208 48 17MB
English Pages 663 [664] Year 2003
Modifying Adjuncts
W G DE
Interface Explorations 4
Editors
Artemis Alexiadou T. Alan Hall
Mouton de Gruyter Berlin · New York
Modifying Adjuncts
edited by
Ewald Lang Claudia Maienborn Cathrine Fabricius-Hansen
Mouton de Gruyter Berlin · New York
2003
Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.
® Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.
ISBN 3-11-017352-2 Bibliographic
information published by Die Deutsche
Bibliothek
Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at . © Copyright 2003 by Walter de Gruyter GmbH & Co. KG, 10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.
Contents Modifying (the grammar of) adjuncts: An introduction Ewald Lang, Claudia Maienborn, and Cathrine Fabricius-Hansen
1
Part A: The argument-adjunct distinction The dual analysis of adjuncts/complements in Categorial Grammar David Dowty
33
Genitives, relational nouns, and argument-modifier ambiguity Barbara H. Partee and Vladimir Borschev
67
Heads, complements, adjuncts: Projection and saturation Manfred Bierwisch
113
Part B: Adjunct placement Syntactic conditions on adjunct classes Werner Frey
163
"Manner" adverbs and the association theory: Some problems and solutions Benjamin Shaer
211
Manner adverbs and information structure: Evidence from the adverbial modification of verbs of creation Regine Eckardt
261
Semantic features and the distribution of adverbs Thomas Ernst
307
Clause-final left-adjunction Inger Rosengren
335
vi Contents Part C: Case studies on wieder/again Process, eventuality, and wieder/again Karin Pittner
365
Competition and interpretation: The German adverb wieder ('again')... 393 Gerhard Jäger and Reinhard Blutner How are results represented and modified? Remarks on Jäger & Blutners's anti-decomposition Arnim von Stechow
417
Part D: Flexibility of eventuality-related modification Event arguments, adverb selection, and the Stative Adverb Gap Graham Katz
455
Event-internal modifiers: Semantic underspecification and conceptual interpretation Claudia Maienborn
475
Flexibility in adverbal modification: Reinterpretation as contextual enrichment Johannes Dölling
511
Secondary predication and aspectual structure Susan Rothstein
553
Real adjuncts in the Instrumental in Russian Assinja Demjjanow and Anatoli Strigin
591
German participle Π constructions as adjuncts Ilse Zimmermann
627
Subject index
651
Modifying (the grammar of) adjuncts: An introduction Ewald Lang, Claudia Maienborn, and Cathrine Fabricius-Hansen
Adverbials are a rich and as yet relatively unexplored system, and therefore anything we say about them must be regarded as quite tentative.
Chomsky (1965: 219) We still have no good phrase structure theory for such simple matters as attributive adjectives [...] and adjuncts of many different types.
Chomsky (1995: 382)
1.
Locating the volume
If we believe the above statements from one competent to tell us, then we would have to concede that there has been no progress in our understanding of what seems to be a "relatively unexplored system" even after 30 years of linguistic endeavour. However seriously these statements are no doubt to be taken, there are also reasons for a moderate optimism. The situation may be spelled out by the various readings of the title of this volume. One aspect of the progress being made is that the focus of attention has widened. Adverbials, though still the heart of the matter, now form part of a much larger set of constituent types subsumed under the general syntactic label of adjunct·, while modifier has become the semantic counterpart on the same level of generality. So one of the readings of Modifying Adjuncts stands for the focus on this intersection. Moreover, recent years have seen a number of studies which attest an increasing interest in adjunct issues. There is an impressive number of monographs, e.g. Alexiadou (1997), Laenzlinger (1998), Cinque (1999), Pittner (1999), Ernst (2002), which, by presenting in-depth analyses of the syntax of adjuncts, have sharpened the debate on syntactic theorizing. Serious attempts to gain a broader view on adjuncts are witnessed by several collections, see Alexiadou and Svenonius (2000), Austin, Engelberg and Rauh (in progress); of particular importance are the contributions to vol. 12.1 of the Italian Journal of Linguistics (2000), a special issue on adverbs, the Introductions to which by Corver and Delfitto (2000) and Delfitto
2
Ewald Lang, Claudia Maienborn, Cathrine
Fabricius-Hansen
(2000) may be seen as the best state-of-the-art article on adverbs and adverbial modification currently on the market. To try and test a fresh view on adjuncts was the leitmotif of the Oslo Conference "Approaching the Grammar of Adjuncts" (Sept 22-25, 1999), which provided the initial forum for the papers contained in this volume and initiated a period of discussion and continuing interaction among the contributors, from which the versions published here have greatly profited. The aim of the Oslo conference, and hence the focus of the present volume, was to encourage syntacticians and semanticists to open their minds to a more integrative approach to adjuncts, thereby paying attention to, and attempting to account for, the various interfaces that the grammar of adjuncts crucially embodies. From this perspective, the present volume is to be conceived of as an interim balance of current trends in modifying the views on adjuncts. In introducing the papers, we will refrain from rephrasing the abstracts, but will instead offer a guided tour through the major problem areas they are tackling. Assessed by thematic convergence and mutual reference, the contributions form four groups, which led us to arrange them into subparts of the book. Our commenting on these is intended (i) to provide a first glance at the contents, (ii) to reveal some of the reasons why adjuncts indeed are, and certainly will remain, a challenging issue, and thereby (iii) to show some facets of what we consider novel and promising approaches.
2.
The major issues tackled
2.1. Interfaces Highlighting the importance of interfaces can be seen as a step forward in view of the fact that most studies on adjuncts, so far, have confined themselves to either a syntactic or a semantic approach. Either type of work ultimately comes to the conclusion that adjuncts somehow resist a clear-cut and satisfactory treatment. This complaint should perhaps be supplemented by adding "... at least within our familiar notion of Core Grammar". Core Grammar is roughly but persistently that system which we are used to claim to be responsible for "the core of a language", which in turn "consists of what we tentatively assume to be pure instantiations of UG" (Chomsky 1995: 19). It is this notion (however tentative) that has governed, and at the same time delimited, most approaches to adjuncts presented so far by suggesting that adjuncts first and foremost have to be incorporated into this very system. The results gained this way were unsatisfactory, see above.
Introduction
3
Now, given that attempts to incorporate adjuncts within what we are used to consider as Core Grammar fail, we are left with two options: (a) revise our notion of Core Grammar, (b) acknowledge that adjuncts more or less belong to the periphery of a language. Choosing (b) implies a reversal of perspective, that is, to look from periphery to core when dealing with adjuncts, which, in turn, does not exclude repercussions on our understanding of Core Grammar. It is this perspective that the volume attempts to take. Among the various interfaces adjuncts are linked with, the volume concentrates on two: the syntax/semantics interface is predominantly addressed in discussions of adverb placement (see Part B); the semantics/pragmatics interface emerges as significant in those papers that treat the notorious under-determinedness of some classes of adjuncts in terms of semantic underspecification (see Part D). Cross-cutting with the interface issues, Part A discusses the (seemingly unavoidable) argument/adjunct distinction from a new - integrative - perspective. The case studies on wieder/again collected in Part C exemplify what is to be gained from an integrative view on adjuncts. Confined to a celebrated, much-discussed field of data, the papers together cover the whole range of syntax/semantics/pragmatics relations that a complete and thorough analysis of (a type of) adjuncts has to account for. 2.2. Reversing the view on adjuncts There are findings and considerations laid down in various papers in this volume that suggest that by reversing the familiar view from Core Grammar onto adjuncts to one looking from adjuncts to Core Grammar, we might arrive at a new and more feasible delineation of the core-periphery border. Part A offers a proposal in this vein. David Dowty's "dual analysis" is a theoretically promising move, the special appeal of which inter alia rests on the phenomena he adduces to illustrate the appropriateness of having a "dual analysis". In brief, Dowty considers adjuncts as representatives of a domain where diachronic fixings take place, that is, a process of change from periphery (adjuncts) to core (arguments, or: complements in Dowty's terminology). We will return to this below. Various papers in Parts Β and D suggest that the adjunct-argument transition (taken in the sense of Dowty) proceeds in a parametrized way. So, in terms of word order, in VO languages like English and Swedish the crucial positions for distinctive adverb placement are (roughly) sentence-initial and sentence-final, while internal positions induce ambiguity, cf. Ernst, Rosengren, Shaer. In an OV scrambling language like German, where the middlefield determines the range of internal adverb positions, the distinction is
4
Ewald Lang, Claudia Maienborn, Cathrine
Fabricius-Hansen
based on more fine-grained conditions, e.g. whether or not the base position of an adverb class precedes or follows that of the internal argument, cf. Frey, Pittner, Eckardt, Maienborn.
2.3. Semantic/syntactic mismatches Almost any treatment of adverbials starts from a long-established classification of adverbials that is somehow based on semantic intuition. So the adverbial subclasses labelled direction, local, duration, frequency, causal, manner etc. are more or less taken to reflect distinct adverbial types. Normally, syntacticians and semanticists make different choices in selecting a subset of these types, by starting their approach with a division into, say, VP- vs. sentential adjuncts or predicates vs. operators, and then concentrate on finding and justifying refined subdivisions below that intuitively assumed level. This given, we are miles away from achieving a list of which semantically recruited types of adverbials match with which classes of adverbials obtainable and/or needed in terms of syntax. Part Β makes an important step towards clarifying semantic-syntactic correspondences by presenting five distinct syntactic classes of adverbial adjuncts each of which is defined in terms of c-command and based on a set of diagnostics. In parallel with, though only loosely related to, the studies on adjunct syntax mentioned in 1.1, recent years have seen a remarkable number of semantic investigations into the field of modifiers within the realm of VPadjuncts, thereby giving an enormous impetus to event semantics. The relevant list includes monographs such as Parsons (1990), Maienborn (1996, 2002), Eckardt (1998), Engelberg (2000), Geuder (2000), Landman (2000), Dölling (2001), Musan (2002), Rothstein (to appear) and collections such as Rothstein (1998), Higginbotham, Pianesi and Varzi (2000), Tenny and Pustejovsky (2000), Dölling and Zybatow (2001). The fact that all of these draw on the classics by Davidson (1967) and Vendler (1967) certainly proves the fertility of the Davidsonian paradigm and explains its enduring popularity. Yet, admittedly, it is also indicative of the amount of unsolved problems we are left with. Viewed from the semantic point of view, the mismatch issue does not merely mirror the deficits we observe on the syntactic side. The crucial point here is to find a balanced way of mapping the range of conceptually discernible types of modifiers onto a reasonable set of ontological entities that figure as their respective target arguments. Maintaining compositionality as a guiding principle, we face the problem of providing grammatical
Introduction
5
evidence for the assumed ontology. The aim thus defined involves the task of justifying to what extent the meaning contributed by modifiers is computed compositionally and what of the interpretation rests on extragrammatical factors. This is what Part D is primarily concerned with. Needless to say, the problems outlined so far cannot be solved at once and simultaneously. In view of the work in progress offered by this volume, an integrative approach to adjuncts will involve several steps in answering the following questions. (Q-l) How can the correlations between the distribution of adjunct classes and their respective interpretations be ascertained and systematized into interface conditions on a more general level? (Q-2) What are, depending on the answers to (Q-l), necessary and sufficient ingredients of a compositional approach to the semantics of modifiers that can account for the whole range of structural ambiguities, underspecified meanings and patterns of reinterpretation typically shown by modifying adjuncts? The task of probing into the argument/adjunct distinction remains a central issue. However, it may change its ranking. In contrast to being considered the natural basis from which to look for answers to (Q-l) and (Q-2), the distinction might turn out to derive from the results obtained wrt. (Q-l) and (Q-2). This line of thinking will now be substantiated by taking a closer look at Parts A-D.
3.
A guided tour through the chapters
The volume as a whole reflects the situation of adjuncts research as outlined in Section 1 by responding to the issues raised in Section 2. The aim of Section 3 is three-fold: (i) to acquaint the reader with the approaches advocated here, (ii) to make the reader aware of the relatedness of the solutions offered, (iii) to invite the reader to take up and continue the issues presented.
3.1. Part A: The argument-adjunct distinction Worked out in the framework of Categorial Grammar, David Dowty's approach comes with a built-in answer to the problem of syntactic/semantic mismatch mentioned in 2.3. above. With respect to 2.2., the paper, based on
6
Ewald Lang, Claudia Maienborn, Cathrine
Fabricius-Hansen
a critical review of what is solid and what is shaky in the commonly assumed complement-adjunct distinction, posits the hypothesis "that a complete grammar (i.e. a grammar covering both core and periphery - the eds.) should provide a dual analysis of every complement as an adjunct, and potentially, an analysis of any adjunct as a complement." (Dowty, this volume). Support for this is provided by (i) a range of synchronic cases that, due to their actual ambiguity, require a dual analysis (i.a. to-Dative constructions, locative vs. dative to, agent phrases in passives, compounds and derived words) and by (ii) taking these cases of ambiguity to reflect stages of the historic development of these constructions. The fertility of this approach can be seen from the impressive list of superficially alike pairs of adjunct and complement constructions in presentday English, cf. Dowty's Table 1 (this volume). The table immediately invites comparison with other languages. Are there cross-linguistically observable patterns of adjunct/complement distribution? Does the division illustrated by the English data receive support from, say, a close cognate like German? Even a brief glance reveals that some of the English cases where adjunct and complement constructions look identical are explicitly differentiated in German, thus lending support to Dowty's analysis. For instance, Fl: purpose infinitives (John sang to impress Mary) and infinitive complements (John attempted to impress Mary) are overtly distinguished in German, cf. John sang um Mary zu beeindrucken vs. John versuchte (*um) Mary zu beeindrucken. Slightly more complicated, though revealing, is the situation illustrated by B1/B2. Dowty classes Mary walked to the park as a directional PP adjunct and John sang to Mary as a Dative complement. In German, however, at least if co-occurring with verbs of motion, directional PPs - as opposed to non-directional PPs - are to be classed as complements. And rightly so, since the adjunct-complement distinction systematically correlates with the Dative/Accusative alternation, cf. Mary rannte im Park (umher) [Dative, local adjunct] vs. Mary rannte in den Park [Accusative, directional complement], quite in parallel to English Mary walked in the park vs. Mary walked into the park; see also the German examples of locative vs. directional PPs adduced in Bierwisch (this volume). So, while the direct German counterpart of B1 would give rise to objections in this particular case, the general strategy in German of explicitly distinguishing adjuncts and complements via case marking is in support of what Dowty intends to show. Moreover, in coupling the celebrated repetitive vs. restitutive readings of wieder/again via word order restrictions with his adjunct-complement distinction, Dowty adds to the issue that is the central theme in Part C.
