrated in (55) for sich öffnen (intransitive open): (55)

/ öffn- / [ +V, -N ] XxXyXe

[ e : [ BECOME

[ OPEN

χ

] ] ]

The event structure

31

of CAUSE and BECOME

The crucial point here is the status of BECOME marked by italicization. 30 What this is intended to indicate is a special type of optionality, in the sense that it can be ignored for conceptual and truth-conditional interpretation, such that (54a) would have more or less the interpretation of (56a) with the SF as indicated in (56b): (56) a.

Das Tor war fur fünf Minuten offen. The door was open for five minutes

b.

3e [ Te < Tu

[ e : [ OPEN [ DEF

χ [ DOOR Χ ] ] ] & [ T e

D 5 MINUTES ] ] ]

There is, however, a crucial difference in interpretation between (54a) and (56a): While (54a) explicitly claims the resulting state to be the effect of a change, (56b) simply states a past situation. 31 This difference would be an automatic effect of the condition (42), if we assume that the presupposed source state is introduced also if the "shadowy" operator does not participate in further interpretation, in other words, if (55) is expanded into (57) in any case, 32 while (42) would of course not apply in (56): (57)

/ öffn- / [ +V, -Ν ] λχ λy λε [{ s z>c e & s: -.[ OPEN χ ] } [e: [ BECOME [ OPEN χ ]]]]

Notice that the presupposed state s immediately precedes e, whether e instantiates an event or a state. This fairly restricted stipulation associated with the status of BECOME yields exactly the two options for reference to eventualities illustrated before. It furthermore carries over to nouns as exemplified in (50). Thus Änderung (change) would have an entry like (58), which refers either to the change or its result, triggering again by means of (42) the presupposed source state: 33 (58)

/ änder-ung / [ +N ] (λχ) λβ [ e : [ BECOME [

DIFFERENT

χ

] ] ]

On this background, it is interesting to note that reference and modification may oscillate between the event (as a whole) and its target state, but not between the event and its presupposed source state. Even adverbials that would conceptually fit the source state can only

30 31

32

33

For the improper argument position λy giving rise to the reflexive anaphor, see note 23 There are, of course, implicatures arising from the delimitation for five minutes, but that is a different issue which I'll leave aside here. For the sake of completeness, it might be noted that the implication required by (41) holds trivially, even if BECOME is ignored. As Änderung is a regular derivation, it presumably does not require a separate lexical entry. (58) furthermore leaves aside the relation to the causative variant of ändern (included in (35) above), which would show up in cases like seine überraschende Änderung der Liste (his changing the list surprisingly). - It should be noted, though, that applying (42) to (58) to introduce the presupposed state would give (ii), if the definition (i) for DIFFERENT given in fn. 22 is taken into account: (i)

[ DIFFERENT χ ] = d c f 3 ( P , s ) [ { s : [ Ρ χ ] }

-. [ Ρ χ ] ]

(ii) / änder-ung / [ +N ] (λχ) X e [ { s z > c e & s : [ P x ] } [ e : BECOME - , [ Ρ χ ] ] ] ] This correctly specifies the (result of the) change as cancellation of some property Ρ that held of χ before.

32

Manfred

Bierwisch

modify the event - as in (59a) - or the target state - as in (59b). Adverbials that would be appropriate only for the source state, as in (59c), are anomalous. (59) a.

Er ist qualvoll gestorben. He died painfully b. Er ist eine halbe Stunde weggegangen. He left half an hour c. ?? Er hat sich seit zwei Stunden hingesetzt. He sat down since two hours

In cases like (60a) the adverbial clearly specifies the duration before the event to which the clause refers. Hence the time interval of the event differs from that specified by the adverbial, exactly as cases like (60b). Thus, the adverbial in (60a) and (60b) does not rely on reference to the target state. (60) a. b.

Er ist nach einer Stunde aufgestanden. He got up after an hour Er ist vor einer Stunde aufgestanden. He got up an hour ago

In general, then, an inchoative verb (or noun, for that matter) refers to one and only one eventuality. This is primarily the event it describes, and secondarily - due to the peculiar, elusive status of BECOME - the target state, but never the source state. Although presupposed and necessary, the source state is not available for reference in SF. The particular, somehow diaphanous character of BECOME may also be supported from the opposite direction, so to speak. One of the criteria for the standard distinction between (a-telic) processes and states on the one hand and (telic) events on the other is their behavior with respect to durational and terminating adverbials. As discussed so far, proper events like come, die, get sick combine freely with temporal delimitations like (with)in a week, but allow durational adverbials only as a specification of the resulting state, such that die for a week is awkward: (61) a.

Hans schlief innerhalb einer Stunde ein. Hans fell asleep within one hour b. Hans schlief für eine Stunde ein. Hans fell asleep for one hour c. Anna starb innerhalb einer Woche. Anna died within one week d. ?? Anna starb für eine Woche ?? Anna died for one week

States and homogeneous processes on the other hand allow durational adverbials, but should resist temporal delimitations, such that be sick within three days would be out. That this is not the case is shown by the acceptability of (62a) and (62c). This does not mean, however, that states and homogeneous processes combine with time-limits, but rather that

The event structure

of CA USE and BECOME

33

states are provided with an initial limitation by what has been called ingressive reinterpretation. 34 (62) a. b. c. d.

Hans schlief innerhalb einer Stunde. Hans slept within an hour Hans schlief eine Stunde lang. Hans slept for an hour Anna war innerhalb einer Woche krank. Anna was ill within one week Anna war eine Woche krank. Anna was ill for one week

Technically, the re-interpretation that provides the required (initial) limitation of the state or process might be considered as introducing the operator BECOME, such that e.g. the SF of schlafen given in (34a) is turned into that of einschlafen in (34b). Similarly, the SF of the copula sein (be) is converted into the SF of the inchoative copula werden (become, get) in (62c). In other words, ingressive reinterpretation of states relies in a way on the improper presence of BECOME, just as the durativization of events is due to its improper absence. Thus the adaptation illustrated in (61) comes out as the inverse to that in (62), both based on the same operator that can be invoked or dropped on demand. These considerations raise the question of whether the optional state reference associated with BECOME is not to be construed as a special case of the more general phenomenon called conceptual shift in Bierwisch (1982) and coercion in Pustejovsky (1995). By conceptual shift, the semantic interpretation of lexical items is adapted to contextual conditions that would otherwise violate conditions of s-selection. 35 Under this perspective, the state reference of inchoatives as e.g. in (61b) is the effect of conceptual shift, just like the ingressive interpretation of duratives in (62a). 36 If this is correct, the particular status of BECOME

34

35

A survey of the extensive discussion about states, processes, events and several types of reinterpretation is given e.g. in Maienborn (2003, chapter 3). For the sake of illustration, consider institute, which refers to a particular social organization in (i), its building(s) in (ii), its personnel in (iii), and something like its general principle in (iv): (i)

The institute appointed a new director.

(ii)

The institute has a new entrance.

(iii) The whole institute went on a long prepared excursion. (iv)

An institute is usually shaped by its head.

In each of these cases, the subject DP refers to a different sort of entities. The variation is, o f course, not arbitrary. It can only select from an organized range of options. It must be emphasized, however, that in spite of systematic restrictions things are more complex than these hints might seem to suggest. 36

Another well known type of adaptation is the wide-spread iterative or habitual interpretation, illustrated in (i) and (ii): (i)

Martin ist den ganzen Tag rein- und rausgegangen.

(ii)

Maria ist jahrelang mit dem Rad nach Hause gefahren.

Martin went in and out all day long For years, Maria went home by bike

34

Manfred

Bierwisch

would merely indicate a systematic possibility for adaptation under contextual conditions. Under this assumption, (63a) would differ from (63b) merely by the presence of BECOME, which (63b) lacks: (63) a.

Auguste Auguste Auguste Auguste

b.

wurde drei Tage krank. became ill for three days war drei Tage krank. was ill for three days

This might correspond to a subtle semantic difference, due to the presupposed source state invoked by BECOME in (63a), but not in (63b). A final problem to be taken up concerns the claim that the argument structure of a verb provides one and only one event position related to BECOME. The problem originates with examples like (64), where both the event and its result seem to be modified, requiring two separate positions in AS: (64) a.

Er ist gestern zehn Minuten in mein Büro gekommen. Yesterday, he came in my office for ten minutes Später wird sich ganz langsam für eine halbe Stunde der Vorhang öffnen. Later on, very slowly the curtain will open for half an hour

b.

Things are fairly uncomplicated in (64a), where the time adverbial yesterday can naturally apply to the state modified by (for) ten minutes. After all, (64a) is almost synonymous with Er war gestern zehn Minuten in meinem Büro. (64b) is more complicated, as very slowly cannot modify the target state, while the event cannot properly be modified by the duration for half an hour. It seems, however, that for half an hour in (64b) does in fact specify the underlying plan of the event, rather than the duration of the result.37 If this is correct, then in both (64a) and (64b) only one eventuality is referred to, viz. the target state in (64a), and the planned event in (64b). This sort of interpretation applies also to cases like (65), where the duration of the target state would be at variance with the time interval between past and present tense - unless for three hours is the destination or intention assigned to the event of leaving. 38

A particular type of iteration, creating a homogeneous process, is involved in cases like (iii): (iii) Er wird langsam größer (und größer). He is slowly getting taller (and taller) 37

38

This gets more obvious if (64b) is contrasted with (i), where the durational adverbial cannot naturally be construed as indicating an intended period, such that the combination becomes deviant: (i) 77 Später wird sich ganz langsam eine halbe Stunde lang der Vorhang öffnen. To the extent to which this sort of reconciliation is blocked or unnatural, the combination of two adverbials becomes deviant: (i)

77 77

Das Wetter änderte sich plötzlich für eine Woche. Suddenly, the weather changed for a week

I owe this observation to an anonymous reviewer.

The event structure of CA USE and BECOME (65)

35

Paul left for three hours. T w o hours are already over.

It might be noted in conclusion that the issues o f event and state reference o f BECOME must not be confused with the intriguing problem o f complex temporal frames and other conditions, as shown in (66): (66)

Vor zwei Jahren in Paris bin ich drei Wochen lang jeden Tag zwei Stunden ins Museum gegangen. T w o years ago in Paris, I went for three weeks every day two hours to the museum

H o w the event reference of inchoatives (and other expressions) is taken up by complex frames o f this sort is a problem o f its own.

7.

T h e e v e n t structure o f CAUSE

A s noted above, CAUSE is a functor relating two propositions, specifying a cause and its effect, respectively. For the sake o f illustration, consider the entry (35), collapsing the causative and inchoative reading o f ändern (change), repeated here as (67): / änder- / [ + V ] λ χ λ y λβ [ e : [ ( [ ACT y] [ CAUSE )[ BECOME [ DIFFERENT Χ ] ] ] ] ]

(67)

The cause is specified by [ ACT y ], the effect is specified by [ BECOME [ DIFFERENT Χ ] ], which determines either a change or - under appropriate conditions - its result. A s borne out by (68a) and (68b), this option carries over from inchoatives to causatives: (68) a. b.

Sie haben den They changed Sie haben den They changed

Plan gestern geändert. the schedule yesterday Plan fur fünf Tage geändert. the schedule for five days

Before looking into the event structure o f CAUSE more closely, I will sketch conditional properties. A widely accepted view on this matter has been proposed (1979). The proposal is based on the notion o f causal factor: φ is a causal factor necessarily implies ψ and ψ would not hold without φ. With this proviso, the truth-condition for CAUSE can be formulated: 39

39

its truthin D o w t y for ψ if φ following

Actually, Dowty provides a more sophisticated characterization of "causal factor", based on the notion of causal dependence of ψ on φ, and φ is a causal factor for ψ, if and only if a sequence of causal dependencies connects φ with ψ. Similarly, the intuitive notion of a more remote causal factor has a more technical characterization in Dowty's original definition: (i) [φ CAUSE ψ ] is true if and only if (i) φ is a causal factor for ψ, and (ii) for all other φ', such that φ' is also a causal factor for ψ, some -^φ-world is as similar or more similar to the actual world than any other —ιφ'-world is.

36

Manfred Bierwisch

(69) [φ CAUSE ψ ] is true if and only if (i) φ is a causal factor for ψ, and (ii) any other causal factor φ' for ψ is more remote than φ. Concerning the event structure related to CAUSE, the properties and dependencies of three eventualities are at issue (cf. (36) above): 40 (70) a. b. c.

the cause e m , characterized by φ the effect β|, characterized by ψ the causation e n , specified as [φ CAUSE ψ ]

As shown in (68), the effect of the causation can be an event or a state, an alternative that need not be due to the particular status of BECOME in the effect-proposition. Thus (71a) is naturally interpreted as the causation of an event, (71b) is the causation of a process, and (71c) indicates the causation of a state: (71) a. b. c.

The truck broke the fence rapidly The truck moved the cart quite a while very slowly The students held the rope straight for at least two hours.

What might be less obvious is the observation that the structure of the effect determines conceptually the homogeneous or non-homogeneous nature of the causation. To put it differently: The causation of an event is an event, while the causation of a state or process is a state or process. This might appear paradoxical at first glance, as one would expect the effect to be determined by the cause, rather the other way round. But notice that here we talk about the interdependence of causation and the effect it gives rise to, rather than the dependence of the effect on its cause. According to this consideration, it would be a natural conclusion that the causation as a whole and its effect are open for the same range of modifiers. As cases like (71) indicate, the modification is just not specialized in this respect. This would be a natural consequence of the assumption that the causation e n and the effect ei are not available for separate event reference. Notice that this accounts automatically for cases like (68), where the event- or state-interpretation of BECOME in ändern carries over from the effect to the causation. So far, the assumption that BECOME should not overtly provide multiple event-reference seems to carry over to CAUSE: causation and effect are not accessed by separate eventvariables. 41 What must be clarified, however, is the status of the cause e m . That cause and

40

41

For the present concerns, the intuitive notion that the cause of an eventuality can be specified as the closest possible causal factor will be sufficient. It should be noted that this is fully in line with Dowty's definition, as he explicitly considers causation as a relation between eventualities, specified by the propositions they instantiate. This applies not only to the temporal structure of the eventualities involved but also to other aspects, for reasons to which we will return shortly. Roughly speaking, the causation en does not exhibit a modality independently from the cause e m .

The event structure

37

of CAUSE and BECOME

effect of a causative verb cannot be temporally distant has already been noted by Fodor (1970). As he observes, (72b) is not an acceptable paraphrase of (72a). 42 (72) a. b.

Floyd heated the glass on Saturday such that it melted on Sunday Floyd melted the glass on Sunday by heating it on Saturday

Notice that the truth conditions for CAUSE given in (69) do not restrict the temporal relation between cause and effect. 43 Their coherence within one eventuality comes out as a natural consequence, however, if we assume that the constituent eventualities do not allow for reference by separate event variables. Strictly speaking, this assumption would require one temporal structure for the causation as a whole. Hence not only the nature of the causation and the effect are interdependent, as already noted, but also that of the causal event and via causation - the effect. In other words, events can only cause events, and processes or states can be causal only for processes or states. This assumption is not as implausible as it might appear in view of the fact that e.g. Floyd broke the glass describes a causal event the effect of which is the state of the broken glass. The effect of Floyd's action, however, is the event described as the glass broke, the target state of which is naturally construed as the result of Floyd's action. Conversely, if the effect cannot be a change, but must be a bare state or process, then according to this assumption the cause must be a process or state as well. This seems to be borne out by examples like (71b) and (71c): The action of the truck in (71b) must be as continuous a process as the motion of the cart it continuously causes. Similarly for the students causing the rope to be straight in (71c). The claim that only the overall eventuality is available for reference or modification in entries with CAUSE (and BECOME) raises a number of problems, the first of which is illustrated in (73). In cases like these, the adverbial apparently modifies just the causal act. Brute force in (73a) and despair in (73b) are neither properties of the effect nor of the causal connection as such. Similarly the slowness in (73c) and the recklessness in (73d) are neither properties of the result nor the causal connection: (73) a. b. c. d.

42

Floyd broke the glass with brute force Elvira closed the shop in despair He sharpened the pencil slowly The enemy's reckless destruction of most of the city

Fodor's argument is directed against pre-lexical syntax, according to w h i c h transitive melt w o u l d have the same underlying structure as cause

to melt. Thus Fodor's point is that this assumption

must be refuted, since (72b) is deviant, while (i) with the putatively s y n o n y m o u s cause to melt is acceptable, providing the basis for an overt separation o f cause and effect: (i) 43

Floyd caused the glass to melt on Sunday by heating it on Saturday.

A s a matter o f fact, D o w t y does not exclude a time course that has the effect preceding the cause, a possibility assumed in certain theories o f m o d e m physics. It might be added that D o w t y ' s causal factor explicitly relies on a sequence o f causally dependent events, such that "direct causation" bec o m e s the borderline case.

38

Manfred

Bierwisch

It would, however, be an artificial abstraction to separate the causing activity from its causal connection. Especially the notion of an activity modified independently from its causal role would not correspond to the conceptual structure imposed on the situation. The despair, for instance, attested in (73b) qualifies Elvira's behavior with regard to the expected or intended effect, rather than her pure performance. Similarly, the brute force involved in (73a) becomes relevant with respect to the causal connection, not the physical act as such. Similarly, what is slow in (73c) 44 or reckless in (73d) can only be determined with respect to the effect. Considerations of this sort, which apply, by the way, to nouns as well as verbs, hold in particular for modifiers of intentionality, as in (74), where the causal activity is qualified as incidental or intentional just with respect to its causal aspect: (74) a. b.

Mary turned the page inadvertently Frank's considerate separation of the different cases

A somewhat different problem arises with respect to the effect-proposition of a causal eventuality. As already noted with respect to (68b), the durative adverbial for five days concerns the situation of the changed schedule, indicating that the effect of the causal eventuality is a state, rather than an event. Otherwise the durative adverbial would be inappropriate. In fact, cases like (68) and (71) were meant to show that the adverbial determines at the same time the structure of the causation and its effect. This observation can only be correct, however, if the characteristics represented by the modifier carry over from the effect to the causal eventuality. This looks plausible with respect to the fence-breaking truck or the ropeholding students in (71). But how could the activity of changing the schedule in (68) be either punctual or durative, depending on the different types of effect? The answer to this puzzle is that the activity is conceptualized differently: Changing the schedule for a certain period imposes a causal condition for the relevant time - just as holding a rope creates a causal condition for a certain period. In other words, to be the source of a condition that holds for a certain time span differs from an otherwise identical situation without this condition, just as a certain activity with a certain intention - say cleaning the table - differs from the same physical movements executed without this goal. Under this perspective, the adverbials in (68) apply naturally to the eventuality as a whole, due to the particular character they impose on the effect. These considerations seem to be supported by the fact that adverbials can hardly specify the effect without automatically involving the eventuality as a whole. Even though the adverbial rasch und eindeutig in (75a) is a genuine qualification of the effected change, it cannot avoid to include the causal connection from which it results. Similarly the causal power of the fire is as partial as its effect in (75b), and even the kids' getting frightened in (75c) is hardly separable from Peter's acting furiously in some way. (75) a.

44

Die Untersuchung klärte die Situation rasch und eindeutig. The inquiry clarified the situation quickly and unequivocally

What counts is obviously not the activity as such - which might depend on the knife or whatever device is used - but the time needed to achieve the result in question. Example (73c) is due to an anonymous reviewer.

The event structure

b. c.

of CAUSE and BECOME

39

Das Feuer erleuchtete die Höhle nur teilweise. The fire lightened the cage only partially Peter hat die Kinder furchtbar erschreckt. Peter frightened the kids furiously

An apparent counter-example to this generalization needs to be clarified, though. Locative adverbials as in (76) can obviously apply only to the effected event or state, as neither the cause nor the causal connection can reasonably be said to be located on the table or in front of the window: (76) a. b.

Hans legte das Buch auf den Tisch. Hans put the book on the table Sie hängten Tücher vor die Fenster. They hang cloths in front of the windows

Notice first that the PPs are directional rather locative. Hence one might argue that they specify a goal rather than a location, which could specify the causal event as well as its effect. A different observation concerns the fact that the PPs are not free adverbials, but directional arguments saturating an argument position, since verbs of location like legen (lay), setzen (set, put), hängen (hang), ziehen (pull, drag), etc. require an (at least implicit) specification of the goal of motion. Constructions like er legte das Buch would in fact be elliptical without some directional complement. 45 Hence lexical entries for these verbs would look like (77), where [ LIE y ] abbreviates the condition that the maximal dimension of y must be horizontal. 46 (77)

/ leg / [ + V , - N ]

λ Ρ λ y λ χ λ ε [ e: [ χ ACT ] [CAUSE [BECOME [[ LIE y ] & [ Ρ y ] ] ] ] ]

[+Dir] On the basis of (77), the directional PP in (76a) saturates the argument position λΡ, such that its Semantic Form applies to the referent of the object das Buch. A rather different problem arises in cases with two (or more) adverbials that seem to affect different (sub)eventualities. Thus in (78) for two hours turns the change into a state, as discussed above, while yesterday places this state as a whole into the day before utterance time. (78)

45

46

Max hat uns gestern fur zwei Stunden geweckt. Yesterday, Max woke us up for two hours

In fact, causative verbs like stellen (put upright) 'inherit' the directional argument from the locative complement o f the corresponding positional verb stehen (stand). Similarly for hang, lay, etc. In view of the above consideration that a directional might indicate the goal of an event, and moreover with respect to the Semantic Form o f directional prepositions, one might adumbrate an entry like (i) instead o f (77), where the PP would specify the event: (i)

/ leg/

[ +V,-N ]

λΡ λ γ λ χ λβ [ e: [ [ χ a c t ] [ c a u s e [BECOME [ l i e y ] ] & [ P e ] ] ] ] [+Dir]

For some discussion o f these alternatives see Bierwisch (1988).

40

Manfred

Bierwisch

While duration and temporal localization, i.e. time specification "from inside" and "from outside" the same eventuality so to speak, are well compatible in (78), this type of reasoning gets in trouble with ordinary cases like (79), where one adverbial is not in the same way compatible with the other: (79) a. b.

Mach in zehn Minuten mal 'ne Weile das Fenster auf. Could you in ten minutes open the window for a while Die Polizei hat um acht zwei Stunden lang alle Ausgänge geschlossen. At eight o'clock the police closed all exits for two hours

In (79a), the durational adverbial for a while identifies the time interval of a state, viz. that of keeping the window open, but the temporal location in ten minutes seems to locate the causal act, rather than the resulting state, relative to the utterance time. The clash between the state holding for two hours and the act taking place at eight ο 'clock is even more obvious in (79b). What needs to be clarified, then, is the question how in these cases the effect of the durative adverbial is reconciled with the temporal location. This question does not arise in (78), because the causation of being awake is easily situated within the time interval yesterday. In (79) however, the temporal specification in ten minutes and at eight ο 'clock must be construed as preceding the duration of the open window and the closed exits, respectively. This situation is by no means unusual, though. Sentences like (80) exhibit essentially the same time structure: (80) a. b.

Wir fahren in zehn Minuten nach Wien. We'll drive to Vienna in ten minutes Um zehn Uhr warteten die Studenten am Eingang. At ten o'clock the students were waiting at the entrance

Here, the adverbials in ten minutes and at ten ο 'clock do not specify the time covered by the process, but rather its beginning, as already noted with respect to cases like (62).47 As observed there, the adverbial in Hans slept within an hour coerces an ingressive interpretation of sleep, by which the state it applies to is turned into the resulting state of a change. In the same way, the temporal delimitation in (79) yields an ingressive interpretation with the causative situation as resulting state. The peculiarity in this case is the fact the state in question is created by the durative adverbial, which turns the change originally involved in aufmachen (open) and schließen (close) into a state, as discussed above. Intuitively, then, the same eventuality is characterized twice, once with respect to its duration, whereby the state is defined in the first place, and once with respect to its beginning, whereby it is separated from the preceding state. Whether and how the ingressive component is to be represented in SF is an issue that has to be left open here. 48 47

48

As a matter of fact, (79a) is ambiguous, as the time interval may determine either the initial or the endpoint of the travel. Under the latter interpretation, in zehn Minuten (within ten minutes) becomes a durational adverbial that specifies the temporal delimitation of a process. The latter possibility is not relevant in the present context. If the ingressive interpretation is indeed to be captured by (some version of) B E C O M E , as adumbrated above in connection with (62), then the SF of (79a) should look roughly like (i), where IMP

The event structure

8.

41

of CAUSE and BECOME

In conclusion

Assuming that propositions, i.e. representations of sets of possible worlds, are to be instantiated by eventualities of different sorts, the considerations and proposals pursued here have tried to motivate four general conclusions. First, verbal as well as nominal descriptions of causative and inchoative situations involve a fairly complex structure of eventualities. A causation involves a cause and an effect, the effect being either a state or a change from source to target state. This does not provide an equally complex instantiation by means of event variables available for positions in the argument structure of lexical items. As a matter of fact, only one event variable seems to be available for a position in the argument structure of lexical items, susceptible for reference, quantification, and modification. 49 Second, with respect to the formal status of the event variable, the alternative between Davidson and Reichenbach seems to be clearly in favor of Reichenbach: Event reference is not a matter of the individual (basic) predicate constants, which would have to be extended from η-place into n+1-place predicates. It rather comes with the operator that provides an instantiation for the proposition it applies to. This is not a purely formal decision, though. It rather reflects the fact that the s-selection, associated with the position in AS, derives ultimately from the role of its variable within SF. Under this perspective, the s-selection of λε in a case like öffnen (open), discussed in (55) and repeated here as (81), differs depending on the presence or absence of BECOME, yielding the restrictions of either event or state restrictions. (81)

/ öffn- /

[ + V , - Ν ] λ χ λ y λβ [ e : [ BECOME [ OPEN χ ] ] ]

These considerations do not only relate to problems of inchoativity, they carry over to causatives and their unaccusative, i.e. decausativized use, as reflected in lexical entries like ändern (change) and many other causative verbs: (82) / a n d e r - / [ + V ] λ χ λ y λ β [ e : [ ( [ ACT y] [ CAUSE )[ BECOME [ DIFFERENT χ ] ] ] ] ]

is a short hand for the operator indicating the illocutionary force of the imperative, heavy brackets enclose the abridged SF of the state represented by 'ne Weile das Fenster aufmachen, and the temporal adverbials specify the duration of the state e and the location of its beginning e': (i)

IMP e '

[ f e ' : [BECOME [ [ e : [ [ACT y o u ] [CAUSE [OPEN [ w i n d o w ] ] ] ] & [ T e 3

I & [ 1 0 MINUTES 3

s o m e time ]

Te' ] ]

This representation would emerge if ingressive interpretation would be the effect of an interpretive template, i.e. a phonologically empty lexical entry with the following provisional characterization: (ii) 49

λ Ρ λ β ' [ e ' : [ BECOME [ Ρ e ] ]

where e' necessarily overlaps with (the initial boundary of) e. It must be emphasized that this restriction concerns only the relevance of event variables for the syntactically controlled compositional SF. Conceptually, however, event reference plays a role in various other respects, among them the presupposed source state required by BECOME. Thus the multiple event reference usually found e.g. in DRS representations such as for instance in Kamp (2001) and the references given there is by no means at variance with the claim under discussion.

42

Manfred Bierwisch

The s-selection mediated by λβ depends on the presence or absence of CAUSE as well as BECOME.

Third, this effect - and the generalizations it supports - is strictly dependent on the fact that there is at most one event reference for each AS. Besides the matters of s-selection already mentioned, this assumption accounts also for the problems related to the requirement of direct causation. It prevents lexical items from covering cases that allow temporal or other distances between cause and effect. This restriction to one event reference applies equally to verbs and nouns, with the further condition that verbs must have a referential eposition, while nouns may have one. 50 The structures leading to these observations give rise to interesting patterns of lexicalization with language particular variations within fairly systematic possibilities. An obvious case in point are the pairs of inchoative and causative verbs, often realized by homonyms like (transitive and intransitive) open, close, break, change, get, etc. but sometimes lexicalized by independent items like kill/die, bring/come, give/receive, and, of course, cause/become. Language-particular and idiosyncratic variations are illustrated by German reflexive inchoatives like sich öffnen (open), sich ändern (change), etc. as opposed to nonreflexive inchoatives brechen (break), schmelzen (melt), which are parallel to their English counterparts. The causative/inchoative alternation carries over to some extent to the event nominalizations, as indicated in (83): (83) a. b.

Die Änderung der Fahrzeiten durch die Bahn the change of the schedule by the company Die Änderung der Windrichtung the change of the direction of the wind

(causative) (inchoative)

The lexicalization-patterns include, of course, also expressions for the resulting state, such that correspondences of the following type emerge:

50

Nouns and verbs as lexical categories are characterized by the fact that their A S has a referential position as its highest (innermost) operator. For verbs, this position must be an e-position, for nouns it may be any sort of individual reference. Discussion and motivation of this aspect of argument structure goes beyond the present scope.

The event

(84) a. b. c. d. e. f.

structure

of CAUSE and

43

BECOME

e :[Ρ χ ]

e : [ BECOME [ Ρ χ ]]

e : [ [ A C T y ] CAUSE [BECOME [ Ρ χ ] ] ]

tot sein be dead wach sein be awake liegen lie berühmt sein be famous haben have wissen know

tot gehen ~ sterben die wach werden ~ erwachen wake up sich legen lie down berühmt werden become famous bekommen ~ kriegen get lernen learn

tot machen ~ töten kill wach machen ~ wecken wake up legen lay berühmt machen make famous geben get ~ give lehren teach

w h e r e Ρ = -> A L I V E f o r ( a ) ,

Ρ = -.SLEEP f o r ( b ) ,

Ρ = HORIZONTAL LOCATION f o r ( c ) ,

Ρ = FAMOUS f o r ( d ) ,

Ρ = HAVE Ζ f o r ( e ) ,

Ρ = KNOW Ζ f o r ( f )

There are wide ranging variations o f quite different types to be observed here. What is worth noting, though, is the recurrent pattern according to which predicates o f different type are turned into characterizations o f states, events, and causations.

References

Bach, Emmon (1986): "The Algebra of Events." - In: Linguistics and Philosophy 9, 1-16. Bierwisch, Manfred (1982): "Formal and Lexical Semantics." - In: Linguistische Berichte 80, 3-17. - (1988): "On the Grammar of Local Prepositions." - In: M. Bierwisch, W. Mötsch & I. Zimmermann (eds.): Syntax, Semantik und Lexikon. 1-65. Berlin: Akademie-Verlag. - (1997): "Lexical Information from a Minimalist Point of View." - In: Ch. Wilder, H.-M. Gartner & M. Bierwisch (eds.): The Role of Economy Principles in Linguistic Theoiy, 227-266. Berlin: Akademie Verlag. - (2002): "A Case for CAUSE." - In: I. Kaufmann & Β. Stiebeis (eds.): More than Words, 327-353. Berlin: Akademie Verlag. - (2003): "Heads, Complements, Adjuncts: Projection and Saturation." - In: E. Lang, C. Maienborn & C. Fabricius-Hansen (eds.): Modifying Adjuncts. 113-159. Berlin: Mouton de Gruyter. Chomsky, Noam (1981): Lectures on Government and Binding. - Dordrecht: Foris. - (1986): Knowledge of Language: Its Nature, Origin, and Use. - New York: Praeger. - (1995): The Minimalist Program. - Cambridge, Mass.: MIT 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. Dowty, David R. (1979): Word Meaning and Montague Grammar. - Dordrecht: Reidel. Fintel, Kai von (1994): Restrictions on Quantifier Domains. - Doctoral dissertation, Amherst. Fodor, Jerry A. (1970): "Three Reasons for not Deriving 'kill' from 'cause to die'", - In: Linguistic Inquiry 1, 429-448.