Introduction
1
Dowty's pair of possessive constructions: Mary's mother (possessive complement) vs. Mary's book (possessive adjunct) is on a par with Barbara Partee and Vladimir Borschev's distinction of relational nouns, that take argumentai Genitives, vs. non-relational nouns, that come with modifying Genitives. Partee and Borschev show that within the internal structure of NP/DPs, the argument-adjunct distinction is at least as complicated as in the VP domain since the status of the "complements" required by relational nouns is still under debate. Based on a critical examination of competing approaches to adnominal Genitives (argument-only, modifier-only, and split analyses), they show that different languages seem to be amenable to different approaches, depending on the constructions considered. Using the Genitive relation as a key diagnostic to examine English and Russian data in parallel, they argue that in the end split analyses might best be suited to account for the fact that, also cross-linguistically, Genitives are sometimes arguments and sometimes modifiers. Manfred Bierwisch, rejecting hybrid notions like "argument-adjuncts" or "obligatory adjuncts", approaches the complement-adjunct distinction from a grammar-internal point of view. He proposes to exploit independently defined syntactic operations to distinguish heads, complements, and adjuncts in the following way. While a constituent X is uniquely identified as a Head by imposing its categorial features onto its projection XP, both complements and adjuncts crucially rest on the notion of 'X discharges a thematic role to Y \ but differ as to the direction in which discharging applies. If a head X discharges a thematic role to Y, Y is a Complement·, if Y discharges a thematic role to the Head X, then Y is an Adjunct. The definitions of complement vs. adjunct thus gained provide the syntactic basis on which their semantic counterparts, viz. Argument vs. Modifier are tackled. The proposal is spelled out by showing its applicability to a wide range of apparently heterogeneous cases of modification. Special emphasis is put on clarifying the coverage and/or competition of extensional vs. intensional modification. Bierwisch's strategy is to extend the scope of data to be treated by intersective modification as far as possible. He argues that assimilating extensional to intensional modification, which amounts to generalizing to the "worst case", is an option to be avoided both on empirical and theoretical grounds.
3.2. Part B: Adjunct placement Facing the choice between assuming an approach that rests on free adjunction of XPs wherever possible (as do e.g. Zwart 1993; Neeleman 1994) and
8
Ewald Lang, Claudia Maienborn, Cathrine Fabricius-Hansen
an approach that draws on an elaborate hierarchy of functional projections to host the full range of adverbials at specified sites (cf. Cinque 1999 and related work), Werner Frey develops what - in various respects - can be called a motivated compromise. First of all, Frey assumes a limited number of fixed base positions to be the crucial condition on which the syntax of adjuncts should be built, but he does not deny that certain movement operations (in particular scrambling) will have to play some part here as well (contra Haider (2000) and at variance with Ernst (2002, this volume)). Second, Frey does not define uniquely fixed positions for a given adjunct type (in the sense of the semantic-based classification mentioned in 2.3. above) but instead allows for an adjunct type to be base-generated in different positions - provided the position at issue meets certain requirements. This leads to the delineation of certain clause-internal areas which in turn yield distributionally ordered classes of adjuncts. Third, the classes thus obtained reflect the interaction of two sorts of constraining factors: (i) the familiar semantic-based inventory of, say, temporal, locative, causal, manner adverbials is assigned a partial order that can be conceived of as anticipating semantic constraints yet to be worked out; (ii) the adjunct classes are strictly differentiated in terms of c-command, both wrt. one another as well as wrt. to internal and/or highest ranked arguments. The precedence and dominance relations among the five adjunct classes identified this way can roughly be depicted as shown below: (1)
Base position areas for adjunct classes: (I) sentence adjuncts > (Π) frame and domain adjuncts > (HI) event-external adjuncts > highest ranked argument > (IV) event-internal adjuncts > (internal arguments) > (V) process-related adjuncts > verb (where ">" denotes c-command)
As will become clear below, the adjunct classes (I)-(V) provide an orientation frame for locating what other papers contribute to the placement and interpretation of adjuncts. Though (I)-(V) have been delimited by distributional criteria within the German middle-field, the names they are given by Frey are indicative of the properties these adjuncts display as semantic modifiers. This is an important step towards clarifying the syntax/semantics interface. It replaces the coarse semantic partition of modifiers into predicates vs. operators by a more finegrained typology which, furthermore, yields a partial reconstruction of the traditional classification of adverbials mentioned in 2.3. above. As a first
Introduction
9
approximation, the correspondences that hold between Frey's adjunct classes and the familiar semantic-based adjunct types can be listed in the following way: Class I:
Class Π:
Sentence adjuncts include attitudinal adjuncts (apparently, anscheinend) subject-oriented adjuncts (stupidly, dummerweise)
and
Frame adjuncts (in the Middle Ages, im Traum) and domain adjuncts (botanically (speaking), scriptwise, finanziell (gesehen))
Class ΠΙ: Event-external adjuncts include causals (due to space limitations, trotz des Regens) Class IV: Event-internal adjuncts include event-related adjuncts like temporals (in a few minutes, gleich), locatives (near you, hinter der Gardine), instrumentais (with a knife, durch Erpressung)·, in addition, mental-attitude adjuncts (willingly, absichtlich) belong to class IV in English and German, notwithstanding much-debated distributional differences Class V:
Process-related adjuncts include, above all, the range of manner adjuncts (carefully, quickly, edgeways, in a soft voice, heftig, auf geschickte Weise)
The adjunct classes (I)-(V) are relevant also to the following papers. Shaer takes up Frey's proposal in elaborating on it wrt. manner adverbs in English that occur in both a "lower", sentence-final, position (= processrelated adjuncts, Class V above) and also in a "higher", VP-external, position (viz. sentence adjuncts, Class I above). Eckardt challenges Frey's ordering of Class V adjuncts wrt. internal arguments by adducing counter-evidence from verbs of creation modified by manner adjuncts. Ernst suggests a semantic explanation for certain distributional restrictions to be observed within the above (syntactically defined) adjunct classes. He attributes the prohibition on right adjunction of non-manner adverbs to a lexico-semantic feature "subjective" that these adverbs embody. Furthermore, the classes distinguished as event-internal (Class TV) vs. process-related (Class V) reappear in Parts C and D as well. They fit in with the detailed analyses of the repetitive vs. restitutive readings of wieder/ again that are presented in Part C, and they serve as syntactic landmarks for the semantic treatment of locative modifiers (Maienborn) and of an extended sample of process-related modifiers (Dölling) in Part D.
10
Ewald Lang, Claudia Maienborn, Cathrine
Fabricius-Hansen
Finally, Zimmermann shows that German participle Π constructions as adjuncts have readings that can be assigned to either Class IV or Class Π and raises the question of how to derive them in a lexicalist framework. Benjamin Shaer, focussing on manner adverbs in English, makes a case for the legitimacy of associating syntactic positions with interpretations by rejecting counter-arguments raised in the literature. He argues that cases which seem to disprove the feasibility of such an approach (e.g. so-called fronted, parenthetical, and afterthought occurrences of certain adverbs) can be separated off and given the special treatment they require. While pleading for an association approach in principle, the paper provides intriguing data to show the low degree of syntactic integration that can be attributed to fronted manner adverbs in English. Shaer's observations are challenging wrt. what has been supposed to fall within the scope of grammar. Regine Eckardt's paper is devoted to linking the syntactic basepositions of certain adjuncts in German with their behaviour in word order variation in terms of information structure. Based on syntactic as well as semantic considerations, she argues for an underlying Adverb-Object-Verb order (AOV) in German for Class V adjuncts (contra Frey's OAV). Evidence is drawn from the observation that indefinite objects occurring to the left of manner adjuncts lack an existential reading: (2)
a. Beate hat vorsichtig einen Drachen verpackt. (AOV: ex. reading) Beate has carefully a kite wrapped, b. #Beate hat einen Drachen vorsichtig verpackt.(OAV: no ex. read)
She concludes that indefinite objects occurring to the left of Class V adjuncts must have been moved to that position triggered by their topicality. Assuming the AOV order, Eckardt can account for an unexplained gap in the distribution of result-oriented adjuncts (a subtype of Class V). The position to the left of a direct object is always unavailable to these adjuncts; cf. (3a). If they occur to the right of a direct object, it makes a difference with what type of verb they are combined. While verbs of creation (3b) do not allow for an existential reading of indefinites, other transitive verbs do, (3c). (3)
a. *Beate hat wasserdicht einen Drachen gebaut/verpackt. (*AOV) Beate has waterproof a kite built/wrapped. b. #Beate hat einen Drachen wasserdicht gebaut. (OAV: no ex. read) c. Beate hat einen Drachen wasserdicht verpackt.(OAV: ex. reading)
The argumentation expounded in this paper shows the heuristic value of the sort of integrative approach to adjuncts advocated in this volume. The
Introduction 11 Eckardt - Frey controversy, being narrowed down to a clear-cut selection of data and spelled out in terms of mutual reference, allows for weighing up the costs and benefits of the alternative solutions that are currently available. Interim balances like this are an important step in coping with the guiding questions (Q-l) and (Q-2) in Section 2.3. above. Thomas Ernst's paper on the High Right-Adjunction of adverbs in VO languages (i.e. adjunction to functional projections above VP) is a case study within the general framework expounded in Ernst (2002). The main issue is to figure out the conditions based on which the class of adjuncts that disallows high right-adjunction can be delineated. While the contrast shown in (4a) vs. (4b) might suggest that it is gradability that bars adverbs from high right-adjunction, (4c) proves that gradability, though relevant, does not suffice. To account for the difference between (4a) and (4c), a further partition within the class of gradable adverbs is needed. (4)
a. * Peter will solve the problem wisely, (in the non-manner reading) b. Peter will solve the problem financially. c. Peter will solve the problem willingly.
According to Ernst, the decisive factor preventing right-adjunction rests on the "subjectivity" of the adverbs involved. Semantically, "subjective" adverbs may be defined as those gradable adverbs that introduce a contextdependent scale onto which the event is mapped according to the speaker's judgement. Hence, the syntactic behaviour of adjuncts regarding rightadjunction is shown to correlate with a specific lexico-semantic feature. Ernst's proposal is, undoubtedly, another step towards delineating interface conditions. The next step will be to spell out the effect of "subjectivity" in syntactic terms. Inger Rosengren's paper aims at explaining the fact that e.g. causal, temporal and locative modifiers (so-called "circumstantials") in VO languages like English and Swedish prototypically appear at the right edge of the clause, whereas in an OV language like German they occur adjoined on top of the VP. In addition, the ordering of these modifiers in English and Swedish exactly mirrors the order they take in German. Rosengren examines several recent explanations proposed within the Minimalist Program and concludes that none of them covers the relevant data wrt. binding, focus, and word order. Following Haider (2000), she suggests having clausefinal circumstantials in VO languages licensed indirectly, viz. by an empty VP-complement of Vo. The solution presented avoids the difficulties that emerge with right-adjunction. Instead, it exploits the different settings of the
12
Ewald Lang, Claudia Maienborn, Cathrine Fabricius-Hansen
verbal head parameter to account for the reversed order in which modifiers appear in VO vs. OV languages.
3.3. Part C: Case studies on wieder/again Based on German wieder and its English counterpart again, the three papers in this part are concerned with the well-known repetitive/restitutive ambiguity. So (5) allows an external or repetitive, event-related, interpretation to be paraphrased as 'John opened the door; he had opened it (once) before' and an internal or restitutive, result-related, reading to be paraphrased as 'John opened the door; the door had been open before'. In spoken language, the two readings of (5) are differentiated prosodically: the repetitive reading comes with (narrow) focus on the adverb, cf. (5a); the restitutive reading has focus accent on the verb, cf. (5b). Furthermore, if wieder precedes the subject or a nominal object as in (6a-c), the repetitive reading is the only possible, or at least the strongly preferred, reading. (5)
(6)
(dass) John die Tür wieder öffnete (that) John the door again opened a. (dass) John die Tür WIEder öffnete b. (dass) John die Tür wieder ÖFFnete
repetitive reading restitutive reading
a. (dass) John wieder die Tür öffnete (that) John again the door opened b. (dass) wieder John die Tür öffnete (that) again John the door opened c. Wieder wurde die Tür geöffnet Again was the door opened
repetitive reading preferred repetitive reading only repetitive reading only
The repetitive/restitutive duality of wieder/again is the most thoroughly debated example of the syntactic-semantic flexibility that (adverbial) adjuncts show, an issue that also forms a major concern of the present volume. In fact, the issue has been subject to discussion since the emergence of Generative Semantics, which to a certain degree was motivated by the externalinternal reading dichotomy itself (see McCawley (1968, 1972) and Dowty (1979)). In view of this, the analysis of wieder/again is a measure of what has, by now, been achieved in the grammar of adjuncts. The controversy primarily concerns the question of where to locate the source of the ambiguity. The two classic options are: (i) in the lexicon, which amounts to assuming lexical ambiguity of the adverb, or (ii) in the
Introduction
13
syntax, which implies that the ambiguity has to be accounted for in terms of structural scope. However, in view of recent developments in syntactic and semantic theorizing, but also due to a large amount of hitherto unnoticed data, additional points of divergence have emerged. Taking stock of the relevant literature, we find basically three types of strategy, including the papers in this volume. (A) Deriving the restitutive reading of wieder/again from the repetitive one Dowty (1979: 261ff.) discusses the view that takes restitutive again to be a complement of the verb it seems to modify and repetitive again to be an adjunct. Such an analysis correctly predicts that, in English, the repetitive/restitutive ambiguity may arise only with again in final position, as in (5), whereas again in a preverbal or sentence-initial position as in (6b) allows for the repetitive interpretation only. As far as its lexical meaning is concerned, the adverb is analyzed as nonambiguous, receiving a single semantic representation which informally can be rendered as in (7), that is, the meaning of again is identified with the semantic contribution of repetitive again as paraphrased above. (7)
again ρ =Def ρ is the case and ρ has been the case before
Within the formal-semantic framework of Dowty (1979), the complementadjunct approach implies a dual and decompositional semantic representation of the relevant verbs (accomplishment and achievement verbs). In order to take again as a complement, the semantic representation of the verb at issue must have a slot (a variable bound by the lambda operator) for such a complement; whereas verbs that are modified by the adjunct again lack such a slot. In the first case, lambda conversion will produce a reading in which the adverb eventually turns up clause-internally, such that it has scope over the result only. Hence, (5b) will be assigned the (simplified) representation (8b). Conversely, if the adverbial adjunct is applied to the saturated verb (5a), we get the repetitive reading shown in (8a). (8)
a. again (CAUSE (john, BECOME (OPEN, the door))) b. CAUSE (John, BECOME again (OPEN, the door))
In the end, Dowty (1979) dismisses the complement-adjunct account on the grounds that it does not capture the restitutive readings in secondary predications (on the latter, see Rothstein and Dölling (this volume)) nor resultati ve readings that emerge with small clause constructions like (9)-(10) below. In his contribution to this volume, however, David Dowty revives the
14
Ewald Lang, Claudia Maienborn, Cathrine
Fabricius-Hansen
complement-adjunct approach to again within his Dual Analysis hypothesis. Under this view, not only again, but also secondary predicates and certain adverbial PPs receive a dual categorization as complements and adjuncts. (9)
John fell asleep during the lecture, but Mary quickly shook him awake again.
(10)
The book had fallen down, but John put it on the shelf again.
(B) Assuming repetitive vs. restitutive wieder/again as separate items The approaches subsumed under (B) have in common that wieder/again is taken to be semantically ambiguous (or polysemous), i.e. they assign (at least) two different semantic representations to the adverb, one for each reading. In other respects, however, they differ from each other to some extent. Thus Dowty (1979: 264) accounts for the distributional differences of again in the two readings by stipulating two adverbs that belong to different syntactic categories: a sentence modifier againι (repetitive) and a VP modifier againi (restitutive). Given an appropriate decomposition of accomplishment and achievement verbs at the semantic level, he assumes a meaning postulate that applies to again2 thereby reducing restitutive again to repetitive again with scope within the modified VP, cf. (8a). FabriciusHansen (1980) treats repetitive wieder as a sentence modifier and introduces restitutive wieder synsemantically by means of a word-formation rule, parallel to the English prefix re- (cf. Dowty 1979: 256, 361). Fabricius-Hansen (2001), however, posits one polysemous lexical item wieder with dual or multiple syntactic class membership and attempts to derive the repetitive from the restitutive meaning rather than the other way round. Kamp and Roßdeutscher (1994), working within a DRT framework, also distinguish two lexical items wieder/again but make no attempt to explain the semantic relationship between the two. Finally, Jäger and Blutner (this volume) also assume genuine lexical ambiguity as does, e.g., Dowty (1979: 264f.), but they take the two adverbs wieder and wieder2 to belong to the same syntactic category. They seek to account for the correlations between adverb positions and adverb interpretations by means of general pragmatic interpretation principles (more on this below). Obviously, accounting for the repetitive-restitutive duality by positing genuine lexical ambiguity is not a very appealing solution as it lacks explanatory power. If this approach were generalized to account for, e.g., event-external vs. event-internal dichotomies (as discussed in Part D), it would lead to an inflation of homonymous lexical entries. In view of these
Introduction
15
disadvantages, a structural explanation in terms of scope should be preferred, on condition that it is theoretically sound and able to account for all relevant data, see (C) below. What has encouraged the adherents of a 'lexicalist' approach, apart from general theoretical considerations (Dowty 1979), is primarily the fact that the structural account (alone) cannot explain the disambiguating effect of focus accent, see (5a, b) above and Jäger and Blutner (this volume). In addition, the repetitive-restitutive duality does not only occur in overt resultative constructions with accomplishment and achievement verbs, as predicted by the decomposition scope account, but also with (non-decomposable) state predicates (Fabricius-Hansen 1983, 2001), as witnessed by (11). (11) a. Der Kapitän ist WIEder nüchtern. 'the captain is sober again' b. Der Kapitän ist wieder Nüchtern. 'the captain has sobered up again'
(restitutive reading)
(C) Assuming only one wieder/again that is based on the repetitive reading The third type of approach unambiguously assigns a repetitive meaning to wieder/again but differs from strategy (A) by advocating a purely structural account of the repetitive-restitutive duality in terms of word order and syntactic scope. This strategy, combined with "prelexical" decomposition, was first introduced in Generative Semantics (McCawley 1968, 1972). Later on it was discussed and rejected by Dowty (1979: 235ff.) in favour of semantic decomposition, see (B). Recently, however, it was taken up again (!) by Arnim von Stechow (1995, 1996, and this volume) in a modernised version of lexical decomposition in syntax (cf. also Rapp and von Stechow (2000), Beck and Snyder (2001)). Under this approach, then, restitutive wieder/ again appears at 'deep structure' as a repetitive wieder/again that modifies a small clause (XP) in the scope of the operator BECOME, whereas repetitive wieder has scope over [CAUSE [BECOME ...]]. Thus, according to von Stechow (1996, this volume), (12) is assigned the syntactic structures shown in (12a) and (12b) for the repetitive and the restitutive readings, respectively. (12)
weil Max das Fenster wieder öffnete because Max the window again opened a. [ AgrS Maxi Ugrothe window2 again [voiceP tl [voice CAUSE [ V P BECOME [χρ t2 OPEN]]]]] (repetitive reading) b. [ A grS Maxi [AgrO the W Ì n d o W 2 [voiceP ti [voice CAUSE [vp BECOME again [ΧΡ Í2 OPEN]]]]] (restitutive reading)
16
Ewald Lang, Claudia Maienborn, Cathrine Fabricius-Hansen
Crucially, the decomposition & scope account hinges upon correlating word order variation with the readings of wieder/again as illustrated in (5) vs. (6), by predicting that the adverb in the restitutive reading cannot have scope over an existentially quantified NP. It is against the background of these major approaches to wieder/again that we will now comment briefly on the papers in this volume. Karin Pittner's contribution favours a scopai treatment of the repetid verestitutive duality by reinforcing the role of syntactic differences to be observed. She argues that restitutive wieder syntactically belongs to the process-related manner adverbs (Class V, cf. 2.3. above) as it is basegenerated below the internal arguments of the verb; whereas repetitive wieder exhibits the distribution of the event-internal adverbs (Class IV) as in its base position it c-commands the internal arguments, cf. (5) vs. (6) above. Apparent counterexamples like (13) below are explained as instances of integration in the sense of Jacobs (1993); see also the discussion between Frey and Eckardt (this volume). Pittner also suggests an explanation for the fact that the repetitive reading of wieder might have developed from its use as a process-related restitutive modifier. Similar considerations were presented by Fabricius-Hansen (2001). Gerhard Jäger and Reinhard Blutner attack the decomposition & scope approach on empirical grounds. They argue that it cannot adequately account for the interplay between word order and focus accent placement, on the one hand, and possible readings of wieder/again, on the other hand. Thus, they claim, it fails to predict restitutive readings of wieder if it precedes an indefinite object NP as in (13a), and it cannot explain the disambiguating effect of the focus accent on wieder shown in (13b), which disallows a restitutive reading. Furthermore, they posit, the decomposition & scope approach wrongly rules out a restitutive reading of wieder/again with wide scope over the indefinite subject in (14). Informants, however, confirm the existence of such a reading on which a Delaware refers to a member of the tribe who moves to the home of his ancestors. In other words, this reading of (14) does not presuppose that the subject referent has previously been in New Jersey in order to re-establish the state that there are Delawares settling in New Jersey. (13)
a. (weil)
Hans wieder ein FENster öffnete
(because) Hans again a window opened b. (weil) Hans WIEder ein Fenster öffnete (14) A Delaware settled in New Jersey again.
Introduction 17 These shortcomings cause Jäger and Blutner to reject the structural account of the restitutive-repetitive duality in favour of an approach that links the assumed lexical ambiguity of wieder/again with the inferential means that are provided by the framework of Bi-directional Optimality Theory (OT). The analysis proposed draws on the pragmatic sources of the distributional differences of repetitive vs. restitutive wieder/again and of the disambiguating role of focus accents. Arnim von Stechow's paper is a direct reply to Jäger and Blutner's, maintaining the essential tenets of the decomposition & scope approach. While conceding a weak point in his own approach wrt. the problematic reading of (14) (as pointed out by Jäger and Blutner), von Stechow proposes a more fine-grained decompositional analysis. He shows that this improved decomposition & scope approach is fully compatible with Jäger and Blutner's pragmatically based OT approach and, hence, can be seen as a serious alternative. Furthermore, von Stechow suggests a new analysis of accomplishment predicates, differentiating verbs like öffnen/open, that have a syntactically visible result state predicate ('be open') from verbs like putzen 'to clean', the result states of which are inaccessible (for most speakers). The prediction is that only the former will allow restitutive readings with wieder. The relevance of this 'visibility parameter', for details cf. Rapp and von Stechow (2000), is confirmed by cross-linguistic evidence presented by Beck and Snyder (2001). In view of the questions raised in Section 2.3. above, the discussion of wieder/again can be summarized as follows: First, if the defining syntactic properties of an adverbial adjunct are determined by its base position, wieder/again must be assigned a dual, or perhaps multiple, class membership not unlike the one needed, e.g., for socalled manner adverbials. Thus, the dual nature of wieder/again seems to be a fact that we cannot get rid of. Second, semantically related adverbials like once more and erneut, abermals in English and German, and 'repetitive' adverbials in many other languages as well, do not occur in the internal position that typically correlates with the restitutive reading, cf. von Stechow (this volume); FabriciusHansen (2001), Beck and Snyder (2001). Thus, unlike those other adverbs expressing repetition, wieder/again display dual class membership as an idiosyncratic property which has to be marked one way or the other. However, it is not evident that an approach that assigns wieder and e.g. erneut the same (repetitive) meaning and which has to block somehow the structural positions that would give rise to a restitutive reading for erneut has more explanatory power than an approach that takes wieder to be polyse-
18
Ewald Lang, Claudia Maienborn, Cathrine
Fabricius-Hansen
mous and synonymous with erneut on one reading. In the end, then, it remains to be seen whether there is any interesting difference between the 'lexicalist' and the scopai account at all. Third, given a compositional theory of non-lexical semantics, the semantic type of the entity to be modified by wieder/again co-varies with the base position or the syntactic class of the adjunct. This is not a trivial matter. If we accept that an adjunct that is assigned one and only one semantic representation can modify different types of entities, we have to account for the interaction between the semantic contribution of the adjunct and the semantic properties of the modified entity in a principled way. That would lend support to an approach in terms of underspecification. Proposals along these lines are made by Maienborn (this volume), but also by Klein (2001) and Dimroth (2002), who assign to wieder/again the meaning "... and not for the first time" thus leaving the rest of the interpretational burden to the context. Fourth, as an alternative to the underspecification approach, which rests on the repetitive meaning of wieder/again, the semantic contribution of wieder/again as a Class IV adjunct (including its use as a contrastive discourse particle) might be derived from its prototypical and more informative use as a process-related Class V adjunct (Fabricius-Hansen 2001). It is an interesting question (to be settled in future research) whether deriving more abstract from less abstract readings by some sort of context-dependent semantic bleaching, rather than the other way round, might develop into a general alternative to the underspecification approaches to modificational flexibility pursued in Part D.
3.4. Part D: Flexibility of eventuality-related modification Graham Katz tackles the issue of sorting adverbial adjuncts by the selectional restrictions they impose on their respective verbal heads. Based on the observation that there are many adverbs that select eventive verbs; cf. (15) but, conversely, no adverbs that exclusively select stative verbs, the paper examines various solutions of how to account for this "Stative Adverb Gap", which is somewhat surprising against the background of a (neo-)Davidsonian approach. (15) a. Eva resembled Max *quickly/*gently ... b. Eva kissed Max quickly/gently ...
Introduction
19
Katz posits that the verb-adverb selection reflects the basic opposition between events and states and draws the conclusion that stative verbs do not induce a Davidsonian eventuality argument. Having no eventuality argument, stative verbs do not provide suitable targets for manner adverbs (15a), nor do they provide reasons for there to be a particular class of stative adverbs. This accounts for the Stative Adverb Gap. According to Katz, cases in which what appear to be manner adverbs may co-occur with stative verbs as in (16) should be analyzed as degree modifiers along the lines of an operator approach. (16) a. Eva knows Maxwell. b. Eva believes this firmly. c. Eva loves Max passionately. Support for this view is provided by the observation that the adverbs at issue strictly select the verb they combine with, not the other way round (i.e. well selects know but not believe, firmly selects believe but not know). For an alternative account that distinguishes genuine stative verbs like resemble, know, believe from state verbs like sit, stand, sleep with consequences for their respective argument structure see Maienborn (2002). Claudia Maienborn provides a semantic analysis for locative modifiers which, in Frey's terms, belong to Class IV and Class V. Special emphasis is put on (i) recognizing, and (ii) accounting for, the interpretations Class V locatives may assume (as opposed to the standard case of event location covered by Class IV adjuncts). (17) Eva signed the contract on the stage. (18) a. Eva signed the contract on the last page. b. The bank robber fled on a bicycle. c. Max jumped around on one leg.
Class IV adj uncts Class V adjuncts (instrumental reading) (manner reading)
In order to capture the whole range of readings that locatives display at Class V level, while sticking to compositionality, Maienborn offers a refined version of the standard Davidsonian account of modification. Under this approach, Class V adjuncts are taken to be semantically underdetermined, and hence flexible to combine with a variety of targets that are conceptually accessible depending on context and world knowledge. The various possibilities to specify the readings at issue are spelled out by means of abduction.