44

Manfred

Bierwisch

Higginbotham, James (1985): "On Semantics." - In: Linguistic Inquiry 16, 547-593. Kamp, Hans (2001): "The importance of Presupposition." - In: Ch. Rohrer, A. Roßdeutscher & H. Kamp (eds.): Linguistic Form and its Computation. 207-254. Stanford: Center for the Study of Language and Information. Kamp, Hans & Uwe Reyle (1993): From Discourse to Logic. Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory. Dordrecht: Kluwer. Kamp, Hans & Antje Roßdeutscher (1994): "Remarks on Lexical Structure and DRS-Construction." In: Theoretical Linguistics 20, 97-164. Maienborn, Claudia (2003): Die logische Form von Kopula-Sätzen. - Berlin: Akademie Verlag. McCawley, James D. (1973): "Syntactic and Logical Arguments for Semantic Structures" - In: Osamu Farjimura (ed.): Three Dimensions in Linguistic Theory. 259-376. Tokyo: TEC Corporation. Parsons, Terence (1990): Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge, Mass.: MIT Press. Pustejovsky, James (1995): The Generative Lexicon. - Cambridge Mass.: MIT Press. Reichenbach, Hans (1947): Elements of Symbolic Logic. - New York: The Free Press. Stechow, Arnim von (1996): "The Different Readings of wieder 'again': A Structural Account." - In: Journal of Semantics 13, 87-138. Vendler, Zeno (1967): Linguistics in Philosophy. - Ithaca, New York: Cornell University Press. Williams, Edwin (1981): "Argument Structure and Morphology." - In: The Linguistic Review 1,81114. Wunderlich, Dieter (2000): "Predicate composition and argument extension as general options." - In: B. Stiebeis & D. Wunderlich (eds.): Lexicon in Focus. 247-270. Berlin: Akademie Verlag.

Stefan

Engelberg

Stativity, supervenience, and sentential subjects

*

The paper explores how verbs like helfen 'help' should be treated within event semantics. These verbs allow both agentive NP-subjects and sentential CP-subjects. Their behavior with respect to adverbial modification reveals that in their agentive variant these verbs refer to events, while in their sentential variant they refer to states. The meaning that sentential helfen conveys is that the beneficiary is in a good disposition and that this state is brought about by what is expressed by the sentential subject. This involves a kind of subjective value statement about what is good for the beneficiary and what is not. The relation of "bringing about" involved here is not mainly one of causal dependence - lacking the typical denseness of causal chains - but one that involves supervenience. Supervenience, a notion widely used in moral theory and philosophy of the mind, allows accounting for the dependence of the rather subjective nature of the resultant state of helfen on particular events which occur in the world. The agentive variant of helfen is derived by embedding the meaning of sentential helfen into an event description.

1.

Introduction

Events have been introduced into the ontology o f semantic theories to deal with quantification data, to account for the semantics o f adverbials by representing them as intersective modifiers, and to explain data in the aspecto-temporal domain. This has been a success story w h e n dealing with verbs referring to concrete, perceptible events in particular. For verbs o f a more abstract or stative nature, w h i c h behave differently with respect to the phenomena mentioned above, event semantics often doesn't have that much to offer. In other words, from a lexical point o f v i e w event semantics works best w h e n ' s o m e b o d y ate an apple in the kitchen'. Since w e don't want to g i v e up the advantages event semantics provides, more efforts have lately been made to deal with abstract and stative verbs in a w a y compatible with the event-based analyses o f constructions involving concrete verbs. This is one o f them. At first sight, a verb like helfen

'help' doesn't s e e m to challenge the basic convictions o f

an event semanticist. Simplifying things a little bit, a sentence like ( l a ) might translate into something like ( l b ) : '

1

I'm grateful to Kate Kearns and Jennifer R. Austin for intensive discussions of this paper, as well as to Cynthia Allen and Avery Andrews for inviting me to Australian National University during which time I was able to write and discuss this paper. Claudia Maienborn, Daniel Hole, and two anonymous reviewers provided valuable comments. I will employ Davidsonian instead of Neo-Davidsonian representations, i. e. representations in which verbal predicates have thematic arguments in addition to an event argument. In Engelberg (2000: 156ff, 2002) it is argued that Neo-Davidsonian predicates fail in sufficiently distinguishing the verb's arguments.

46

Stefan Engelberg

(1)

a. b.

Rebecca half Jamaal in der Küche. 'Rebecca helped Jamaal in the kitchen' 3e[HELP(rebecca,jamaal,e) & AGENT(rebecca,e) & BENEFlClARY(jamaal,e)& iN(the-kitchen,e)]

That this account of the semantics of helfen might be too simple becomes evident when we look at sentences such as those in (2), where instead of an animate NP-subject we find sentential subjects which might (2a) or might not (2b) report events. 2 It is obvious that the meaning of the sentences in (2) cannot be accounted for in any way similar to (lb): 3 (2)

a. b.

Dass Rebecca sein Motorrad repariert hatte, half Jamaal sehr. 'That Rebecca had fixed his motorbike helped Jamaal a lot' Dass Jamaal so gut aussah, half Rebecca sehr. 'That Jamaal was so good-looking helped Rebecca a lot'

The semantic analysis of verbs selecting sentential subjects and the relationship between their sentential and their agentive variant are the topic of this paper. At the end we will see that representations like (lb) are inappropriate for both the sentential and the agentive variant. The analysis will focus on helfen 'help' and partly on some other related verbs like gefährden 'endanger', verbessern 'improve', erleichtern 'facilitate', and verschlechtern 'make worse' (3). I will call these verbs 'dispositional verbs' since they convey that somebody is - in a wider sense - brought into some disposition. (3)

a. b. c. d.

Dass sie die Tribüne verlegt hatten, gefährdete die Zuschauer. 'That they had relocated the grandstand endangered the spectators' Dass sie das Problem lösen konnte, verbesserte ihre Stellung in der Firma. 'That she could solve the problem improved her position in the company' Dass sie seinen Computer repariert hatte, erleichterte ihm die Arbeit. 'That she had fixed his computer, facilitated his work' Dass sie gefeuert wurde, hat ihre Lage noch verschlechtert. 'That she was fired, made her situation even worse'

The paper proceeds as follows: In section 2 , 1 will discuss the behavior of the agentive and the sentential variants of dispositional verbs with respect to adverbial modification and thereby show that they differ in their referential properties. In section 3, the nature of the particular resultant state that is part of the meaning of dispositional verbs will be investigated. Section 4 discusses how this state is brought about. In particular, it will be argued that the relation between this state and what is expressed in the subject is an evaluative one that involves a relation of supervenience. Section 5 will deal with the semantic derivation of

2

3

There is well-founded scepticism that these CPs really occur in subject position. For easy reference and because this syntactic issue does not really matter for what I have to say about these structures, I will still call them 'sentential subjects'. Though I will mainly discuss German examples, most claims hold for the corresponding English sentences, too.

Stativity, supervenience, and sentential subjects

47

sentences containing dispositional verbs and the question how the agentive and the sentential variant of these verbs are related. Section 6 presents the conclusion.

2.

Stative versus eventive variants of dispositional verbs

The purpose of this section is to show that helfen in its agentive variant is an event verb while in its sentential variant it shows all properties of state verbs. Evidence for this assumption comes from four kinds of adverbials: agent-oriented adverbials, locative adverbials, temporal adverbials, and degree adverbials. 4

2.1. Agent-oriented event adverbials Instrumental, manner, and other agent-oriented event adverbials occur freely with agentive helfen, while they yield ungrammatical constructions with sentential helfen: (4)

a.

Rebecca hat Jamaal mit ihrem neuen Trockentuch / fröhlich geholfen. 'Rebecca helped Jamaal with her new dish towel / happily' b. * Dass Rebecca das Geschirr abgetrocknet hat, hat Jamaal mit dem neuen Trockentuch / fröhlich geholfen. 'That Rebecca dried off the dishes helped Jamaal with the new dish towel / happily'

One might want to explain the data in (4b) by the fact that both adverbials are agent-related and no agent surfaces as the subject of helfen. However, adverbials of this sort are usually also licensed when the agent remains implicit:5 (5)

a. b.

Das Geschirr wurde mit dem neuen Trockentuch / fröhlich abgetrocknet. 'The dishes were being dried off with a new dish towel / happily' Jamaal wurde mit einem neuen Trockentuch / fröhlich geholfen. 'Jamaal was being helped with a dish towel / happily'

4

It might be more appropriate to speak of adverbs or adverb phrases instead of adverbials in some o f the following cases (particularly with respect to degree modifiers). The relationship between these three terms is rather problematic (cf. Austin et al. 2004). In this paper I use 'adverbials' as a cover term.

5

English differs from German in this respect. In German the adverbials express a relation between the event and the agent w h o is left implicit, while in Englisch the adverbials are often more related to the overt subject referent whether this is the agent or not. That might be the reason why the corresponding English sentences sound rather odd. Even though it has to be admitted that not all instances of modification of implicit agents sound completely acceptable in German, there is a clear difference to the fully unacceptable (4b).

48

Stefan Engelberg

I will therefore assume that the ungrammaticality of (4b) arises mainly because helfen in its sentential variant does not refer to an event.

2.2. Locative and temporal adverbials The situation presented by locative modifiers is different: both variants of helfen allow locative modifiers but these are not interpreted in the same way. Locative modifiers with agentive helfen as in (6a) localize the event the subject referent is involved in. (6a) says that whatever helpful action Rebecca performed happened in the kitchen. With sentential helfen the interpretation of locative modifiers is less simple. In (6b) the modifier certainly does not localize what Rebecca did - she might have fixed the pipes in the basement. It rather seems to convey that what Rebecca did helped Jamaal with whatever he was doing in the kitchen. However, the interpretation of locatives with sentential helfen is not confined to the localization of some additional implicit event. In (6c) the locative rather conveys the meaning that the scarf helped her (e. g., to not catch a cold) while she was on the grandstand. (6)

a. b. c.

Rebecca half Jamaal in der Küche. 'Rebecca helped Jamaal in the kitchen' Dass Rebecca die Wasserrohre repariert hatte, half Jamaal in der Küche. 'That Rebecca had fixed the water pipes helped Jamaal in the kitchen' Dass sie einen Schal mitgenommen hatte, half Rebecca auf der Stehtribüne. 'That she brought a scarf helped Rebecca on the grandstand'

Maienborn (2001, 2003) has argued that locatives with stative expressions do not localize an event-like entity but function as frame-setting modifiers which "provide a semantically underspecified domain restriction for the overall proposition" (Maienborn 2003). I assume that the locative adverbials in (6b,c) perform this frame-setting function, while the locative in (6a) simply predicates over an event argument. As frame-setting modifiers, locatives usually allow for a temporal interpretation involving the holder of the state. This is why in (6b) and (6c) the locative can be reinterpreted as 'it helped him when he was in the kitchen / when she was on the grandstand' while (6a) does not imply that she helped him when he was in the kitchen. He could have been somewhere else at that time. Different interpretations with sentential vs. agentive helfen are also displayed by temporal adverbials, which are known to be compatible with activities and statives (Vendler 1957): (7)

a. b.

Rebecca half Jamaal eine Zeitlang. 'Rebecca helped Jamaal for some time' Dass Rebecca so gut Gadakhisch gelernt hatte, half ihr eine Zeitlang. 'That Rebecca had learned Gadakhian so well helped her for some time'

As w e would expect with event-referring verbs, in (7a) the adverbial relates to the time of the event Rebecca was engaged in, but it does not refer to the length of the helping effect. (7b), on the other hand, says that the helping effect of Rebecca having learned Gadakhian held for some time. This clearly points to a stative use of helfen.

Stativity, supervenience,

and sentential

49

subjects

2.3. Degree modifiers Degree modifiers are usually found with statives and are also available for a number of activity verbs. Thus, it is no surprise that we find them with both variants of helfen·. (8)

a. b.

Jamaal half Rebecca sehr / ein bisschen / nicht viel. 'Jamaal helped Rebecca very much / a little bit / not much' Dass sie so gut Gadakhisch gelernt hatte, half ihr sehr / ein bisschen / nicht viel. 'That she had learned Gadakhian so well, helped her very much / a little bit / not much'

But degree modifiers can also be used to show the difference between agentive and sentential helfen. Using the diagnostics developed in Maienborn (2003) it can be shown that the modifer ein bisschen 'a little bit' can either serve as an event modifier indicating that the run time of the event was rather short or as a degree modifier indicating the degree to which a certain property holds.6 With event predicates the modifier always has an eventive reading (9a) and sometimes, depending on the particular predicate, also a degree reading as in (9b) where it either indicates the duration of the event or the amount of sweat transpired. With statives, on the other hand, ein bisschen never predicates over the duration of the state, having only the degree reading (9c). (9)

a. b. c.

Sie joggte ein bisschen. 'She jogged a little bit' Sie schwitzte ein bisschen. 'She sweated a little bit' Sie mochte ihn ein bisschen. 'She liked him a little bit'

(eventive reading) (eventive and degree reading) (degree reading)

Applying this procedure to the two variants of helfen, the results are clear. Combined with agentive helfen, ein bisschen is ambiguous. (10a) can either mean that she helped him for a little while (that seems to be the dominant reading) or that the effect of what she was doing was of a moderate degree of helpfulness. In (10b) only the degree reading is available. (10) a. b.

6

Rebecca hat ihm ein bisschen geholfen. 'Rebecca (has) helped him a little bit' Dass sie seinen Computer repariert hat, hat ihm ein bisschen geholfen. 'That she fixed his computer (has) helped him a little bit'

Maienborn (2003) employs this test mainly in order to show that there are two kinds of state expressions, Davidsonian states (sleep, sit, wait) which behave like event predicates and Kimian states (love, be hungry, be blond) which yield negative results when subjected to the usual tests for determining event reference of predicates. It should be noted that not all event predicates allow modification by ein bisschen. The phrase is restricted to certain kinds of homogenous event predicates.

Stefan Engelberg

50 3.

The resultant state of sentential helfen

If whatever is conveyed by the sentential subject A helps the beneficiary Β then A brings about a certain state which holds for B. This state - as has been shown in the preceding section - is what sentential helfen refers to.7 Our investigation into the meaning of sentential helfen and other dispositional verbs will proceed in two parts. In this section I will give an account of the nature of this state and in section 4 I will discuss how it is brought about.

3.1. Referential state arguments Having established the Stative nature of the sentential variant of helfen, the question arises as to how this observation can be dealt with in semantic terms. A solution favored by a number of semanticists is to enrich our ontology by state entities (cf. Asher 1993, Maienborn 2003). Although arguments in favor of state arguments are scarce they might suffice to defend this view. There are some facts about temporal modification that provide support for state arguments. Not only do stative expressions allow durative adverbials, we also find them with event predicates where it is usually assumed that they are one-place first-order predicates functioning as intersective modifiers. If we don't want to be forced to adopt a completely different analysis for statives it would be handy to have an argument the durative adverbials can predicate over. Since all kinds of agent-related adverbials as well as locative ones do not predicate over states, the motivation for the assumption of state arguments coming from adverbial modification remains rather thin. A further argument concerns anaphoric reference. In (11) das 'that' must pick up a more concrete entity than a fact or a proposition since it is modified by a temporal adverbial. States seem to be the right sort of entity here. (11) a. b.

Sie war müde, und das einige Stunden lang. 'She was tired and that (lasted) for several hours' Dass sie Gadakhisch konnte, half ihr, und das zumindest für die ersten Wochen. 'That she knew Gadakhian, helped her, and that at least for the first weeks'

That stative verbs can be nominalized is another hint that there are state arguments. At least, no other ontological sort seems to be appropriate for the referential argument of his love of thimbleberries, the help with the homework or his everlasting tiredness. On the other hand, states are very restricted when it comes to quantification. Although examples like (12) might suggest countability of state entities, I'm doubtful that in uttering these sentences we are really quantifying over states. It rather seems that what we want to say is that on two occasions (another dubious ontological sort) Jamaal was tired and on

7

This is clearly different from change-of-state verbs, which refer to events and imply a result state. In she emptied the bottle the state of the bottle being empty after the event can be inferred, but that doesn't make the verb a stative one.

Stativity, supervenience, and sentential subjects

51

several occasions it helped him that he was a good dancer. Thus, the quantificational adverbials are similar to the frame-setting locatives discussed in section 2.2. (12) a. b.

Jamaal war zweimal müde. (from Maienborn 2003) 'Jamaal was tired twice' Dass Jamaal so ein guter Tänzer war, hat ihm mehrmals geholfen. 'That Jamaal was such a good dancer, helped him several times'

This intuition is corroborated by the fact that state nominalizations (13a) are not prone to combine with numerical quantifiers, in contrast to event nominalizations (13b): (13) a. * drei Lieben / Hilfen / Müdigkeiten 8 'three loves / helps / tirednesses' b. drei Sprünge / Küsse / Läufe 'three jumps / kisses / runs' Thus, states are dubious entities: they have temporal properties, they can be referred to, but they can hardly be counted. For the time being and without particular enthusiasm, I will assume that states are what expressions containing sentential helfen predicate over. I assume though that states are not as concrete as objects and events but rather something of a partly concrete and partly abstract nature. Approaches like Asher (1993) and Maienborn (2003, forthcoming) are probably on the right track. Asher (1993:15ff) assumes that there are a number of entities of a more or less abstract nature that are construed for efficient cognitive and natural language processing while they might not play a role in the ontology of the world we are talking about. Maienborn (2003, forthcoming), following Asher (1993), makes use of Kim's (1969, 1976) approach to event ontology and conceives of a state as an abstract object for "the exemplification of a property Ρ at a holder χ and a time t." 9 If this is understood in the spirit of Kim (1976: 161), two states characterized by [x,P,t] and [>', Q, t'] should be identical iff χ = y, t = t' and Ρ = Q. In extensionalizing over the holder of the state, I think, this approach accounts for the fact that different descriptions of a holder of a state do not influence our judgment about the identity of states. That is, (14a) and (14b) are different propositions but the state she was in described in (14a) is certainly the same as in (14b) if Rebecca is Jamaal's girlfriend. Therefore, if we extend (14a) as in (14c), i. e. pick up the state by that and introduce a durative adverbial as a predicate over states, (14c) implies (14d): (14) a. b. c. d.

Rebecca Jamaal's Rebecca Jamaal's

was tired girlfriend was tired was tired ... and that persisted for the whole afternoon girlfriend was tired ... and that persisted for the whole afternoon

8

The nouns can, of course, be reinterpreted, e. g. as 'three men/women I loved', 'three people who were engaged as helpers' and perhaps 'three different kinds of tiredness'.

9

This was originally proposed as an event concept by Kim (1976), but as I have argued in Engelberg (2000: 235ff), for most linguistic purposes it is too fine-grained a concept for events.

52

Stefan Engelberg

The property Ρ is constitutive for the state. If we assume TIRED to be the property being at stake in (14), then Rebecca's being TIRED is a different state from Rebecca's being EXHAUSTED at the same time since EXHAUSTED would be a property different from P. This renders states as more fine-grained entities than events, where we would not principally exlude that KlCKing the ball over the line and HAMMERing the ball into the goal necessarily describe different events. 10

3.2. States and domains So far we have established that helfen 'help' refers to an event e in its agentive variant and to a state s in its sentential variant. What is it that helfen and other dispositional verbs actually say about this state? The effect state s embedded in the meaning of dispositional verbs is of a rather unspecific nature: if a helps x, x ' s life becomes good in a certain sense, if a endangers χ, χ is in a certain sense at risk of experiencing something negative, if a improves χ, χ becomes better in a certain sense. The notion "in a certain sense" is crucial here and it has to be inferable from the context in which sense the effect state holds. E.g., having been asked Have you heard anything about Rebecca lately? it would be a rather strange thing to answer Yes, she helped Jamaal or Yes, she endangered Jamaal, at least in a situation were no previous knowledge on part of the hearer pertaining to the help and endangerment can be presupposed. The two most probable reactions to such an answer would be to ask How? or In what sense? The first of these reactions would require information about how this state has been brought about (more about this in section 4), the second reaction invites the specification of the just mentioned notion of helping or endangering "in a certain sense". Every evaluation of the truth of a sentence containing a dispositional verb requires a domain to be determined with respect to which the effect state holds. Imagine the following dialogue (15) a. b.

Jamaal: "Our organization helps Gadakhs in rural communities by sending them used clothes and blankets." Rebecca: "That's nonsense. It would help them if it would grant them loans to rebuild and modernize the local textile workshops."

In (15a), the domain might be determined by how much the Gadakhs will have to freeze during the next winter. In that sense, clothes and blankets will be of some help. If the domain is chosen differently, as in (15b), the truth of what has been said can be challenged. Choosing the long-term development of rural Gadakh economies as the domain of help, sending used clothes and blankets is not a help since it might devastate the local textile workshops. Thus, the function of these domains is to restrict the application of the proposition. In this sense they are comparable to "frames" in the sense of Jacobs (2001: 655f) or "domains" as treated by Ernst (2004). The domains can of course be made completely or par10

This at least is the most straightforward interpretation for the properties in Kim's approach. There are different possibilities of interpreting Kim's constitutive properties, though; cf. Engelberg (2000: 236).

Stativity, supervenience,

and sentential

subjects

53

tially explicit. Domain adverbials (16a), prepositonal phrases (in German in particular those headed by bei, (16b)), or certain kinds of subordinate clauses (16c) can contribute to the establishment of a particular domain. (16) a. b.

c.

Dass Jamaal in der Lotterie gewonnen hat, hat ihm finanziell geholfen. 'That Jamaal won the lottery helped him financially' Dass Jamaal in der Lotterie gewonnen hat, das hat ihm bei der Finanzierung des Projekts geholfen. 'That Jamaal won the lottery, helped him in financing the project' Was das Finanzielle angeht, so hat es Jamaal geholfen, dass er in der Lotterie gewonnen hat. 'As far as financial matters are concerned, it helped Jamaal that he won the lottery'

I will further assume that some predicates are obligatorily interpreted with respect to domains. E. g. help somebody in contrast to smell something always has to be relativized to a domain." An interpretation of a sentence with respect to a domain has a number of consequences for the inferences we may draw. As should have become clear by now, a proposition which is true with respect to one domain might be false with respect to another. If winning the lottery helped Jamaal financially (he could settle all his open bills) it still might not have helped him healthwise (he died as the consequence of heavy partying after having cashed in his win). It has sometimes been claimed that a proposition relativized to a domain does not entail the same unrelativized proposition (cf. Jacobs 2001: 656). This would mean that from (17a) we cannot infer (17b): (17) a. b.

That they stopped sending used blankets helped the Gadakhs economically, That they stopped sending used blankets helped the Gadakhs.

I assume that words like helfen are always interpreted with respect to a domain, i. e. there is no unrelativized interpretation for (17b). Instead, there is always the urge to construct a domain from the context. The unwillingness to infer (17b) from (17a) is obviously due to the fact that it is difficult to establish any specific domain here. This makes the sentence hard to interpret. We might be less reserved in accepting the inference from (18a) to (18b), though, which is probably due to the fact that the relevant domain suggests itself rather easily. But since even (18b) can be interpreted with respect to other domains, e. g. with respect to environmental issues, there is indeed no inference from a proposition with an explicitly given domain to the same proposition without an explicit domain. (18) a. b.

That they attracted companies which provided many new jobs helped the region economically. That they attracted companies which provided many new jobs helped the region.

" Alternatively, we might assume that for predicates like smell something a domain like 'with respect to sensory perception' / 'olfactorily' is lexically pre-installed by default.

54

Stefan

Engelberg

Furthermore, a sentence with respect to a domain dom doesn't entail the same sentence with respect to a domain that "includes" dom or is included by dom. If a financial allocation by the ministry helps the linguistic department with respect to its finances (its budget is balanced now) it neither follows that it helps the department in financing a new lecturer position (the money might be restricted to the acquisition of technical equipment) nor does it follow that it helps the department in its general structural development (the allocation is tied to the department's commitment to develop a new program in applied linguistics that nobody wants).

3.3. Parameters involved in the effect state of helfen We are now able to sum up the ingredients that go into the characterization of the state a dispositional verb like helfen 'help' refers to: • • • • •

a holder of the state: Λ: the degree to which the effect state holds: d a time T(S) at which the state holds and which overlaps with the reference time of the finite verb: t a property that describes the state: GOOD (for χ) a domain with respect to which the holding of the state is evaluated: dom

Equipped with these basics about the nature of the effect state s of dispositional verbs, we can describe 5 for some of the dispositional verbs (where A stands for whatever is expressed by the subject): (19) a. b.

c.

helfen 'help': [A brings about something such that...] with respect to a domain dom, χ is in a state .v at time t characterized as GOOD to a degree d.n verschlechtern 'make worse' [A brings about something such that...] with respect to a domain dom, χ is in a state 5 characterized as BAD to a degree d at time t. gefährden 'endanger': [A brings about something such that...] with respect to a domain dom, Χ is in a state 5 characterized as BEING AT RISK TO EXPERIENCE SOMETHING NEGATIVE to a degree d at time t.

I won't claim to have presented here a satisfying account of all that could be said about the resultant state of dispositonal verbs, but I assume, I have identified the parameters that are crucial for its semantics, in particular the stative argument s, the domain dom with respect to which sentences containing the verb have to be evaluated, and the degree d to which the state holds.

12

In section 4 I will argue that it is actually the comparative form of this predicate that characterizes the state, and not the positive form.

55

Stativity, supervenience, and sentential subjects

4.

Supervenience and the sentential subject of helfen

4.1. Causal dependency and evaluative dependency Having established that helfen-type verbs in their sentential variant refer to a state s which is somehow brought about by what is expressed by the subject, it remains to be seen (i) what kind of entity is denoted by the subject and (ii) what kind of relation connects this entity and the state. As to (i), restricting myself to sentential subjects, the entity in question is a proposition or - more precisely - a fact, since its truth is presupposed. In the following, I want to address the second question by showing that the relation between ρ and s involves different forms of complex reasoning. (20) a.

b. c.

d.

Dass die Organisation gebrauchte Decken und Kleider geschickt hat, hilft den Gadakhen. 'That the organization has sent used blankets and clothes, helps the Gadakhs' Dass Rebecca zu Hause geblieben ist, hat Jamaal geholfen. 'That Rebecca stayed at home, helped Jamaal' Dass Rebecca die Bretter hochgehalten hat, hat Jamaal beim Aufstellen des Bücherregals geholfen. 'That Rebecca held the boards up helped Jamaal with putting upthe bookshelf Dass es ein warmer, trockener Abend war, half den Organisatoren des Festivals 'That it was a warm, dry evening helped the organizers of the festival'

If we evaluate the truth of sentence (20a) with respect to the domain of how much the Gadakhs freeze, the following reasoning seems to be triggered: (21) a. b. c.

The organization sent used clothes and blankets. (=p) The Gadakhs froze to a certain degree for a particular time at a particular place. The Gadakhs are in a state that is better than the state they were in had ρ not been the case (with respect to the domain).

dep dep

"A dep 4- B" is to be read as "B depends on A" which should indicate an asymmetric relation involving counterfactuality. It is intended as a cover term for more specific relations. In particular, I want to suggest that the two dependency relations in (21) are different: the relation between (21a) and (21b) is one of causal dependency while the one between (21b) and (21c) is one of evaluative dependency. The intuition behind this differentiation is that the relation between (21a) and (21b) is one that is established between entities in - so to speak - the same "sphere of reality", while the one between (21b) and (21c) mediates between two different "spheres" in relating a subjective evaluation to whatever fact is being evaluated. I will not try to give an explicit semantics here for different dependency relations, but I will at least sketch what the difference between typical cases auf causal and evaluative dependency are:

56

Stefan Engelberg

(i)

Causal relations hold between events. This at least conforms to a majority view among linguists and philosophers. Thus, the relation between (21a) and (21b) can be constructed as involving a causal dependency between a sending and a freezing event. The relation involved between (21b) and (21c), on the other hand, does not seem to be one between events since (21c) does not contain any event predicates. 13

(ii)

Causal relations are based on generalizations about certain types of event cooccurrence. Whether we accept a causal statement as true depends on how we think the world works. We know that putting on clothes has a certain effect on the regulation of our body temperature and, moreover, we assume that this relation does not depend on some subjective view about the world. It is rather based on generalizations about how the occurrence of an event of one type leads to the occurrence of an event of another type. Furthermore, in typical cases of causation we are even inclined to say that it inevitably leads to the occurrence of this second event. There are no generalizations of this sort involved in the way we link (21b) to (21c) since the assertion that the world is good for the participant expressed by the dative NP, which is part of any Ae//e«-statement, involves a rather subjective judgement based on the values of the speaker. Even if we have already fixed the domain dom to 'freezing' as we did in (21), it is still a matter of judgement whether freezing is good or not. Putting on clothes when it is cold outside will reduce feelings of freezing, but whether less freezing is good - such that providing people with warm clothes can count as a help - depends on a very subjective view. Benjamin Franklin is said to have spent cold winter mornings outside undressed because he considered that beneficial for his health. Him you wouldn't have helped with a blanket.

(iii)

Causal relations typically correlate with temporal sequence. I. e., if we assume that e' causes e2 then the run time of e' precedes the run time of e2 (at least partially). This implication doesn't seem to hold for the evaluative relation underlying dispositional verbs, i. e., the time (21b) pertains to does not seem to precede the time (21c) pertains to. In fact, a temporal relation does not even seem to be at stake here.

(iv)

Causal relations form chains. The most important difference between causal and evaluative dependency is that causal dependency may lead to an unfolding of a chain of temporally and spatially connected events, e. g.:

(22) a. b. c. d. e.

13

The organization sent used clothes and blankets. ( = p ) Somebody transported the blankets to Gadakhia.

caus. caus. caus. caus.

depends depends depends depends

on on on .y),e3s,[HiGHER(5 q η H(w 0 ) * 0

The expected course of events in our world is such that a car comes to a halt if one applies the brakes. (46b) says precisely that q is an expected event if ρ is an expected event. Given the validity of sub-clause (46b), the consecutive connective so dass exhibits the semantic form (SF) in (47), and is thus also reduced to the general form of the conditional in (31). (47)

3.3.2

Consecutive: p, so dass q H(wo) = Everything which belongs to the normal course of events. SF(/.VO dass/) = λρ λq [ [Vw: H(w 0 ) A q(w)] p(w) ]

Concessive connectives

Concessive connectives behave in a similar fashion, though with some differences. Namely, these express the fact that the event designated by ρ is not expected as part of the normal course of events, if the event designated by q has taken place. Hence, if q is compatible with assumptions concerning the normal course of events, then ρ is not compatible with these assumptions. The condition in (49) says exactly this: If the intersection beween q and the background H(w 0 ) is non-empty (i.e. there is compatibility), then the intersection between ρ and H(w 0 ) is empty, since ρ is not expected if q has taken place. A condition on the use of obwohl (although) may therefore be given as (49) (cf. König 1991, 1994): (48)

Der Wagen fuhr weiter, obwohl Paul gebremst hatte. The car continued moving, even though Paul had applied the brakes.

(49)

If q, then normally not p.

It does not belong to the expected course of events that a car continues moving when the brakes are applied. (49), however, has simply the status of a condition on the use of obwohl (ιalthough). (48) expresses the fact that (49) does not obtain, for (48) says that Paul's applying the brakes did not bring the car to a halt, contrary to expectations. Applying the brakes and not bringing the car to a halt are, hence, not placed in any consecutive sequence in (48). On the one hand, then, the use of obwohl leads to a revision of what is otherwise expected and, on the other hand, serves to establish a connection between an event ρ and an unexpected alternative to q. For the sake of simplicity, I shall use the negation —>p to denote

Horst Lohnstein

130

this unexpected alternative to p. However, this captures only one of the possible alternatives. The real picture is more complicated. The reconstruction of the semantics of obwohl thus requires two pieces of information. Firstly, the expectations expressed through (49), and secondly, the revision to these expectations expressed through the use of the oiwoW-construction, one which modifies assumptions concerning the normal course of events in such a way that a standard implication between q and —>p obtains, which we once again recognize to be a variant of (31). (50)

Concessive: p, obwohl q H(w 0 ) = Everything which belongs to the normal course of events. NB {obwohl) = (49) SF{obwohl) = λρ Xq [[Vw: H(w 0 ) A q(w)] —>p(w)]

The sub-clause NB {obwohl) on the use of obwohl is (49), and the semantic form S¥{obwohl) of obwohl is identical to the semantic form of the consecutive connective in (47) - disregarding the negation. The essential distinguishing criterion, however, is in my opinion the condition (49). This short outline has sought to show that choice and parametrization of the background constitute a relevant part of the specification of the semantics of clause-connectives. The next section will show that the structure in (31) may be subject to considerably more abstract conditions.