20
Ewald Lang, Claudia Maienborn, Cathrine Fabricius-Hansen
Johannes Dölling, applying the framework elaborated in Dölling (2001), proposes a general approach to cope with the flexibility of interpretation to be observed with, e.g., Class IV and Class V adjuncts and secondary predicates. Dölling's proposal to account for the semantic underspecification of those adjuncts rests on the way he treats the verbs they are linked with. The latter enter the representation as one-place predicates that obligatorily undergo various steps of structural enrichment in the course of semantic calculation. The first step consists in furnishing the underspecified basic Semantic Form of the verb with variables to the extent that is needed for linking the modifiers. In the next step, the meaning of a verb and its modifier is composed by means of abductive parameter fixing. The three-level approach advocated here is shown to also account for the familiar cases of coercion but is meant to cover the whole range of modifying adjuncts in a unified way. Dölling illustrates this claim by showing how secondary predications can be treated. In this framework, the distinction of depictive vs. resultative secondary predicates is made only on the purely semantic level of parameter fixing, that is, abstracting away from morpho-syntactic and other possible structural differences. The attractiveness of this approach on the conceptual side has to be weighed against the requirements of fully-fledged interface conditions. Susan Rothstein's paper on secondary predicates shares with Dölling's the aim of analyzing depictive and resultative predications in a general way that brings out their differences on the basis of what they have in common, structurally as well as semantically. Depictive and resultative predicates are both analyzed as aspectual modifiers in terms of event summation which, in turn, is augmented by a constraining relation called TPCONNECT (short for: Time-Participant Connected). TPCONNECT holds between two events ei, and an individual y iff eiand e 2 share the same run time and also share y as a participant. While depictives require TPCONNECT to relate the event argument of the secondary predicate to the event introduced by the matrix verb (19a), resultati ves are TPCONNECTed with the culminating event of the matrix verb (19b). (19) a. John, drove the car drunk, b. Mary painted the house¡ red¡
... TPCONNECT(ei, e 2 ) ... TPCONNECT(cul(e0, e2)
The approach also accounts for a number of facts that have remained unexplained so far. For instance, based on the central fact that a resultative reading is possible only when the predicate is predicated of an incremental theme, the approach predicts that subject-oriented resultati ves may occur provided the subjects are incremental themes. Hence we find subject-
Introduction 21 oriented resultatives with passive (20a) and unaccusative (20b, c) verbs, but not with unergative verbs (20d): (20) a. b. c. d.
The house¡ was painted red¡ The riveri froze solidi Mary ι grew up smarti *John¡ ran tired¡
The analysis of secondary predications presented here is extended and elaborated at monograph length in Rothstein (to appear). It may be rewarding to compare this approach to secondary predication with the one pursued by Dolling (this volume). Assinja Demjjanow and Anatoli Strigin's contribution is important for at least two reasons. First, focussing on Russian adjunct-DPs in the Instrumental case they make us aware of the role of morphology and case marking in the grammar of adjuncts, which has been neglected under the predominance of adverbial adjuncts that come as PPs. Free Instrumental is shown to be the standard case for adjuncts in Russian, just as Free Dative is the preferred case for adjunct-DPs in German - facts like these put typological investigations on the agenda. Second, in addition to being the adjunct case, the Instrumental in Russian covers a wide range of other functions due to which it has been assigned "peripheral status" by Jakobson (1936/1990), and has been claimed to be extremely polysemous by Wierzbicka (1980), who assigns the Instrumental seventeen discernible meanings. Hence the adjunct-DPs in the Instrumental per se are semantically underdetermined; cf. the selection in (21). (21) a. On exal poezdom. He drove train-msir 'He was going by train' b. Do reki on sei dorogoj. To river he went TO&d-instr 'To the river, he went on the road' c. Rebënkom on bolel.
(Instrumental of Transport)
(Instrumental of Path)
(Temporal Instrumental)
child-i'nsir he ill.PAST
'He was ill as a child' The specific semantic interpretations of these adjunct-DPs obviously are determined by the respective context including world knowledge about the situation type involved. Assuming a unified syntactic small clause analysis
22
Ewald Lang, Claudia Maienborn, Cathrine Fabricius-Hansen
for the adjunct-DPs at issue, Demjjanow and Strigin propose a device of semantic interpretation that is based on abduction. Though being confined to three typical uses of the Free Instrumental, the proposal is meant to be extended to all non-idiosyncratic uses of the Instrumental case in Russian, including secondary predications. The adjuncts investigated by Ilse Zimmermann come as German Participles Π that form the lexical heads of adnominal attributes (22) or of adverbial phrases (23): (22) der seit zwei Wochen verreiste Nachbar the since two weeks away neighbour 'the neighbour who has been away for two weeks' (23) Das Fleisch bleibt, im Römertopf gegart, schön saftig. the meat stays, in.the chicken brick roasted, nice(ly) juicy 'Roasted in the chicken brick, the meat stays nice and juicy' In terms of morpho-syntax, German Participles Π are conceived of as nonfinite verb forms that project into reduced sentence-like structures: (i) they preserve the argument structure of the underlying verb, but have no position for the subject; (ii) they lack access to ForceP, TenseP, and MoodP, but they can undergo passivization and perfectivization, and (iii) they can convert to adjectives at word or phrase structure level. Being deprived of standard linkers, German Participles Π as adjuncts make semantically underspecified modifiers. Zimmermann analyses them as one-place predicates, the integration and interpretation of which is accounted for by means of modification templates that provide for the linking conditions needed. Two of these templates are discussed in more detail: MODI, which i.a. induces the propositional connector &, accounts for intersective modification as represented by attributive adnominal adjuncts like (22) and by event-related adjuncts of Frey's Class IV, whereas MOD2, which induces a relational non-Boolean parameter C, does so for frame adjuncts like (23) thereby making Participle Π constructions comparable to Frey's Class Π adjuncts. For an alternative account of the semantic integration of frame adjuncts, see Maienborn (2001).
4.
Outlook
Our attempt to present the volume as an interim balance of current research on adjuncts would be incomplete without drawing some conclusions for the
Introduction
23
future. In Section 2.3., we defined an integrative approach to adjuncts and modification by two guiding questions concerning interface conditions (Q1) and compositionality (Q-2). In Section 3, we intended to outline what the papers, both individually and jointly, offer in coping with these questions. In the present section we will, maintaining (Q-l) and (Q-2) as guidelines, point out which well-known crucial issues have been left untouched, recall what new problems have been raised, and at the same time suggest what the next steps towards an integrative approach to adjuncts might look like.
4.1. Adjuncts and "integration" Obviously, the volume shows a predominance of studies delving into adverbial adjuncts. This is not surprising but may be seen as a joint result of the syntactic tradition surveyed in Section 1 and the impact of event semantics on modification studies. As a consequence, the wide range of adnominal adjuncts will have to be kept on the agenda. Furthermore, the fact that most of the studies in this volume deal with adjuncts that fall within the Classes I-V given in Section 3.2. may indicate that these classes somehow define the domain of prototypical adjuncts. This domain, in turn, forms a scale of syntactic integration decreasing from Class V to Class I. What is at stake now is to address adjunct-like structures such as parentheticals or afterthought constructions that, due to being less integrated than Class I, have been left out of consideration. Future research will have to spell out the notion of "integration" in syntactic as well as semantic terms and to assess its role in grammar, not the least wrt. the core - periphery issue. The volume suggests questions and search strategies that might be helpful in this respect. Here is an example. Dowty's approach rests on data that, while justifying the "dual analysis" by synchronically available adjunctcomplement pairs (cf. 3.1.), by and large indicate a directed diachronic change from adjuncthood to complementhood. This calls for a confirmation by crosschecking to what extent we find diachronic data that show the opposite move, that is, adjuncts as degenerated arguments. Finally, in order to work out the integration issue, we will have to consider further levels of structure, first and foremost those of information structure and prosody, which in this volume are merely mentioned when used for diagnostic purposes. Linking adjunct studies with information structure is needed both on discourse level and on categorial level. A question addressing the latter is this: how come that seemingly all subclasses of sentential adverbs (= Class I), that is, mental-attitude adverbs (unfortu-
24
Ewald Lang, Claudia Maienborn, Cathrine Fabricius-Hansen
nately, leider), epistemic adverbs {probably, vermutlich), subject-related adverbs (stupidly, dummerweise) etc. are not focussable? An exception is the small subclass of "confirmators" (indeed, actually, tatsächlich, wirklich) which, if stressed, serve as the carriers of verum focus. The next question is in what way the non-focussability observed with these subclasses of sentence adverbs may be correlated with other restrictions they are subject to: they cannot occur within the scope of negation, cannot be conjoined etc., in short: they resist any configuration that induces a semantic contrast.
4.2. Adjunct classes and morphology Several papers in the volume (Eckardt, Ernst, Frey, Maienborn, Shaer, Zimmermann) deal with the problem raised by having identical items in distinct adjunct classes, e.g. stupidly as a manner adjunct (Class V) and as a subjectrelated sentence adjunct (Class I) or recently as an event-internal adjunct (Class IV) and as a frame adjunct (Class Π). This raises the question of whether in the field of adjuncts there are correlations between class membership and morphological marking, and if so, whether the possible correlations come in clusters. The fact that deadjectival manner adverbs in English, Russian and Romance are overtly marked by -ly, -o, and -ment(e), respectively, might suggest that it is manner adverbs that form the basic inventory, from which certain subsets might move up in the scale of base positions to also become members of Class Π or I. German shows an entirely different picture: de-adjectival manner adjuncts (dumm 'stupidly', schwer 'heavily', sicher 'safely') are morphologically non-distinct from predicative adjectives, both primary and secondary (Max ist dumm, Max trinkt sich dumm), hence manner adverbs appear as unmarked. However, in Class Π and Class I, lexical adjuncts are distinctly marked: e.g. as domain adjuncts finanzmäßlK 'financially', gesundheitlich 'as to health', or as subject-oriented adjuncts dummerweise 'stupidly', schwerlich 'hardly', sicherlich 'certainly'. Adding the detail that Class I adverbs like dummerweise go back to the manner PP in dummer Weise 'in a stupid manner' makes clear that it may be rewarding to look for language particular interactions of adjunct syntax and morphological adjunct marking. Another case in point is the adjectivization of temporal adverbs by the suffix -ig in German (damals —> damalig, gestern —> gestrig) with its semantic consequences as discussed by Bierwisch (this volume). While this is but a first step towards parametrizing the grammar of adjuncts, a series of steps is needed to address adjuncts, and hence the core periphery issue they raise, from a typological perspective.
Introduction
25
4.3. Adjuncts and modification In support of an integrative view on adjuncts, the present volume no doubt offers some insights and quite a number of suggestions on what the interrelation between adjunct syntax and modifier semantics might turn out to be. The general impression we are left with is that of a few-to-many mapping. Take, once again, the adjunct Classes I-V as defined by their base positions on purely syntactic criteria. Unfortunately, the rigidity of this syntactic classification cannot be uniquely mapped onto an equally clear-cut inventory of modifier types, instead, we have to reckon on a variety of interpretations which (practically) each of the classes has access to. The six papers collected in Part D are devoted to exactly this issue. As for now, the conclusion thus reads: sticking to compositionality as a guiding principle, the best we might expect from the syntactic adjunct classes is the filtering effect they impose on the interpretations they may receive as modifiers, while leaving much space for further differentiations and subtleties yet to be discovered, in short: for the interpretational flexibility that has become the trademark of adjuncts. Hence, the challenge that will guide future research in the field of modifier semantics consists in delimiting the scope of admissible variation and in unveiling the constraints it is subject to. Readers who might find this somewhat abstract are invited to answer an apparently simple question such as "What do all manner adverbiale have in common semantically (except their name)?" To conclude: precisely because adjuncts and modifiers have been placed at the periphery of grammar, they deserve to be moved into the centre of grammar research.
5.
Acknowledgements
On the way from the Oslo conference to the date of its publication, the present volume has been accompanied by helpful people whose generosity is hereby gratefully acknowledged. The whole enterprise would not have been possible without the generous funding of the Norwegian Research Council. Additional financial support was granted by the Norwegian Academy of Science and Letters, the Faculty of Arts of the University of Oslo, the Centre of General Linguistics, Typology and Universale Research (ZAS, Berlin) and the Humboldt University of Berlin. Next, we wish to thank the publisher Mouton de Gruyter for accepting a volume of this size; special thanks are due to Ursula Kleinhenz and
26
Ewald Lang, Claudia Maienborn, Cathrine
Fabricius-Hansen
Wolfgang Konwitschny for continuously and patiently taking care of the various stages the book project had to run through. Thanks also to Artemis Alexiadou and Tracy A. Hall for including the volume in the Interface Explorations
series.
The volume has greatly benefited from the fact that the majority of contributors engaged in mutual refereeing, which created an atmosphere of vivid discussion and continuous revising. The list of referees to be thanked includes: Manfred Bierwisch, Helen de Hoop, Johannes Dolling, Regine Eckardt, Werner Frey, Bart Geurts, Hubert Haider, Gerhard Jäger, Manfred Krifka, Sebastian Löbner, Irene Rapp, Susan Rothstein, Arnim von Stechow, Anatoli Strigin, Chris Wilder, Ilse Zimmermann. As is well-known, the technical side of the editing business is no less exhausting than matters of content. We are indebted to Torgrim Solstad for bridging the gap between incompatible text processors. That the whole chaos could finally be condensed to an orderly formatted and - as we hope quite readable manuscript is, above all, the merit of two wonderful people. Sabine Krämer will certainly be awarded the best formatter medal for her meticulous and untiring work in preparing the camera-ready copy. The fact that 18 Englishes from native and non-native speakers could be unified to something that comes close to royal English is a miracle performed by Philippa Cook.
References Alexiadou, Artemis 1997 Adverb Placement: A Case Study in Antisymmetric Syntax. Amsterdam: Benjamins. Alexiadou, Artemis, and Peter Svenonius (eds.) 2000 Adverbs and Adjuncts. (Linguistics in Potsdam 6.) Potsdam, Universitätsbibliothek. Austin, Jennifer, Stefan Engelberg, and Gisa Rauh (eds.) in progress The Syntax and Semantics of Adverbials. Amsterdam/Philadelphia: Benjamins. Beck, Sigrid, and William Snyder 2001 The Resultative Parameter and restitutive again. In Audiatur Vox Sapientiae, Caroline Féry and Wolfgang Sternefeld (eds.), 48-69. Berlin: Akademie Verlag. Chomsky, Noam 1965 Aspects of the Theory of Syntax. Cambridge, Mass: The MIT Press. 1995 The Minimalist Program. Cambridge, Mass: The MIT Press.
Introduction
27
Cinque, Guglielmo 1999 Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Corver, Norbert, and Denis Delfitto 2000 Introduction: On adverbs and adverbial modification. Italian Journal of Linguistics 12(1): 3—11. Davidson, Donald 1967 The logical form of action sentences. In Essays on Actions and Events, D. Davidson (1980), 105-122. Oxford: Clarendon Press. Delfitto, Denis 2000 Adverbs and the Syntax/Semantics Interface. Italian Journal of Linguistics 12(1): 13-53. Dimroth, Christine 2002 Fokuspartikeln und Informationsgliederung im Deutschen. Dissertation, Max-Planck-Institut für Psycholinguistik, Nijmegen. Dölling, Johannes 2001 Systematische Bedeutungsvariationen: Semantische Form und ¡contextuelle Interpretation. (Linguistische Arbeitsberichte 78.) Leipzig. Dölling, Johannes, and Tatjana Zybatow (eds.) 2001 Ereignisstrukturen. (Linguistische Arbeitsberichte 76.) Leipzig. Dowty, David R. 1979 Word Meaning and Montague Grammar. Dordrecht: Reidel. Eckardt, Regine 1998 Adverbs, Events and Other Things. Issues in the Semantics of Manner Adverbs. (Linguistische Arbeiten 379.) Tübingen: Niemeyer. Egg, Markus 2000 Deriving and resolving ambiguities in wieder-sentences. Ms. Engelberg, Stefan 2000 Verben, Ereignisse und das Lexikon. Tübingen: Niemeyer. Ernst, Thomas 2002 The Syntax of Adjuncts. (Cambridge Studies in Linguistics 96.) Cambridge: Cambridge University Press. Fabricius-Hansen, Cathrine 1980 Lexikalische Dekomposition, Bedeutungspostulate und wieder. In Perspektiven der lexikalischen Semantik, D. Kastovsky (ed.), 26—41. Bonn: Bouvier. 1983 Wieder éin wieder? Zur Semantik von wieder. In Meaning, Use, and Interpretation of Language, R. Bäuerle, C. Schwarze, and A. v. Stechow (eds.), 97-120. Berlin: Walter de Gruyter. 2001 Wi(e)der and Again(st). In Audiatur Vox Sapientiae, Caroline Féry and Wolfgang Sternefeld (eds.), 101-130. Berlin: Akademie Verlag. Geuder, Wilhelm 2000 Oriented adverbs. Issues in the lexical semantics of event adverbs. Dissertation, Universität Tübingen.
28
Ewald. Lang, Claudia Maienborn, Cathrine Fabricius-Hansen
Haider, Hubert 2000 Adverb placement - convergence of structure and licensing. Theoretical Linguistics 26: 95-134. Higginbotham, James, Fabio Pianesi, and Achille C. Varzi (eds.) 2000 Speaking of Events. New York, Oxford: Oxford University Press. Jacobs, Joachim 1993 Integration. In Wortstellung und Informationsstruktur, M. Reis (ed.), 63-116. Tübingen: Niemeyer. Jakobson, Roman 1936/1990 Beitrag zur allgemeinen Kasuslehre. TCLP IV. English version: Contribution to the general theory of case: General meanings of the Russian cases. In Roman Jakobson, On Language, Linda R. Waugh et al. (eds.), 332-385. Cambridge: Harvard University Press. Kamp, Hans, and Antje Rossdeutscher 1994 DRS-construction and lexically driven inference. Theoretical Linguistics 20: 166-235. Klein, Wolfgang 2001 Time and Again. In Audiatur Vox Sapientiae, Caroline Féry and Wolfgang Sternefeld (eds.), 267-286. Berlin: Akademie Verlag. Laenzlinger, Christopher 1998 Comparative Studies in Word Order Variation: Adverbs, Pronouns and Clause Structure in Romance and Germanic. Amsterdam: Benjamins. Landman, Fred 2000 Events and Plurality: The Jerusalem Lectures. Dordrecht: Kluwer. Maienborn, Claudia 1996 Situation und Lokation: Die Bedeutung lokaler Adjunkte von Verbalprojektionen. Tübingen: Stauffenburg. 2001 On the position and interpretation of locative modifiers. Natural Language Semantics 9: 191-240. 2002 Die logische Form von Kopula-Sätzen. Berlin: Akademie Verlag, (in press) McCawley, James D. 1968 The role of semantics in a grammar. In Universals in Linguistic Theory, E. Bach and R. Harms (eds.), 124-169. New York: Holt, Rinharr & Winston. 1972 Syntactic and Logical Arguments for Semantic Structures. Indiana University Linguistic Club. Musan, Renate 2002 The German Perfect. Its Semantic Composition and its Interactions with Temporal Adverbials. Dordrecht: Kluwer.
Introduction
29
Neeleman, Ad 1994 Scrambling as a d-structure phenomenon. In Studies on Scrambling: Movement and Non-Movement Approaches to Free Word-Order Phenomena, N. Corver and H. van Riemsdijk (eds.), 387-429. Berlin/New York: Mouton de Gruyter. Parsons, Terence 1990 Events in the Semantics of English. A Study in Subatomic Semantics. Cambridge, Mass: The MIT Press. Pittner, Karin 1999 Adverbiale im Deutschen. Untersuchung zu ihrer Stellung und Interpretation. Tübingen: Stauffenburg Verlag. Rapp, Irene, and Arnim von Stechow 2000 Fast "almost" and the Visibility Parameter for functional adverbs. Journal of Semantics 16: 149-204. Rothstein, Susan to appear Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Rothstein, Susan (ed.) 1998 Events and Grammar. Dordrecht: Kluwer. Stechow, Arnim von 1995 Lexical decomposition in syntax. In The Lexicon in the Organization of Language, U. Egli et al. (eds.), 81-118. Amsterdam: Benjamins. 1996 The different readings of wieder "again": A structural account. Journal of Semantics 13: 87-138. Tenny, Carol, and James Pustejovsky (eds.) 2000 Events as Grammatical Objects. Stanford, CA: CSLI Publications. Vendler, Zeno 1967 Linguistics in Philosopy. Ithaca, New York: Cornell University Press. Wierzbicka, Anna 1980 The case for surface case. Ann Arbor: Karoma Publishers. Zwart, C. J. W. 1993 Dutch syntax: A Minimalist approach. Ph.D. dissertation, University of Groningen.
Part A: The argument-adjunct distinction
The dual analysis of adjuncts/complements in Categorial Grammar David Dowty
Abstract The distinction between COMPLEMENTS and ADJUNCTS has a long tradition in grammatical theory, and it is also included in some way or other in most current formal linguistic theories. But it is a highly vexed distinction for several reasons, one of which is that no diagnostic criteria have emerged that will reliably distinguish adjuncts from complements in all cases - too many examples seem to fall into the crack between the two categories, no matter how theorists wrestle with them. In this paper, I will argue that this empirical diagnostic "problem" is, in fact, precisely what we should expect to find in natural language, when a proper understanding of the adjunct/complement distinction is achieved: the key hypothesis is that a complete grammar should provide a DUAL ANALYSIS of every complement as an adjunct, and potentially, an analysis of any adjunct as a complement. What this means and why it is motivated by linguistic evidence will be discussed in detail.
1.
Preliminaries: phenomena and theory
1.1. The pre-theoretic notions of 'adjunct' vs. 'complement' We will begin with some basic, intuitive, characteristics that have motivated linguists to draw the adjunct/complement distinction over the years, whatever their theory (if any) of these categories. That is, we start from common pre-theoretic notions of how adjuncts differ from complements, and proceed to build a formal account that, as a first goal, satisfies these: •
Syntax: An adjunct is an "optional element", while a complement is an "obligatory element".
•
Semantics: An adjunct "modifies" the meaning of its head, while a complement "completes" the meaning of its head.
To try to spell out more concretely what these entail, I propose the following restatement of them; I have chosen this particular way of formulating
34
David Dowty
them because it will help us to better see how the theory presented below does satisfy them, but I believe this formulation is still consistent with linguists' pre-theoretic notions. •
An adjunct is "optional" while a complement is "obligatory":
-
A constituent Y in a phrase [XY] (or in [YX]) is an ADJUNCT if and only if (i) phrase X by itself (without Y) is also a wellformed constituent, and (ii) X (without Y) is of the SAME syntactic category as phrase [XY]. (X is in this case the HEAD of the phrase [XY].) Then, a constituent Y in [XY] is a COMPLEMENT if and only if (i) X by itself (without Y) is not well-formed, or else (ii) if it is grammatical, then X standing alone does not have the same category as in [XY] (and does not have exactly the same meaning as it has in [XY]).
The caveat in (ii) is needed to allow for elliptical complements, which this criterion might otherwise class as adjuncts; see more just below. •
An adjunct "modifies" the meaning of its head while a complement "completes" its head's meaning: -
-
-
If Y is an adjunct, the meaning of [XY] has the same kind of meaning (same logical type) as that of X, and Y merely restricts [XY] to a proper subset of the meaning/denotation of X alone. Where Y is a complement in [XY], (i) the meaning of X by itself, without Y, is incomplete or incoherent. Else, (ii) X must be understood elliptically - the hearer must imagine/infer some context-dependent or anaphoric meaning of the general kind of Y to "fill in" the semantic slot that X requires semantically.1 (For example both eat lunch and eat alone are grammatical VPs, but the latter must be understood as "eat something or other", so lunch is a complement, not an adjunct.) Also, the same adjunct combined with different heads affects their meaning in the "same" way semantically (e.g. walk slowly vs. write slowly). But the same complement can have more radically different effects with different heads (e.g. manage to leave vs. refuse to leave).
The dual analysis of adjuncts/complements
in Categorial Grammar
35
There are, to be sure, a number of well-known problematic cases of adjuncts and complements that do not quite fit these characterizations (for example, intensional adjuncts like utter in utter fool), but I still maintain that these general, pre-theoretic characteristics are the first and most basic properties that a linguistic account of adjuncts vs. complements should capture.
1.2. Some Categorial Grammar fundamentals Although the hypothesis of the dual analysis of complements as adjuncts could possibly be formulated within other current grammatical frameworks, it is the theory of Categorial Grammar 2 (henceforth: CG) that offers a particularly direct and compelling way of implementing this hypothesis: because of the tight connection between syntactic analysis and compositional semantics in CG, stronger than in any other current theory, we can show within CG that many of the semantic properties of the argument/modifier distinction follow directly from the syntactic CG characterization of adjunct/complement (and vice-versa). For this reason, we need to explain some assumptions, familiar within CG for a long time now (cf. Venneman and Harlow 1977), as to how the basic distinction is to be made in that theory; these are stated further below in (2). But for this, in turn, we first need to review the way categories are named and are combined to form constituents in CG: (1) a. Standard definitions of syntactic categories: these include both PRIMITIVE CATEGORIES, denoted by simple symbols (usually only these three: S, Ν (common nouns), and NP), and COMPLEX CATEGORIES, formed (recursively) from a pair of more basic categories by "/" and e.g. S/NP, NP\S, S/S, S/(NP\S), etc.) b. How groups of syntactic categories are put together to form constituents: A/B+B => A. ("Where A and Β stand for any categories, a category with a name of the form 'A/B' will combine with a category named Έ ' , to its right, to form a phrase '[Α/Β B]' of category Ά ' ".) Cf. a (nearly) equivalent phrase structure rule A —> Α/Β B. (This rule-schema is called the Functional Application Rule Schema, also known as Slash Elimination and as the "L-rule for / ".) Note that where the slash direction is reversed, (A/B vs. B\A) the left-right order in which the two constituents are combined is to be reversed: Β + B\A A.
36
David. Dowty
c. Semantic interpretation via the CURRY-HOWARD ISOMORPHISM:3 compositional meaning is uniquely and rigidly determined by syntactic structure; the only two possibilities are (semantic) functional application (for Slash-Elimination) and functional abstraction (for Slash-Introduction). In other words, all other kinds of compositional semantic effects, within a construction, must be attributed to meanings of one or more lexical items in the construction (usually, the head), not to compositional semantic rules specific to the construction. This can be viewed as the semantic counterpart of what has been called the "Radical Lexicalism" (Karttunen 1989) that CG demands. d. Categorial Grammar derivation trees have traditionally been conceived of as built up from the leaves of the tree (words) "upward" to the root node, rather than generating a tree from the top node (the root) downward as in PS grammars. Hence, the category that would "dominate" two constituents in PS terms is called the RESULT CATEGORY. This different viewpoint on derivations does not ultimately make any theoretical difference at all, but I will adopt the bottom-up terminology in this paper.
2.
C o m p l e m e n t vs. adjunct in Categorial G r a m m a r
2.1. The traditional characterizations Now, we show how a formal but general definition of complement and adjunct can be made in CG in a way that generalizes across all kinds of syntactic categories. (These definitions can be traced back to Venneman and Harlow 1977.) The distinction is NOT made in terms of phrase-structure configurations and does not mention specific syntactic categories. Rather, the definitions of "head", "complement" and "adjunct" are METACATEGORIAL DEFINITIONS: they use A and Β as variables over category
names, and the characterization depends on the relationship between the relative form of two category names that enter into a combination. Since both grammatical function and semantic interpretation in CG are fully determined by the form of a category name and the category name it is combined with, it is perfectly natural that these meta-categories are specifiable this way in CG.
The dual analysis of adjuncts/complements in Categorial Grammar (2) a. A Head-Complement structure is defined in CG as any syntactic combination of two constituents having the form [Α/Β B] (or else [B B\A]), where A and Β are any categories with the condition that A and Β are not the same category: here, A/B is the Head4 and Β is the Complement. b. A Head-Adjunct structure is defined in CG as any combination of two constituents having the form [A A\A] (or else [A/A A]), where A stands for any category; here, A is the Head and A\A is the Adjunct. Head-Complement structure
A/B
Head-Adjunct structure
Β
AVA
Head Compi. Semantics: Head {Compi')
Head Adjunct Semantics: Adjunct' {Head)
For example, all of the cases in (3) fit the characterization of Head-Complement structures: (here, VP is a notational abbreviation for N\S) (3)
VP VP/NP
VP NP
I
VP/S
I
eat sandwiches those women Semantics: eat' {sandwiches') those' {women')
believe Mary-left believe' {Mary-left')
and (4) shows examples of adjunct constructions: (4)
VP VP
S VP\VP
walk slowly Semantics: slowly' {walk')
s/s^^^
s
clearly John-sings clearly' {John-sings')
sings to-please-Mary to-please-Mary' {sing')
It is immediately clear why the obligatoriness of complements is captured: since the category of the head by itself is not the same as the category of
37
38
David Dowty
[head + complement], the head alone cannot fill the same grammatical slots as the [head + complement] phrase can fill; likewise semantically, the meaning of the head alone is not the same semantic type as that of the phrase, hence the meaning of the head alone is "incomplete" without the complement meaning and cannot yield a meaning of the required semantic type for the phrase as a whole. Conversely, it should be easy to see how it does follow from the characterization of Head-Adjunct structures that adjuncts are "optional" in both syntax and semantics.