3.4 Temporality und adversion: während

{while/whereas)

Während {while) in German normally has a temporal interpretation, under which the timespan denoted by the main clause is embedded in the time-span denoted by the währendclause. This temporal relationship is diagrammed in (51b), where the brackets correspond to the clauses in (51a). The left or right end-points of the two time-spans may coincide, though this is not necessary. (51) a. b.

Während [Otto die Suppe löffelt], {isst Fritz ein Sandwich}. While [Otto is eating the soup], {Fred eats a sandwich}. 4 f } 3 t

Other temporal connectives, such as sobald {as soon as), permit an interpretation with a similar temporal relationship, but place restrictions on the coincidence of the end-points. When sobald connects two clauses in the present tense, the two left-hand end-points ([ and {) coincide, as shown in (52): (52) a. b.

Sobald [Otto die Suppe löffelt], {isst Fritz ein Sandwich}. As soon as [Otto eats the soup], {Fred eats a sandwich}. -H ) 3 t

In contrast, if the clause introduced by the connective is in the perfect, the right-hand endpoint of the subordinate-clause time-span coincides with the left-hand end-point of the

Sentence connection as quantificational structure

131

time-span of the superordinate clause, and this point is the single element which the two clauses share: (53) a. b-

Sobald [Otto die Suppe gelöffelt hat], {isst Fritz ein Sandwich}. As soon as [Otto has eaten the soup], {Fred eats a sandwich}. Ε Η ) t

A variant of this kind is not possible with während, since there is no interaction with the tense of the connected clauses. Accordingly, the time-period over which the truth-value of the proposition dass Otto die Suppe gelöffelt hat (that Otto has eaten the soup) holds includes the time-period over which the truth-value of the proposition dass Fritz ein Sandwich isst (Fred eats a sandwich) holds. The evaluation of the main clause is thus restricted to the time-span given by the subordinate clause. This may be represented as a quantificational structure of the familiar form as in (54), where i is a variable for time-spans, and where the subordinate clause again restricts the quantifier which takes the main clause as its scope: (54)

[Vi: Suppe-löffeln(i, Otto)] Sandwich-essen(i, Egon)

Accordingly, während (while) has a lexical representation as in (55), where i is a member of the set of time-spans: (55)

^(/während!)

= λρ Xq [ [Vi: p(i)] q(i) ]

However, während may also have an adversative interpretation (cf. (56)) which is similar to the interpretation of coordinative aber (but) (cf. (57) and Clement & Thümmel 1996): (56)

Otto löffelt die Suppe, während Fritz ein Sandwich isst. Otto is eating the soup, whereas Fred is eating a sandwich.

(57)

Otto löffelt die Suppe, aber Fritz isst ein Sandwich. Otto is eating the soup, but Fred is eating a sandwich.

The question now is how this relation of contrast arises for während (while, whereas). Lang & Umbach (2002) have formulated conditions on alternatives and contrast, based on the observations of Lang (1977) concerning coordinative links. Here, two linguistic objects are said to form a contrast when they can be subsumed under a Common Integrator (CI) and when they are semantically independent, i.e. when they do not subsume each other. In concrete terms, an adversative interpretation arises, or is excluded, only under quite specific conditions. The following examples illustrate cases of excluded adversion, where capital letters indicate primary stress. (59) a. # Otto löffelt die Suppe, während Otto ein Sandwich isst. Otto is eating the soup, whereas Otto is eating a sandwich.

132

Horst Lohnstein

b. #Otto löffelt die Suppe, während Fritz die Suppe löffelt. Otto is eating the soup, whereas Fred is eating the soup. c. #Otto mag Kartoffelsuppe, während Fritz die Schuhe putzt. Otto likes potato-soup, whereas Fred is polishing his shoes. d. #HEUte löffelt Otto die Suppe, während FRITZ morgen ein Sandwich isst. Otto is eating the soup toDAY, whereas FRED is eating a sandwich tomorrow. e. #HEUte löffelt Otto die Suppe, während Fritz HEUte ein Sandwich isst. Otto is eating the soup toDAY, whereas Fred is eating a sandwich toDAY. (59a) shows that the construction of a contrast with respect to subjects fails when these subjects are identical. (59b) shows the same for predicates. The main clause in (59c) contains an individual-level predicate; the subordinate clause, a stage-level predicate. Clearly, there must be a certain equivalence in type between the predicates so that the Common Integrator is recoverable. In (59d), the constituent which bears the focal stress in the main clause is different from that which bears the focal stress in the subordinate clause. There are clearly conditions on the Common Integrator here regarding the compatibility of the focussed constituents. The focus semantic value of the one focussed constituent must be contained within that of the other focussed constituent. Furthermore, as (59e) makes clear, the focus values of the focussed constituents must be different in order for a contrastive interpretation to be induced. The focus value of a predicate such as Suppe löffeln (eat soup) is the set of all alternatives to this predicate, cf. (60a-c). But it is also necessary to take into account the types of the predicates: if they are of the same type, identical sets of alternatives result, as in (60d); if they are of different types, then distinct sets of alternatives result, as in (60e): (60) a. b. c. d. e.

|[Suppe löffeln]| F = {Supgejoffeln,Sandwich essen,Cocktail trinken,...} |[eat soup]|F = {eat soup, eat a sandwich, drink a cocktail,...} |[Sandwich essen]| F = {Sandwich essen,Suppe löffeln,Cocktail trinken,...} |[eat a sandwich]| F = {eat a sandwich, eat soup, drink a cocktail,...} |[Kartoffelsuppe mögen]| F = {Kartoffelsuppe mögen, intelligent sein,...} |[like potato-soup]| F {like potato-soup, be intelligent,...} |[Suppe löffeln]| F = |[Sandwich essen]|F |[eat soup]| F = |[eat a sandwich]| F F |[Kartoffelsuppe mögen]| * |[Sandwich essen]| F |[like potato-soup]IF Φ |[ eat a sandwich]| F

The focus semantic value | [Suppe löffeln] | F of the stage-level predicate Suppe löffeln (eat soup) is the set of all of its alternatives (cf. (60a)). For each of these alternatives, it must be the case that its focus-value gives rise to the same set of alternatives (cf. (60b) and 60d)). Since stage-level predicates and individual-level predicates have different properties, their focus values cannot coincide (cf. (60e)). As Btiring (1997) suggests, a sentence consists of a focussed part and a non-focussed part, and a contrastive topic may optionally appear in the latter. To illustrate this, let us consider the question Was macht Otto? (Was is Otto doing?), to which (61a) is an appropriate and complete answer. The bracketed expression [...]F arks the focus constituent which

Sentence connection as quantiflcational

structure

133

appears in this context and whose semantic value consists in the set of alternatives to löffelt die Suppe (is eating soup) (cf. (60a)). (61) a. b.

Otto [löffelt die Suppe,] F Otto [is eating the soup] F während [Fritz] x [ein Sandwich isst]F. whereas [Fred]T [is eating a sandwich] F .'

The additional subordinate clause in (61b) introduces a contrastive topic [,..]T which in turn points to a set of alternatives to this topic expression (cf. (62)). (62)

I [Fritz] Γ = {Fritz, Otto, Maria, Clara, Helga,...}

The topic Otto of the main clause must, on the one hand, be a member of the set of alternative topics but cannot, on the other hand, be identical to the topic of the während-clause. The sentence (63) cannot, therefore, express a contrast: (63) # Otto löffelt die Suppe, während Otto die Suppe löffelt. Otto is eating the soup, whereas Otto is eating the soup. In order to formulate the condition on the contrastive relation, let us consider the focusvalues. The focus-value of the main clause (61a) Otto löffelt die Suppe (Otto is eating the soup) is the question Was macht Otto? (Was is Otto doing?). Following Hamblin (1976), the meaning of this question can be characterized as the set of possible answers. For (61), this corresponds to the following set: (64)

|[(61a)]|F ={Otto is eating a sandwich, Otto is eating the soup, Otto is licking an ice-cream, Otto is drinking a cocktail,...}

The topic-value of the während-clause in (61b) während Fritz ein Sandwich isst (whereas Fred is eating a sandwich) is given by the alternatives for its topic position in the set of propositions in (64), so that the following set of questions results (cf. Biiring 1997): (65)

|[(61a)]|T = { {Fred is eating a sandwich, Fred is eating the soup, Fred is licking an ice-cream, Fred is drinking a cocktail,...,...} {Otto is eating a sandwich, Otto is eating the soup, Otto is licking an ice-cream, Otto is drinking a cocktail,...,...} {Mary is eating a sandwich, Mary is eating the soup, Mary is licking an ice-cream, Mary is drinking a cocktail,...} {Clara is eating a sandwich, Clara is eating the soup, Clara is licking *an ice-cream, Clara is drinking a cocktail,...} }

Combining these results, one obtains for the construction of a contrast the condition that a set of propositions from the topic-value of the während-clause must be the focus-value of the main clause. This can be stated as in (66), which says that there must be a set Q of propositions in the set of topics ρ which is equal to the focus-value of q:

134

Horst Lohnstein

(66)

[ [3Q: Q e TOPIC(p)] FOCUS(q) = Q ] (Q is a variable for sets of propositions, TOPIC(p) is the topic-value of ρ and FOCUS(q) is the focus-value of q.

The essential property of the adversative reading for the connective während appears now to consist in the fact that this connective exhibits a semantic structure which conforms to the condition of contrast in (66). If one compares the form for temporal während with the form for adversative während, one can easily see that the two are structurally isomorphous. (67) a. b.

λρ λς [ [Vi: p(i) ] q(i) ] λρ λq [ [3Q: Q e TOPIC(p) ] FOCUS(q) = Q ]

(temporal) (adversativ)

The difference between the two forms lies solely in the type of variable bound by the quantifier, which is subject to the following parametrization: In the case of the temporal reading, it is time-spans which are quantified over and, in the case of the adversative reading, it is sets of propositions which are quantified over. If one now chooses clause-combinations where the conditions on contrast do not hold, as in the examples (68a-c), then the ambiguity disappears, leaving exclusively temporal readings: (68) Fritz löffelt die Suppe, Fred is eating the soup, a. während es regnet, while it is raining.' b. während die Sitzung andauert, while the meeting is still on. c. während die Gesetze gültig sind, while the laws are still in force. Both under the temporal interpretation and under the adversative interpretation, während introduces the condition that the semantic object supplied by the main clause must be contained in the semantic object supplied by the subordinate clause. One may confirm from the forms for temporal and adversative während in (67), not only that their abstract structures are identical, but also that both can be represented as parametrised variants of the general form (31).

4.

Conclusion

In this contribution, I have presented an analysis which reconstructs the essential semantic aspects of clause-connectives in terms of a quantificational structure which places the two events signified by the respective propositions in a specific relationship. I demonstrated that clause-connectives are formed from the essentially invariant structure (31) repeated here

Sentence connection as quantificational (31)

structure

135

λρ λ ς [ [ 0 P w t: H(w 0 ) Λ p(w,t)] q(w,t)]

w h o s e individual components are subject to parametrization, thereby accounting for the variation between the connectives. For the semantic categories temporal, conditional, causal, consecutive and concessive, it was shown how this parametrization is to be undertaken in order to derive the relevant interpretive effects.

References

Biiring, Daniel (1997): The Meaning of Topic and Focus - The 59th Street Bridge Accent. - London, New York: Routledge. Biiring, Daniel & Katharina Hartmann (1997): "Doing the Right Thing." - In: The Linguistic Review 14, 1-42. Clement, Daniele & Wolf Thiimmel (1996): "Während als Konjunktion des Deutschen." - In: G. Harras & M. Bierwisch (eds.): Wenn die Semantik arbeitet. 257-276. Tübingen: Niemeyer Eisenberg, Peter (1999): Grundriss der deutschen Grammatik - Der Satz. - Stuttgart, Weimar: Metzler. Hamblin, Charles (1976): "Questions in Montague-English." - In: Β. Partee (ed.): Montague Grammar. 247-259. New York: Academic Press. Haegeman, Liliane (2004): "Adverbial clauses, the left periphery and implications for the syntax of topics." Ms. Universite Charles de Gaulle - Lille III (GURT 2004 handout). König, Ekkehard (1991): "Concessive Relations as the Dual of Causal Relations." - In: D. Zaefferer (ed.): Semantic Universals and Universal Semantics. Berlin: Foris. - (1994): "Concessive Clauses." - In: R.E. Asher (ed.): The Encyclopedia of Language and Linguistics. Oxford: Pergamon Press. Kratzer, Angelika (1978): Semantik der Rede. Kontexttheorie - Modalwörter - Konditionalsätze. Königstein/Taunus: Scriptor. - (1991): "Conditionals." - In: A. von Stechow & D. Wunderlich (eds.): Semantik. Ein internationales Handbuch der zeitgenössischen Forschung. 650-656. Berlin, New York: Walter de Gruyter. Lang, Ewald (1977): Semantik der koordinativen Verknüpfung. - Berlin: AkademieVerlag. Lang, Ewald & Carla Umbach (2002): "Kontrast in der Grammatik: spezifische Realisierungen und übergreifender Konnex." - In: Linguistische Arbeitsberichte 79, 145-186. Lewis, David (1986): "Causation. " - In: Philosophical Papers Vol II. New York: Oxford University Press. 159-240. Lohnstein, Horst (2000): Satzmodus - kompositionell. Zur Parametrisierung der Modusphrase im Deutschen. - Berlin: Akademie Verlag (= studio grammatica 49). - (2001): "Sentence mood constitution and indefinite noun phrases." - In: K. von Heusinger & Κ. Schwabe (eds.): Theoretical Linguistics 27.2/3. Special Issue "NP Interpretation and Information Structure", 187-214. - (2004): "Variable und invariante Strukturmerkmale von Satzkonnektoren." - In: H. Blühdorn, E. Breindl & U. Waßner (eds.) Brücken schlagen. Beiträge zur Konnektorensemantik. Berlin, New York: de Gruyter.

Section II:

Event nominals

Artemis

Alexiadou

Gerund types, the present participle and patterns of derivation*

This paper is concerned with certain aspects of the distribution of the -ing morpheme in English. In particular the distinct syntactico-semantic contexts in which this morpheme is found are discussed. The question to be addressed concerns the properties of these environments and how their structural built up makes reference to different types of events.

1.

Introduction

A s is well known, there are several environments in which -ing appears in English. These are illustrated below. 1 A first environment where -ing is found is the present participle, w h o s e primary function consists o f combining together with the auxiliary B E to form the progressive (1). Further functions o f the present participle include its use in reduced relatives (2), its use as the complement o f verbs like begin, perception verbs, and its use as an attributive modifier (3). (1)

John is destroying the manuscript.

(2)

People destroying forests will be punished severely.

(3)

the destroying effect o f pollution

A second environment where -ing is found is within verbal gerunds as in (4): 2 (4)

' 2

John's destroying the book annoyed everybody

Earlier versions of this paper were presented at the Workshop on Event Arguments in Syntax, Semantics and Discourse, in February 2003 at the 25th Annual Meeting of the German Linguistic Association in Munich. Many thanks to the participants of the workshop for their comments. 1 am indebted to the editors of this volume as well as to two anonymous reviewers for helpful suggestions. I do not discuss instances of -ing on e.g. adverbs and prepositions. Presumably these derive from either nominal or verbal uses of -ing. I do not discuss differences between Acc-ing and Pos,s-/«g here, (i) illustrates the two subtypes of the verbal gerund. A crucial difference between the two is that the agent bears accusative in (ia), while it is accompanied by the possessive's' in (ib): (i) a. Him collecting the money led to a disaster, b. His collecting the money led to a disaster.

Artemis Alexiadou

140

A third environment is the so called nominal gerund (or mixed nominalization), as in (5a). Examples such as the one in (5a) often seem clumsy to native speakers of English, but they seem to be acceptable especially in cases where the corresponding derived nominal also exists (5b): (5)

a. b.

John's destroying of the book annoyed everybody, John's destruction of the book annoyed everbody.

Forthly, -ing is found within nominals that in contrast to the ones in (4) and (5) do not seem to be accompanied by any NPs, as the ones in (6): (6)

a. b.

They did the destroying. a good living, a strong craving, a beating

Superficially then English seems to have at least four ing affixes: a nominal one as in (6), a deverbal one that appears in mixed nominalizations/nominal gerunds, a suffix that occurs within verbal gerunds, and finally a verbal suffix to form present participles. What is interesting about this distribution is the fact that we find the same form in distinct semantic as well as syntactic environments. Hence the question that arises is how we can explain this homophony. An additional issue that needs to be addressed concerns the exact characterization of these environments in terms of event structure in which this affix can appear. The discussion of the present paper is presented in terms of Distributed Morphology (Halle & Marantz 1993 and subsequent work); relevant assumptions drawn from this framework are presented as the analysis proceeds. The paper is structured as follows. In sections 2 and 3 I discuss in detail the differences between nominal and verbal gerunds, assuming that the properties of (1) and (6) are relatively well understood. In section 4 I proceed to my analysis of the patterns introduced in this section. Section 5 concludes my discussion.

2.

On the similarities and differences between nominal and verbal gerunds

Leaving the status of the examples in (1) and (6) aside for the moment, the distinction between the gerunds in (5a) and (4) has been the subject of some controversy in the literature (see Chomsky 1970, Houston 1985, Harley & Noyer 1998 and others). On the one hand, both gerund types occur in nominal positions, as subjects or objects of clauses. On the other hand, they both bear the same semantic relationship to the DPs that accompany them as their corresponding verbs do, e.g. John destroyed the boot, non-deverbal nominals, e.g. book, do not have such properties. That is both the verb destroy and the two nominals bearing the form destroying refer to an event of destroying, where John is the agent, and the book is the item being destroyed. Assuming that obligatory presence of internal arguments makes these nominals pattern together with Grimshaw's (1990) complex event nominals, questions concerning the exact analysis of such nominals arise. However, note that there is

Gerund types, the present participle

and patterns

of

141

derivation

a difference with respect to the obligatory presence of both arguments between the two gerunds. In fact verbal gerunds are mostly similar to verbs in that both the subject and the object are obligatory, while, as Grimshaw, citing Lebeaux (1986), notes, for nominal gerunds it is only the presence of an object that is obligatory (7): (7)

a. b. c. d.

the felling *(of the trees) they felled *(the trees) the destroying *(of the city) They destroyed *(the city)

Grimshaw (1990: 50, her (6))

To the extent that presence of argument structure is licensed in these contexts, our analysis of these patterns should be able to capture this difference. In addition to that, there are several differences between the two types that have previously been discussed in the literature, and can be summarized as follows. First, as can be seen in (5a) and (4), verbal gerunds take accusative complements, while the complements of the nominal gerunds are introduced by the preposition of. Second, verbal gerunds can be modified by (lower) adverbial modifiers. 3 However, sentence adverbs, i.e. higher adverbs, are not found within these gerunds. Most importantly, adjectival modification is not possible with verbal gerunds. In contrast, nominal gerunds take adjectival modifiers and do not license adverbial modification (8-9). (8)

a. Pat disapproved of me/my quietly leaving the room before anyone noticed. b. * Pat's fortunately collecting the money rescued the operation. c. * the carefully restoring of the painting took six months.

(9)

a. His prompt answering of the question b. * His prompt answering the question

Under the standard assumption that such adverbs are VP modifiers, while adjectives are noun modifiers, this contrast suggests that the verbal gerund contains at least a VP (Abney 1987, Kratzer 1994 and others), while the nominal gerund has a nominal internal structure. In addition the observation that higher adverbs are not licensed within verbal gerunds suggests that the internal structure of the gerund cannot be as 'big' as an IP (assuming that sentence adverbs attach to IP or other high functional heads). A third related difference is the fact that John's in (4) cannot be replaced by any determiner, while this is possible for the nominal gerund (10b): (10) a. * that/the criticising the book annoyed us b. the destroying of the manuscript annoyed the author Moreover, the gerund cannot form the plural while this is possible for the mixed nominalization:

3

The term lower adverbial modifiers refers to what is traditionally called VP modifiers.

142

Artemis Alexiadou

(11) a. many readings of the poem b. * E m m a ' s readings the poem Forth, no auxiliary can be present within nominal gerunds, while this is possible with verbal gerunds (12a): (12) a. John's having criticised the play annoyed us. b. * John's having criticised of the play annoyed us. To conclude, the differences between these two constructions suggest that there is a gradience in verbal-ness, in the sense that the nominal gerund behaves internally like a nominal, and the verbal gerund behaves internally like a verb. Due to this behavior, Chomsky (1970) was among the first to point out that the two types should not be subject to the same analysis. In the framework of that time, verbal gerunds were analysed as transformations of clauses, done in the syntax, while nominal gerunds were viewed as forms that are constructed in the lexicon. The distributional differences go hand in hand with a number of interpretive differences. In the next section I discuss these, drawing from Fraser (1970) and Zucchi (1993).

3.

Interpretive differences between nominal and verbal gerunds

Vendler (1967), Fraser (1970) and Zucchi (1993) among others have pointed out that there are subtle meaning differences between the two gerund types. In particular the verbal gerund is interpreted as having a propositional reading, while the nominal one is interpreted as having an eventive reading. Vendler referred to nominal gerunds as perfect nominals, while he referred to verbal gerunds as imperfect nominals. This meaning contrast is illustrated in the data below: (13) a. b.

John's riding of his bicycle The destroying of the building

(14) a. b.

John's riding his bicycle His figuring out the problem

(13) can be paraphrased as in (15): (15) a. b.

The activity, namely John's riding of his bicycle The act, namely, the destroying of the building

On the other hand, (14) can be paraphrased as in (16). As can be seen from these examples, the propositional use is synonymous with the use of a corresponding that-clause.

Gerund types, the present participle and patterns of derivation (16) a. b.

143

The fact that John rode his bicycle The fact that he figured out the problem.

What is the relation between the two syntactic forms, nominal and verbal gerund, and the two semantic forms, propositional and eventive (see also Zucchi 1993, Hamm 1999 for a recent discussion)? Vendler (1967) points out that the question whether we are dealing with a propositional or an eventive use of the gerund has a great deal more to do with the context in which it occurs (Vendler's container) than the syntactic form of the gerund itself. Although the verbal gerund is almost always usable in contexts in which a propositional reading is natural, and the nominal gerund is always usable in contexts in which an eventive reading is natural, the forms mix quite freely in many contexts. On the other hand, Fraser (1970: 84), as well as Zucchi (1993), argue that the differences in interpretation are made visible in a number of environments. In particular, co-occurrence restrictions are not the same for the two types, which can be explained by the fact that these two types differ in interpretation. One cannot photograph, participate in, or be present at a fact, only an activity. Conversely, one cannot write down, acknowledge or be struck by an activity, only a fact. The differences can be illustrated by the following set of examples, Fraser's (5) and (6): (17)

John's driving the car was considered a major comeback.

(18)

The breaking of the window was an act of violence.

(19) * The breaking of the window was written down (20)

John's riding his bicycle was written down.

In view of the fact that derived nominals can almost always replace nominal gerunds it seems reasonable to separate the two types semantically as well as syntactically from the verbal gerunds. As Zucchi (1993) argues extensively, the semantics of English nominalization reveals a basic ontological distinction between events, represented by derived nominals and nominal gerunds, and propositional entities represented by verbal gerunds. This is definitely a result that our morpho-syntactic approach to nominalization must express, in view also of the fact that for the purposes of morphology, as Harley & Noyer (1998) point out, the existence of a derived nominal inhibit productive nominal gerunds but not verbal gerunds. The productive form of the nominal gerund is available only when the derived nominal has a specialized meaning, as shown in (21): (21)

admiration destruction of mixture of

? admiring of ? destroying of mixing of

Mixture has the specialized meaning 'the resulting substance', hence the formation of -ing to produce the nominal meaning 'the activity o f is not blocked by mixture.

144

Artemis

Alexiadou

In the next section I come to my analysis of these patterns. Before I do that, it is necessary to introduce my basic assumptions concerning the syntax-morphology interaction in the domain of word formation.

4.

Analysing -ing

4.1. Background In recent years, and especially within the framework of Distributed Morphology (DM), (Halle & Marantz (1993), Marantz (1997), strictly transformational analyses to word formation have been re-introduced. It is argued that syntactic structure is constructed freely on the basis of abstract categories defined by universal features. In this theory, there no longer remains a Lexicon in which morpho-phonological expressions having related argument structures can be related. Importantly, unlike other theories, DM explodes the Lexicon and includes a number of distributed, non-computational lists as Lexicon replacements (narrow lexicon containing roots, Vocabulary and Encyclopedia listing special meanings of roots) see Marantz (1997), Embick & Halle (to appear). What is of interest to us here is the assumption that what we call "Lexical" categories, i.e. verbs and nouns, are reinterpreted as (category neutral) roots plus some functional layers, see Marantz (1997), Harley & Noyer (1998), Embick (2000), Alexiadou (2001), and others. In other words, the category of the neutral root is determined by its environment of insertion; if it is inserted in the domain of F, where F is verbal, the outcome is a verb; if F is nominal, then the outcome is a noun (22): (22) a.

V root = verb

Vroot = noun

The point is that different functional heads play the role of a verbalizing vs. a nominalizing functional structure see e.g. Alexiadou (2001), cf. Marantz 1997, Borer (2001): (23)

TP vP VFORM = verb form

145

Gerund types, the present participle and patterns of derivation

(24)

DP

V FORM = noun form In particular, Marantz (2000) argues that ν and η 'bring' verbal as opposed to nominal properties to abstract roots. In the domain of nominal formation, the referential index for nouns is introduced by D (Longobardi 1994). As seen above, Τ does not directly embed roots. Presumably for reasons of temporal event-structure, Τ must embed something that is an eventuality. Assuming that pieces of morphology are represented as occupying syntactic terminals, and hence that the relationship between syntax and morphology is direct, Marantz (2000) argues that heads of the same type can attach both to roots and to other functional heads, as illustrated in (25), (25) represents morphological structures: (25)

a.

VBREAK

η

b.

η

ν

My analysis of the patterns examined in the previous sections capitalizes on this intuition, see also Alexiadou (2001) for further discussion. In particular, I propose that the differences between the various -ing nominals are the result of various possibilities for attaching -ing. More specifically, -ing can attach directly to the root yielding a nominal with no event implication, or it can attach to a number of functional heads yielding a structure with event implication, and, as we will see later on, argument structure. At this stage it is necessary to present my assumptions concerning the semantic import of the various functional heads that could be involved within nominalizations. Two such heads play an important role: vP and AspectP. In the literature see e.g. Kratzer (1994), Chomsky (1995) and others, the functional category ν (or Voice or Event Phrase (Travis 1991)) is associated with the following properties: (I)

a. b. c. d. e.

is the locus of agentivity, i.e. of features relevant to the licensing and interpretation of external arguments. bears Case features for the object (Burzio's Generalization results from a and b). bears features related to eventivity. bears features related to the licensing of a manner component (manner adverbs). comes in two types: one that introduces an external argument, and one that does not. Both can be eventive.

Hence for all nominals that receive an event interpretation a vP must be present.

146

Artemis

Alexiadou

A further projection that can be present and contributes to event interpretation is AspectP. Aspect contains features that relate to the semantic properties of the event denoted by a predicate; for instance, perfective for a completed event, imperfective for an ongoing event. The presence of AspectP is linked to the licensing of lower adverbs (Alexiadou 2001). (26)

The structure in (26) is atomic in the sense that it has a certain level of independence. As a result, this domain is found embedded within different categories, verbs, participles. Moreover, variation on the type and number of projections contained within the structure of a 'word' gives us different types of nominals not only across languages but also within a language. In particular, the variation depends on (i) the number of functional projections included in the structure, i.e. whether Aspect, or ν are present or not and (ii) the type, i.e. the feature specification of the projections, e.g. whether ν is transitive or not. The number of projections concerns nominal as well as verbal projections, i.e. whether NumberP is contained or not, and whether TP is included or not. In what follows, I show how this can be used to derive the different interpretations and behavior of the -ing nominals by capitalizing on the possibility of attaching -ing high similarly to -ity in (25a) or low, similarly to -ity in (25b).

4.2. The high vs. low attachment of nominal -ing Starting with the nominals in (6) I propose that they have the following structure (the structures below are morphological structures): (27) a. b.

the reading η VREAD

ing

In this case -ing attaches directly to the root. Hence we have no event interpretation of the nominal and no arguments can be present. This follows from the independently motivated assumptions that event structure is interpreted here in terms of functional layers and it is only the presence of event structure that licenses argument positions, i.e. roots do not come with arguments. (28) illustrates the structure for nominal gerunds; table 1 relates the properties associated with them to the functional material they contain:

Gerund types, the present participle and patterns of derivation

(28) a. b.

147

the reading of the manuscript D Numb

ing VREAD Properties

nominal gerund

article

V

adjectival modification

V

adverbial modification

*

overt subject accusative object

^(gen) *

Functional Structure D + NumberP no AspectP Spec,DP ν [-ext arg]

Table 1

As can be seen from the table, ν is present but it does not come with an external argument. I assume, following crucially Abney (1987), that John in John's reading of the manuscript is generated in Spec,DP, which is a thematic position in English. Evidence for the view that adverbial (manner) modification requires the presence of Aspect comes from -er nominals. -er nominals do not contain Aspect, but do include a v, and hence a manner component; see Alexiadou (2001) for details. Consider (29): (29)

the beautiful dancer

(29), on one reading, means the person that dances beautifully (Larson 1998 for discussion). VBEAUTIFUL can be included in the structure. In the environment of functional layers that include Aspect, it will be spelled out as an adverb. If Aspect is not included, it will be spelled-out as an adjective. Turning to verbal gerunds, van Hout & Roeper (1998) proposed to consider that -ing is basically an Aspectual head, exactly the one that we find with the participle. This affix then raises to a nominal head, as illustrated in (30). The main argument for this analysis is that gerunds, much like the participle, refer to ongoing events; cf. (30). However, this account misses an important similarity between the gerund and the other types of nominals, as one could not possibly argue that the nominals in (6) and (4a) receive an aspectual reading. Moreover, as is well known the progressive cannot be formed for Stative verbs. In contrast we can form the gerund of a stative verb, as shown in (31).

148

Artemis

(30)

Alexiadou

Ν' Ν -ing

TP ^ ^ ^ ^ Τ AspP Spec

Asp' AspP

EventP Spec

Event' Event

VP

(31) a. * I am knowing the answer b. His knowing the answer Furthermore, dialects of English distinguish between the two -ings. In particular older speakers of Scots still differentiate the present participle and the gerund (verbal noun). In older Scots the present participle was written /an(d)/ and the gerund / i n / . These pronunciations are still used. (32) a. b.

Present Participle: He wis aye stravaigan aboot. He was always roaming around, Gerund: He's fond ο stravaigin aboot. He likes roaming around.

Hence we have evidence coming from two areas that the gerund -ing is a nominal head. Since, however, gerunds do refer to ongoing events and allow adverbial modification, we do have evidence for the presence of AspectP within the verbal gerund. (33) illustrates the structure assumed for verbal gerunds:

Table 2 relates the functional material contained in (33) to the properties associated with verbal gerunds:

Gerund types, the present participle

Properties

149

and patterns of derivation

verbal gerund

Functional Structure

article

*

D contains's', no NumbP

adjectival modification

*

No NumbP

V

AspectP

adverbial modification overt subject accusative object

^ (gen) V

SpecJDP movement from Spec, vP ν [+extemal argument]

Table2 It appears to be the case that a crucial difference between the two gerund types concerns the fact that nominal gerunds contain number, while verbal gerunds lack number. Moreover, the latter contain Aspect, while the former lack Aspect. Note here that while pluralization is not possible for the verbal gerund, it is possible for the nominal one, e.g. *Johrt's readings

the book vs. John's readings of the book. This analysis captures the different distribution and interpretation of the nominal and the participle in English. What has not been answered thus far is the propositional reading related with the verbal gerund. The suggestion made in Chomsky (1970) that gerunds contain a sentence level, i.e. a TP, cannot be maintained in view of the fact that modal and speaker oriented adverbials cannot be contained within the verbal gerund. An option would be to assume that the propositional reading is not linked to a terminal in the syntax. On this view, the reading associated with the verbal gerunds is thus a result of the combination D + nP, much like D + CP gives rise to propositional/fact readings across languages. There is nothing in the syntax that encodes that, but the interpretative component gives this meaning to this particular construction. This structure is not available in the case of nominal gerunds, as in this case NumbP intervenes, blocking the factual interpretation.