2.2. When a head has both complement and adjuncts Two further predictions follow immediately from these characterizations which correspond to old observations about adjuncts vs. arguments: (i) multiple adjuncts (an unlimited number) can accompany the same head (indicated by the dotted line in the diagram in (5)), while only a fixed number of complement(s) can accompany a head (viz. just the one (or two, etc.) subcategorized by the particular head), and (ii) when both complement(s) and adjunct(s) accompany the same head, the complement must generally be "closer" to the head, with the adjunct(s) "outside" the complement. The reasons for these predictions can be seen from this schematic derivation tree and example: (5) Both complements and adjuncts of the same head: VP A A Α/Β
I
Head
A\A A\A
Β
I
Adjunct
Complement
VP
Adjunct
VP
VP/NP
I
eat
VP\VP VP\VP
NP
quickly
with a fork
I
sandwiches
Because the addition of an adjunct to a head leaves the result category the same as the head's category, one can continue to add on more and more adjuncts at will: this is because the highest phrasal category in the tree will always be the same category (here, A) as the one below it. But the combination of a complement (here, B) with a head (A\B) produces a different result category from that of the head (result is A), hence each specified compie-
The dual analysis of adjuncts/complements
in Categorial
Grammar
39
ment must be added exactly once, never more than once. Also, the adjuncts) can be added only after the complement, because the category with which the adjunct can combine is not present until the complement has been added, thus explaining why adjuncts (in this category configuration) occur "outside" complements - and of course the observation that complements typically occur closer to their head than adjuncts do.5 Similar predictions of course follow from X-Bar Theory in other syntactic frameworks - but usually only as a result of stipulating the principles of X-Bar Theory separately from the underlying definition of a phrase-structure grammar. The important point here is that here these predictions already follow simply from the bare fundamentals of CG theory, together with our definition of adjunct and complement.
2.3. Subcategorized adjuncts Traditional grammar has sometimes viewed adjunct and complement as fixed sets of syntactic categories - for example, Adjective and Adverb were considered adjunct categories, once and for all, and Noun (Phrases) were considered complement categories. But more recently it has been recognized that adjectives and even adverbs do in certain contexts appear to behave like complements. Some examples are given in (6): the verb tower seems to take a locative PP as a complement, and the verbs treat and behave take adverbs as complements: (6)
a. The campanile towers over the Berkeley campus/into the sky. *The campanile towers. b. He always treated me fairly. *He always treated me. (OK only with different meaning for treat)6 c. Johnny behaved badly. * Johnny behaved. (OK only with different meaning for behave)
The term Subcategorized Adjunct has been used for such cases.7 Notice that the CG account of adjuncts above, in not treating "adjunct" as a fixed set of categories, does already offer an interesting way of characterizing a subcategorized 'adjunct'. In a head-complement configuration, [Α/Β B], the complement Β can be any category whatsoever, including one that is an adjunct category in other configurations: viz., where Β = C\C. Also, C can be equal to A here, so that Β = A\A. This possibility is illustrated in (7),
40
David Dowty
where (7a) is the typical configuration in which VPYVP (the category of (verb-phrase) adverbs) occurs as an adjunct. But (7b) shows the case where an adverb occurs as a complement: (7) a.
normal adjunct structure: A AVA I β
A α b.
Head-Complement structure with "subcategorized adjunct":
example:
VP AJ (A\A)
I
α'
A\A
VP/(VP\VP)
Iβ
behave
I
VF\VP
I
badly
In fact, we now adopt the category configuration in (7b) as the basis of our definition of Subcategorized Adjunct. Many readers will notice that (7b) is reminiscent of Type Raising, a rule (or theorem) in most versions of CG which can at any time convert a category A to B/(A\B). We do not, however, intend such type raisings to count as heads of subcategorized adjuncts. There are two differences: The adjuncttaking head has the more complex category "in the Lexicon", it is not derived syntactically from the less complex category. Thus, "reanalysis of adjuncts" is a change that happens within the Lexicon, in one way or another. Whereas the semantics of true type-raising constructions is strictly determined by the rule itself, the semantics of the head (functor) is not completely rule-predicted and can be obtained only from the lexical entry: Type Raising of α: a' λ / (/(a')] Adjunct reanalysis of α: a ' => α", α "Φ λ f [/"(a')] This semantic difference is in fact a very important characteristic of argument-to-adjunct reanalyses as presented in this paper, as we will see in the following sections.
The dual analysis of adjuncts/complements
3.
in Categorial Grammar
41
The Dual Analysis Hypothesis
3.1. A Case study: Locative vs. dative to Perhaps the best way to begin to see motivation for the dual analysis is to examine a (very) familiar case where the same prepositional phrase has different meanings with different verbs: English PPs headed by to which sometimes have directional, sometimes non-directional meaning. The directional readings, which are systematic and perfectly compositional, are exemplified in (8a)-(8c): (8)
a. Mary kicked the ball to the fence. b. John pushed the desk to the wall. c. Sue slid the paperweight to the edge of the table.
In these cases, the transitive verb always denotes an action performed on the direct-object's referent, and the to-PP always adds the information that the object of to denotes the new location at/near which the direct object referent ends up as a result of the action performed on it. Such examples can readily be constructed with dozens of transitive verbs of motion. However, (9a)-(9c) are different semantically from the above: (9)
a. Mary explained the memo to John. b. Mary rented the apartment to John. c. John offered a glass of tomato juice to Mary.
(9a) does not mean that the memo itself came to be at/near John, but only that the information contained in the memo came to be more fully understood by John, as a result of Mary's explanation. In (9b), however, neither the apartment nor its "semantic content" changes location: rather, because the verb is rent, we understand that a kind of temporary ownership of the apartment is acquired by John (subject to the conditions of the rent agreement). With offer, neither the glass of tomato juice nor its ownership changes location or possession - what happens is that Mary acquires the option to take possession of the tomato juice, if she so chooses. Examples of such "ambiguity" can be reproduced with many other prepositions (locative remove it from the table vs. non-locative learn it from the doctor), and in other languages. We want to reexamine it here in detail anyway, to delve into the reasoning behind the two best-known ways to try to solve it.
42
David Dowty
3.1.1. First approach: "Abstract thematic roles" Gruber (1965), Jackendoff (1972, and later papers by Jackendoff), Fillmore (1968), and others urged us to analyze the preposition meanings in (8) as well as (9) so that all signify the same thematic role (or abstract deep case), called GOAL. The meaning of GOAL is broad enough to represent both literal change in physical location (directional) in (8), and abstract change in some property not involving literal motion, thus no ambiguity in to need be postulated at all. The same is done with SOURCE and LOCATION, so all nonlocative "changes of state" marked by prepositions are reduced to abstract versions of locative prepositions; this approach has been called the 'Localist Hypothesis' by Anderson (1971). This idea gained wide support. Yet what Gruber and Jackendoff do not ever fully explain to us is how, exactly, the semantic component of the grammar determines which kind of meaning GOAL has in which example. After all, kick the ball to the fence cannot mean that the fence acquires possession of the ball, any more than (9a) can mean that the memo itself moved to John's location. The situation is actually worse than this: the various abstract instances of GOAL differ semantically from each other in unpredictable ways. With explain, the GOAL apparently means "transfer of the information contained in something to NP, but in a more intelligible form". With rent, the GOAL does not mean "transfer the information in the apartment", nor conversely can GOAL with explain refer to a change in possession of (something). With offer, GOAL refers to a transition in an option to acquire, but neither a transfer in information content nor a change in possession. (There are even examples of to that do not refer to a transition into a state at all, but rather the avoidance of such a transition: refuse a hearing to the prisoner, deny requests to all of them.) Thus (as has been recognized by the critics of Jackendoff and Gruber for some time), the abstract element GOAL is not really a semantic element that can play any consistent, useful part in the compositional semantics of all sentences involving to: GOAL is merely a label for a class of cases which may intuitively seem somehow related, but for which we still do not have a real semantic analysis.
3.1.2. Second approach: Ambiguity between adjunct and syntactic marker Logicians, and many semantically conscious linguists, have long regarded the various non-locative occurrences of prepositions as purely grammatical
The dual analysis of adjuncts/complements in Categorial Grammar markers, with the verb of the sentence being the sole semantic source of the multi-place relation being expressed: Mary gives the book to John is thus represented logically using a 3-place relation give: give (m, the-book, j) A currently popular syntactic implementation, then, is to postulate an ambiguity in every relevant preposition (to, from, at, off of, on, onto, etc.) between (i) a meaning-bearing literal locational preposition, and (ii) a syntactic artifact, a (semantically vacuous) idiosyncratic "case marker", "case marking preposition". This permits us to give a correct account of sentences with non-locative PPs, but it is ultimately satisfactory? Note that this approach fails to make any connection in the grammar or semantics between locative to and abstract "dative" to, between locative from and abstract "SOURCE"; it leaves it entirely as a grammatical accident that example after example of prepositions and morphological cases, in language after language (though not in every language), shows this synchronicity.8 Ultimately, this connection must have its origin in the psychology of language acquisition or cognition itself: Clark and Carpenter (1989) show that many English-speaking children make several systematic "errors" in acquiring the ways that "SOURCE" is expressed in English, which taken together, imply unmistakably that children are at some stages working with an underlying concept of "SOURCE" of just the Gruber-Jackendoff kind.
3.1.3. The Dual Analysis: Case-marking-to as a reanalysis of directional adjunct-to Is there no way to better describe the case-localist connection in terms of grammar, or must grammar theorists sit back until some other field (psychology of language?) solves this difficult problem? In fact, I think we can improve on the formal theoretical side of the problem significantly, and the first important step is the DUAL ANALYSIS HYPOTHESIS. The idea behind the dual analysis view can be thought of (for now, anyway) as the claim that the locative adjunct analysis of all occurrences of to, from and other locative prepositions is a PRELIMINARY ANALYSIS which serves language-learners as a semantic "hint" or "crutch" to figuring out the idiosyncratic correct meaning of the complement analysis for the non-locative instances: a preliminary adjunct analysis of the to-PP (as locative) as in (10a) gives way to a complement analysis of to-PP structure as in (10b):
43
44
David Dowty
(10) a. adjunct structure:
b. complement reanalysis:9
VP VP
VP\VP
speak¡ ('VF\VP)/NP I to
NP
(VP\VP)/NP
I Mary
I to
NP
I Mary
Sem: (to' (Mary))(speak' ¡) Sem: speak'2 (to' (Mary')) The semantic interpretation of John speaks to Mary, under the adjunct analysis as in (10a) is "John speaks, and the result of this action is that John ends up in a location next to Mary" - not the real intended meaning of the sentence of course, but a rough "hint" for the learner who has not grasped the spea&-to-constraction. Note that the complement interpretation of speak2 (in (10b)) cannot be the same as speak\ in (10a): rather, it takes the changeof-place /¿»-"adjunct" as its semantic argument, and its meaning is something like "speak, with the intention that the verbal content of what one is saying will end up at a certain place (to-Mary) and will be understood there"; in other words, the proper way to interpret to Mary here is now built into the meaning of speak2.
3.2. The cognitive 'trade-off between adjuncts and complements But why should languages really need an adjunct analysis as a "preliminary step" toward a complement analysis, anyway? If we step back and reflect on the communicative advantages of each, as opposed to the language-learning advantages of each, we can see that there is a trade-off between the two analyses. If we focus on the effort required from the learner of a language, then an adjunct analysis offers the advantage of yielding more quasi-multi-place predications at a lesser load on lexical memory - because they are semantically compositional. Suppose the lexicon of a language has η different intransitive verbs (say, 100 verbs) and m different prepositions that can form adjuncts (say, 10 prepositions), then compositional syntactic and semantic rules automatically produce (η x m) different two-place predications (= 1.000 in this case), all of which have distinct meanings. By contrast, if the
The dual analysis of adjuncts/complements in Categorial Grammar
45
learner had to express all these two-place predications by learning individual transitive verbs, she would need to learn 1.000 different lexical items. But adjunct analyses achieve this advantage at the cost of a limitation on the range of meanings that can be expressed. If we focus on the semantic expressivity of the language, then lexical two-place predicates (verbs taking an object as well as a subject) have an advantage over the quasi-two-place predications derived by adding adjuncts: Though there may be 1000 (m x n ) of the adjunct-derived meanings, these meanings are all limited, in a way that the lexical meanings are not, to what is produced by a consistent compositional semantic rule that combines a verb meaning with a preposition meaning.10 Each lexical two-place predicate can express any imaginable (humanly 'processable') two-place semantic relation. Thus, we achieve greater expressivity at the cost of a larger burden for the language learner. This is just the trade-off we saw with to: we can compositionally generate lots and lots of adjunct-derived locative two-place semantic relations with little effort (walk to, drive to, swim to, walk from, drive from, swim from, etc.) but none of these can correctly express the semantic relation lexicalized in speak to, rent to and offer to, which instead must be learned as individual items. However, by allowing the language learner to access the adjunct analysis as a fruitful preliminary "clue", one would soften the learning burden. If some multi-place relations like speak to, rent to look superficially the same as an adjunct structure, then the learner will be led through the preliminary step automatically. This "trade-off' may not be a very earth-shaking idea for locative-to vs. "dative" to, but note that my claim here is that this same trade-off applies to all parallel cases of an adjunct vs. a superficially similar complement - for example, infinitive adjuncts (e.g. sing to please Mary) vs. infinitive complements (try to please Mary), and the other cases in Section 6.
3.3. Extension to adjuncts/complements in general Against this background of the need for dual analyses of derived words and collocations, I believe we can better understand what is being claimed about dual analyses of complements and adjuncts. Virtually all complements have a dual analysis as adjuncts, and any kind of adjunct can potentially receive an analysis as a complement. The dual analysis is often hard to recognize because:
46
David Dowty
-
Often, the adjunct analysis serves merely as a mnemonic for the complement analysis, and/or as an aid to learning what the complement means when the structure is first acquired by the language learned.
-
Or else only a few members of a pattern have salient complement readings, while adjunct readings appear in most cases.