4.3. The participial -ing Given what I have said so far, I cannot assume that the participial -ing is of the same type as the nominal one. (34) demonstrates the participial structure and the table in 3 relates the properties associated with the participle to the functional categories it contains.

Artemis Alexiadou

150 Properties

participle

fimctional structure

article

*

noD

adjectival modification

*

no NumbP

adverbial modification

V

AspectP

overt subject

V

vf+extemal argument]

accusative object

V

ν [+extemal argument]

tense/agreement

*

no TP

Table3 The structure of the Modern English progressive contains a version of (34), which is actually an adjectival construction embedded under BE. As is clear from the above representations I assume that we have two -ings: the aspectual one, and the nominal one. These two have different distribution and different meaning. It so happens that it is a historical fact about English that in the standard dialect the two have the same form. In particular, Old English ended in -ende has already become -inde in late West Saxon. Then the final /e/ or schwa got lost in levelling. By the mid 15th century, -ind has been replaced by -ing in the south of England. As a result, a form like irnende is now running (cf. Houston 1991, Moore, Meech & Whitehall 1935). Because of these morphophonological changes, the participial and the nominal constructions contain identical forms.

4.4. Realisation of arguments As has become evident from the approach outlined here the properties of 'argument realisation' are attributed solely to functional layers. What this means is that the presence of an argument in the context of word formation is not a result of inheriting the argument taking properties of the corresponding verb. Rather the presence of certain functional layers leads to the projection of arguments. This perspective is in general agreement with the view that lexical representations are not sufficient to account for argument structure (e.g. Borer 2001). The intuition here is that the syntactic structure gives rise to a template which in turn determines the interpretation of arguments. This view can be schematically summarized as in (35). (35)

syntactic structure —• event structure —> interpretation of arguments

(35) also says that the presence of a special type of functional layers is responsible for the event reading that certain nominals have. Here the assumption is that these functional layers represent a set of abstract features that relate to event semantics. Crucially, the presence of the arguments will be licensed by this event structure. Most importantly the presence of ν (independently of its type, +external argument or -external argument) brings about an internal argument. This view contrasts with a more lexicalist approach to event structure, as shown in (36) from Alexiadou et al. (2004:11):

Gerund types, the present participle and patterns of derivation

(36)

151

Lexical Semantics —• Predicate Argument Structure —> lexical/syntactic structure

Rather the view followed here sees event structure as belonging to syntax proper, i.e. event interpretations arise only in the presence of a number of functional layers that are related to event readings. Note that my analysis of nominalizations does not make reference to any event arguments of the type discussed in Grimshaw's work. In fact presence of temporal/eventive interpretation is not sufficient to explain the properties that characterize the group of complex event nominals, e.g. nouns such as trip and event have event readings, but do not license argument structure, (see also Borer 2001). Hence the relevant distinction between different types of nominals is based on the presence or absence of argument structure, interpreted here in the sense of (35).

5.

Conclusion

In this paper I discussed the fact that English seems to have at least four ing affixes: a nominal one, a deverbal one that appears in mixed nominalization, a suffix that occurs within gerunds, and finally a verbal suffix to form present participles. I pointed out that we find the same form in distinct semantic as well as syntactic environments, and I described the semantic differences in detail. I proposed that while the nominal and gerund suffix are really nominal, the one found in the participle is not, but it is rather aspectual in nature. Hence we have two -ings, one in the nominal domain, and one in the participial domain.

References

Abney, S. (1987): The English noun phrase in its sentential aspect. - Doctoral dissertation, MIT. Alexiadou, A. (2001): Functional structure in nominals: nominalization and ergativity. - Amsterdam: John Benjamins. Alexiadou, A, E. Anagnostopoulou & M. Everaert (eds.) (2004): The unaccusativity puzzle: explorations of the syntax-lexicon interface. - Oxford: Oxford University Press. Borer, H. (2001): The form, the forming and the formation of nominals. - Ms, USC. Chomsky, N. (1970): "Remarks on nominalization." - In: R. Jacobs & P. Rosenbaum (eds.): Readings in English Transformational Grammar. 184-221. Waltham, Mass.: Ginn and Company. Chomsky, N. 1995. The Minimalist Program. Cambridge: MIT Press. Embick, D. (2000): "Features, Syntax and Categories in the Latin Perfect." - In: Linguistic Inquiry 31, 185-230. Embick, D. & M. Halle (to appear): Word formation: aspects of the Latin conjugation in Distributed Morphology. - Berlin: Mouton de Gruyter. Fräser, B. (1970): "Some Remarks on the Action Nominalization in English." - In R. Jacobs & P. Rosenbaum (eds.): Readings in English Transformational Grammar. 83-98. Waltham, Mass.: Ginn and Company. Grimshaw, J. (1990): Argument Structure. - Cambridge, Mass.: MIT Press.

152

Artemis

Alexiadou

Halle, M. & A. Marantz. (1993): "Distributed Morphology and the Pieces of Inflection." - In K. Hale & S.J. Keyser (eds.): The View from Building 20. 111-176. Cambridge, Mass.: MIT Press. Hamm, F. (1999): Modelltheoretische Untersuchungen zur Semantik von Nominalisierungen. - Habilitationschrift, University of Tübingen. Harley, H. & R. Noyer. (1998): "Mixed nominalizations, short verb movement and object shift in English." - In Proceedings of NELS 28. Houston, Α. (1985): Continuity and change in English morphology: the case of variable -ing. Doctoral dissertation, University of Pennsylvania. - (1991): "A grammatical continuum for ING." - In: P.Trudgill & J.K. Chambers (eds.): Dialects of English. 241-257. Longman: London. Hout, A. van & T. Roeper. (1998): "Events and Aspectual Structure in Derivational Morphology." In: MIT Working Papers in Linguistics 32, 175-220. Kratzer, A. (1994): "The event argument and the semantics of Voice." Ms., University of Massachusetts at Amherst. Larson, R. (1998): "Events and modification in nominals." - In: D. Strolovitch & A. Lawson (eds.): Proceedings from Semantics and Linguistic Theory (SALT) VIII. Cornell University, Ithaca, NY. Lebeaux, D. 1986. "The Interpretation of derived Nominals." - In: Proceedings of CLS 22, 231-247. Longobardi, G. (1994): "Reference and proper names: a theory of N-movement in syntax and LF." In: Linguistic Inquiry 25, 609-665. Marantz, A. (1997): "No escape from Syntax: don't try morphological analysis in the privacy of your own lexicon." - In: A. Dimitriadis et al. (eds.): University of Pennsylvania Working Papers in Linguistics, vol. 4.2, 201-225. Proceedings of the 21st Annual Penn Linguistics Colloquium, 1997. - (2000): Roots and Derivational Morphology: Recapturing the Lexical/Syntactic Distinction. - Ms. MIT Moore, S. & S.B. Meech & H. Whitehall. (1935): Middle English dialect characteristics and dialect boundaries. - University of Michigan Language and Literature Studies. Travis, L. (1991) "Inner Aspect and the Structure of VP." Ms., McGill University. Vendler, Z. (1967): Linguistics in Philosophy. - Ithaca: New York: Cornell University Press. Zucchi, A. (1993): The Language of Propositions and Events. - Dordrecht: Kluwer.

Ingrid Kaufmann *

Referential arguments of nouns and verbs

Since Grimshaw (1990), it has been generally assumed that argument realization is associated with the event structural properties of event nominalizations. In this paper, I take up Grimshaw's original assumption that these properties are related to the status of the referential argument of event nominalizations. Based on a corpus study of German nominalized infinitives (NIs), I claim that argument realization patterns cannot be explained by the assumption that complex event nouns (CENs) inherit the argument structure of the base verb, while simple event nouns (SENs) do not. Whether arguments are to be realized or not, instead depends on whether the event nominal is used referentially or non-referentially. I argue that referentially used event nominalizations are functional nouns in the sense of Löbner (1985), while generically used event nominalizations refer to event-types. The observation that SENs, but not CENs may be used referentially although no argument is realized, follows from their referential properties: as nouns in general, SENs can be used as count nouns, whereas CENs cannot. Adopting the analysis of Chierchia (1998), I propose that referentially used NIs are derived from kind-referring NIs by 'predicativization', i.e. by an operation that derives the property of being an instance of the relevant event-type.

1.

C o m p l e x e v e n t n o u n s and s i m p l e e v e n t n o u n s

Following Grimshaw (1990), most current analyses of event nominals distinguish between nominalizations that allow or even require the realization of arguments, and those that do not. Grimshaw observed a correlation between argument realization and a number o f other grammatical properties. Some o f these are related to the referential properties o f event nominals. According to Grimshaw, event nominals that require the realization o f the internal argument o f the base verb may not occur with the indefinite article and do not pluralize (see (1)). Furthermore, aspectual adjectives like frequent and constant can modify the singular form of the nominal. (1)

a. The shooting o f rabbits is illegal. b. * The shooting is illegal. c. * a shooting of rabbits d. * the shootings o f rabbits e. The frequent shooting o f rabbits is illegal.

The ideas presented in this paper were developed in the project "Valenz im Lexikon" which was part of the research program "Theorie des Lexikons" (SFB 282), supported by the German Science Foundation (DFG). I would like to thank Kerstin Blume, Miriam Butt, Veronika Ehrich, Stefan Engelberg, Joachim Jacobs, Anja Latrouite, Sebastian Löbner, Judith Meinschäfer, Albert Ortmann, Irene Rapp, two anonymous reviewers and the editors of this volume for discussion and helpful comments.

154

Ingrid

Kaufmann

In contrast, as illustrated in (2), event nominale that do not require argument realization occur with the indefinite determiner and pluralize, and aspectual adjectives like frequent can modify only the plural form of the nominal. (2)

a. The examination took a long time. b. an examination c. the examinations d. * The frequent examination is annoying e. The frequent examinations are annoying.

Grimshaw suggested that these properties are connected with the referential properties of the event nominals. According to her analysis, some event nominalizations inherit the argument structure of the verb, including the event argument Ev} These nominals refer to complex events ('complex event nominals', CENs) and share the aspectual properties of the base verb. Grimshaw contrasts CENs with another type of event nominals that do not inherit the thematic arguments of the verb. The referential argument of these nominals is not an event argument Ev, but the ^-argument of count nouns. 2 Depending on their interpretation, Grimshaw refers to these nominals as 'result nominals' (RN) or 'simple event nominals' (SEN). RNs differ from both SENs and CENs in that they denote objects, not events (e.g., The examination/exam was long/on the table). I will not be concerned with RNs in this paper. According to Grimshaw, event denoting nominals can either be unambiguously classified as CENs or SENs, or allow either reading. English /wg-nominalizations are CENs, while root nominalizations such as walk, talk are SENs. Affixes like e.g., -ation, -ment are ambiguously specified as introducing either Ev- or Ä-arguments. Nominals derived by these affixes thus can be either CENs or SENs. One of the problems of Grimshaw's analysis is that the data are not as clear as her classification suggests, since argument realization is very flexible. Whether arguments are realized or not depends largely on the context. Although some of Grimshaw's data have been disputed, there is a consensus that with respect to event structure and argument realization, two classes of event nouns are to be distinguished. However, her assumption that the differences in event structure are related to the nature of the referential argument has been abandoned, presumably because it is not clear what kind of difference is encoded by Ev and R. After all, both SENs (i.e. event nominals with an Λ-argument) and CENs (i.e. event nominals with an £V-argument) refer to events, whatever their grammatical properties are. Instead, the verbal properties of CENs are often taken as evidence for a syntactic derivation, while the nominal properties of SENs are explained by their lexical character. Borer (1997:13), for instance, points out that Grimshaw's CENs are always deverbal, while there are at least some non-derived SENs (e.g., trip, race, journey, event). Borer takes this as evidence that a VP projection must be part of the representation of CENs (thus modelling the verbal character), but not of SENs, given that they do not necessarily have to be based

2

Grimshaw assumes that the argument structure inherited by event nominals is that of passive forms, i.e. the external argument of the verb is implicit. Grimshaw refers to R and Ev as external arguments rather than referential arguments. I adopt Bierwisch's (1990) position that in the case of nouns the external argument and the referential argument are not distinguished.

Referential arguments of nouns and verbs

155

on verbs. In her view (as in many other analyses), the status of the referential argument is not relevant for the syntactic properties of the nominal. In this paper, I take up Grimshaw's distinction of R- and AVevent nominals and suggest an ontological explanation of the correlation between the nature of the referential argument on the one hand, and the necessity of argument realization on the other. Note, however, that the distinction of R- and £V-event nominals that I propose does not correspond to Grimshaw's distinction of CENs and SENs. Instead, I argue that the nature of the referential argument is determined by the derivational process. /?-event nominals are genuine nouns, derived by nominal affixes or root nominalizations. £V-event nominals are verbal nouns, which inherit the referential properties of their verbal base. My distinction between R- and £V-nominals thus corresponds to Borer's distinction of lexically and syntactically derived event nominals, although my assumptions about argument realization differ from Borer's. The crux of the explanation presented here is as follows. A comparison of the relations that hold between the referents of nominal and verbal expressions and the referents of their thematic arguments reveals a crucial difference between nouns and verbs. Verbs characterize events that involve participants. No individual event exists independently of its participants, since the participants are the constituting parts of the event. The identification (or spatiotemporal individuation) of a particular event thus depends on the identification of the participants involved. I assume that this dependency has an ontological consequence, namely, that there is no given domain of individual events from which the referent of a verbal expression can be chosen. Instead, events are construed as soon as their defining components are identified. Prototypical nouns, in contrast, presuppose the existence of their referents, i.e. there is a domain of individuals from which the referent of a nominal expression can be chosen. Since even the referents of relational nouns such as father of, sister of are not constituted by the referents of their thematic arguments, there is no (ontological) dependency relation. Depending on whether an event nominalization is categorized as verbal or nominal, it is subject to the ontological conditions that hold for its category. 3 Taking up Grimshaw's initial idea, I therefore assume that /?-event nominals share the referential properties of nouns due to their categorial classification, while £V-event nominals share the referential properties of verbs, although there is no 'real world' difference between the referents of R- and £V-nominals. The aim of this paper is two-fold. On the one hand, I want to establish the difference between Ev- and Λ-event nouns, on the other, I want to show that the argument realization of £V-event nouns is actually determined by their referential properties instead of their argument structural properties. Once the interaction of the referential properties and argument realization is established, the different behaviour of £V-event nominals and Λ-event nominals with respect to argument realization follows. The claim is based on a corpus analysis of nominalized infinitives in German, which are £V-event nominals according to Grimshaw's criteria. The paper is organized as follows. Section 2 first lays out the relevant properties of deverbal nominalizations in German and places them in the context of the classification 3

Given that in the real world there is no difference between events referred to by verbal or nominal event nominals, the distinction is conceptually motivated rather than ontologically, as far as event nominals are concerned. However, since the ontological differences between the referents of nouns and verbs motivate the distinction, I will keep the term.

156

Ingrid Kaufmann

proposed for English by Grimshaw. Section 3 discusses the generalizations which emerge from a corpus study of German nominalizations and shows that argument realization of German nominalized infinitives (NIs) is determined by semantic and pragmatic conditions rather than by argument structural requirements. Different conditions apply when NIs are used referentially or non-referentially. In section 4, it is argued that the conditions on argument realization can be accounted for if a distinction is made between NIs that refer to specific situations and NIs that refer to situation types. The former can be classified as functional nouns in the sense of Löbner (1985). For these NIs, argument realization is necessary in order to identify their referent. Section 5 deals with the relation between the two uses of NIs. Adopting the analysis of Chierchia (1998), I suggest that bare NIs refer to event-types and that functional NIs are derived from them by 'predicativization'. Section 6 concludes this paper.

2.

German nominalization patterns

In German, three main types of event nominalizations can be distinguished: nominalized infinitives, which roughly correspond to English /«g-nominalizations, nominalizations derived by nominal affixes (e.g., -ung), which correspond to English nominalizations derived by -ation, -ment, and, finally, root or stem nominalizations such as Feier (verb stem: feier) 'celebration', Fall (verb stem: fall) 'fall'. According to Grimshaw's criteria, root nominalizations are either SENs or RNs, but never CENs. As illustrated in (3) they can occur with the indefinite article, they pluralize, they do not allow the realization of the object argument of the base verb, and modifiers like permanent can only occur with the pluralized noun. (3)

a.

Das war eine schöne Feier! 'That was a nice celebration!' b. Nächste Woche wird es drei Feiern geben. 'Next week there will be three parties.' c. * die Feier ihres Geburtstags 'the celebration of her birthday' d. die ständigen Feiern/*die ständige Feier 'the permanent parties/the permanent celebration

In this paper, I am not concerned with root nominalizations, but focus on nominalized infinitives (NIs), although wwg-nominalizations are referred to occasionally. Similar to English /wg-nominalizations, German NIs preserve the verbal infinitive morphology {-en). No nominal affix is involved in the derivation. I argue in section 4 that the preservation of the infinitive morphology is crucial for the grammatical behavior of Evevent nominals. As illustrated in (5), wrcg-nominalizations and NIs differ as to whether argument realization is required or not. Durchsuchen 'search through' is a verb with an obligatory object (see (4a,b)). While without any further context the NI Durchsuchen is

Referential

arguments

of nouns and

verbs

157

highly marked if the object of the base verb remains implicit (5a), Durchsuchung can be realized with or without argument (5b). (4)

a.

Die Polizei durchsuchte die Räume unter strengen Sicherheitsvorkehrungen. 'The police searched the premises with strict security measures.' b. # Die Polizei durchsuchte unter strengen Sicherheitsvorkehrungen, the police searched with strict security measures

(5)

a.

b.

Das Durchsuchen #(der Räume) fand unter strengen Sicherheitsvorkehrungen statt.4 the searching (of the rooms) took place under strict security measures 'The search (of the premises) took place under strict security measures.' Die Durchsuchung (der Räume) fand unter strengen Sicherheitsvorkehrungen statt. 'The search (of the premises) took place under strict security measures.'

Furthermore, NIs behave as English /«g-nominalizations in that they have no plural but allow the combination of the singular noun with modifiers like permanent, frequent etc. (see (6a,b)). 5 t/«g-nominalizations, in contrast, may behave as SENs or CENs according to Grimshaw's classification: the object of the base verb may remain implicit, pluralization is possible, and only the pluralized noun can occur with modifiers like permanent unless the argument is realized (see (6c,d)). 6 (6)

a. * die Durchsuchen the searchings b. das ständige Durchsuchen der Räume 'the permanent searching of the premises' c. die Durchsuchungen 'the searches'

In the examples, nominalizations are marked with italics, arguments of nominalizations are under5

6

lined. Note that for most NIs, there is not even a morphological plural. NIs that are lexicalized as result nominale, however, can be pluralized, e.g., die Essen 'the dinners', lit. 'the eatings', die Rennen 'the races', lit. 'the runnings'. Ehrich (2002) points out that German wwg-nominalizations never require argument realization and do not permit event control. She therefore questions whether wrcg-nominalizations are to be classified as CENs at all.

158

Ingrid

d.

Kaufmann

die ständigen Durchsuchungen ("der Räume) die ständige Durchsuchung #(der Räume) 7 'the permanent searches of the premises/the permanent search of the premises'

The examples above show that - at least out of context - NIs of obligatorily transitive verbs require the realization of the object of the base verb while wwg-nominalizations do not. However, as will be shown in the following section, German NIs do not behave as homogeneously with respect to argument realization as one would expect of CENs according to Grimshaw's original classification.

3.

Argument realization of nominalized infinitives in German

This section lays out the patterns of argument realizations found for NIs in German. The examples are taken from the COSMAS corpus of the 'Institut fur deutsche Sprache' (IdS) in Mannheim. 8 I have collected NIs of 110 transitive and intransitive verbs. Among the intransitive base verbs are unergatives and unaccusatives, with human as well as nonhuman subjects. Depending on the base verb, the frequency of the NIs differs largely. For some NIs there are more than 10,000 instances (e.g., Erkennen 'recognizing', for others there are less than 10). In their diversity, the examples given here are representative of the overall patterns that were found.

3.1 Data As the examples in (7)-(9) and (10)-(12) show, NIs may appear with and without arguments, contrary to what is generally expected with respect to CENs. Whether an argument is realized or not does not depend on the arity of the base verb or - i n the case of intransitive base verbs- on whether the argument is external or internal. Examples with two realized arguments are very rare (less than 0.5%) and are not shown here.

7

An anonymous reviewer asked why die ständige Vertretung lit. 'the permanent representation', i.e. 'the embassy' is acceptable. As far as I can see, Vertretung is a result noun here ("the bureau of the diplomats who permanently represent X"). Since the construction is not acceptable when permanent is replaced by an other adjective (#die häufige Vertretung 'the frequent representation') I assume that ständige Vertretung is lexicalized.

8

COSMAS provides different kinds of text archives. The archive that I used comprises written data from Austrian, German and Swiss newspapers (494.230.731 words, 272.115.224 words and 161.129.048 words, respectively), some novels and some other sources. See www.idsmannheim.de.

Referential arguments of nouns and verbs

159

(7)

NIs with one argument: base verb unaccusative a. Jedermann ist es freigestellt, auf das Steigen und Fallen einer Aktie zu spekulieren. 'Everyone is free to speculate about the rise and fall of a share.' b. Er bedauerte sein Ausscheiden aus dem aktiven Dienst [...]. 'He regretted his standing down from active duty.' c. Der Unfallmechanismus ist dabei jeweils ein harmloser Sturz, ein Stolpern des Trägers, dem eine unsanfte Stauchung der Wirbelsäule durch die Last nachfolgt. 'The course of the accident is always a harmless fall, a tripping of the carrier, for whom an uncomfortable sprain of the spine through the burden follows.'

(8)

NIs with one argument: base verb unergative In ihrem Versteck hörte sie das Schreien der Menschen und die Schüsse. a. 'In her hiding place she heard the screaming of the people and the shots.' Durch das Weinen eines der Jungen aufmerksam geworden, schritt ein Zeuge b. gegen den deutlich alkoholisierten Mann ein. 'Alerted by the crying of one of the boys, one witness took steps against the clearly alcoholized man.' Im sonst so betulichen Reichstag ist nicht einmal das Atmen der Abgeordneten zu vernehmen. 'In the otherwise so leisurely Reichstag not even the breathing of the delegates can be heard.'

(9)

NIs with one argument: base verb transitive a. Beim Aufbrechen der Kasse blieb der 22-Jährige mit den Fingern stecken. 'During the breaking open of the cash register, the 22-year old got his fingers stuck.' Die Besonderheit dieser Technik liegt im schnellen Brennen des glasierten Gegenstandes. 'The speciality of this technology lies in the quick burning of the glazed item.' Darin mahnt das Amt, "stärker auf das regelmäßige Reinigen der Gehwege zu c. achten". 'In this the council reminds (the community), "to take greater care of the regular cleaning of the sidewalks".'

(10) NIs without arguments: base verb unaccusative a. Da ist zuallererst schon das richtige Fallen wichtig (...) 'There, first of all the right (kind of) falling is important.' b. M B B in München sichert bei Kündigung zum Ende des Erziehungsurlaubs innerhalb von drei Jahren nach dem Ausscheiden das Recht auf Wiedereinstellung [...]zu. ' M B B in Munich guarantees the right to be re-employed within 3 years after leaving in case employment was terminated towards the end of the extended maternity leave.'

160

Ingrid

c.

Kaufmann

Aber Vorsicht! Absichtliches Stolpern wird leicht als sexuelle Belästigung ausgelegt! 'But beware! Purposeful tripping can easily be interpreted as sexual harassment.'

(11) NIs without arguments: base verb unergative a. Ein paar Tage nach der Geburt kam ein Hirte auf der Suche nach einem ausgerissenen Schaf in die Nähe des Poyas und hörte das Schreien. Ά few days after the birth, a shepherd looking for a fugitive sheep came into the vicinity of the poya and heard the screaming.' b. Boulevardzeitungen hatten das Weinen auf Dianas nervliche Belastung zurückgeführt. 'The yellow press had attributed the crying to the pressure of Diana's nerves' c. Gläserne Stille trat ein, nur noch das Atmen war zu hören. 'Glassy silence fell, only the breathing was to be heard.' (12) NIs without arguments: base verb transitive a. Der Austausch von alten Heizkörpern gegen moderne erleichtert der Hausfrau auch die tägliche Arbeit durch bequemeres Reinigen. 'The exchange of old heaters for modern ones eases the house wife's daily chores through more comfortable cleaning.' b. Die Schlösser wurden in einen normalen Sicherheitsbeschlag - ein äußeres Metallschild, das ein Schloss gegen gewaltsames Aufbrechen schützen soll - eingebaut. 'The locks were built into normal security plating - an external metal plate that is supposed to protect a lock against violent breakage.' c. Denn der Werkstoff Ton ist gefahrlich: Leicht modellierbar und ohne Schwierigkeiten durch Brennen zu konservieren, ist er auch vom Laien zu bewältigen. 'Because the material clay is dangerous: easy to form and to conserve by burning without difficulties, it can also be worked with by the lay person' If the base verb is transitive, in most of the cases either the object of the base verb is realized or no argument is realized at all. 9 There are some NIs, however, that may occur with the subject of the base verb. The base verbs of these NIs are e.g., verbs of emotion like love and hate.m (13) a.

9

10

Erstaunlich, in welchem Maße das Leben und Lieben (und Hassen) des englischen Königshauses [...] dem Operngenre Stoff geliefert hat. 'Amazing to what extent the living and loving (and hating) of the English royal family [...] has provided material for the opera genre.'

Although there are differences between the individual NIs, the cases in which the object of the base verb is realized make up approximately two thirds of the examples. Other NIs of transitive verbs that occur only with the subject argument are Ansprechen 'addressing', Angreifen 'attacking\ Akzeptieren 'accepting' and Erkennen 'recognizing'.

Referential arguments of nouns and verbs b.

161

Die archaischen Geschichten erzählen vom Kampf zwischen Mutter- und Vatergöttern. Das verzweifelte Lieben und Hassen eines Vater-Tochter-Paares schildern [...] 'The archaic stories tell of the battle between maternal and paternal gods. The desperate loving and hating of a father-daughter pair are described [...]'

These data show that the argument realization of German NIs cannot be derived solely by the subcategorization properties of the base verb. The next section goes on to formulate the conditions which are relevant for the (non-)realization of arguments.

3.2 Conditions on argument realization Whether an argument is realized or not depends on a number of factors. The first criterion is whether the NI is used referentially or non-referentially. In the following, I talk of a 'referential use' if the nominal refers to an individual event, and of a 'non-referential use' otherwise. Since the conditions on argument realization of NIs of transitive verbs are more complex, I restrict the discussion to NIs of intransitive verbs. The behavior of transitive verbs follows from the same analysis. In the majority of the cases with non-referential uses of NIs, no argument is realized. Non-referentially used NIs can be predicative or generic. On the assumption that generic NIs refer to event-types (as proposed by Chierchia 1998, see section 5), argument realization can be regarded as a means to restrict the reference to a more specific event-type. This assumption provides a natural explanation for the realization of the experiencer instead of the theme in (13): an event-type that abstracts over all situations in which a particular person has a certain emotion seems to be at least as natural as an event-type of having a certain emotion towards a particular individual. There are two ways in which an event-type can be restricted by argument realization. In both cases, the properties of the participants are involved. Firstly, the reading of NIs that allow arguments with different selectional properties can be specified ('conceptual specification' in the sense of Bierwisch (1982), e.g., das Heulen des Windes 'the howling of the wind' vs. das Heulen der Wölfe 'the howling of wolves') or even disambiguated (e.g., das Anziehen der Hose 'the putting on of trousers' vs. das Anziehen der Bremse 'the applying of the brake' vs. das Anziehen der Preise 'the rising of the prices'). Secondly, in the case of NIs that select or strongly prefer only one type of argument (e.g., Klagen 'moaning' of humans), the event-type can be further restricted by narrowing down the set of possible participants to a certain subset (e.g., das Klagen der Frauen 'the moaning of the women', das Klagen der Bürokraten 'the moaning of the bureaucrats'). In the first case, argument realization serves to exclude competing event-types, while the specification of the participants in the second case narrows down the event-type. Examples in which the argument realization excludes a competing interpretation are given in (14a) and (15a), where the default reading would be that the breathing is done by some human, although semantically other readings are also available. As shown in (14a'), the default reading arises if the argument remains implicit. Examples in which the default reading is narrowed by a specification of the participants are given in (14b) and (15b).

162

Ingrid Kaufmann

(14) a.

b.

(15) a. b.

Sie passen sich den Körperformen besonders gut an und erleichtern auch das Atmen der Haut [...] 'They adapt particularly well to body forms and also ease the breathing of the skin [...].' Sie passen sich den Körperformen besonders gut an und erleichtern auch das Atmen. 'They adapt particularly well to body forms and also ease the breathing.' [...] eine Sensorschicht, in der [...] Elektromagneten sogar das Atmen von Personen registrieren, die auf dem Boden liegen. '[...] a sensory layer in which [...] electromagnets even register the breathing of people who are lying on the floor.' Dies ist f...] das Ein- und Ausatmen der Welt, in der wir leben [...]. 'This is [...] the breathing in and out of the world in which we live [...]' Heute sind die Grabschätze nur noch im Museum zu sehen, weil sie [...] vor der Feuchtigkeit, die beim Ausatmen der Besucher entsteht, geschützt werden müssen. 'Today the tomb treasures can only be viewed in the museum, because they [...] have to be protected from the moistness which arises from the breathing of the visitors.'

If the NIs are used referentially, the argument is realized in the majority of the cases. There are two exceptions, however. First, the argument can remain implicit if the participant has been mentioned in the immediate context, as in (16). In (16a), the context is that the political FDP party failed to make the crucial 5% necessary to remain in government. The high topicality of the participant 'FDP party' is preserved by the mentioning of the FDP delegates in the sentence preceding the NI, so that the relevant argument of Fallen 'falling' can be identified as the FDP party. In (16b), it is clearly the team mentioned in the previous clause whose elimination (.Ausscheiden) took place." (16) a.

11

Als die FDP-Abgeordneten vor vier Jahren ihre Landtagsbüros räumten, sollen Tränen geflossen sein. So niederschmetternd war das Fallen unter die FünfProzent-Marke. 'When the FDP delegates packed up their congressional offices four years ago, tears are supposed to have flowed. So deep a blow was falling under the five percent figure.'

High topicality is not the only factor that is relevant for the contextual identification of a participant. Besides world knowledge, information structure presumably plays an important role in that the information about the individuals already mentioned in the previous context, and their involvement in other events, helps to identify the relevant participant.

Referential arguments of nouns and verbs b.

163

Sicherlich gut, daß die Mannschaft [...] noch einmal eine hervorragende Leistung brachte und die natürliche Enttäuschung über das Ausscheiden doch noch etwas milderte. 'Probably good that the team [...] managed an outstanding performance once more and was able to somewhat alleviate the natural disappointment about the elimination.'

Second, arguments of NIs of verbs of emission need not be realized, even if no participant can be identified, as in (17). Verbs of emission are special in that the event can be perceived (and thus spatiotemporally individuated) even if the participants involved in the event are not identifiable. As such, neither explicit realization of arguments need take place, nor has the participant to be mentioned in the context. 12 (17) a.

b.

c.

Als eine Funkstreife der Polizei gegen 3.40 Uhr [...] ein verdächtiges lautes Knacken vernahm, schöpfte sie Verdacht. 'When a police patrol heard a suspicious loud cracking around 3.40 a.m., they became suspicious.' Das Klirren und Dröhnen, die lauten Schreie, Anweisungen und Werftsirenen sind kaum noch zu hören, [...] 'The clinking and booming, the loud screams, instructions and dockyard sirens are hardly audible any more, [...] ' [...] ein kurzes Aufblitzen im Unterholz, die Explosion einer Mine, ein Schatten in einem Tunnelversteck. '[...] a short flash in the undergrowth, the explosion of a mine, a shadow in a tunnel hiding place'

The table in (18) shows the distribution of NIs with and without arguments in referential and non-referential uses. The only criterion that is not reflected by the distribution is the contextual presence of the argument. The other criteria, namely referential vs. nonreferential use, selectional restrictedness of the argument, and the perceptibility of the event clearly influence the argument realization. The referential use registered in the table was illustrated by the examples in (7b), (8a,b) and (9a) (argument realization) and (16) and (17) (no arguments). The non-referential uses displayed in the table were exemplified by the sentences in (14) and (15). 13 When they occur with arguments they encode a more specific event-type than the bare NI would. In the second column, some detailed information about the types of arguments are given, in order to allow for a better illustration of the disambiguating function of argument realization. In order to show that in the non-referential use the types of syntactically unrealized arguments are less diverse, the type of the implicit (default) argument is explicitly noted in the fourth column.

12

N o t e that the NIs in (17) cannot be classified as result nominals, since they cannot be pluralized.

13

Whether the reading of an NI is referential can be tested by adding attributes such as 'that I w a t c h e d ' . Generic readings can be identified by adding modifiers such as 'which is a bad habit' (cf. Krifka et al. 1995).

164

Ingrid Kaufmann

(18) Argument realization of some NIs of intransitive verbs selectional restrictedness

non--referential use arg. no arg.

referential use arg. no arg.

Brennen 'burning'

e.g., eyes, skin, house, sun, passion

13

3 (skin)

6

-

Fallen 'falling'

e.g., flap, trousers, voices, humans, frontier, curtain

11

12 (humans)

7

3

Schlagen 'hitting'

clock, wheels, bell, door, oars

24

3

16

4

Schreien 'screaming'

e.g., rooster, baby, fans, pigs, handicapped, children

14

2 (humans)

24

5

Ausbrechen 'breaking out'

e.g., fire, disease, volcano, animals, war, cars

20

12 (humans)

9

Atmen 'breathing'

e.g., humans, skin, screen

4

22 (humans)

10

Ausatmen 'exhaling'

e.g., film, world, tuba, humans

5

11 (humans)

5

Ausscheiden 'leaving'

humans

9

66

6

Stolpern 'stumbling'

humans, squirrel, political party, government

6

93 (humans, government)

15

10

Weinen 'crying'

humans

24

146

57

32

Klirren 'clinking'

e.g., chime, ice, bracelet

7

9 (glass)

25

37

-

-

4 -

For the majority of the NIs listed in (18), the argument is omitted more frequently in the non-referential uses, while argument realization takes place in most of the referential uses. The only exceptions are the NIs of the emission verb klirren 'to clink' which occur more frequently without an argument not only in the generic, but also in the referential use. As mentioned above, I assume that this is a consequence of the fact that events of sound emission can be recognized without the identification of their participants. Further evidence for this assumption comes from the referential uses of Schlagen without argument realization: In two of the examples, Schlagen is the head noun of the direct object of a perception verb, in the other two cases Schlagen is modified by the adjectives loud and muffled. The same holds for many of the non-referential uses without arguments of Weinen 'crying' and Schreien 'screaming'. If the base verbs allows for arguments with different selectional properties and has no preferred default reading, it is more likely that the corresponding nominalization will exhibit argument realization in the non-referential use (fallen 'fall', brennen 'burn', ausbrechen 'break out', schlagen 'hit', schreien 'cry'). This is because the necessity for eventtype specification is greater when the selectional properties are less restricted.

Referential arguments of nouns and verbs

165

Considering these data, argument realization is clearly not obligatory in the case of NIs. I propose that arguments are realized for two reasons. One is to specify an otherwise unclear reading of the nominal. Another is to identify a specific event by identifying one of its participants. 14 The assumption that argument realization serves to identify a specific event in the referential use is motivated by two observations. First, argument realization is not necessary if the participant is mentioned in the context, and second, argument realization is not necessary if there is an observable 'manifestation' of the event, as in the case of verbs of emission. The two different purposes roughly correspond to the non-referential use (reference to an event-type) and the referential use (reference to a specific event), respectively. Since identification of a specific event is irrelevant for non-referentially used NIs, argument realization is only motivated by the necessity to specify or differentiate the reading. As a consequence, no argument is realized as long as the selectional restrictions of the NI are 'homogeneous' enough and no narrowing of the reading is intended. In the case of the referentially used NIs, the predominant motivation for argument realization is the identification of a participant in order to identify the event. Thus, even if the selectional restrictions of the NI allow only for one type of argument, argument realization takes place. Specification of the reading plays a secondary role, although it might be one of the reasons why it is the object argument of NIs of transitive verbs that is realized in most of the cases. The two uses of NIs thus differ in that no argument realization is the unmarked option in the non-referential use, while argument realization is the default in the referential use. In the remainder of this paper, I sketch an analysis for the two uses of NIs. In the next section, I argue that referentially used event nominalizations are functional nouns in the sense of Löbner (1985). The differences between referentially used wng-nominalizations and NIs follow from their referential properties. Section 5 deals with the relation between referentially and non-referentially used NIs. Adopting the analysis of Chierchia (1998), I assume that 'functional' NIs are derived from kind-referring NIs by 'predicativization', i.e. by an operation that derives the property of being an instance of the relevant situation type.

14

One of the reviewers noted that reference tracking does also play an important role with respect to argument realization: arguments may stay implicit if the NI refers to an event that is already introduced in the discourse (see (i)). (i) ???(Er knipste wieder und wieder). Das viele Aufnehmen machte mich ganz verrückt. Although I have to confess that I have not analyzed the information structure o f the data yet, I do not think that examples like (i) are incompatible with the explanation offered here. If one motivation for argument realization is the necessity to identify a participant in order to identify the event, this does not mean that there are no other strategies to identify the event. My analysis predicts that if the NI refers to an event that is already established in the discourse, argument realization should be no longer necessary. Whether the referential uses o f NIs without arguments can be accounted for by this assumption, has to be answered by future work.

166 4.

Ingrid

Kaufmann

Referential NIs as functional nouns

4.1 Nominal vs. verbal arguments As already mentioned above, I assume that the argument realization of NIs (i.e. £v-event nominals) is determined by their referential properties. If this is correct, the differences between £V-event nominals and Λ-event nominals with respect to argument realization should also follow from different referential properties. According to Grimshaw, event nominals with an /^-argument are ordinary count nouns. £V-event nominals, in contrast, inherit the referential argument of the verb. The difference between referential properties of R-event nominals and £V-event nominals is therefore best established by comparing the prototypical lexical expressions that display these arguments: object denoting count nouns and verbs. Verbs characterize events that involve participants. No individual event exists independently of its participants, since the participants are the constituting parts of the event. The dependence of events on their participants is a consequence of their ontological properties: while objects are spatiotemporally constant, events are not. They are defined only by the temporary relatedness of their participants, and their spatial localization depends on the spatial localization of their participants. The identification of a specific event thus depends on the identification of the participants involved.15 This view on events corresponds to the position of Kim (1976), according to which metaphysically, events are not primitive objects, as assumed by Davidson (1967), but derived objects. According to this view, events are exemplifications of properties, displayed by certain individuals at a certain time.16 Therefore, in order to refer to a specific event by means of a verbal expression, identification of the participant by means of argument realization is necessary. In contrast to verbs, prototypical object denoting nouns are one-place predicates. If nouns have more than one argument, the referential argument and the internal argument represent individuals that exist independently of each other. Thus, the identification of the referent of the thematic argument is not in the same way essential for the identification of the referent of the nominal as in the case of verbs, since the referents of relational nouns are not constituted by the referents of their internal arguments. In terms of argument structure, this difference between verbs and nouns is reflected by the distinction of the referential event argument and the thematic arguments in the case of verbs, and the non-distinction of the referential argument and the (thematic) external argu15

16

One of the reviewers noted that the existence of an event can also be stated by quantification. I will argue in section 4.2 that this device is restricted to finite verbs. In an unpublished paper, Friederike Möllmann discusses a number of problems of this approach, which are a consequence of the lack of an event argument in the semantic representation of the verb. Moltmann proposes to deal with those problems by conceiving of events as 'truth-makers'. Perhaps her concept of a truth maker could be taken to be the interpretation of the semantic relation INST ('instantiates') proposed by Bierwisch (e.g., Bierwisch 1990). INST relates the event argument to the prepositional verbal representation. Although I am aware of the fact that the question of how an event argument can be related to the semantic representation of verbs is far from trivial, I leave it for future work.

Referential arguments of nouns and verbs

167

ment in the case of nouns. The difference also can be taken as a conceptual motivation of the feature specification for the major parts of speech introduced in Wunderlich (1996). Wunderlich proposes to replace the arbitrary categorical features [N] and [V] by the features [dep] '(referentially) dependent' and [art] 'articulated (argument structure)'. According to this classification, nouns and verbs share the feature [-dep] but differ in that verbs are specified as [+art] and nouns as [-art]. 17 With respect to argument realization, the difference between nouns and verbs is responsible for the greater flexibility of the argument realization of nouns. Since the existence of the referent of an object denoting relational noun can be verified independently of the referent of its internal argument, the identification of the latter is not always necessary in order to identify the first. With respect to the properties of their referents, event nominals should group with verbs rather than with nouns: The referents of their internal arguments participate in the event represented by their referential (and external) argument. Thus, in order to refer to a specific event by means of an event nominal, the participants must be identified either via argument realization or by contextual information. The fact that arguments of £V-event nominals have to be realized is therefore a consequence of the relation between their referential and thematic arguments. Since there cannot be any ontological difference between the events referred to by Revent nominals and those referred to by £V-event nominals, the relation between the events and their participants must be the same. Thus, the question why Λ-event nominals may be used referentially even if their participants are not identified by argument realization is yet unanswered. The only property that Λ-event nominals, but not £V-evcnt nominals could share with prototypical nouns, is a conceptual one: I want to claim that the characteristic property of nouns is that the spatiotemporal existence of their referents is presupposed at least for some possible world (i.e. there is a domain of quantification). This holds for events as well as for other objects. While this presupposition is ontologically motivated in the case of prototypical nouns, it is motivated only by class membership in the case of Λ-event nominals. The presupposition of the spatiotemporal existence of a set of possible referents is a 'grammatical' property associated with the categorical information of nouns. If a noun is derived by means of a nominal affix, it inherits this property automatically. 18 Thus, the nature of the difference between Grimshaw's Λ-event nominals and £V-event nominals is basically a conceptual one, although associated with their categorical properties. There is no need to identify the participants of Λ-event nominals by means of argument realization, since the existence of events of the relevant type is presupposed. In the case of £V-event nominals, however, the identification (or individuation) of an individual event depends on the identification of one of its participants. Thus, at least one argument has to be realized. If the differences between Ev- and Λ-event nominals depend on the categorial information of the affix, no difference between a CEN and an SEN reading of nouns derived by ation (or -ung) can be made. This is not problematic, however, if the relevant difference between Ev- and Λ-event nominals does concern the thematic arguments. Note that accord17

Ehrich (2002) assumes that event nominalizations are hybrid categories and that their verbal and nominal properties correlate to the extent to which they instantiate the properties associated with the feature [art].

18

Note that this holds for root nominalizations as well.

168

Ingrid

Kaufmann

ing to the view on R- and i'v-evcnt nouns outlined here, there is no need to assume that Evevent nouns inherit the argument structure of the base verb while Λ-event nouns do not. Although Λ-event nominals share the relevant properties of prototypical nouns, this does not mean that they may have no thematic arguments beside the referential argument. Whether the thematic arguments are realized or not depends on the use of the nominal and the range of its interpretations. Departing from Grimshaw's analysis, I assume that the presupposed existence of a set of referents is the crucial property of inherent nouns, root nominalizations and nominalizations derived by nominal affixes. Only verbal nouns have an £V-argument that is not associated with a presupposed set of referents. Having an £V-argument restricts the use of event nominals in that argument realization is required in certain contexts, while argument realization of Λ-event nominals is optional.19 Whether a thematic argument is realized or not thus does not depend on whether the nominal inherits the argument structure of the verb or not, but only on semantic requirements. One of the semantic requirements is the identification of a participant.

4.2. Accounting for default argument realization I thus claim that the argument realization of £V-event nominalizations is a consequence of the relation between their (verbal) referential argument and their thematic arguments and that this relation reflects the specific relation between the event and its participants. If this is correct, the question arises why the argument realization of NIs differs from the argument realization of verbs. Verbs, other than NIs, may be used referentially although no argument is realized, as in the case of impersonal passives like German Gestern wurde viel getanzt (lit.: 'Yesterday was much danced'). The reason why argument realization is necessary for NIs, but not for verbs, lies in the different devices by which the referential arguments are bound. Verbs require not only temporal specification, but also information about the 'validity' of the situation, which is provided by sentence mood. Since (verbal) infinitives lack this morphology, they are referentially dependent, i.e. their referential argument has to be bound to the event argument of another verb. £V-event nominalizations are derived from non-finite, i.e. referentially dependent verb forms. Since the existence of the referents of nominals is taken for granted (at least in some 'possible world'), there is no nominal inflectional device comparable to mood.20

19

20

This is probably a simplification, since at least one subclass of ff-event nominals are subject to restrictions with respect to argument realization. As Ehrich (2002) observes, root nominalizations like Tritt 'kick' do not allow the realization of the theme argument (*der Tritt des Hundes 'the kick of the dog', see also (3c)). It thus seems that these nominals are not relational at least syntactically, if not semantically. The fact that mood in contrast to verbal categories like tense, aspect, and voice is absent from event nominals cross-linguistically, has been observed by Comrie & Thompson (1985): "We might expect the verbal category of mood to appear in action nominals [= event nominals, I.K.] as well, but in fact, as we are not aware of languages where mood is retained in action nominals (indeed, mood in any non-finite verbal form seems relatively rare), we are led to assume that

Referential arguments of nouns and verbs

169

Suppose that it is the indicative or 'realis' mode that confirms the existence of the situation characterized by a verb. What happens then if a referentially dependent NI is realized in a nominal context where its referential argument can neither be bound to another event argument, nor be referentially anchored, since there is no suitable grammatical device?21 I want to propose that under these circumstances, referential anchoring is only possible if the situation is validated by its relation to another individual, whose existence can be taken for granted, namely one of its participants. The reason why argument realization is sufficient to anchor the event argument of NIs is that NIs are functional nouns in the sense of Löbner (1985, 1998). Löbner (1985, 1998) defines functional nouns as a subclass of relational nouns for which the relation that determines their reference is a function. Functional nouns such as father, weight, president unambiguously identify one referent, since there is a one-to-one relation between the referent of the functional noun and the referent of its argument. Löbner furthermore assumes that independently of their basic meaning, functional nouns can also be used as ordinary count nouns (i.e. as nouns that classify objects and logically are one-place predicates). This is illustrated by (19), where the functional nouns Vater 'father' and Bürgermeister 'mayor' are used as count nouns. (19)

Heute sind zehn Väter und sogar ein Bürgermeister unter uns. 'Today, there are ten fathers and even a mayor among us.'

The reason why functional nouns can be used as count nouns is that although the unambiguous identification of the referent of a functional depends on the referent of its argument, the physical existence of the referent does not. Thus, if the referent is already identified as in (19), a predicative (i.e. count noun) use of the functional noun is available. Referentially used event nominalizations can be classified as functional nouns: there is a one-to-one relation between the referent of the event nominal (the event) and the referents of each of its arguments, i.e. the participants of the situation. Interestingly, even if more than one participant is involved, the event can be unambiguously identified by its relation to only one of its participants. This follows from the specific properties of events as spatiotemporal objects, and the part-whole relation between them and their participants: a participant can be involved only in one event of a certain type at a certain time.22 No matter who

21

12

as far as mood is concerned action nominals simply do not retain this verbal category." (Comrie & Thompson 1985:360) Probably PPs like beim Schlafen 'while (lit.: 'at-the') sleeping', nach dem Aufstehen 'after getting up' can be analyzed as constructions in which the referential argument of the NI is bound to the referential argument of the modified verb. Note that this does not rule out the possibility that a participant is also involved in another event (characterized by another relation between the participant and this event) at the same time. The question of how thematic roles can help to decide whether descriptions of events taking place at the same location describe the same event or distinct events is discussed in Carlson (1998). Carlson assumes that an event has at most one entity playing a given thematic role. Thematic roles thus function as event-individuators in that they help to decide whether two descriptions may describe one and the same event or not. Note, however, that Carlson's notion of event-individuation is not the same as mine, since he adopts the Davidsonian view on events as primitives.

170

Ingrid Kaufmann

the other participants are, the spatiotemporal properties of the relevant event are determined as soon as the spatiotemporal properties of one participant are identified. As Löbner (1985, 115f) puts it, thematic roles such as 'agent o f , 'theme o f are not only functional concepts, but are also unambiguous in the reverse direction. This is the reason why even event nominalizations of transitive verbs generally occur with only one argument. 23 Although both £V-event nominalizations and Λ-event nominalizations are functional nouns, Λ-event nominalizations (i.e. nominalizations that are derived by nominal morphology) do not require argument realization while £v-event nominalizations do. In other words, Λ-event nominalizations can be used as count nouns while £V-event nominalizations cannot. This difference follows on the assumption that the existence of the referents of Revent nominalizations is presupposed as for other nouns, while it is not for the referentially dependent AV-event nominalizations. As the examples in (20) illustrate, a spatial PP modifier is sufficient to identify a specific situation referred to by an wftg-nominalization. This means that the referent of an ungnominalization is an individual that can be identified by its localization properties. In the case of NIs on the other hand, modification by a spatial PP is not sufficient to identify the referent, since the localization properties of the referents of NIs cannot be determined if their participants are not identified, as shown in (21). (20) a. b.

Die Untersuchung 'The examination Die Untersuchung 'The examination

unseres Lieblingspatienten hat heute 2 Stunden gedauert. of our favorite patient took 2 hours today.' im kleinen Behandlungsraum hat heute 2 Stunden gedauert. in the small treatment room took 2 hours today.'

(21) a.

Das Untersuchen unseres Lieblingspatienten hat heute 2 Stunden gedauert. 'The examining of our favorite patient took 2 hours today.' d. # Das Untersuchen im kleinen Behandlungsraum hat heute 2 Stunden gedauert. 'The examining in the small treatment room took 2 hours today.'

Although both ««^-nominalizations and NIs represent functional concepts in their basic reading, they differ in that only the former can be used as count nouns, like other functional

23

Although this means that in principle any argument could be realized in order to identify the referent of the event nominal, there are restrictions that probably follow from the internal structure of the event. As in the case of wwg-nominalizations (Ehrich & Rapp 2000), there is a strong preference for the realization of the theme argument in the case of NIs of accomplishment and achievement verbs. According to Ehrich (2002) ««^-nominalizations of transitive process verbs differ from NIs in that they also allow the realization of the agent while NIs do not. Note that this is true if the agent is realized as a genitive NP, but not if it is realized as a durch-PP: *das Vernehmen des Richters vs. das Vernehmen durch den Richter. As an empirical study on argument realization of NIs by Blume (2004) shows, constructions in which the agent is realized as a genitive NP while the theme is implicit are judged as totally ungrammatical, while constructions in which the agent is realized as a possessive pronoun are less marked. Blume draws the conclusion that there is a restriction that only themes of transitive NIs can be realized as a genitive NP. Thus, the grammaticality difference between ««g-nominalizations and NIs seems to be motivated by a linking restriction rather than by a general restriction on argument realization.

Referential arguments of nouns and verbs

171

nouns. Since they also differ in that ««g-nominalizations are derived by means of a derivational affix while NIs are not, it is reasonable to adopt Grimshaw's assumption that the difference is due to the nominal affix. In contrast to Grimshaw (1990), however, I do not assume that the two types of event nominals refer to different types of entities. Instead, I assume that they differ with respect to the ontological status of their referents: Since in the case of genuine nouns, the existence of entities of the relevant type is presupposed, no additional device is necessary to 'validate' the existence of the referent of an ungnominalization in a specific context. As argued above, this does not hold for the referents of genuine verbs and therefore not for infinite verbal forms and their nominalizations either. Whatever the nature of the nominalization device is, it does not change the referential properties of infinitives.

5.

Relating the referential and non-referential use of NIs

As already mentioned above, I assume that generic NIs refer to event-types (i.e. kinds) while referentially used NIs are functional nouns. Whether an NI is a kind-referring expression or not can be tested via verbal predicates like ist eine schlechte Angewohnheit 'is a bad habit' (cf. Krifka et al. 1995). In German, kind-referring NIs can be either bare infinitives (or even infinite VPs), or infinitives with a definite article, or full NPs. (22) a. b.

(Das) Rauchen ist eine schlechte Angewohnheit. 'Smoking is a bad habit.' Das Rauchen (von Zigaretten) ist eine schlechte Angewohnheit. 'The smoking (of cigarettes) is a bad habit.'

Referentially used NIs, on the other hand, can only be full NPs (unless the participant can be identified by contextual information as discussed in section 3). (23) a.

Ich habe das Rauchen der Schüler auf dem Schulhof beobachtet. Ί watched the smoking of the pupils on the playground.' b. * Ich habe (das) Rauchen auf dem Schulhof beobachtet. Ί watched smoking on the playground.'

Since bare infinitives and infinitives plus definite article may occur as complements of verbs but display no other nominal properties (e.g., modification by adjectives, genitive complements), I assume that they are more basic as far as their derivational status is concerned. The fact that they are restricted to the event-type reading suggests that the referential use of NIs requires a further derivational step. Chierchia (1998) has proposed an analysis of how the use of nouns as kind referring expressions and as predicates can be related. He assumes that kinds can be derived from properties and vice versa. Predicative common nouns then are 'predicativizations' of kinds that encode the property of being an instance of the relevant kind. Kinds, on the other hand, can

172

Ingrid Kaufmann

be construed as 'nominalizations' of predicative common nouns, so that a kind is the largest member of the extension of the relevant property in a given world. Taking into account that bare NIs refer to event-types, it is reasonable to assume that nominalizations of infinitives derive kind-referring expressions. According to Chierchia (1998), kinds can be intuitively characterized as the totality of all their instances. In the case of event-types, this means that the thematic arguments of the base verb are generically bound. Since there are no open argument slots, argument realization is not necessary. However, more specific event-types can be derived by making explicit the selectional properties of one of the arguments by means of argument realization. By 'predicativization', NIs with unbound arguments can be derived from the kindreferring NIs. These are the functional nouns that can be used to refer to specific situations or as predicates. According to Chierchia, properties that are derived from kinds have a special property: Since the property of being an instance of a kind does not differentiate between singular and plural instances, predicates derived from kind-referring expressions have a mass denotation (Chierchia 1998:353). He assumes that mass nouns differ from count nouns in that they apply to pluralities as well as to singular entities. As mass nouns, nominal predicates derived from kind-referring expression thus do not pluralize, because they are already lexically pluralized. This is one of the characteristic properties of £V-event nominalizations (cf. (Id), (6c)). Maybe the fact that the anchoring of the referential argument of £V-event nominalizations has to be mediated by the identification of one of the participants of the event also can be explained by the assumption that Zsv-event nominalizations have a mass denotation. In a way, the identification of a participant has the same effect as a measure phrase or classifier: The measure defines a unit for counting and thereby a means to identify atomic individuals. Due to the one-to-one relation between the event and (each of) its participants, the identification of a participant identifies an event individual as well. To summarize, on the assumption that referentially used NIs are functional nouns, the fact that they require argument realization while generically used NIs do not, can be explained by adopting Chierchia's approach to kind-referring expressions. The need of argument realization follows from semantic, not argument structural requirements.

6.

Conclusion

The aim of this paper was to investigate the nature of the difference between the referential arguments of Grimshaw's 'simple event nominals' and 'complex event nominals'. I have argued that argument realization of German NIs is better accounted for by semantic conditions than by argument structural requirements. Furthermore, the relevant conditions can be motivated if one preserves Grimshaw's assumption that complex event nominals inherit the verbal event argument, and thus differ from event nominals derived by nominal affixes. I have proposed that the difference between Ev- and Λ-arguments is basically an ontological one: i?-arguments refer to event individuals, whose existence is independent of the referents of the thematic arguments, while the referents of £ ^-arguments are constituted by mediation

Referential arguments of nouns and verbs

173

o f the referents o f the thematic arguments. Obligatory argument realization is related to the nature o f the £V-argument: in nominal contexts, the referential anchoring o f the Evargument depends on the identification o f at least one o f the participants o f the situation. Finally, I have tried to show that the relation between the different uses o f NIs can be derived by adopting Chierchia's (1998) analysis o f nominals.

References

Bierwisch, Manfred (1982): "Formal and Lexical Semantics." - In: Linguistische Berichte 80, 3-17. - (1990): "Event Nominalizations. Proposals and Problems." - In: Acta Linguistica Hungarica 40, 19-84. Blume, Kerstin (2004): Nominalisierte Infinitive. Eine empirisch basierte Studie zum Deutschen. Tübingen: Niemeyer. Borer, Hagit (1997): "The morphology-syntax interface: Α study in autonomy." - In: W.U. Dressler, M. Prinzhom & J.R. Rennison (eds.): Advances in Morphology, 5-30. Berlin: Mouton de Gruyter. Carlson, Greg (1998): "Thematic Roles and the Individuation of Events." - In: S. Rothstein (ed.): Events and Grammar, 35-51. London: Kluwer. Chierchia, Gennaro (1998): "Reference to kinds across languages." - In: Natural Language Semantics 6, 339-405. Comrie, Bernard & Sandra A. Thompson (1985): "Lexical Nominalization". - In: T. Shopen (ed.): Language Typology and Syntactic Description Vol. III., 349-398. Cambridge: CUP. Davidson, Donald (1967): "The logical form of action sentences." - In: N. Rescher (ed.): The Logic of Decision and Action, 81-95. Pittsburgh: University of Pittsburg Press. Ehrich, Veronika (2002): "On the verbal nature of certain nominal entities." - In: I. Kaufmann & Β. Stiebeis (eds.): More than Verbs. A Festschrift for Dieter Wunderlich, 69-89. Berlin: Akademie Verlag. Ehrich, Veronika & Irene Rapp (2000): "Sortale Bedeutung und Argumentstruktur: wng-Nominalisierungen im Deutschen". - In: Zeitschrift für Sprachwissenschaft 19, 245-303. Grimshaw, Jane (1990): Argument Structure. - Cambridge, Mass.: MIT Press. Kim, Jaegwon. (1976): "Events as property exemplifications." - In: Myles Brand & D. Walton (eds.): Action Theory. Dordrecht: Reidel. Krifka, Manfred et al. (1995): "Genericity: An Introduction." - In: G. Carlson & F.J. Pelletier (eds.): The Generic Book, 1-124. Chicago: UCP. Löbner, Sebastian (1985): "Definites." - In: Journal of Semantics 4, 279-326. - (1998): Definite associative anaphora. Ms. Düsseldorf. Moltmann, Friederike (undated): "Nominalizations, Events, and Other Concrete Objects." www.philosophy.stir.ac.uk/staff/Friederike_Moltmann_files/friederike.html Wunderlich, Dieter (1996): "Lexical Categories." - In: Theoretical Linguistics 22, 1-48.

Section III: Events in composition

Angelika

Kratzer .

*

Building resultatives

The paper argues that resultative constructions consisting of a verb and what looks like an adjective are marginal cases of serialization that are possible in German or English (but not in Romance, for example) because in those languages, adjectival roots can enter syntactic derivations without inflection. This property allows those roots to combine with the same unpronounced causative suffix [CAUSE] that is responsible for zero-derived verbs like dry or empty in English. The semantic characteristics of resultatives are derived from the lexical properties of the suffix [CAUSE] in interaction with the operation of Event Identification, which semantically combines the 'adjectival' and the verbal part of the construction. The English VP to drink my teapot dry, for example, describes events that are both drinking events and events of causing my teapot to be dry. This identification of two events accounts for the impression that the causative relation expressed by resultatives is a relation of 'direct' causation. It also makes it possible for the direct object in a resultative to be perceived as an argument of the verb, even though it never truly is, a conclusion I defend in spite of a mount of apparent counterevidence.

1.

The construction: Adjectival resultatives

Some natural languages allow their speakers to put together a verb and an adjective to create complex predicates that are commonly referred to as "resultatives". Here are two runof-the-mill examples from German: 1 (1)

die Teekanne leer trinken the teapot empty drink 'To drink the teapot empty'

(2)

die Tulpen platt giessen the tulips flat water Water the tulips flat

Resultatives raise important questions for the syntax-semantics interface, and this is why they have occupied a prominent place in recent linguistic theorizing. What is it that makes this construction so interesting? Resultatives are submitted to a cluster o f not obviously

1

Thank you to Claudia Maienborn and Angelika Wöllstein for organizing the stimulating DGfS workshop where this paper was presented, and for waiting so patiently for its final write-up. Claudia Maienborn also sent comments on an earlier version of the paper, which were highly appreciated. (2) is modeled after a famous example in Carrier & Randall (1992).

178

Angelika Kratzer

related constraints, and this fact calls out for explanation. There are tough constraints for the verb, for example: (3)

a.

Er hat seine Familie magenkrank gekocht. he has his family stomach sick cooked. 'He cooked his family stomach sick.'

b. * Er hat seine Familie magenkrank bekocht, he has his family stomach sick cooked-for And there are also well-known restrictions for the participating adjectives (Simpson 1983, Smith 1983, Fabb 1984, Carrier & Randall 1992): (4)

a. * The chef cooked the food blackened, b. * The joggers ran themselves exhausted.

As for semantic interpretation, resultatives are a species of causatives, where the causal relation is always direct causation. Where does this particular interpretation come from? It doesn't seem to be contributed by either the adjective or the verb alone. Do we have to conclude, then, that constructions all by themselves can introduce meaning components as specific as causal relations? Looking beyond German and English, we observe that not every language has resultatives: The Romance languages don't, for example. Does this mean that there is a 'resultative parameter'? In this paper, I will develop an analysis of (mostly German) resultatives with the above questions in mind. I will argue that the causal relation in resultatives is carried by an unpronounced affix attached to the adjective. Resultatives do not force us to assume that syntactic constructions or semantic composition rules can introduce non-logical meaning components like causal relations. I will also work towards a hypothesis that links the constraints for the verbs and adjectives in German resultative constructions to more general properties that might eventually tell us what it takes for a language to have resultative constructions to begin with. In interaction with a standard event semantics, simple morphological properties conspire to allow or disallow resultatives and force them to behave the way they do. Before embarking on a systematic exploration, a word of caution is in order. Resultative constructions have to be distinguished from depletives, which involve a verb and an adjective, but do not have a causal interpretation. (5)

J'ai connu Marie heureuse. I have known Marie happy

Depictive Construction (Legendre 1997: 45)

Resultatives built from verbs and adjectives must also be distinguished from directional particle constructions, which have a causative interpretation, but do not involve adjectives. (6)

den Abfall rausbefördern. the garbage out-transport 'take the garbage out'

Directional Particle + Verb

Building resultatives

179

Inclusion of directionals in discussions of resultatives has obscured important generalizations that emerge clearly once we restrict our enterprise to resultatives built from adjectives. Nevertheless, directionals have interesting properties of their own that merit investigation and invite comparison with other types of resultatives. I will discuss one characteristic property of directional particles in section 3, but I won't be able to develop an analysis of directionals here.

2.