3.4. A second case study: Agent phrases in passives In the case of the dual analysis just discussed, individual verbs differ fairly sharply as to whether they ultimately take adjunct or complement to. In other cases to be discussed below, a single verb may still permit, in "adult" speech, both an adjunct reading and a complement reading equally, or else a whole construction may prefer the adjunct reading almost exclusively, or the complement reading almost exclusively - the last possibility being illustrated by Agent Phrases in Passives. My proposal is that all these possibilities should be treated formally via dual analyses, with it being left to psycho-linguistics to determine exactly how these cases differ in mental processing. The Agent Phrase of a Passive (by Mary in John was visited by Mary) has been frequently analyzed as an adjunct, but just as frequently analyzed as an instance of "prepositional case marking", i.e. the fry-phrase is a complement of the passive verb, but by has no independent meaning of its own, it is merely the marker that passive verbs subcategorize for. The dual analysis of fry-phrases will provide a second useful case study because it differs from the dative ίο-phrase in several ways; notably, it involves a syntactic/morphological construction, not just single verbs, and more importantly, it shows how the dual analysis is motivated by diachronic and typological facts, not just "thought experiments" in language acquisition. One reason to believe that Passive Agent Phrases are possibly adjuncts is that the meaning borne by the fry-phrase in a passive, as in (11) seems intuitively very similar to that of other fry-phrases as in (12) that do not accompany a passive verb and hence must necessarily be analyzed as adjuncts: (11)
John was touched by Mary.
The dual analysis of adjuncts/complements in Categorial Grammar (12)
This book is by Frege. A dress by Chanel. She sent him a letter by courier. He washed the dishes by hand. She died by her own hand. cf. Cheating by students is punishable with expulsion. (Keenan (1985): NB cheating here is not from a passive verb.)
Note the èy-phrases in (12) all seem to entail a semantically-related sentence that is a true passive: for This book is by Frege, compare "This book was written by Frege"; for She sent him a letter by courier, compare "A letter was delivered by courier", and so on. Nevertheless, it has been recognized in the semantics literature for some time that a semantically correct adjunct analysis of agent phrases in passives is either impossible or else very difficult (and has not been achieved in any case; cf. Thomason (1974), Cresswell (1985), Dowty (1979)). For one thing, pairs like (13) show that an adjunct analysis cannot be extensional but must be handled intensionally in some way, while a complement analysis never requires this complication: (13) a. This chair was sold (at noon today). b. This chair was bought (at noon today). Since (13a) is true if and only if (13b) is true, it follows that the predicate is bought has the same denotation as is sold. But if so, then the result of applying an extensional adjunct to one predicate must be equivalent to the result of applying that same adjunct to the other: if by John is such an adjunct, then bought by John must be equivalent to sold by John, which, of course, is a wrong result. If by John is a complement rather than an adjunct, then this problem does not arise, cf. (22), (23) below. (See the above-cited references and Dowty (1989) for comment.) It is sometimes claimed that a "NeoDavidsonian" analysis in terms of events can circumvent this problem, cf. Parsons (1990), but as argued in Dowty (1989) and Dowty (2001), this will not really work. This approach appeals to an abstract Thematic Role 'AGENT', but this fails for the same kind of reason that we saw with GOAL earlier: there is no possible semantic definition of AGENT that is independent of the particular verb meaning that it occurs with. Notably, Passives of stative verbs occur with Agent Phrases (in English and other languages), but these are not "Agents" in a semantic sense, rather they are "Experiencers": (14)
This rumor has now been heard by almost every voter, and it is believed by many of them.
47
48
David Dowty
The only correct way to identify what the fry-phrase refers to here is to appeal to the meaning of the active verbs hear and believe, and this demands a complement analysis of the fry-phrase. The account in terms of a dual analysis begins with the assumption (well-justified by cross-language typological studies) that the agentless passive is the most basic form of passives - they occur in more languages than agentive (or "full") passives occur in, while there are no languages with only the agentive passives but no agentless passives. The agentless passive can be analyzed adequately and very simply as a detransitivizing, "relationreducing" operation on transitive verbs: (15) Passive as a detransitivizing operation: (Agentless) Passive: Lexical Rule: a e vp/np —> PST-PRT(a) e vp^a»] Semantic Interpretation: a' —> Àx3y[a'(x)(y)] (16)
[John was touched^ John,'np
[was touched]vp wasvp / psip
touched/^ touch,vp/np
(17) Translation of (16): 3x [touch' (John')(x)] Assuming that fry-phrase adjuncts (as in (12) above) exist in the language already, then the meaning of a full passive can be approximated, without any addition to the syntax, by adding a fry-phrase adjunct to an agentless passive: (18)
[John was touched by Mary]5 John„p
[was touched by Mary]vp [was touched]vp was^Pìip"
~~toùchedps,p toucha
[by Mary] vp \ vp by(vp\vp)/np
Mary„p
The dual analysis of adjuncts/complements
in Categorial
Grammar
49
(19) Translation of (18): by'(Mary')(Ày[3xtouch' (y)(x)])(John') "John was touched, and Mary was a causal factor in this event." As the paraphrase in (19) suggests, the meaning of by' here, which is the adjunct meaning, does not produce the correct meaning of the actual English passive sentence but only approximates it, and of course it also cannot possibly serve as the final analysis of full passive for the reasons cited above (and it is important to note in this regard (cf. below) that many languages exist in which agent phrases are not found with passives of stative verbs, only active verbs). And so, I argue, the adjunct analysis serves as a preliminary step through which the complement analysis is reached. That analysis is: (20) (Reanalyzed) Passive (as yielding 2-place predicate): syntactic rule: α e vp/np PST-PRT(a) e PstP/np^, semantic rule: α' —» λγλχ[α'(χ)(γ)] In this rule, PstP stands for the category of past participles (semantically the same type as the VP category), and I have incorporated the further simplification of the category of passive touched from PstP/(VP\VP) to PstP/NP^y] (see note 9). (21)
[John was touched by Mary]s John,np
[was touched by Mary]vp wasVi,/ PstP
[touched by M a r y W touched pslp/ touchVp/np
[by Mary]nP[by)
np
lbyi
by
"Plby]'"P
Mary,np
(22) Translation of (21) is equivalent (by λ-conversion) to: touch'(John'XMary') For passive agent phrases, typological data about the distribution, form, and semantic restriction on agent phrases cross-linguistically imply that there are observable diachronic manifestations of the reanalysis hypothesis as just sketched above. The following typological generalizations about passive agent phrases were observed in (Keenan 1985: 247):
50
David Dowty
(23)
a. Some languages exist in which only agentless passives occur, though no languages apparently exist in which agentive passives occur but no agentless passives. b. In many languages, passives of stative and other "not highly transitive" verbs are ungrammatical. c. Either instrumental case or a preposition with instrumental meaning is (almost) always used to mark the agent of a passive in natural languages, according to Keenan (1985: 261). (Actually, it seems that prepositional phrases with SOURCE meaning sometimes appear instead, e.g. English from, German von.)
These typological generalizations are just what we would predict if the dual analysis/reanalysis of agent phrases is given the following diachronic interpretation, as five (possibly hypothetical) stages in the development of passives in a language: (24)
Hypothesized stages in the development of passives with agentphrases:
1. Passive is a relation-reducing (detransitivizing) rule (Dowty 1982). Only the 'agentless forms' of passive sentences appear in the language; what will become agent phrases (fry-phrases, in English) only occur as instrumental adjuncts of non-passive VP's (send the package by airmail or the like). 2. Agent Phrases occur as adjuncts (with instrumental/source meaning) to passive verbs; as instrumental agent phrases would not make sense with stative and other non-volitional and non-causative passive verbs, agent phrases never occur with them. 3. Agent-phrases are reanalyzed as complements of passive verb phrases, thus leading to step 4: 4. The agent-phrase-marking preposition (by in English) or instrumental case marking is reanalyzed as a marker of grammatical function (without independent semantics), a so-called case-marking preposition and does not contribute any meaning per se to the compositional semantics of the sentence. Passive is now an argument-permuting rule ("relation-changing" rule), i.e. one that yields the same 2-place lexical meaning as the active verb but with subject and non-subject arguments interchanged. 5. Passives of stative verbs, other non-causati ves, etc. now occur.
The dual analysis of adjuncts/complements in Categorial Grammar 51 Before going on to the next section, we may note that our two case studies of adjunct reanalysis - directional PPs and full passives - differ in that the first involves replacing lexical items (verbs) one by one by their reanalyzed counterparts, whereas the second involves a reanalysis of the lexical rule for passives. I intend both cases to fall under the rubric dual analysis hypothesis as used here. The original passive lexical rule produces individual verbs as outputs, and the outputs of the revised lexical rule are also individual verbs: the first vs. second version of each passive verb stand in the same kind of relationship to each other as the two versions of a verb taking a ίο-phrase as adjunct vs. complement. (This point may be clearer if considered under the view of lexical rules in which they operate "off-line" to create possible lexical items, and then these possible additions can be transferred, one at a time, into the actual, or "on-line", lexicon.)
4.
Predictions about word order and meaning of adjuncts vs. complements
4.1. Infinitive adjuncts vs. infinitive complements In English and typologically similar languages, adjuncts in general can often occur in various positions within the clause, while superficially similar complements have more restricted distributions. Specifically, complements in English always follow verbal heads, as for example (25). (25) a. John sang to please Mary. {to please Mary is adjunct) To please Mary, John sang. John, (in order) to please Mary, sang for hours. b. John tried to please Mary. (to please Mary is complement) *To please Mary, John tried. *John, (in order) to please Mary, tried for hours. (acceptable only if tried is taken to have an elliptical complement distinct from try.)
4.2. Subcategorized adverbs Subcategorized adverbials, are distinguished from true adjunct adverbials in just this way:
52
David Dowty
(26) a. They criticized him harshly, (adjunct) They harshly criticized him. b. They treated him harshly. (subcategorized adverb) *They harshly treated him.11 Allowing multiple syntactic positions for adjuncts can be accomplished in various ways in CG; one way is to give adjuncts multiple category membership (e.g. S/S vs. S\S for sentential adjuncts). Another is to introduce them as permutable constituents within a multi-modal CG, allowing adverbs to obey the logic of LP. But once we observe the (independently verifiable) typological demand in English that complements always follow their heads, then the implication for adjunct reanalysis is this: -
An adjunct can be reanalyzed as a complement (in English) only when it follows its head; the same adjunct in any other syntactic position cannot be reanalyzed.
4.3. Position of repetitive vs. restitutive again ('wieder') At this point it is interesting to look at the word order possibilities for the repetitive (external) reading of English again and German wieder vs. those for the restitutive (internal) again and wieder (Dowty 1979; Fabricius-Hansen 1983). This ambiguity is present in a sentence like Mary shook John awake again: the external (or repetitive) reading entails that this was the second time that Mary had shaken John awake; the internal (or restitutive) meaning only entails that John became awake for a second time as a result of Mary's shaking him, not that she shook him for a second time, i.e. Mary has merely restored the state of John's awakeness. Dowty (1979: 260-264) proposed that this meaning difference results from a complement vs. adjunct ambiguity (an ambiguity in the verb's category), not an ambiguity in the adverb per se12, so this predicts that the availability of both readings will depend on word order. Fabricius-Hansen (1983 and elsewhere) and others have argued for a different analysis of this ambiguity, thus not immediately predicting any word order sensitivity. In fact, the two readings are indeed limited by syntactic position, just like infinitives and subcategorized adverbs (treat harshly). When again occurs to the right of the verb, both readings for again are available; in any other position, only the external (repetitive) reading exists:
The dual analysis of adjuncts/complements in Categorial Grammar 53 (27) a. Mary shook John awake again. (both readings) b. Again, Mary shook John awake. (only repetitive reading) Mary again shook John awake. (only repetitive reading) c. When the power failed, the satellite entered the atmosphere again, (both readings) d. When the power failed, the satellite again entered the atmosphere. (only repetitive reading) Arnim von Stechow (1996 and p.c.) has noted that the restitutive reading of wieder in German is only available when wieder appears in a syntactic position in which a verbal complement can appear in German, while the repetitive reading is available for wieder in any position in which German allows an adverb. Thus for both English and German, the syntactic predictions of the reanalysis hypothesis are met for again/wieder. See Dowty (1979, 1993, to appear) for more data and details.
5.
Arguments for simultaneous multiple analyses from historical linguistics
The postulation of simultaneous multiple analyses has often been regarded with suspicion within the methodology of modern linguistic theory - a sign of a "missing generalization" at least, and always deemed inferior to a proposed alternative that appeals only to a single analysis. In spite of this, several papers over the years have argued explicitly for multiple syntactic analyses, even when there is little or no detectable accompanying semantic ambiguity. A few of these are: -
Hankamer (1977) "Multiple Analyses" Kroch (1989) "Reflexes of Grammar in Patterns of Language Change" Ladusaw and Dowty (1988) and Bresnan (1982), 'Syntactic Control' of complements vs. 'Real-World Control' of actions and objects, and the unexpected acceptability of He was promised to be allowed to leave.
But many historical linguists have long accepted the idea that multiple analyses must be assumed to be available to a single generation of speakers in order to fully explain the facts of language change. One clear, explicit statement of the reasoning behind this deserves to be quoted in full here:
54
David Dowty 4.4.3. Multiple analyses during actualization During the period of actualization, a single input structure continues to have multiple analyses in the grammar of the individual speaker. For descriptive purposes it is convenient to recognize three stages to reanalyses: Stage A, Input: The input structure has all of the superficial characteristics of the input analysis. Stage B, Actualization: The structure is subject to multiple analysis: it gradually acquires the characteristics of an innovative analysis, distinct from that of Stage A. Stage C, Completion: The innovative structure has all of the superficial characteristics of the innovative analysis. Reanalysis is the transition from Stage A to Stage B. Stage Β is the period of actualization, and the speaker makes both (or many) analyses, which may be related to each other in different ways at different times. Stage Β typically consists of multiple changes, reflecting the characteristics of the particular construction in the particular language. It may be noted that the gradualness of change is due in part to the duration of actualization in some changes. Some reanalyses may not reach Stage C; they are never completed, in the sense that all the characteristics of the innovative analysis may not be acquired. It has often been assumed, especially in the description of change in individual languages, that in reanalysis the period of multiple analyses is only transient, and that the innovative analysis rapidly replaces the earlier analysis. There are at least three kinds of evidence that multiple analyses continue to be available in individual grammars for some time, though that time of course is different for different changes. Evidence comes from the possibility of multiple reflexes, from variation and conflicting data, and from the possibility of reversibility of change. (Harris and Campbell 1995: 8Iff.)