Deriving the Direct Object Restriction from a raising analysis

Resultatives have long been known to have a special relationship with direct objects (Simpson 1983). This relationship is reflected in Levin & Rappaport Hovav's (1995: 34) Direct Object Restriction, which says that a result phrase "may be predicated of the immediately postverbal NP, but may not be predicated of a subject or of an oblique complement". Where could such a generalization come from? In the best of all possible worlds it would fall out from an analysis of resultatives and would not have to be stipulated. To have a concrete example to look at, take example (1) from above, repeated here as (7): (7)

die Teekanne leer trinken the teapot empty drink 'To drink the teapot empty'

In an influential paper (Hoekstra 1988), Teun Hoekstra argued that resultatives project a raising structure. Within the framework I want to assume here, this seems to be the only option for (7). The DP die Teekanne ('the teapot') is an argument of leer ('empty'), not of trinken ('drink'). That DP must therefore start out within the projection of leer, assuming that, in the syntax, arguments have to start out within the projection of their heads. Making the additional assumption that the unique (non-eventuality) arguments of adjectives are internal, rather than external arguments, no further small clause or predication structure has to be built. In fact, as I will argue shortly, no functional structure can be built on top of the AP headed by leer. This lack of functional structure, I will suggest, forces a raising analysis like the one sketched in figure 1:

180

Angelika

Kratzer

trinken die Teekanne

leer

Raising forced by case needs

Compounding via head movement Figure 1

The analysis I am proposing for (7) is a raising analysis like Hoekstra's. Technically, it's not the usual small clause analysis, though. I am not assuming, as did Stowell (1983), that external arguments of adjectives originate within the projection of their head. My adjectives do not have external arguments. They couldn't, since they lack voice, which, at least according to the story told in Kratzer (1996), is responsible for the introduction of external arguments. Voice inflection is by necessity verbal inflection, hence can only build verbal projections. From its base position within the projection of the adjective leer, die Teekanne moves up into the functional structure of trinken to check its accusative case features, possibly triggering incorporation of leer into trinken. This is how die Teekanne becomes a direct object of the compound leer trinken ('empty drink') and acquires some of the typical properties we associate with direct objects. A raising analysis is plausible enough for cases like (7). But can it be defended for all kinds of resultatives? If it could, the Direct Object Restriction would follow, since on a raising account, the unique (non-event) argument of the adjective becomes a derived direct object, not a subject or oblique object. The most pressing question we have to look into, then, is whether a raising analysis is indeed viable for all resultatives. Suppose it is. Resultatives should then always come with an originally intransitive verb, that is, with a verb that does not start out its life in syntax with a direct object of its own. This prediction looks quite wrong. The literature is full of examples where resultatives seem to be able to combine with transitive or unaccusative verbs. Nevertheless, I will argue that a raising analysis is at least correct for all those resultatives that are built with the help of adjectives. Even in this qualified version, the claim goes against widely held beliefs. Usually, a raising analysis is taken to be untenable for the kind of resultatives we see in (8), which are from Carrier & Randall (1992:115). Those resultatives seem to be able to combine with transitive or unaccusative verbs: (8)

a. b.

The gardener watered the tulips flat, The pond froze solid.

I will show in the following section that in spite of many apparent counterexamples, there really aren't any good cases of adjectival resultatives that combine with transitive or unac-

Building resultatives

181

cusative verbs. This result will pave the way for a raising analysis for all kinds of adjectival resultatives, including those exemplified by (8a) and (8b).

3.

N o transitives or unaccusatives

Look at the following German transitive / intransitive alternations:2 (9)

a.

Er hat gekocht. he has cooked 'He cooked.' b. Er hat seine Familie magenkrank gekocht, he has his family stomach sick cooked 'He cooked his family stomach sick.' c. Er hat *(seine Familie) bekocht, he has his family(acc.) cooked-for 'He cooked for his family.' d. * Er hat seine Familie magenkrank bekocht. he has his family stomach sick cooked-for

(10) a.

Sie haben geschossen. they have shot. 'They shot.' b. Sie haben ihn tot geschossen, they have him dead shot c. Sie haben *(ihn) erschossen, they have him(acc.) shot-dead 'They shot him dead.' d. *Sie haben ihn tot erschossen. they have him dead shot-dead

(11) a.

b.

Sie haben (unser Geld) geraubt. they have our money robbed 'They robbed our money.' Sie haben uns arm geraubt, they have us poor robbed 'They robbed us poor.'

The star in front of the (d) examples means that the sentence is ungrammatical as a resultative construction. In some cases, the sentence is acceptable when the adjective can be understood as a depictive or a manner adverb.

182

Angelika Kratzer

c.

Sie haben *(uns) they have us(acc.) 'They robbed us.' d. * Sie haben uns arm they have us poor

beraubt, robbed-from beraubt. robbed-from

(12) a.

Sie hat gebetet. she has prayed 'She prayed.' b. Sie hat dich gesund gebetet, she has you healthy prayed 'She prayed you healthy.' c. Sie hat *(dich) angebetet, she has you(acc.) adored 'She adored you.' d. * Sie hat dich gesund angebetet. she has you healthy adored

(13) a.

Sie hat gequasselt. she has babbled 'She was babbling away.' b. Sie hat uns tot gequasselt, she has us dead babbled 'She babbled us dead.' c. Sie hat *(uns) bequasselt. she has us(acc.) babbled-at 'She was babbling at us.' d. * Sie hat uns tot bequasselt. she has us dead babbled-at

(9) to (13) illustrate the behavior of intransitive / transitive verb alternants in adjectival resultative constructions. In each case, the intransitive verb is acceptable, but its transitive counterpart is not. That transitivity is the relevant property, and not telicity, is shown by (9), (12) and (13).3 The verbs bekochen ('cook for'), anbeten ('adore') and bequasseln ('babble at') are as atelic as the simplex verbs they are derived from. They do not characterize a target state as part of their meaning. They also combine happily with durational phrases. This means that we cannot exclude the ungrammatical resultatives in (9) to (13) by appealing to some version of what has been called "Tenny's Generalization", requiring that no simple or complex predicate can characterize more than one target state (Tenny 1987, Giannakidou & Merchant 1998). The ban on transitives in resultative constructions would 3

The notion of transitivity needed to account for the data in (9) to (13) doesn't coincide with Levin's (1999) notion of 'core transitive verb'. For Levin, the core transitive verbs are those that have a "causative event structure", hence are causatives. But at most one of the essentially transitive verbs in (9) to (13) is a causative (erschießen, 'shoot dead'). The direct objects of the other verbs denote the benefactor, source, or goal of the events described by the verb.

Building

183

resultatives

be entirely expected if adjectival resultatives were uniformly raising constructions. A raising analysis for resultative constructions assumes that the direct objects we see in those constructions are always syntactically derived. The verbs appearing with such resultatives should never have an obligatory direct object of their own, then. They should never start out as transitives or unaccusatives. Unfortunately, there are lots of examples that seem to upset the generalization suggested by (9) to (13). Here is a rather nasty looking batch: (14) a.

b.

(15) a.

b.

(16) a.

b.

(17) a.

b.

(18) a.

b.

(19) a.

b.

Sie they 'They Sie they

haben *(den Fußballplatz) beleuchtet. have the soccer field illuminated illuminated the soccer field.' haben den Fußballplatz hell beleuchtet, have the soccer field bright illuminated

Sie they 'They Sie they

haben *(die Patienten) ausgezogen. have the patients undressed undressed the patients.' haben die Patienten nackt ausgezogen, have the patients nude undressed

Sie haben *(das Fleisch) angebraten. they have the meat at-roast 'They briefly roasted the outside of the meat.' Sie haben das Fleisch braun angebraten, they have the meat brown at-roasted Sie haben *(den Laster) beladen. they have the truck loaded 'They loaded the truck.' Sie haben den Laster schwer beladen, they have the truck heavy loaded Sie they 'They Sie they

haben *(die Wand) bemalt. have the wall painted painted the wall.' haben die Wand blau bemalt, have the wall blue painted

Sie haben they have 'They rolled Sie haben they have

*(den Teig) ausgerollt. the dough out-rolled out the dough.' den Teig dünn ausgerollt, the dough thin out-rolled

The examples in (14) to (19) have verbs that are obligatorily transitive, just like the verbs in (9) to (13). Yet all of those verbs seem to be able to participate in adjectival resultative

184

Angelika Kratzer

constructions. Fortunately, the threat of (14) to (19) is not too hard to divert. Look at the following set of short dialogues: (20) a.

b.

c.

d.

e.

f.

Wie how 'How Wie how 'How Wie how 'How Wie how 'How Wie how 'How Wie how 'How

soll ich den Fußballplatz beleuchten? Hell. shall I the soccer field illuminate bright shall I illuminate the soccer field? Bright.' soll ich die Patienten ausziehen? Nackt, shall I the patients undress nude shall I undress the patients? Nude.' soll ich das Fleisch anbraten? Braun, shall I the meat at-roast brown shall I roast the meat? Brown.' haben sie den Laster beladen? Viel zu schwer, have they the truck loaded much too heavy. did they load the truck? Much too heavy' haben sie die Wand bemalt? Blau, have they the wall painted blue. did they paint the wall? Blue.' haben sie den Teig ausgerollt? Dünn, have they the dough out-rolled thin did they roll the dough? Thin.'

The data in (20) suggest that the apparent adjectives in (14) to (19) do not have to be parsed as adjectives, but might also be parsed as adverbs. In German, manner adverbs and predicative adjectives look exactly alike, and this makes it hard to keep the two apart in certain cases. In (20a), hell ('bright') relates to the intensity of the illumination. When you illuminate a place brightly (note the use of the English adverb), it becomes bright. Likewise, when you undress completely (again, note the adverb in English), you end up nude. Different ways of roasting meat can be distinguished by their effect on the meat's outside color. When you load a truck heavily, the truck ends up with a heavy load, and that makes the truck heavy, too. Using different paints to paint walls can be seen as different ways of painting walls. If you roll out the dough just a little bit, it ends up being thick. The more you roll it and the more pressure you apply, the thinner it gets. In English, too, adverbs can appear without the suffix ~ly in certain cases, hence might look exactly like adjectives. Here are some examples from the Longman grammar (Quirk, Greenbaum, Leech & Svartik 1985, sections 7.6 to 7.9): (21) a. b. c. d. e. f. gh. i.

Danger, go slow. Did you have to wait long? She cut her hair short. The flowers smell sweet. Don't talk daft. She pays her rent regular. They played real good. He spoke to John sharp. Speak clearer.

Building

j. k. 1. m. n. o.

resultalives

185

The car went slower and slower. They played the game clean. She travels light. The food tastes marvelous. That's easier said than done. He behaves even worse than his brother.

It is not impossible, then, that even in English, apparent adjectives in a resultative construction might sometimes be parsed as adverbs. Returning to German, there is a sharp contrast between (20a-f) and the following examples, which are completely unacceptable on the intended readings: (22) a. * Wie hat er seine Familie gekocht? (Magenkrank.) How has he his family cooked? (Stomach sick.) b. * Wie haben sie ihn geschossen? (Tot.) How have they him shot? (Dead.) c. * Wie haben sie uns geraubt? (Arm.) How have they us robbed? (Poor.) d. * Wie hat sie dich gebetet? (Gesund.) How has she you prayed? (Healthy.) e. * Wie hat sie uns gequasselt? (Tot.) How has she us babbled? (Dead.) The apparent adjectives in (14) to (19), then, are adverbs. That adverbs can have resultative interpretations of their own is shown by the following examples: (23) a. b. c. d. e. f. gh. i. j· k.

Finely chopped parsley Nicely wrapped presents Heavily loaded trucks Brightly illuminated rooms Beautifully manicured nails Wonderfully arranged flowers Perfectly grown oak trees Magnificently painted ceilings Coarsely grated carrots Thinly spread layers of mayonnaise Illegibly written notes

Not all adverbs in (23a-k) have exactly the same interpretation as the corresponding adjectives would have. When a present is wrapped nicely, for example, it has a nice wrapping. It might still be rather ugly. Likewise, manicuring nails beautifully does not necessarily make your nails beautiful. There is only so much that nail polish can do. Nevertheless, adverbial and resultative interpretations often coincide, or are so close that they are hard to distinguish. The line between adjectives heading resultative phrases and adverbs is often very hard to draw, and in German, the systematic absence of any overt morphology for those adverbs blurs the boundaries even more.

186

Angelika Kratzer

I take it, then, that the data in (14) to (19) do not truly challenge the generalization that transitive verbs cannot co-occur with adjectival resultatives. The apparent adjectives in (14) to (19) can all be parsed as adverbs. Sometimes, describing a result brought about by an action can be seen as describing a way the action was performed. In those cases, it seems, adverbs can be easily confused with resultative adjectives, not only in German, but possibly also in English. Transitivity is not always easy to diagnose. Sometimes, eyeballing the situation is not good enough, and more serious tests are needed. Take the following sentences: (24) a.

b.

(25) a.

b.

(26) a.

b.

(27) a.

b.

Sie haben ?(die Äpfel) gepflückt. they have the apples picked 'They picked the apples.' Sie haben die Bäume kahl gepflückt, they have the trees bare picked 'They picked the trees bare.' Sie haben ?(dasAuto) gekauft. they have the car bought 'They bought the car.' Sie haben den Laden leer gekauft, they have the shop empty bought 'They bought the shop empty.' Sie haben ?(die Suppe) gelöffelt. they have the soup eaten with a spoon 'They ate the soup with a spoon.' Sie haben den Teller leer gelöffelt. they have the plate empty eaten with a spoon 'They emptied the plate with a spoon.' Sie haben they have 'They built Sie haben they have 'They built

?( die Garage) gebaut. the garage built the garage.' das Grundstück voll gebaut, the plot full built the plot full.'

The resultatives in the (b)-sentences of (24) to (27) are all of the kind that seem to require a raising analysis. What is picked in (24b) are not the trees, what is bought in (25b) is not the shop, what is scooped up with a spoon in (26b) is not the plate, and what is built in (27b) is not the plot. The surface objects, then, cannot be arguments of the verbs in the (b)sentences of (24) to (27). Nevertheless the verbs that come with those resultatives might wrongly be classified as obligatorily transitive based on the kind of contexts given in the (a)-sentences. Upon closer scrutiny, we find that all of those verbs have at least some intransitive uses. They do not require a direct object, for example, when they are 'reduplicated' to produce an iterative interpretation. We have:

Building resultatives (28) a. b. c. d.

Sie they Sie they Sie they Sie they

187

pflückten und pflückten. picked and picked kauften und kauften, bought and bought löffelten und löffelten. ate (with a spoon) and ate (with a spoon) bauten und bauten, built and built

When we look at the alternating verb pairs in (9) to (13) above we find that the intransitive alternant can, but the transitive alternant cannot stand on its own when 'reduplicated'. As before, telicity cannot be what rules out those cases, since in (29), (32), and (33), the transitivized version of the verb is as atelic as its underived counterpart. (29) a.

Er he b. * E r he

kochte und cooked and bekochte cooked-for

Sie they b. * Sie they

kochte. cooked und bekochte, and cooked-for

(30) a.

schössen und shot and erschossen und shot-dead and

(31) a.

raubten und robbed and beraubten robbed-from

raubten. robbed und beraubten, and robbed-from

(32) a.

betete und prayed and betete an und adored and

betete. prayed betete an. adored

(33) a.

quasselte und quasselte. babbled and babbled bequasselte und bequasselte, babbled-at and babbled-at

Sie they b. * Sie they Sie she b. * Sie she

Sie she b. * Sie she

schössen. shot erschossen, shot-dead

It is clear, then, that the verbs in (24) to (27) have at least some intransitive uses. The same is true of the verbs in the following resultatives that are not usually given a raising analysis, and where an adverbial analysis is at best marginally possible: (34) a.

Sie hat die Tulpen platt gegossen. she has the tulips flat watered 'She watered the tulips flat.'

188

Angelika Kratzer

b.

Sie goss und goss, she watered and watered c. * Wie hat sie die Tulpen gegossen? Platt. how has she the tulips watered flat 'How did she water the tulips? Flat.'

(35) a.

Sie haben den Tisch sauber gewischt. they have the table clean wiped 'They wiped the table clean.' b. Sie wischten und wischten, they wiped and wiped c. * Wie haben sie den Tisch gewischt? Sauber. how have they the table wiped clean 'How did they wipe the table? Clean.'

When apparent resultative adjectives can be parsed as adverbs, the surface objects in those sentences are arguments of their respective verbs. It follows that what was illuminated in (14b) from above, for example, was the soccer field and what was loaded in (17b) above was the truck. On the other hand, if we want to defend a raising analysis for all resultative constructions with adjectives, we have to struggle with the fact that the surface direct objects we find in those constructions are never semantic arguments of their verbs, and this brings up the question of just how the clearly present inferences about the watering of the tulips in (34a), for example, or the wiping of the table in (35a) would come about. In order to see that such inferences can be accounted for even on a raising analysis, we have to look at the semantic interpretation of resultatives in some detail. I will do so in the following section. The issue is a very important one. Carrier & Randall (1992) criticized Hoekstra's raising analysis as being unable to derive the fact that a sentence like (34a), for example, implies that the tulips became flat as a result of having been watered. The criticism is unfounded. The desired inference falls out from the semantics, as we will see. Like transitivity, unaccusativity is an elusive property and needs to be diagnosed with care. If a raising analysis is right for all types of (adjectival) resultatives, we shouldn't find any truly unaccusative verbs participating in that construction. Yet Levin & Rappaport Hovav (1995:39, examples (19a-e)) present the following examples: (36) a. b. c. d. e.

The river froze solid. The prisoners froze to death. The bottle broke open. The gate swung shut. This time the curtain rolled open on the court of the Caesar. (Olivia (D. Bussy), Olivia, 35.)

Typical uses of the English verb freeze are illustrated in (37a-c): (37) a. b. c.

It was freezing. I was freezing. The water froze yesterday.

Building

189

resultatives

The German counterpart of freeze in the translations of (37a-c) shows mixed unaccusative/unergative behavior, as indicated by the use of the auxiliaries sein ('be') versus haben ('have'): 4 (38) a.

b.

c.

Es hat gefroren, it has frozen 'The temperature was below freezing.' Ich hab' gefroren. I have frozen Ί was cold.' Das Wasser ist gestern gefroren, the water is yesterday frozen 'The water froze yesterday.'

The German translation of swing, schwingen, also shows mixed unaccusative/unergative behavior. In standard German, it takes the auxiliary haben5, hence behaves as an unergative. Examples (36c,e) cannot be done away with so easily. Those are not isolated cases either. Any of the following verbs can be used in a resultative construction: (39)

Burst (open), pop (open), fly (open), tear (open), rip (open), crack (open), slide (open).

The counterparts of intransitive roll, break, burst, pop, fly, tear, rip, crack, and slide are all unambiguously unaccusative in German. There is something special about open, though. Here, too, a look at German is revealing. German has an adjective offen and (what looks like) a separable prefix or verb particle auf both meaning 'open'. Offen has attributive and predicative uses. When used attributively, it takes adjectival agreement morphology. Auf only allows predicative uses in Standard German. However, according to the Duden (Die Zweifelsfälle der deutschen Sprache), attributive uses with adjectival inflection occur in certain varieties of colloquial German. The Duden gives die aufe Flasche ('the open+Agr bottle') and ein aufes Fenster ('an open+Agr window') as examples. Auf and open are interchangeable in predicative uses of the kind illustrated in (40) to (42): Die the Die the

4 5

Tür door Tür door

ist is ist is

auf. open offen. open

The (c)-example was provided by Claudia Maienborn, personal communication. It is listed as taking the auxiliary haben in Kempcke's (2000) Wörterbuch Deutsch als Fremdsprache. My own dialect admits both sein and haben in cases like The pendulum swung back and forth.

190

Angelika Kratzer

(41) a.

b.

(42) a. b.

Sie she 'She Sie she 'She

hat die has the left the hat die has the left the

Die the Die the

Tür door Tür door

Tür aufgelassen. door open-left door open.' Tür offengelassen, door open-left door open.'

bleibt auf. stays open bleibt offen, stays open

I perceive a slight difference between auf and offen in resultative constructions: (43) a.

Die Wunde the wound 'The wound b. ?Die Wunde the wound

(44) a.

Sie she 'She b. ?Sie she

(45) a.

b.

(46) a.

b.

Wir we 'We Wir we 'We

ist aufgeplatzt. is open-burst burst open.' ist offengeplatzt, is open burst

hat die Tür aufgebrochen. has the door open-broken broke the door open.' hat die Tür offengebrochen, has the door open-broken haben die have the hammered haben die have the hammered

Türen mit den Fäusten aufgehämmert. door with the fists open-hammered the doors open with our fists.' Türen mit den Fäusten offengehämmert, door with the fists open-hammered the doors open with our fists.'

Dieser Wunderhund kann this miracle dog can 'This miracle dog can even Dieser Wunderhund kann this miracle dog can 'This miracle dog can even

sogar Türen aufbellen. even doors open-bark bark doors open.' sogar Türen offenbellen, even doors open-bark bark doors open.'

(43) has an unaccusative verb. The verb in (44) is obligatorily transitive. In both cases, I find offen marginally - but only marginally - possible. 6 The verbs in (45) and (46) have unergative uses. Offen is as acceptable as auf there. The data suggest that there is insecurity

6

For Claudia Maienborn, (43b) and (44b) deserve two question marks.

Building

191

resultatives

about the categorial status of both auf and offen. Normative grammar tells us that auf is a separable prefix or verb particle, and offen is an adjective. But speakers' intuitions seem to blur the differences between the two. The appearance of adjectival inflection shows that auf is treated as an adjective by some speakers, hence is assimilated to offen. The marginal acceptability of offen in resultatives built with transitive or unaccusative verbs in my own idiolect might then indicate that after overcoming some initial resistance, I can categorize offen as a verb prefix or particle. The CHILDES database (MacWhinney 2000) documents the case of a child who consistently uses offen in resultative constructions with transitive verbs. Interestingly, each of those uses of offen is translated as auf in the transcriber's glosses. Wagner's Carsten corpus (Wagner 1985) has three occurrences of offen machen ('open make'. We find in line 392, for example: Warum hast du meine Buchse offen gemacht ('Why did you open my pants?'). The transcriber adds the gloss Hose aufgemacht ('pants opened'), translating the dialect word Buchse as standard Hose, as well as glossing the child's offen as adult standard auf.1 The most interesting use of offen in the Carsten corpus occurs in line 4032, where it comes with what is clearly an essentially transitive verb ('cover'): hab i wieder offen gedeckt ( Ί have again open covered', that is, Ί have again uncovered it'). The adult standard version of the child's verb offendecken is aufdecken, and that's the translation given by the transcriber. Those data suggest that Carsten might have started out with the hypothesis that offen is a particle. There is not a single occurrence of attributive offen in Carsten's speech in the CHILDES database, nor in the Clahsen or Miller corpora. In the Wagner corpus, attributive uses of offen only appear with Frederik, who is already 8;7 years old. Attributive adjectives are inflected in German, and this property is likely to play a role in triggering a recategorization. Since English does not have overt adjectival agreement morphology, open may live on as a hybrid in the adult language. There may indeed be something special about open, then. It may be able to act like a particle. That English open is capable of the behavior of a particle is shown by examples (47) to (50) below, which are based on paradigms constructed by Joe Emonds and At Neeleman. The behavior of open in (47) and (48) matches the behavior of the verbal particles out and off in (49) and (50):8 (47) a. b. c. d.

7

8

The The The The

children children children children

cracked the nuts open. cracked open the nuts. cracked open and ate the nuts. collected and cracked open the nuts.

Here are the other examples: Line 1494: Wie soll ich denn offen [% 'auf ] machen! (How shall I open it?). Line 1498: Auch ein Löffel damit offen machen [% meint er möchte den Joghurtbecher mit einem Löffel öffnen]. (Means he wants to open the yogurt container with a spoon.) Examples (47c,d) and (48c,d) are modeled after examples from Neeleman (1994). Neeleman discusses V-V coordination as a major argument for the existence of complex predicates, and gives the examples John cut open and ate a watermelon yesterday and John bought and cut open a watermelon yesterday (p. 33). For the relevant properties of verbal particles, see Emonds (1972, 1985), Fräser (1976), and Neeleman (1994). Examples (50a,b) are from Emonds (1985:253).

192

Angelika Kratzer

(48) a. b. c. d.

The The The The

police police police police

broke the door open. broke open the door. broke open and removed the door. found and broke open the door.

(49) a. b. c. d.

We We We We

threw out threw the threw out identified

(50) a. b. c. d.

You You You You

shouldn't shouldn't shouldn't shouldn't

the documents. documents out. and shredded the documents. and threw out the documents. put off such tasks. put such tasks off. put off and neglect such tasks. neglect and put off such tasks.

In contrast, the resultatives in (51) to (53) seem more reluctant to follow the pattern of (47) to (50): (51) a. They painted the barn green. b. ? They painted green the barn. c. ? They painted green and sold the barn. d. ? They bought and painted green the barn. (52) a. He wiped the desk clean. b. ? He wiped clean the desk. c. ? He wiped clean and repainted the desk. d. ? He examined and wiped clean the desk. (53) a. They watered the tulips flat. b. ? They watered flat the tulips. c. ? They watered flat or picked the tulips. d. ? They picked or watered flat the tulips. The difference between (47) and (48) on the one hand, and (51) to (53) on the other suggest that particles can, but adjectives cannot that easily form PF-visible compounds with verbs. Using a technique from Neeleman (1994), the (c) and (d) examples of (47) to (53) have coordinations where one of the conjuncts is a simple transitive V. Assuming parallelism coordination conjoins equals with equals - we can reason that, if those examples are good, the other conjunct has to be a transitive V as well. If they are bad, there is likely to be a problem with compounding. So far so good, but why do we only find question marks, rather than stars in (51) to (53)? If in English, particles can form PF-visible compounds with verbs, but adjectives can't, shouldn't the (b), (c), and (d) examples of (51) to (53) be outright unacceptable, rather than merely marginal? The likely reason for the question marks - rather than the expected stars - in (51) to (53) is the marginal availability of alternative syntactic parses that do not involve compound verbs. Take the (b) examples in (47) to (53). They could all be parsed as not-so-perfect

Building

resultatives

193

cases of H e a v y N P Shift. T h e direct object would then be extraposed, and there w o u l d be no c o m p o u n d verb. A n analysis of particle constructions along those lines w a s in fact proposed by Richard K a y n e (1985). (54)

He wiped

t,

I

clean

[the desk],

f

T o discourage such a parse, I deliberately kept the direct objects short and light. A s for the (c) and (d) e x a m p l e s in (47) to (53), they also h a v e alternative parses. A s N e e l e m a n observes, they could b e not-so-perfect Right N o d e Raising constructions. (55) is a typical Right N o d e Raising example: 9 (55)

His s e c o n d h a n d quotations distort, m o r e than they represent his authors.

In (55), the object his authors is shared b y t w o verbs, distort and represent. A better sounding Right N o d e Raising version of, say, (52c) f r o m a b o v e could n o w look as follows: (56)

H e first w i p e d clean, and then repainted the desk he had inherited f r o m his grandfather.

A Right N o d e Raising parse does not involve c o m p o u n d verbs (see e.g. Phillips (1996) for a syntactic analysis). It c o m e s with a f e w strings attached, however. It requires a particular intonation, including f o c u s on both verbs, as s h o w n in recent work by Lisa Selkirk 1 0 . T h e alternative parses, then, are expected to b e marginal and dispreferred out of context. W h e n sentences needing unusual intonation conditions are presented in written f o r m and without contextual support, it takes s o m e effort to c o m e up with the discourse conditions that w o u l d m a k e them acceptable. In contrast to (51) to (53), the (b), (c), and (d) e x a m p l e s of (47) to (50) do not need special intonation conditions or h e a v y objects to b e j u d g e d acceptable. T h i s suggests that they d o not have to rely on marginal H e a v y N o u n Phrase Shift or Right N o d e Raising parses, but allow fully acceptable V - V - c o n j u n c t i o n parses at PF, with one of the c o n j u n c t s being a c o m p o u n d verb. It seems, then, that in English, particles do indeed d i f f e r f r o m adjectives in being able to f o r m PF-visible c o m p o u n d verbs. W e n o w seem to h a v e the a r g u m e n t w e were after: English open can be optionally parsed as a v e r b particle. T h e fact that open c o m b i n e s with transitive and unaccusative verbs, then, does not underm i n e the generalization that adjectival resultatives are impossible with transitive and unaccusative verbs. Let m e s u m m a r i z e w h a t this section has accomplished. I have been building a case in favor of a raising analysis for all adjectival resultatives. A raising analysis for adjectival resultatives is attractive since it automatically accounts for the tight connection between direct objects and resultatives. T h e adjective in a resultative construction is predicated of the v e r b ' s direct object, because the a d j e c t i v e ' s a r g u m e n t has b e c o m e that very s a m e direct object. It has m o v e d out of the projection of the adjective into the projection of the verb. It

9

N Y Times Book Review, Lisa Selkirk personal communication.

10

Lisa Selkirk personal communication.

194

Angelika Kratzer

could only do so because the verb didn't have a direct object of its own. Such an analysis predicts that only verbs that have unergative uses can appear in this construction. Transitives and unaccusatives are barred. Attractive as it may be, we saw that the proposed raising analysis faced a long and frightening list of apparent counterexamples, even after setting aside directional constructions, which require a different analysis." I examined the problem cases one by one. I argued that in some cases, what looked like an adjective could be parsed as an adverb. In other cases, an apparently transitive or unaccusative verb turned out to be unergative or had at least some unergative uses. And in at least one case, we saw that an adjective might have crossed over into the particle domain. The results I obtained for German and English should make us all skeptical about reports of transitivity or unaccusativity from lesser known languages. You need to apply complicated diagnostics to identify those properties. There are still a number of issues that have to be attended to for the proposed analysis to be even remotely plausible. The most pressing one has to do with Carrier & Randall's objection to the idea that in resultatives like those in (57) the verbs water and wipe are used intransitively, and that, consequently, the surface direct objects are not semantic arguments of those verbs. (57) a. b.

The gardener watered the tulips flat, The butler wiped the table clean.

For (57a), we want to derive that the tulips became flat as the result of having been watered, and (57b) should be shown to imply that the table became clean as a consequence of having been wiped. In the following section, I will argue that those inferences follow from the semantics of resultative constructions.

4.

Interpreting resultative constructions

I want to begin the semantic investigation of adjectival resultatives by looking at our old teapot example, which is a good starting point, since in that case, nobody would claim that the teapot should be an argument of drink: (58)

die Teekanne leer trinken the teapot empty drink 'Drink the teapot empty'

In Rappaport Hovav & Levin (2001), the authors retract the Direct Object Restriction, prematurely, I think. The apparent counterexamples mostly involve directionals.

Building resultatives

195

Resultative constructions like (58) are causative constructions. 12 They are "concealed causatives", though, in the terminology of Bittner (1999), since there is no visible carrier of causative meaning. All we seem to have in (58), for example, is an adjective, an intransitive verb, and a DP. Dowty (1979) proposed that the causal relation in concealed causatives is introduced by a construction specific interpretation rule working in tandem with a syntactic rule combining a transitive verb with an adjective to produce a verb-adjective compound. More recent work on semantic composition has done away with construction specific interpretation rules in favor of a very restricted set of general composition principles that apply freely. We have to explore the question, then, whether resultatives are likely to involve a freely available composition principle that introduces a causal relation. To have an example to look at, let us try to build up the meaning of (58). Here is the beginning of a derivation within a system of indirect interpretation, where a translation function Τ assigns expressions of an extensional typed λ-calculus to expressions of natural languages: 1. 2. 3. 4.

Τ (trinken) = λ ε ^ α ί ο η (e) & drink(e)] Τ (die Teekanne) = the teapot Τ (leer) = λχ ε λ β ^ Ι ^ Ι ε ^ ) & empty (x)(s)] Τ ((die Teekanne) leer) = λΒ,,Ι^Ιε^) & empty(the teapot)(s)] from 2. & 3. by Functional Application.

In the λ-calculus used here, variables of type e (in this case just "x") range over individuals, variables of type s ("e" and "s") range over eventualities, including events proper and states. I have used the variable "s" (not to be confused with the type s) as a convenient way to remind us that we are dealing with states. Available composition operations include Functional Application, Predicate Abstraction, and conjunction operations for predicates, including Event Identification. The computation displayed here catches the syntactic derivation at a stage where the direct object die Teekanne is still within the projection of the adjective leer. The denotation of the AP die Teekanne leer is a property of states that is true of any state that consists in the teapot's being empty. In order to compute the denotation of the VP die Teekanne leer trinken, the denotation of the AP, which is a property of states, and the'property of being a drinking action have to be combined. If we try to do this via Event Identification, we will end up with the empty property, and the computation will crash. There is no eventuality that is both a state and an action. Following the spirit of Bittner (1999), we might salvage our semantic computation by positing a type shift that allows a property of states to combine with a property of events by introducing a causal interpretation. Since an operation that turns adjectival roots into the corresponding causatives is independently needed to derive the meanings of verbs like flatten, blacken, and so on, let us assume (at least temporarily) that there is a type shifting operation that maps properties of states into the denotation of the corresponding causative. Within the present framework, this type shifting operation would correspond to a freely available composition operation that maps properties of states into properties of events proper. Not fussing about the distinc-

12

See e.g. Lakoff (1972), Dowty (1979). Dowty (1979) is the classic work on causatives in the formal semantics tradition. Parsons (1990) discusses causatives within an event semantics. Bittner (1999) is a recent study in cross-linguistic semantics.

196

Angelika Kratzer

tion between presuppositions and truth-conditional content, we might consider the following shift: (59)

Causative Shift Ρ

=>

λε 5 3s s [state(s) & event(e) & P(s) & CAUSE(s)(e)]

The predicate ' C A U S E ' in (59) will have to do the main work. ' C A U S E ' is meant to stand for a causal relation. Which one? It doesn't have to be the same relation as the relation expressed by the English verb cause, a fact discussed by many writers on the topic. 13 Nor does it have to be the same relation as the one expressed by make. Overt causatives like make or cause express what is usually referred to as 'indirect causation', a relation that allows for possibly very long causal chains connecting the mentioned cause to the mentioned effect. The causal relation in concealed causatives is always direct, a generalization that Bittner (1999) states as follows: (60)

Bittner's Generalization If a causal relation is syntactically concealed (only its arguments are overtly expressed), then it is semantically direct (no intermediate causes).

What does it mean that there are no intermediate causes? A promising starting point for a possible answer comes from an observation in Carl Ginet's (1990) book On Action.14 Adapted to our teapot example, Ginet's observation is that we have to distinguish between a drinking action that causes the teapot to be empty, and a drinking action that is an event of causing the teapot to be empty. Put more generally, the distinction is between an event c that causes an effect e and an event c that is a causing of an effect e. In the second case, e is a part of c, in the first case, it is not. Here is an illustration. Suppose my drinking all the water in the well causes your teapot to be dry. The reason is that, without any water left, there just isn't any more tea to be had. This is a case of indirect causation. In such a situation, I did not drink your teapot dry. In Ginet's terms, my action was an event that caused your teapot to be dry. It was not an event of causing your teapot to be dry. Not every part of the causal chain leading from what I did to the deplorable state of your teapot was part of my drinking activities. There was that extra stretch - the part of the chain that led from the emptiness of the well to the dryness of your pot that is not part of my drinking. Let me make this more precise.

13

14

The (1986) postscript to "Causation" in David Lewis' Philosophical Papers discusses some important differences between the 'CALSE' needed here and English cause. See also Dowty (1979) and Bittner (1999). "...it is natural to think that what is designated by a phrase of the form "S's causing E" is an event or episode that includes Ε as a part, as well as the causal relation between Ε and whatever S did to cause E; so S's causing R to become offended (by voting against a proposal) must be distinct from S's voting against a proposal, because the first has a part, R's becoming offended, that the second lacks." Ginet (1990:59). On p. 60, Ginet denies explicitly the following identity statement from Davidson (1980:58): "Doing something that causes a death is identical with causing a death."

Building

resultatives

197

We might start with a primitive notion of causation between events, or else adopt the one in Lewis (1973), and see where it takes us. We would begin with a definition of causal dependence. Let e and c be two distinct actually occurring events in our universe of events E.15 Then e depends causally on c just in case e wouldn't have occurred if c hadn't. In this first step, then, causal dependence is reduced to counterfactual dependence. I will not introduce the machinery needed to define counterfactual dependence. An intuitive understanding will do for our purposes, and the technical details can be found in Lewis' work. Still following Lewis, the relation 'e is caused by c' is defined as the transitive closure of the relation of causal dependence. 16 The causation relation we are after is the inverse of the relation 'e is caused by c'. There are now certain convex subsets of Ε that are linearly ordered by the causation relation we have just defined. Those sets are causal chains. All members of a causal chain are connected by the causation relation, then. By requiring causal chains to be convex, we make sure that no relevant intermediate causes can be skipped. More formally, the requirement says that whenever there are events c and e in a causal chain C, and there is an event c' in Ε that is caused by c and causes e, then c' must be in C as well. Since Ε only contains contextually relevant events, irrelevant intermediate causes can be ignored. A maximal element of a causal chain C is an event in C that does not cause any of the others in C. Likewise, a minimal element of a causal chain C is an event in C that is not caused by any of the others in C. If causal chains have maximal or minimal members at all, they always have unique ones. We now have the main ingredients we need. (61) a.

Events of causing other events An event c is an event of causing an event e iff c is the sum of all the members of some causal chain with maximal element e.

b.

Events that cause other events An event c is an event that causes an event e iff c is the minimal element of some causal chain with maximal element e.

(61a), rather than (61b), gives us the meaning of 'CAUSE' according to what I am proposing. To see what the definition does, let us go back to the type-shifting rule (59). The expression 'CAUSE(s)(e)' appearing there is intended to mean that e is an event of causing s, and not that e is an event that causes s. We can now continue with the abandoned computation from above, and apply Causative Shift to line 4. Here is the result: 5.

15

16

λ β ^ , , ^ ε ^ ) & event(e) & empty (the teapot)(s) & CAUSE(s)(e)]

The definition of causal dependence I am about to present is not quite the one Lewis gives as his definition of causation. My definition only defines causation as a relation between actual events. This simplification seems to be justified for our purposes. See Lewis (1973) and the postscript in Lewis (1986) for why causal dependence and causation have to be distinguished. In a nutshell, the situation is as follows: Causal dependence is defined as counterfactual dependence. Counterfactual dependence is not a transitive relation, but causation is. Here is a definition of the notion 'transitive closure': Where R is a binary relation, then Trans(R) is the smallest set satisfying (i) and (ii): (i) R is a subset of Trans(R), (ii) if and are in Trans(R), then is, too.

198

Angelika Kratzer

Causative Shift has mapped the property of states that is true of any state that consists in the teapot's being empty into the property of events that is true of any event that is an emptying of the teapot - an event of causing the teapot to be empty, that is. Applying Event Identification to step 5 yields the denotation for the VP in our teapot example: 6.

Τ (((die Teekanne) leer) trinken) = Xes5ss [action(e) & drink(e) & state(s) & empty(the teapot)(s) & CAUSE(s)(e)]

The denotation of (58), then, is the property of actions that is true of any action that is a drinking and is also a completed event of causing the teapot to be empty. Here is the important part: A drinking event is identified with a completed event of causing the teapot to be empty. We can now reason that if an action of drinking is identical to a completed action of causing the teapot to be empty, then what was drunk is bound to be the content of the teapot. A similar piece of reasoning can be applied to examples like (61): (61)

The butler wiped the table clean.

According to the analysis of resultatives I have just proposed, the VP of (61) describes a property of actions that is true of any action that is a wiping activity and is also a completed action of causing the table to be clean. We can again infer that if a wiping activity was identical to a completed action of causing the table to be clean, then what was wiped was bound to be the table. This is how a raising analysis of sentence (61) can account for the inference that the table was wiped, even though the DP the table does not start out as an argument of wipe. The semantics delivers that result free of charge, given certain plausible assumptions about the extensions of verbs like 'drink' or 'wipe' which I will address shortly. The key concept was the subtle distinction between an event that causes the table to be clean and an event of causing the table to be clean. I suggested that the linguistically significant difference between direct and indirect causation is closely related to that important but easy to overlook distinction that Carl Ginet drew attention to. Before moving on, I want to reflect on what we have to assume about the extension of predicates like drink or wipe for my account of resultatives to work out. Take again the example where my drinking all the water in the well causes your teapot to be dry. We saw earlier that in such a situation, I did not drink your teapot dry. This is a clear judgment, which should come out right on our analysis. Suppose e is my drinking all the water in the well, and s is your teapot's being dry, or just a part of your teapot's being dry, the very beginning part, say. Then causal chains leading from e all the way to s are not in the extension of the predicate drink under normal assumptions. For such causal chains to be in the extension of drink, the state s would have to be part of a drinking event. Borrowing imagery from Parsons (1990), s would have to overlap the culmination part of a drinking event, which is not the case in our scenario. To have another example, take my walk to the post office this morning. That walk consisted of a development part, the walking, and a culmination part, which we can think of as an initial part of the state of my being at the post office. Likewise, the development part of your wiping the table clean is the wiping activity, while the culmination part is an initial segment of the state of the table's being clean that your action brought about.

Building

resultatives

199

All of this fits well with what others have said about resultatives. Following Levin & Rappaport Hovav (1995), Wechsler (1997), and Rappaport Hovav & Levin (2001), Rothstein (2001) proposes a semantic analysis of resultatives based on the intuition that "what resultatives do is give information about the state initiated by the culmination point of an event" (Rothstein 2001:158f). An essential feature of Rothstein's account, which is preserved in mine, is that the directness of the causal relation in resultatives crucially depends on what event-plus-state complexes we are willing to admit to the extension of a simple activity predicate like drink ,17 Drink does not require the events it describes to culminate, nor does it impose any conditions on culmination parts. But this does not mean that drink cannot have event-plus-state complexes in its extension. It is those event-plus-state complexes that constrain the resultatives it might combine with. On such an approach, the notions of 'direct' and 'indirect causation' that are reflected in the causative vocabulary of natural languages are intimately tied to the extensions of predicates. The direct causation interpretation of adjectival resultatives, for example, is produced by identifying the events described by the participating verb with causal chains leading to a state described by the adjective. Which of those causal chains qualify as 'direct' is determined by the denotation of the verb. 'Directness' is not an inherent property of the causal chains themselves, then.

5.

A n affix instead of a type shift

The semantics for resultatives I presented in the last section yields the right result and meets Carrier & Randall's challenge, but, as is, it still has a blemish that it shares with Bittner's analysis. The Causative Shift operation introduces an unorthodox kind of composition principle. Intuitively, Causative Shift contributes 'lexical meaning'. Restrictive systems of composition principles or type shifts shouldn't include operations of this kind. A plausible constraint for possible semantic composition operations or type shifts is that they be 'logical' or 'topic neutral' in the sense of van Benthem (1.986). Causative Shift is not a topic neutral or logical operation and is therefore an unlikely candidate for a semantic composition principle or type shift. If there is no composition principle like Causative Shift, and there is no overt lexical item contributing the causal relation in resultatives, the only remaining option seems to be that the causal relation in resultatives must be introduced by an

17

Rothstein (2001: 159) has a composition operation, Resultative Conjunction, which construes the events described by a resultative construction as the set of sums consisting o f a state s described by the adjective and an event e described by the verb such that the culmination part of e is part of s. Given the notion o f ' c a u s a l chain' I introduced earlier, w e can define the notion 'culmination part o f as follows: (i) A state s is a culmination part of an event e iff e is the sum of some causal chain with maximal element s. (ii) A state s is the culmination part o f an event e iff s is a culmination part o f e and for all s': if s' is a culmination part of e, then s' < s. Since events do not determine unique causal chains they are the sums of, an event can have more than one culmination part, even if any given causal chain can only have one maximal element.

200

Angelika Kratzer

invisible lexical item. Minimally, such an item could be an unpronounced morpheme consisting of an interpretable feature [cause]. If it is to have the same semantic effect as Causative Shift, [cause] should have the following denotation: (62)

T([cause]) = λΡ λε 5 3s s [state(s) & event(e) & P(s) & CAUSE(s)(e)]

According to (62), [cause] introduces an event argument, but crucially, doesn't introduce any other argument, a causer argument, for example. It is clear that it shouldn't do so if it is to be used in resultative constructions. Interestingly, Liina Pylkkänen has argued that causative heads should never introduce causer arguments (Pylkkänen 2002). As Pylkkänen points out, her claim might initially seem implausible since it appears to be a universal fact that causative verbs have causer arguments that the non-causative stems they are derived from lack. It would thus be natural to assume that causatives are derived via the addition of a head that also adds a causer argument. However, Pylkkänen shows in her dissertation that the apparent universal generalization is wrong. Causativization does not always add a causer argument. According to Pylkkänen, "what universally distinguishes causative verbs from their non-causative counterparts is a syntactically implicit event argument ranging over causing events. Specifically, I will argue that all causative constructions involve the head CAUSE which combines with non-causative predicates and introduces a causing event to the semantics..." (Pylkkänen 2002:75). As for causer arguments, they would now be introduced by independent heads, [active] voice heads, for example. If the causative interpretation in adjectival resultatives is brought about by an unpronounced affix, we might be able to understand certain constraints for those constructions. In his (1984) dissertation, Nigel Fabb observes that the -ing participles of experiencer verbs have all the properties of adjectives, except that they can't appear in resultatives: (63) a. b. c. d.

* I cooked it disgusting. * I brewed it soothing. * She knocks herself frightening. * He dances himself embarrassing.

Similar cases, including some involving the -ed participle suffix, are reported in Simpson (1983), Smith (1983), and Carrier & Randall (1992). 18 (64) a. b. c. d. e. f. g. h.

18

* The maid scrubbed the pot shined / shining. * The jockeys raced the horses sweating. * She knocks herself frightening. * The chef cooked the food blackened / charred. * The joggers ran themselves sweating / exhausted. * The kids laughed themselves sickened. * The chef cooked the kitchen walls blackened. * The tourists walked their feet blistered.

All examples in (64) are quoted from Carrier & Randall (1992:184).

Building

resultatives

201

If [cause] is carried by a derivational affix, we expect that affix to be submitted to constraints that go beyond mere interpretability. 19 Derivational affixes can only attach to bases that satisfy certain conditions, which may be phonological, morphological, syntactic or semantic in nature. English derivational affixes, for example, have been claimed to fall into two classes, referred to as "class I" and "class II" in Siegel (1974). Selkirk (1982) distinguishes between Root and Word affixes. More recently, Jennifer Hay proposed a "complexity based" account for the ordering of derivational affixes (Hay 2000, Hay & Plag 2004). The basic idea behind Hay's analysis is that a derivational affix that can be easily "parsed out" should not occur inside one that cannot. Such an analysis seems to imply that unpronounced derivational affixes should never occur outside of pronounced ones. Unpronounced derivational affixes are bound to be the best possible instances of derivational affixes that cannot be "parsed out". If that is so, the ungrammaticality of (63) and (64) follows. If [cause] is carried by an unpronounced derivational affix, it should not be able to appear outside of the suffixes -ing and -ed, which are highly separable - if indeed they are derivational affixes at all. If -ing and -ed are not derivational affixes, but pieces of inflection, a derivational affix like [cause] is even less expected to attach outside of -ing or -ed. Whatever the right account of affix ordering may turn out to be in the end, a morphological explanation for the ungrammaticality of (63) and (64) would not be expected if the causal interpretation of adjectival resultatives was brought about by Causative Shift. Such a type shift would only see the semantic types of the input adjectives. As long as the denotations of the adjectives in (63) and (64) are of the same type as the denotations of simple adjectives (admittedly not a necessary assumption), Causative Shift would not be able to distinguish between good adjectival resultatives and the cases in (63) and (64). Even if we tried to constrain the application of Causative Shift, we would still have to explain why typeshifting operations should be constrained in that particular way. On the other hand, if [cause] is a feature carried by an affix, constraints of the kind exemplified by (63) and (64) are expected to follow from independently needed accounts of the ordering constraints for affixes. The feature [cause] whose meaning is defined in (62) does not only appear in adjectival resultative constructions. In English, it seems to be responsible for turning adjectives into verbs, without outward sign of the change, as in (65a), or with the help of a pronounced suffix, as in (65b): 20 (65) a. b.

Empty, dry, clean, cool, dim, dirty .... Flatten, shorten, blacken, sweeten, stiffen...

If there are pronounced or unpronounced affixes that carry [cause] and are thus capable of turning stative roots into eventive predicates, an important question pops up immediately. Why is (66a) grammatical, but (66b) is not? (66) a. The gardener watered the tulips flat. b. * The gardener watered the tulips flatten.

19 20

Pesetsky (1995) pioneered investigations based on reasoning of this kind. See Levin (1993:28) and Levin & Rappaport Hovav (1995:95f) for longer lists.

202

Angelika Kratzer

If one and the same feature [cause] is involved in adjectival resultatives like (66a), as well as in verbal causative constructions like (66b), why do we see a morphological difference in the shape of the two causatives? Here, too, a crucial ingredient for a possible answer might come from the work of Liina Pylkkänen. Pylkkänen (2002) argues that while causative features and voice features are different features, they may be "bundled together" in some languages, and thus get spelled out together. If that is so, the suffix -en offlatten might not merely spell out the feature [cause]. It might spell out a bundle of features, including [cause] and a voice feature, which could be [active] or [non-active] in this case, thus generating transitive and intransitive alternants. That both transitive and intransitive verbs have a voice feature is in line with Alexiadou & Anagnostopoulou (2004:119), who report that there is a recent trend in the literature to assume voice inflection to be present with transitives and intransitives. Representatives of this trend are Harley (1995), Collins (1997a), and Embick (2004). That even the intransitive alternants of the verbs in (65a,b) are causatives is argued in Chierchia (2004, circulated since 1989) and Levin & Rappaport Hovav (1995). The events described by inchoatives and anticausatives can then be seen as internally, rather than externally caused in the sense of Levin & Rappaport Hovav. The intransitive VP in the sentence The sauce thickened, for example, would describe events of causing the sauce to be thick(er), where the prominent cause for that event can be (but doesn't have to be) linked to properties inherent in the sauce itself. A commitment to a prominent external cause like an agent or a force of nature would be contributed by [active] voice. Chierchia's and Levin & Rappaport's proposals have the interesting consequence that the commonly posited BECOME operator becomes superfluous in the decomposition of inchoatives, causatives, and anticausatives. Those three types of verbs are all causatives. They differ with respect to voice. It is quite plausible, then, to assume that the suffix -en of the verbs in (65b) spells out [cause] and a voice feature together in both the transitive and the intransitive alternants. What we would still want to know, however, is why the adjectival root in resultative constructions can combine with an affix carrying [cause], but not with an affix carrying a bundle consisting of [cause] and a voice feature. What is so special about voice features? In contrast to [cause], voice features are inflectional features according to Kratzer (1996). It might be, then, that for some still unknown reason, an adjectival root + [cause] compound is prevented from putting on any kind of inflectional morphology in a resultative construction. Why should that be so, however? A possible clue comes from an important observation by William Snyder (Snyder 1995, 2001). Snyder found that there is a correlation between the availability of resultatives and the availability of productive root compounds, both across languages and in the course of language acquisition. In the following section, I will try to connect Snyder's observation to an explanation for why sentences like (66b) are so bad.

203

Building resultatives 6.

Resultatives and serialization

I want to begin this section by reflecting one more time on the stepwise derivation of adjectival resultatives. At some point in the syntactic derivation, we might have built a structure of the following kind (neglecting any kind of DP-movement), going through different stages in the spirit of Chomsky (2001):

Stage 1

Merge leer Merge die Teekanne Interpretation: Combine the translations of die Teekanne and leer: Xs empty(the teapot)(s)

Stage 2

Merge [CAUSE] Interpretation: Head movement:

Stage 3

Merge trinken Interpretation:

Combine the translations of [CAUSE] and its sister node: Xe3s [empty (the teapot)(s) & CAUSE(s)(e)]. [cause] attracts leer to satisfy its affixal needs.

Combine the translations of trinken and its sister node: λεΞβ [empty (the teapot)(s) & drink(e) & CAUSE(s)(e)].

In stage 1, a head is combined with an argument. The result in turn provides an argument for [cause]. The root leer can now incorporate into [cause] to satisfy that feature's need for affixation. The merging pattern is disrupted in stage 3, where a head is introduced that embeds a constituent that is not its argument. This, I suspect, is a problem, even though it does not lead to uninterpretability. Let me explain why. Technically, the structure we have built looks like the closest analogue to a serial verb construction we might find in languages like English or German. Two independent eventive lexical predicates are piled on top of each other: trinken and /eer+[cause]. If we had chosen to continue the derivation beyond stage 3, the two predicates would have had to share a subject and all pieces of verbal inflection. Since there is no overt marker of coordination or subordination, the standard criteria for serial verb constructions are satisfied (Dechaine

204

Angelika Kratzer

1993, Collins 1993, 1997b). Interestingly, Gruber (1990), Collins (1997b), Gruber & Collins (1997), Nishiyama (1998), have all argued that there is a fundamental structural similarity between serial verb constructions and V-V compounds: V-V compounds of the kind we find in Chinese, Japanese, or Φ Hoan, for example, are syntactically derived from serial verb configurations, and serial verb configurations of the kind found in the Kwa languages involve compounding in the covert syntax. That is, in a serial verb construction with two verbs, for example, the lower verb inaudibly incorporates into the higher verb. If those authors are right, we can assume that whatever forces compounding for serial verb constructions can be assumed to force compounding for adjectival resultatives as well. What might that force be, though? One possibility I can think of is that there might indeed be something illicit about stage 3 in the derivation above: A second eventive head is merged that cannot take its sister constituent as its argument. Complex predicate formation may be required, then, to eliminate an illicit embedding configuration via 'clause union'. In our case, overt or covert head movement adjoins /eer+[cause] to the left of trinken. There are no semantic consequences of that step. While still speculative, the 'elimination of an illicit embedding configuration' hypothesis fits well with recent work by Marcin Morzycki (Morzycki 2001, forthcoming). Morzycki has argued that most adverbial phrases are not really modifiers, but arguments of functional heads. According to Morzicki's program of Mediated Modification, the apparent modifier interpretation of many different kinds of adverbial phrases is in fact the semantic contribution of the functional heads whose specifier positions those adverbials occupy (Cinque 1999). Morzycki's work has the consequence that locatives, as well as temporal, manner, and other types of adverbials, are all arguments of specialized heads. From this perspective, the configuration created in stage 3 of the derivation is highly anomalous. When the projection of a head is built, the creation of head argument configurations seems to be the driving force: Merge a head a, give it an argument b. In the next stage, merge a head c that takes a+b as an argument. Then continue by merging a head d that takes a+b+c as an argument, and so on. If there is a violation, overt or covert head movement offers a remedy by bringing about 'clause union'. Languages differ as to whether or not they unite the clauses overtly. Suppose that some such story can be told. How might that story help us explain the ungrammaticality of (66b)? Baker (1996, 2003) mentions and uses a generalization that he labels "Li's Generalization" (Li 1990) or "Proper Head Movement Generalization": (67)

The Proper Head Movement Generalization It is impossible to move from a functional category into a lexical category. Baker (2003:306)

As is, (67) is a mere generalization, and we would eventually want to have an explanation for it. For the time being, however, let us suppose that the generalization is true and not worry about how to make it follow from more general principles. 21 Turned into a constraint, 21

Following up on a suggestion by David Lebeaux (p.c. to Mark Baker, reported in Baker 2003:306), we might consider the possibility that the PHMG follows from an architecture where the lexical part and the functional part of a head's extended projection are built separately, and are fused at some point in the derivation (Lebeaux 1988). Movement of lexical heads into functional heads

Building resultatives

205

(67) excludes (66b) on the assumption that the suffix -en of flatten spells out [cause] and a voice feature together. Supposing that the features are spread out at the relevant stage of the derivation, we are committed to the following head movement chain when deriving (66b), assuming that we are dealing with the intransitive version o f f l a t t e n : (68)

Step 1 Step 2 Step 3

FLAT => FLAT+[cause] FLAT+[cause] => FLAT+[cause]+[non-active] FLAT+[cause]+[non-active] => FLAT+[cause]+[non-active] + WATER

In step 2, FLAT+[cause] adjoins to a functional head, the voice head [non-active]. In violation of the Proper Head Movement Generalization, the head created by step 2 adjoins to a lexical head again in Step 3. Assuming that [cause] is a derivational affix, no problems are expected to come up in the derivation of adjectival resultatives of the kind we have been examining. In those constructions, simple stative roots that are commonly used to build adjectives have chosen to become eventive by attaching the unpronounced derivational affix [cause]. The result is a causative with the looks of an adjective. In section 3 , 1 argued that a raising analysis is a possibility for all adjectival resultatives, by removing major apparent counterexamples. I have now spelled out such an analysis. What is still missing, though, is an explanation for why adjectival resultatives must have a raising analysis. What is it that forces us to project a raising structure, and thus prevents us from using transitive or unaccusative verbs in adjectival resultatives? Why couldn't a language learner come up with a parse of the following kind, for example (Bowers 1997, 2001) 22 ? VP

Figure 3

would now be part of the process that fuses the two types of structures after the lexical part o f the projection has been inserted into the functional part. Lebeaux's proposal is attractive since the PHMG would fall out from the very architecture of the syntactic engine. 22

Bowers posits a Pred(ication) head on top o f the adjective, though, rather than [cause]. He argues for a control structure for transitive resultatives, and a raising structure for intransitive resultatives.

206

Angelika Kratzer

In the structure of figure 3, the argument given to the adjective flat is an unpronounced pronoun that is anaphorically related to the verb's direct object, the tulips. In this way, the adjective and the verb can share an argument. There are requirements for the pronoun in this configuration. It shouldn't need case, for example, since there is only one objective case available. That case should go to the tulips. The pronoun posited in figure 3 must be PRO, then, rather than pro, which needs case like any overt pronoun. But PRO doesn't really fit the bill either. The known occurrences of PRO (primarily control infinitives in English) all occur in environments where a fair amount of functional structure intervenes between it and its antecedent. On accounts such as Finer (1984, 1985), Borer (1989) and Hale (1992), that functional structure is essential. It includes agreement morphology, which is responsible for establishing the anaphoric relationship between PRO and its antecedent: PRO enters a local agreement relation with an agreement head, which in turn enters a local agreement relation with the agreement head of the 'controlling' DP. There is no direct anaphoric relation between PRO and its 'controller', then. An apparently non-local anaphoric relation between two DPs is produced by a local anaphoric relationship between two inflectional heads. This type of account has interesting consequences beyond control infinitives. Finer originally developed it for switch reference phenomena. In Kratzer (1998) I invoked the same agreement mechanism for the analysis of sequence of tense, de se attitude reports, and apparent bound variable interpretations of indexicals. If this is the right way of thinking about PRO, the structure in figure 3 is ruled out for principled reasons. PRO needs agreement morphology to identify its features via an anaphoric relationship with a functional head. However, if the adjective flat projected agreement morphology, the constraint behind the Proper Head Movement Generalization would block clause union via incorporation, hence the derivation would crash in stage 3. A raising analysis is forced, then. A control analysis is not an option. The constraints for the participating verbs follow.

7.

Resultat!ves across languages

How come not all languages have resultatives? Famously, the Romance languages don't, for example. What is it that German and English can, but Romance languages can't do when it comes to resultatives? On the account of adjectival resultatives that I have been developing, a root that we usually associate with adjectives managed to become eventive, crucially without the help of any inflectional morphology. Why can't the same happen in Romance? Take French. In French, as in other Romance languages, both attributive and predicative adjectives must have agreement morphology. Manner adverbs usually appear with the suffix -ment, but there are also a few simple 'adjectives' that have adverbial uses, including court ('short'), doux ('sweet'), haut ('high'), and has ('low') (Grevisse 1964:311). Inflectionless adjectival roots are very rare, and there do not seem to be any adjectival resultatives. Here are two apparent exceptions from Legendre (1997:46f):

Building

resultatives

(69) a.

Pierre a peint les murs enblanc. Pierre has painted the walls in white 'Pierre painted the walls white.' II lui a coupe les cheveux court, he him/her has cut the hair short. 'He cut his/her hair short.'

b.

207

In (69a), the 'adjective' appears after a preposition, which suggests that it is used as a noun. (69b) is one of those resultative constructions where the apparent 'adjective' can be parsed as an adverb ('How did he cut his hair?'), and, as expected, court does indeed have purely adverbial uses.23 What is highly relevant for my plot is that when French uses adjectives within compounds, the adjectives like to be fully inflected, thus confirming Snyder's generalization:24 (70) a. b. c. d.

e. f.

Union chretienne - democrate Union Christian (fem.sing.) - Democratic (sing.) Les partis sociaux - democrates the parties social (masc.pl.) - democratic (pi.) Une jeune femme sourde - muette a young woman deaf (fem.sing.) - mute (fem.sing.) Les dames courtes - vetues the ladies short (fem. pi.) - dressed (fem. pi.) 'The short-skirted ladies' Gilberte Swann (nouvelle riche, snob) Gilberte Swann (new (fem. sing.) rich, snob) Herbes aromatiques fraiches cueillies herbs aromatic (fem. pi.) fresh (fem. pi.) picked (fem. pi.) 'Freshly picked aromatic herbs.'

In German A-A compounds only the second adjective can be inflected. Inflecting the first one, too, would result in severe ungrammaticality. We see a solid difference between German and French, then. Except for a few frozen cases, French adjectival roots cannot stand alone. They even inflect in compounds. In contrast, German adjectival roots appear bare in all predicative constructions, as manner adverbs, and within compounds. Suppose, then, that in contrast to German, French adjectives are fully inflected at the starting point of a syntactic derivation. There would never be a stage, then, where they appear without inflection. This would make it impossible to merge [cause] in stage 2 above. The derivational affix [cause] could not satisfy its affixal needs, since it could not attach outside of inflectional morphology. As for German adjectives, there would only be roots at 23

Grevisse (1964:311) quotes A. France's example Ses idees s'arretaient short'), a clearly adverbial use of court.

24

Grevisse forms of nes filles Grevisse

court ('His ideas stopped

(1964). All examples in (70) are the result o f a Google search. In some cases, uninflected the first adjective came up, too: Cette ft He sourd-muette ('This deaf-mute girl'). Des jeucourt-vetues ('Short-skirted girls'). That last form of court-vetu is the one sanctioned by (1964:318), w h o refers to a 1901 decree o f the French Minister of Education.

208

Angelika

Kratzer

the beginning of a syntactic derivation.25 Those roots could be inserted into different types of functional structure, then. Combined with nominal inflection, they could modify nouns. Combined with certain verbal functional heads, they could become adverbs, as Morzycki has proposed. With the help of a copula, they could hook up with verbal inflectional morphology. After attaching [cause], they could participate in serialization, or become fully inflected causative verbs. German adjectival roots can in fact do all those things. The possibilities for French adjectives are much more limited. They seem to be stuck with what looks like nominal inflection. That inflection is even present in adverbial forms ending on the suffix -ment, which attaches to the feminine form of adjectives, as in heureusement ('fortunately'), legerement ('slightly'), etc. The situation in some Italian dialects is even more dramatic. Adverbs have basically disappeared in southern Italian dialects south of the Gaeta-Rieti-Termini line. Inflected agreeing adjectives are used instead: (71)

I ligna sicchi addümanu boni the (masc. pi.) wood dry (masc. pi.) burn good (masc. pi.) 'Dry wood burns well.' Rohlfs (1969:243); glosses and translation are mine.

More cross-linguistic research is needed to follow up on the consequences of the suggested difference between German and French (Romance). As emphasized by Catherine Fabricius Hansen (personal communication), Norwegian presents an interesting test case for bold and (quite possibly) premature claims about the connection between resultatives and adjectival inflection. The following data come from Afarli (1984:33): (72) a. b.

c.

Vi vaska golvet reint. we washed the floor clean. Vi vaska rein(t) golvet. we washed clean the floor 'We washed the floor clean.' Golvet er reinvaska. the floor is clean-washed 'The floor is washed clean.'

Neuter agreement (obligatory) Neuter agreement (optional) Compound

The good news is that Norwegian seems to use uninflected adjectives for overt incorporation, as illustrated in (72b,c).26 The potential problem is covert incorporation. The agree-

25

26

Marantz (1997) and Borer (2004) are constructional proposals in this spirit. However, what we seem to be seeing in French is, that the most radical form of syntactic constructivism might not always be realized. Also, the lack of productive verb serialization in Indo-European languages suggests that verbs in those languages might start their life in syntax with at least some piece of inflection. For an interesting explanation for why the incorporated version should appear as vaska rein in (72b), rather than rein vaska, as in (72c), see Collins (2002). According to Collins, the order vaska rein would be produced by left-adjoining vaska to the first functional head above VP (my [active] voice, his v). The adjective rein could then in turn be left-adjoined to [active], and would thus end up sandwiched between vaska and [active]. In the adjectival passive construction (72c), there is no

Building

resultatives

209

ment morphology we see in (72a,b) would have to be assumed to be a mere PFphenomenon that does not interfere with the covert incorporation of the adjective. This is not an impossibility, but without investigating the complete pattern of adjectival agreement in Norwegian, I feel I cannot proceed any further. For the time Ijeing, I let FabriciusHansen's challenge stand. I have presented an analysis of adjectival resultative constructions that explains the peculiar cluster of restrictions they are submitted to. A verb and an adjective can only come together under very special conditions. First and foremost, the adjective has to find a way to become eventive. The suffix [cause] can help. Together with Event Identification, [cause] produces a direct causation interpretation. The adjective's object needs case, and this prevents the verb from taking a direct object of its own. The adjective's object becomes the joint object of the verb-adjective pair. Being unergative, the verb can't embed the phrase projected by its adjectival mate. This forces clause union, and a complex predicate is born. Both parties have to pay a price, though. The verb can't be transitive or unaccusative. The adjective must be bare. That's in a nutshell the analysis of resultatives I have proposed. I have used resultatives as a probe into the architecture of the syntax-semantics interface. The analysis of resultatives I have come up with generates expectations about serialization. In a serial verb construction, a stack of VPs is interpreted via successive applications of Event Identification. Consequently, there are tight constraints on what kind of verbs can participate in the construction. Most run-of-the-mill event descriptions are not compatible with each other: I can laugh while dancing and move while sleeping, but no laugh can be a dance, and no sleep can be a move. On the other hand, a watering event can be an event of causing the tulips to be flat, and a drinking event can be an event of causing your teapot to be empty. As long as VPs can describe such causing events without the help of inflection, we should find causal interpretations in serial verb constructions. We saw that in German and English, the availability of an unpronounced derivational suffix [cause] seems to produce a marginal case of serialization. What other types of event identifications might be possible in principle? A walking event could be identified with an event that has a particular purpose, for example, like buying a refrigerator or talking to my boss. If VPs could describe such events without the help of inflection, we would expect to find serial verb constructions with purpose interpretations. We should be looking for inflectionless VPs with meanings corresponding to English in order fo-infinitivals, then. More generally, the major constraints on verb serialization should be jointly determined by the operation of Event Identification and the expressive possibilities for bare VPs. The next step is now to find out what inflectionless VPs can mean in the languages of the world.

voice, even if adjectival participles can be constructed in the syntax (see Kratzer 2000). Only reinvaska should be possible, then - rein can only incorporate into (that is, left-adjoin to) the verb.

210

Angelika

Kratzer

References

Afarli, Τ. Α. (1984): "Norwegian Verb Particle Constructions as Causative Constructions." - In: Working Papers in Scandinavian Syntax. Trondheim. Alexiadou, A. & Anagnostopoulou, E. (2004): "Voice Morphology in the Causative-Inchoative Alternation: Evidence for a Non-Unified Structural Analysis of Unaccusatives." - In: A. Alexiadou, E. Anagnostopoulou & M. Everaert (eds.): The Unaccusativity Puzzle. Explorations at the SyntaxLexicon Interface, 114-136. Oxford: Oxford University Press. Baker, M. C. (1996): The Polysynthesis Parameter. - New York: Oxford University Press. - (2003): Lexical Categories. Verbs, Nouns, and Adjectives. - Cambridge: Cambridge University Press. Benthem, J. van (1986): Essays in Logical Semantics. - Dordrecht: D. Reidel Publishing Company. Bittner, M. (1999): "Concealed Causatives." - In: Natural Language Semantics 7, 1-78. Borer, Η. (1989). "Anaphoric AGR." - In: O. Jaeggli & K. J. Safir (eds.): The Null Subject Parameter, 69-109. Dordrecht: Kluwer Academic Publishers. - (2004): Structuring Sense. - Oxford: Oxford University Press. Bowers, J. (1997): "A binary analysis of resultatives." - In: R. C. Blight & M. J. Moosally (eds.): Texas Linguistic Forum 38: The Syntax and Semantics of Predication, 43-58. University of Texas at Austin. - (2001): "Predication." - In: M. Baltin & C. Collins (eds.): The Handbook of Contemporary Syntactic Theory, 299-333. Oxford: Blackwell Publishers. Carrier, J. & J. H. Randall (1992): "The Argument Structure and Syntactic Structure of Resultatives." - In: Linguistic Inquiry 23(2), 173-234. Chierchia, G. (2004): "A Semantics for Unaccusatives and its Syntactic Consequences." - In: A. Alexiadou, E. Anagnostopoulou & M. Everaert (eds.): The Unaccusativity Puzzle. Explorations at the Syntax-Lexicon Interface, 22-59. Oxford: Oxford University Press. Chomsky, N. (2001): "Derivation by Phase." - In: M. Kenstowicz (ed.): Ken Hale. A Life in Language, 1-52. Cambridge/Mass.: The MIT Press. Cinque, G. (1999): Adverbs and Functional Heads. A Cross-Linguistic Perspective. - New York: Oxford University Press. Collins, C. (1993): Topics in Ewe Syntax. - Doctoral dissertation, MIT. - (1997a): Local Economy. - Cambridge/Mass.: The MIT Press. - (1997b): "Argument Sharing in Serial Verb Constructions." - In: Linguistic Inquiry 28, 461-497. - (2002): "Multiple Verb Movement in φ Hoan." - In: Linguistic Inquiry 33(1), 1-29. Davidson, D. (1980): Essays on Actions and Events. - Oxford: Clarendon Press. Dechaine, R.-M. (1993): Predicates Across Categories. - Doctoral dissertation, Amherst. Dowty, D. R. (1979): Word Meaning and Montague Grammar. - Dordrecht: Reidel Publishing Company. Duden (1972): Zweifelfälle der Deutschen Sprache. - Mannheim, Bibliographisches Institut. Embick, D. (2004): "Unaccusative Syntax and Verbal Alternations." - In: A. Alexiadou, E. Anagnostopoulou & M. Everaert (eds.): The Unaccusativity Puzzle. Explorations at the Syntax-Lexicon Interface, 137-158. Oxford: Oxford University Press. Emonds, J. (1972): "Evidence that Indirect Object Movement is a Structure Preserving Rule." - In: Foundations of Language 8, 546-561. - (1985): A Unified Theory of Syntactic Categories. - Dordrecht: Foris. Fabb, N. A. J. (1984): Syntactic Affixation. - Doctoral dissertation, MIT. Finer, D. (1984): The Formal Grammar of Switch Reference. - Doctoral dissertation, Amherst. - (1985): "The syntax of switch- reference." - In: Linguistic Inquiry 16(1), 35-55.

Building

resultatives

211

Fräser, B. (1976): The Verb-Particle Combination in English. New York : Academic Press. Giannakidou, A. & J. Merchant (1998): "Why Giannis can't scrub his plate clean: On the absence of resultative secondary predication in Greek." - 3rd International Conference on Greek Linguistics, University of Athens, Athens. Ginet, C. (1990): On Action. - Cambridge: Cambridge University Press. Grevisse, M. (1964): Le bon usage. Grammaire Frangaise. 8 th ed. - Paris- Librairie A. Hatier. Gruber, J. (1990): Thematic Form and the Head Conflation Parameter. - Ms., University of Benin, Nigeria. Gruber, J. & C. Collins (1997): "Argument Projection, Thematic Configurationality, and Case Theory." - In: A.-M. Di Sciullo (ed.): Projections and Interface Conditions, 130-154. Oxford: Oxford University Press. Hale, K. (1992): "Subject Obviation, Switch Reference, and Control." - In: R. K. Larson et al. (eds.): Control and Grammar, 51-77. Dordrecht: Kluwer Academic Publishers. Harley, H. (1995): Subjects, Events, and Licensing. - Doctoral dissertation, MIT. Hay, J. (2000): Causes and Consequences of Word Structure. - Doctoral dissertation, Northwestern University. Hay, J. & I. Plag (2004): "What Constrains Possible Suffix Combinations? On the Interaction of Grammatical and Processing Restrictions in Derivational Morphology." - In: Natural Language and Linguistic Theory 22(3), 565-596. Hoekstra, T. (1988): "Small Clause Results." - In: Lingua 74, 101-139. Kayne, R. (1985): "Principles of Particle Constructions". - In: J. Gueron, H.-G. Obenauer & J.-Y. Pollock (eds.): Grammatical Representation, 101-140. Dordrecht: Foris. Kempcke, G. (2000): Wörterbuch Deutsch als Fremdsprache. - Berlin: Walter de Gruyter. Kratzer, A. (1996): "Severing the External Argument from its Verb." - In: J. Rooryck & L. Zaring (eds.): Phrase Structure and the Lexicon, 109-137. Dordrecht: Kluwer. - (1998): "More Structural Analogies Between Pronouns and Tenses." - In: D. Strolovitch & A. Lawson (eds./· Proceedings of SALT VIII at MIT, 92-110. Ithaca: CLC Publications. - (2000): "Building Statives." - In: Lisa J. Conathan et al. (eds.): Proceedings of the 26th Annual Meeting of the Berkeley Linguistics Society, 385-399. Berkeley. Lakoff, G. (1972): "Linguistics and Natural Logic." - In: D. Davidson & G. Harman (eds.): Semantics of Natural Language, 545-665. Dordrecht: Reidel. Lebeaux, D. (1988): Language Acquisition and the Form of the Grammar. - Doctoral dissertation, Amherst. Legendre, G. (1997): "Secondary Predication and Functional Projections in French." - In: Natural Language and Linguistic Theory 15, 43-87. Levin, B. (1993): English Verb Classes and Alternations. A Preliminary Investigation. - Chicago: University of Chicago Press. - (1999): "Objecthood: An event structure perspective." - In: CLS 35, Part 1. Chicago, Chicago Linguistic Society: 223-247. Levin, Β. & M. Rappaport Hovav (1995): Unaccusativity. At the Syntax-Lexical Semantics Interface. - Cambridge/Mass.: The MIT Press. Lewis, D. (1973): "Causation." - In: Journal of Philosophy 70, 556-567. - (1986): Philosophical Papers, Volumes I and II. - Oxford: Oxford University Press. Li, Y. (1990): "X°-Binding and Verb Incorporation." - In: Linguistic Inquiry 21, 399-426. MacWhinney, B. (2000): The CHILDES Project: Tools for Analyzing Talk. 3rd edition, vol.2: The Database. - Mahwah/N.J.: Lawrence Earlbaum Associates. Marantz, A. (1997): "No Escape from Syntax.: Don't Try Morphological Analysis in the Privacy of your Home." - In: U. Penn Working Papers in Linguistics 4(2), 201-225.

212

Angelika

Kratzer

Morzycki, M. (2001): "Interpreting Measure DP Adverbials." - In: K. Megerdoomian & L. Bar-El (eds.): Proceedings ofWCCFLXXat the University of Southern California, 442-455. Somerville: Cascadilla Press. - (forthcoming): Modifier Interpretation and Functional Structure. - Doctoral dissertation, Amherst. Neeleman, A. (1994): Complex Predicates. - Utrecht: OTS-Publications. Nishiyama, K. (1998): "V-V Compounds as Serialization." - In: Journal of East Asian Linguistics 7, 175-217. Parsons, T. (1990): Events in the Semantics of English. A Study in Subatomic Semantics. - Cambridge/Mass.: MIT Press. Pesetsky, D. (1995): Zero Syntax. - Cambridge/Mass.: MIT Press. Phillips, C. (1996): Order and Structure. - Doctoral dissertation, MIT. Pylkkänen, L. (2002): Introducing Arguments. - Doctoral dissertation, MIT. Quirk, et al. (1985): A Comprehensive Grammar of the English Language. - London: Longman. Rappaport Hovav, Μ. & B. Levin (2001): "An Event Structure Account of English Resultatives." - In: Language 77(4), 766-797. Rohlfs G. (1969): Grammatica storica della lingua Italiana e dei suoi dialetti. Sintassi e formazione delle parole. - Torino: Giulio Enaudi. Rothstein, S. (2001): Predicates and their Subjects. - Dordrecht: Kluwer Academic Publishers. Selkirk, E. O. (1982): The Syntax of Words. - Cambridge: MIT Press. Siegel, D. (1974): Topics in English Morphology. - Doctoral dissertation, MIT. Simpson, J. (1983): "Resultatives." - In: L. Levin, M. Rappaport & A. Zaenen (eds.): Papers in Lexical Functional Grammar. Bloomington: Indiana University Linguistics Club. Smith, C. (1983): "A Theory of Aspectual Choice." - In: Language 59, 479-501. Snyder, W. (1995): Language Acquisition and Language Variation. The Role of Morphology. - Doctoral dissertation, MIT. - (2001): "On the Nature of Syntactic Variation: Evidence from Complex Predicates and Complex Word-Formation." - In: Language 77, 324-342. Stowell, T. (1983): "Subjects Across Categories." - In: The Linguistic Review 2(3), 285-312. Tenny, C. L. (1987): Grammaticalizing Aspect and Affectedness. - Doctoral dissertation, MIT. Wagner, K. R. (1985): "How Much do Children Say in a Day?" - In: Journal of Child Language 12, 475-487. Wechsler, S. (1997): "Resultative Predicates and Control." - In: Texas Linguistic Forum 38.

Daniel Hole

Reconciling "possessor" datives and "beneficiary" datives Towards a unified voice account of dative binding in German

This paper argues for a neo-Davidsonian voice approach in the spirit of Kratzer (1996, 2003) to the syntax and semantics of dative arguments in German, and against syntactic or lexical theories of possessor raising. The voice approach is developed in some detail for so-called "possessor" datives, and later on extended to "beneficiary" datives. It is argued that, in both cases, the local contribution of the dative argument has to be kept strictly separate from an effect that arises across a distance. Locally, "possessor" and "beneficiary" datives encode (intended) affectees. By way of the compositional process of Variable Identification, which is modelled after, and extends, Kratzer's (1996) Event Identification, the dative argument co-refers with another argument in the local c-command domain, viz. a possessor in the case of "possessor" datives, and a participant in a purposive predication in the case of "beneficiary" datives. Locality constraints on Variable Identification are reviewed, and it is tentatively concluded that intervening CP boundaries and prepositional nouns limit the application of Variable Identification. The costly assumption of a principle like Variable Identification is further justified by its applicability to other voice phenomena, such as the middle voice of Ancient Greek, or reflexivity.

1.

Introduction: Dative arguments, neo-Davidsonianism and voice

This paper has two main objectives. The first goal is to get closer to an analysis of what all dative arguments in German have in common. The second goal is to characterize the mechanisms behind the syntax and semantics of dative arguments as phenomena belonging to a suitably defined dimension o f voice. I use the term 'voice' to characterize the linguistic tie-up between the predicate and the arguments/adjuncts o f a predication. Following work by Kratzer (1996, 2 0 0 3 ) and others I conceive o f this tie-up as largely syntactic. On this view, verbs are not voice-derived in the lexicon to acquire special argument structural properties. Instead, voice heads above VP add the syntactic structure and the thematic role information needed to accommodate (further) participants in the linguistic rendering o f a state of affairs. This w a y of implementing voice phenomena may be seen as a specific execution o f a neo-Davidsonian program which builds complex predications from the conjunc-

It took me a long time to write this paper. More colleagues and friends than I could possibly mention individually have discussed the ideas with me that are laid out in it. I wish to thank them all. Claudia Maienborn and two anonymous reviewers made me start all over again, for good reasons. Joachim Jacobs has been a source of encouragement, and of frank criticism. Daniel Büring, finally, took the time to understand the aim of my project, and helped me set up the abstraction part of the semantics as it is presented in this paper. None of the aforementioned people is responsible for its shortcomings.

214

Daniel Hole

tion of several smaller predications that all contribute to the characterization of a single event. The first - possibly surprising - step in the argument will be to establish so-called "free" datives as the primary objects of investigation and to sort out the datives of true ditransitive verbs as irrelevant to the development of a voice account of dative arguments (§2). §3 constitutes the core of the paper. So-called "possessor" datives are given an analysis which derives their semantic and syntactic behaviour from the inherent properties of an affectee voice head. In spelling out these properties I make recourse to Dowty's (1991) concept of Proto-Role entailments, and to a suitably generalized version of Kratzer's (1996) composition principle of Event Identification. A number of locality effects discussed in §4 illustrates the strictly local nature of the involved mechanisms. §5 compares the proposal defended here with other proposals in the literature. "Beneficiary" datives are discussed in §6, and I sketch a way to decompose beneficiary semantics as (intended) affectedness plus purposivity. This paves the way for a treatment of the datives of §3 on a par with "beneficiary" datives. In §7, the middle voice of Ancient Greek and reflexivity are briefly discussed, and I demonstrate that the generalized version of Event Identification introduced in §3, viz. Variable Identification, can be used to account for these phenomena, too, if they are to be given a neo-Davidsonian analysis in Kratzerian terms.

2.

R e d e f i n i n g " f r e e " d a t i v e s as t h e p r i m a r y o b j e c t o f i n v e s t i g a t i o n

Analyses of the syntax and semantics of datives will typically take prototypical ditransitive verbs as a point of departure, such as verbs meaning 'give', 'show', or 'tell'. An analysis is developed for some of these verbs and the syntax and semantics of their dative arguments, and then other verbs that may take dative arguments are considered. Verbs that may, but need not take dative arguments thus typically end up having a secondary status in comparison with true ditransitive verbs. The present study adopts the exact opposite as its research strategy. Free or optional dative arguments are considered basic for the argumentation, and obligatory dative arguments enter the perspective only secondarily. In fact, the present paper concentrates entirely on structures with dative arguments that have predictable and grammatical counterparts without dative arguments. How can such a step be justified? The basic tenet of the proposal defended below is that what licenses a dative argument in a sentence is a voice phenomenon and is therefore akin to passivization, middle voice formation in Ancient Greek, and even reflexivization. This view will be spelled out and substantiated in some detail in the remainder of this paper. If the licensing of datives is to be analyzed as a voice phenomenon, the research strategy should be in accordance with this basic assumption. When investigating voice phenomena, linguists will start out from structures that display the voice contrast under scrutiny. Passive structures, for instance, are compared with active structures, and an analysis of the principled correspondence between as many active-passive pairs as possible is developed. To be sure, there are some verbs that are only used in the passive, but have no active counterpart.

Reconciling "possessor" datives and "beneficiary" datives in German

215

An English example is be born. Such passives are traditionally called 'deponents' or 'passiva tantum' (from Latin tantum 'only'). Similarly, Greek verbs that are only used in the middle voice, but have no active counterpart, are media tantum, and an English verb like behave oneself is a reflexivum tantum. Starting out from a verb like geben 'give' in an analysis of the German dative should, on the voice view of datives, be just as dispreferred, or even absurd, as starting out from a verb like be born if one aims at a principled account of the passive, because give is a dative-voice tantum verb. Assuming a voice approach to dative arguments in German, I therefore take it to be a licit strategy to set up the system for all and only the free datives of German. §§3-6 constitute first steps in this direction.

3.

Implementing "possessor" datives

The voice account of dative licensing will be set up for structures in which the referent of the dative argument is, among other things, a possessor. Two examples are provided in (1), and we interpret these examples in such a way that Ede possesses the shin in (la) and the scalp in (lb) (or that the shin and the scalp are Ede's body-parts). 1 (1)

a.

b.

Die Paula trat dem Ede gegen das Schienbein, the Paula kicked the.DAT Ede against the shin 'Paula kicked Ede in the shin.' Dem Ede juckt die Kopfhaut. the.DAT Ede itches the.NOM scalp 'Ede's scalp itches.'

It was just stated that 'Ede possesses the shin [...] and the scalp' in (1). This is not to be taken to mean that the dative argument as such encodes the possessor relation. Even though analyzing the datives as expressing a relation of possession (of the shin, of the scalp) in sentences like (1) is probably the prevalent option in the literature among researchers of the most diverse theoretical persuasions (cf., e.g., Gallmann 1992, König & Haspelmath 1998, or Wunderlich 1996, 2000), I defend the claim that the dative argument is an affectee in (la), and an experiencer in (lb). I defer the discussion of diverging proposals to §5; a more precise characterization of the affectee notion and the experiencer notion will be given in §3.3.1.

In (I), definite articles are used preceding the proper names. This is done to make the differences in case marking visible. From now on, only the determinerless proper names of the standard language will be used. Dative arguments will, however, be italicized throughout for better perspicuity.

216

Daniel Hole

3.1 The basic idea As said above, I assume that the dative argument is not itself a possessor argument in (1). How, then, do we reconcile the fact that the dative argument denotes an affectee (cf. (2a)) with the intuition of possession or a part-whole relationship that is clearly felt to hold between the dative referents in (1) and the shin and the scalp, respectively? (2b) sketches a way to do this. (2)

a. b.

Intuition of affectedness of the dative referent: local effect, induced by a voice head tying an affectee to the encoded eventuality Intuition of possession/of a part-whole relationship: non-local effect, induced by the dative argument "binding" the possessor variable of the more deeply embedded noun

The major aim of the present larger section (§3) will be to make more precise what "binding" is taken to mean in (2b). 2 1 will couch my proposal in a neo-Davidsonian event semantics in the style of Kratzer (1996, 2003). The sub-section to follow will familiarize the reader with Kratzer's version of event semantics. We will then be able to implement the proposed analysis in §3.3.

3.2 Kratzer's (1996, 2003) neo-Davidsonian event semantics In a neo-Davidsonian event semantics like Kratzer's (1996, 2003), verbs are predicates of events, i.e. they take a referential event argument, just as a nominal predicate takes a referential argument. The thematic relations expressed by arguments of the verb and by prepositional phrases may then be conjoined with the basic predication of the event. For instance, a sentence like Paul is dancing in the ballroom may, in an event semantics, be paraphrased as T h e r e is an event of dancing, and Paul is the agent in this event, and the event takes place in the ballroom'. Kratzer's project over the past decade has been to tie a special version of a neo-Davidsonian event semantics as closely as possible to syntax. With one exception to be discussed below, arguments are also licensed independently of the verb in the syntax. It is not the verb itself that brings along thematic relations or that projects enough structure to accommodate all arguments of a sentence, but special functional heads which build up structure above VP. Kratzer (1996) has implemented this idea in some detail for what she calls the VoiceP, a thematic role head introducing agent arguments of transitive verbs into the structure right above VP (syntacticians often use Chomsky's 1995 concept of little ν instead). Let us see in more detail how the derivation of the sub-inflectional structure of a transitive predication proceeds in Kratzer's system. A verb like wash will have a lexical entry as

2

In the generative literature, Gueron (1985) seems to mark the starting point for a tradition that models "possessor" datives as datives that really "bind" possessor variables on some understanding of the term 'binding'. More on this proposal and the related one by Vergnaud & Zubizarreta (1992) will be said in §5.3.

Reconciling "possessor" datives and "beneficiary" datives in German

217

in (3), with the predicate ' w a s h ' on the right side of the equation meant as an abbreviation of the truth-conditions of eventualities of washing, whatever they may be. (3)

OwashYl

^x?-e.wash(x)(e)

(3) means that wash takes the internal argument as an argument, and an event argument, but no external argument. If we bind the variables of the verb existentially, w e get as a paraphrase: 'There is an event of washing something'. Note that this lexical entry is, in fact, not at all Davidsonian with respect to the theme argument. A radically Davidsonian implementation would rather provide a lexical entry for wash along the lines of (3'). (3')

QwashYl = λ ε ^ Β β Ι ^ β )

In Kratzer's implementation of event semantics, internal arguments are the only arguments that may be pre-specified in the lexical entries of verb roots for their thematic involvement into the denoted eventuality. On her view, they must be. The gist of the argument in Kratzer (2003) is as follows: All simple predicates in natural languages must be cumulative. This is taken to be a universal. Stated for thematic role predicates, this comes out as in (4a) (cf. Kratzer 2003, ch. 3: 8), a natural language paraphrase is provided in (4b). (4)

a. b.

VeVe'VxVyVR < e . < s , t »[natural« c .< s ,»,>(R)&R(xXe)&R(y)(e')]->R(x+y)(e+e')] 'If an individual stands in a natural (thematic) relation to an eventuality, and a different individual stands in the same relation to a second eventuality, then the sum of the two individuals also stands in that relation to the sum of the two eventualities.'

Kratzer claims that this universal holds true of thematic relations like the agent relation, but not of the putative theme relation. Moreover, she states that it is impossible to find a common core of all the different relations that have traditionally been subsumed under the theme relation. Her example to show that the theme relation is not cumulative involves the planting of a rose bush. Different individuals perform different actions that all add up to the planting of the rose bush. One person digs the hole, another person puts the rose bush in the hole, a third person adds manure, and so on. All these people acting in sub-events of the planting event count as agents in the super-event of planting the rose bush. On the other hand, the referents of the theme arguments of the actions of digging the hole, putting in the rose bush, adding manure, etc. do not add up to a single super theme. The planted rose bush doesn't consist of the addition of the rose bush, the hole, the manure etc. If theme arguments display an idiosyncratic behaviour in the semantics, they should also be treated idiosyncratically by the grammar. In our context, being treated idiosyncratically

means not being introduced by a regular thematic role head, but being pre-specified for thematic properties in the lexical entry of the verb root. Put differently: Only arguments with thematic roles that are in some sense anomalous - e.g., non-cumulative - need to be assigned their thematic role directly by the verb. Many internal arguments are exceptional in this sense, and that is why they are pre-specified in the verb. On this view, the notorious difficulties encountered when generalizing over the thematic roles of internal arguments of verbs like praise, avoid, imagine, meet etc. (see Kratzer 2003, ch. 3: 10 or Levin 1999)

218

Daniel Hole

vanish, because as internal arguments they are exempt from falling under any broad thematic role category. We can now start with the derivation of a sentence as in (5). (5)

a.

Paul washed Mary's shirt.

b. SpecVoiceP Paul

wash-Mary's shirt

If we apply the denotation of wash in (3) to the denotation of the direct object, we get the VP denotation in (6), a function from events to truth values of type (s is Kratzer's type for eventualities, which subsumes events and states). (6)

O V P n = λε.λνϋδΐι Mary's shirt(e)

Kratzer's denotation of the Voice head is spelled out in (7). It is a function from individuals to a function from events to truth values, i.e. it is of type ; it takes an individual and an event and checks whether the individual stands in the agent relation to the event. (7)

OVoice°n = Xata.Agent(x)(e)

Neither of (6) or (7), the daughter nodes of Voice', is of the right semantic type to combine with the other daughter node by Functional Application. Voice 0 would require a term denoting an individual as its argument, which is not the type that the VP has, and the VP would require an event to combine with. This is not the type of Voice 0 . The way out that Kratzer proposes is Event Identification. 3 Event Identification is a special rule of composition. Its general format is given in (8a), (8b) applies it to our case. (8)

Event Identification (Kratzer 1996:122) a. f g => h > acts as a 'event delimiting argument' (in the sense of Tenny 1994) with respect to the verbal predicate. The following axiom then characterizes [+quantity], Stative and telic predicates. It establishes a morphism from degrees to quantities. 14 The axiom concerns only predicates that have an associated degree scale, with a specified maximal degree, and degrees intermediary between 0 and dmax. (56)

QUANTITY:

Vd Vy [ Quantity(d,y) Λ 0 < d < d lrax -> 3! y° [ [y° = Max [ y' | y ' < 0 y A Quantity(d max , y') ] ] A d / d max = Prop (y°, y) ] The left part of the axiom's consequent says that if the degree associated with the quantity role Quantity{d, y) is between 0 and dmal, then there is a unique, greatest part y° o f y such that Quantity{dmm, y°) (Max acting as a maximizing function). 15 Thus, if an apple is half eaten, then there is a unique, greatest part of the apple that is completely eaten. Likewise, if a table is half wooden, then there is a unique, greatest part of the table that is completely wooden. The right part of the axiom's consequent, d / dmax = Prop (y°, y), means that y° is in the same proportion to y as d is in proportion to dmax. Thus, if an apple is half eaten, then half the apple is eaten. The axiom does not concern atomic predicates, since their associated scale is reduced to {0,1}. It should be noted here that the notion of atomicity is a lexical semantic issue, by and large; it amounts to the possibility or impossibility of a distributive/incremental reading of the verb with respect to a certain argument noun phrase. Thus, rent the car is atomic, while paint the car is not. This is reflected in the content of the Semantic-Type-1/2 predicates in e.g., (52)-(54): rent requires an object of e.g., the vehicle type, whereas paint requires its object to be of the material surface type. Only in the latter case is the predicate non-atomic. See Caudal (1999) for a detailed discussion of these phenomena and a treatment within a formal lexical semantic framework. Finally, we should stress that the part-of relation on objects is to be applied differently when the quantity argument is a definite plural. In that case, indeed, it can be understood in a manner similar to Link's (1998) i-part operator ('individual-part'), which returns a subset of a collection of individual entities. This reading arises if the degree expression involved has 'wide scope', i.e., if Yannig ate his pancakes partially means that Yannig ate part of the

14

15

Naumann (1996) noted that Krifka's (1992) homomorphisms between objects and events need not apply directly to the denotation of some patient noun phrase, but rather to some facet of its denotation - cf. peel the apple, which does not involve the whole apple but merely its skin. In (56), y is therefore built from the relevant facet or dimension of the quantity argument at stake. For the sake of simplicity, we ignore issues related to so-called event and object uniqueness. Indeed, Krifka (1998) uses homomorphisms like the following, where θ0 shows 'uniqueness of objects', and ΘΕ shows 'uniqueness of events': Vy Ve [ 0 o (y,e) A e ' < E e - » B!y' [ y ' < 0 y A 6 0 ( y ' , e ' ) ] ] Vy Ve [ 9 E (y,e) A y ' < 0 y - > 3!e' [e'< E e λ 0 E (y',e')] ] These notions are relevant to the treatment of predications such as read the novel, where several parts of the same novel can be read several times in a telic reading event.

Types of degrees and types of event

structures

293

set of pancakes, and not that he ate part of each pancake-this would be the 'narrow scope' reading of the degree expression. In the latter case, the part-of relation on objects is to be understood as Link's (1998) q-part operator ('quantity of matter-part'), which returns a subpart of an individual entity. This also accounts for examples such as The tourists left partially: the agent noun phrase acts as a quantity argument, with an associated scale {0,.. ,,dmax}cz\, dmax being the cardinality of the set of tourists.

5.5

Changes-of-state

We see changes-of-state as events developing along a degree scale. Non-atomic telic events are endowed with complex changes-of-state: they go through different intermediary degrees of development. On the contrary, atomic telic events are endowed with simple changes-ofstates, reduced to two degrees: 0 and 1. We capture the notion of change-of-state in axiom (57), which makes sure that if a verbal predicate Ρ describes an event e with degree d, every initial part e' of e is associated with a unique lesser degree d' (cf. (57)a), and vice versa (cf. (57)b). A part is said to be initial if it satisfies the INI predicate, cf. (58): 16 (57)

BECOME :

VP [ Become (P) MAP-ED(P) Λ MAP-DE(P) ] a.VP[MAP-ED(P)