6. Evidence for the adjunct "origin" of most complements in English Probably one of the most compelling arguments for dual analyses in English comes from the very large set of pairs of cases where (i) an adjunct construction is found that parallels a complement construction exactly, at least in "surface" syntax, (ii) the two parallel constructions can be shown to have the same kinds of semantic similarities and differences between adjunct and complement already discussed above, and (iii) the same syntactic differences also occur (i.e. word order possibilities). Because of space limitations, all I can do here is enumerate a representative list of these pairs, with examples for each pair, cf. Table 1 below.13
The dual analysis
of adjuncts/complements
in CategoriaI
Grammar
55
It will have to be left as an exercise for the reader (i) to find more examples for each pair of constructions, (ii) to verify that the allowable word orders are usually broader for the adjunct than for the complement case, (iii) to figure out the (regular) adjunct meaning of each case, and (iv) to verify that the "specialized" meanings of the complement examples do in fact differ (sometimes subtly) from the corresponding regular adjunct meaning. The case of complement vs. adjunct genitives is worth special comment, all the more so in this context because of the interesting connections between it and Partee and Borschev's paper on genitives in this volume. It has been widely recognized for years that possessives (and genitives) have a different semantic function when they combine with relational nouns (friend, mother, top, etc.) than with non-relational nouns (team, dog, table, etc.). This idea has been thoroughly investigated (independently) by Barker (1991, 1995) and by Partee (1997) (based on unpublished work by Partee from 1983 and developed in subsequent papers). The reading (normally) found with relational nouns (Mary's mother) is called LEXICAL, INTRINSIC (Barker) or INHERENT (Partee), and that with non-relational nouns (Mary's book) is called EXTRINSIC (Barker) or FREE (Partee), or MODIFIER. The meaning of the extrinsic possessive is quite broad but is also context dependent - for example, John's team could mean, depending on the context in which it is uttered, either "the team that John plays on", or "the team that John owns", or "the team that John cheers for", or "the team that John placed a bet on today".
56
David Dowty
Table 1. Examples of corresponding adjunct vs. complement in English
Al. A2. A3. Β1. B2.
C. D. E.
Fl.
F2.
ADJUNCT CONSTRUCTION
COMPLEMENT CONSTRUCTION
Adjective adjuncts to VPs John left work exhausted. Adj. adjuncts to transitive Vs13 John ate the meat raw. "Repetitive" again (adjunct) Again, Mary shook John awake. Directional PP adjuncts to intr. Vs Mary walked to the park. 13 Directional PP adjuncts to tr. Vs John threw the ball to the fence.
Adjective complements to VPs John arrived alone. Adj. complements to transitive Vs 13 John's attitude made Sue unhappy. "Restitutive" again (complement) Mary shook John awake again. Dative complements to intr. Vs John sang to Mary. Dative complements to tr. Vs John threw the ball to Mary. John threw Mary the ball. W/iA-marked complements John loaded the truck with hay. Other intr. and tr. PP complements I learned it from a doctor. Agent phrases of passives (final) [ = fry-phrase as complement of passive verb]
Instrumental wä/i-adjuncts John swept the floor with a broom. Other intr. and tr. PP adjuncts I took it from the box. Agent phrases of passives (early stage) [ = fry-phrase as instrumental adjunct] 'Rationale' purpose infinitives John sang (a song) (in order) to impress Mary. Gapless object-controlled 13
G.
H.
I.
infinitive adjuncts Joe hired her to fix the sink. Gapped non-subject-controlled infinitive adjuncts Sue bought it to read on the plane. Adjective-modifying gapped infinitive adjuncts It is available to figure your tax with . Possessive adjuncts of non-relational nouns Mary's book (team, etc.) A book (team) of Mary's
Infinitive complements of verbs John attempted to impress Mary. Infinitive complements of transitives Joe persuaded her to fix the sink. (None?)
"Tough"-complements It is hard to figure your tax with Possessive complements of relational nouns Mary's mother (enemy, etc.) the mother (enemy, etc.) of Mary
The extrinsic/free reading, it has been proposed, has a meaning such that Poss Noun is uniformly "the unique Noun that stands in some contextuallydetermined but salient relation to Poss"; it is up to the hearer to figure out
The dual analysis of adjuncts/complements in Categorial Grammar 57 exactly what kind of relation is intended, though the relation of 'ownership' is probably the most common. If so, this extrinsic meaning can be semantically analyzed as an adjunct reading in my sense. The intrinsic/inherent possessive (Mary's mother, mother of Mary) differs in that the nature of the relation between Possessor and Noun is determined by the relational noun (so it is of course different for each relational noun). Thus in terms of this paper's hypothesis, the inherent genitive must be a complement of the relational noun, not an adjunct. The syntax of these two kinds of possessives and genitives is different from the other complement/adjunct cases above: the pre-nominal possessive is the one case I know of where a complement can precede its head (in English), e.g. Mary's mother. But the two readings do differ syntactically in the post-nominal position, albeit in a subtle way: the so-called "double genitive", as in a book of Mary's is only found with extrinsic (adjunct) genitive meaning:14 note that #The mother of Mary's sounds quite odd, which is because mother is relational. Conversely, the post-nominal genitive with no possessive suffix occurs only with inherent/intrinsic (relational) readings (The mother of Mary) and not with non-relational heads (#A book of Mary sounds odd). (Cf. also Partee and Borschev's paper in this volume.) The significance of all the cases A-I in Table 1 can be summarized this way: If it is important to the grammatical structure of a language (and important to the learners of the language) to distinguish adjuncts from complements, why should the grammar of English have dozens of cases in which an adjunct construction and a complement construction look superficially exactly alike? This seems rather counter-productive. But, if it somehow helps the language learner that each complement construction should look so similar to an adjunct construction as to be initially "mistaken" for one, then this is exactly the distribution of data that we should expect!
7.
Dual Analysis is a more complex matter than just reanalysis in language acquisition
We can better understand that there are broader implications of the dual analysis hypothesis by digressing for a moment to examine the semantics of compounding and other word formation rules.
58
David Dowty
7.1. Compounds and derived words In the history of the study of compounding in generative transformational grammar, linguistic theory has alternated repeatedly between deriving compounds by grammatical rule (Lees 1960; Levi 1975) and arguing that derived compounds are not derived grammatically but are only listed individually "in the lexicon"; the latter position is supported by pointing to the idiosyncrasies of the meanings of individual examples that cannot possibly follow from any general rules (Chomsky 1970, and in a different sense, also Downing 1977). The mistake that I think has usually been made in this debate is the assumption that if compounds (etc.) are listed individually in the lexicon (and each is assigned a meaning there), then there cannot also be a rule that derives meanings of compounds as the default case. Instead, I believe that a speaker's knowledge of her/his language includes both ways of deriving meanings for most compounds. This is best shown with English Adjective-Noun compounds. All English speakers, I propose, know that any compound of the form "Adj-Noun" has associated with it a general, rule-predicted meaning paraphrasable as "Noun that is Adj". Thus a blackberry must be "a berry that is black", a bluebonnet is "a bonnet that is blue", and so on. But at the same time, speakers are perfectly aware that "berry that is black" (etc.) is not the real meaning of blackberry·, that is rather "a certain species of bush that produces edible black, tiny berries in clusters." Other examples: (28) example:
predictable meaning:
real meaning:
a.
big shot
"shot that is big"
b.
blackboard
important or influential person surface made for writing on with chalk, often black in color fine sand mixed with water that sucks down an object resting on its surface
"board that is black" c.
quicksand "sand that is quick"
To deny that speakers know there is some elementary sense in which soft drink means "drink that is soft" is to deny an obvious facet of speakers' knowledge of their language, notwithstanding the fact that they also know a "real" or "correct" meaning for such compounds.
The dual analysis of adjuncts/complements in Categorial Grammar
59
W h y should languages have such double meanings for compounds? With only a moment of reflection, the answer is obvious, I believe: the "predictable" meaning of a compound: -
gives the hearer a "clue" or "hint" to the compound's real meaning upon first encountering the compound serves as a mnemonic for more easily retrieving that real (and individually learned) meaning from memory when the compound is encountered again later.
(Try as a mental exercise to imagine what English would be like if all compounds were replaced by mono-morphemic words that had to be learned individually, without any morphological clues: English would be far harder to learn!) On encountering the compound software for the first time, a speaker at least has a clue, from its derivational meaning ("wares that are soft"), where to start guessing what the real meaning might be. That is, one does not necessarily assume for an initial period of time that it really literally means "ware that is soft" and then correct that assumption later: more likely, a person realizes already at first hearing that software must have a much more specific, probably technical meaning. What exactly is the relationship between the two meanings of a pair in the speaker's mind? What should it be in a linguist's grammar? T h e first question is no doubt highly interesting for psycholinguistics and the psychology of memory, but I doubt that much can be specified about this relationship in linguistic theory - nor should we try to. What w e can and should do is simply specify that there are two kinds of meanings for each: (i) a predictable but only approximate meaning (and the rule that gives it from the meanings of the parts), and (ii) an individually-learned meaning for it just like the individually-learned meanings of all monomorphemic words. Other kinds of derivational word formation also show the need for dual analysis: it is intuitively felt by all speakers of English that all derivations of VERB+able have a uniform approximate meaning: "capable of being verb+ed" - so that washable means "capable of being washed". A t the same time, speakers know that many such forms have a more specific actual meaning: readable does superficially mean "capable of being read", but its actual meaning is something narrower. 1 5 M y general point in making these observations about word formation is to argue that the two analyses for each instance of word formation are almost certainly not simply a matter of the lexicalized analysis replacing the preliminary analysis, then disappearing forever; rather the preliminary, semantically compositional analysis is still employed, in some subtle psycho-
60
David Dowty
logical way, in on-line processing - though in a way that only connectionism or some other future theories of the psychology of language can explain. If this is plausible, then simultaneous on-line processing is just as plausible for the "dual" complement-adjunct analyses. So what the dual analysis hypothesis accomplishes (for both domains) is to allow theorists to formalize - for now - the two endpoints of a complex psycholinguistic "continuum", a psychological phenomenon where we are not ready to try to formalize the intermediate points. I have argued that being able to acknowledge and isolate these endpoints, within a formal linguistic theory, improves our understanding of the phenomena of "adjunct" and "complement".
Notes 1.
2.
3. 4.
Admittedly, the difference between elliptical complements and adjuncts is hard to establish empirically for certain individual examples. However, we will see later on in this paper why the indeterminacy of some particular examples is in fact just what the dual analysis view predicts. Since the audience for this paper includes readers unfamiliar with recent versions of Categorial Grammar (or type-logical syntax, as these are called), my presentation here is deliberately kept informal and simple, e.g. phrase structure trees will often be used in place of natural deduction or Gentzen sequent derivations. For a systematic and detailed introduction to Categorial Grammar, see Carpenter (1998) or Morrill (1994). However, readers with more extensive knowledge of type-logical syntax should keep in mind that everything presented here is intended to be formulatable more precisely. To treat the problems in this paper, the Associative Lambek Calculus (L) will suffice, and all theorems of L will therefore hold. For a larger fragment of English, one should choose a multi-modal system, to be able to treat both hierarchical and "flat" natural language constituents correctly, and to include both wrapping (Dowty 1996) and occasional free word order. The syntactic features mentioned below can be treated (conservatively) by introducing them only on the primitive types; the result is that the number of primitive types is large but still finite, and since no new provision is needed for features in the logical rules (SlashElimination and Introduction), the logic of L remains intact. See Carpenter (1998) for an exact account of the so-called Curry-Howard Isomorphism. This characterization of "Head" has been criticized because it appears that certain heads would incorrectly be classed as adjuncts, even when morphological features are taken into account. For example, in (i) John can help wash the car the form of the verb help is determined by its head (can), i.e. must bear the inflectional feature [BASE]. In turn, help itself governs a complement VP of
The dual analysis of adjuncts/complements in Categorial Grammar
5.
61
form [BASE], so its fully specified category is "VP[basc]/VP(base|". (In other Germanic languages this situation arises more frequently than in English.) The first step to understanding this problem is distinguishing agreement from government in CG. This can be done by (i) incorporating features into category names (though only on the primitive categories, so the logic of L remains fully intact), and (ii) assigning category membership of many words to category schemata, not just to fully specified categories. An item that should AGREE in all features with its VP 'head' (i.e., an adjunct modifying this VP) will be assigned to category VPa\VPa, where a stands for a range of features (finiteness, number, inflectional class, etc.) with feature values specified for these, as long as the corresponding feature values are the same in both occurrences of a. A word that GOVERNS a morphological category has a fixed feature specification on its argument category, e.g. VP/NP[ACC) for transitive verbs. The verb help agrees with features of its 'head' (is helping wash dishes, has helped wash dishes, etc.) but governs the feature [BASE] on its complement, so it would belong to category VPa/VP[BASEj· Actually, we do not really need to refer to the notation VP a \VP a to define agreement in adjuncts: since features play a role in distinguishing one category from another, these cases are just an instance of A\A. Similarly, VPa/VP[BASE] is not really an instance of A/A but a kind of A/B. Once we have such morphologically-specified categories and a treatment of category schemata like VP a \VP a in the Lexicon, then there is no real purpose for the notion of "head" to play in the syntax itself. Even if the particular inflectional form of help in this example is VPbase/VPbase, all the morphosyntactic and grammatical properties are still correctly described in this and related sentences with help. If for convenience we want to speak of "heads" in the traditional sense, we can do so via implicit reference to the lexeme a word belongs to in the Lexicon, not the particular inflectional form that realizes the lexeme in a given sentence. A further caveat is needed here. While it is clear from the tree above why a complement Β should appear closer to the head than the adjunct A\A does, what about the case of a phrase [[Α/Β (A/B)\(A/B)] Β] in which there is an "inner" adjunct that would seem to have to appear closer to the head than the complement Β? We can fix this problem by relativizing the characterization of adjunct (and complement) to the particular configuration in which it appears: (i) A\4 is an adjunct in a phrase A iff A is an instance of the configuration [A A\A]; we may also say that A\A is a adjunct to A or, if A is phrasal, an adjunct to the head of A. (ii) ß is a complement in phrase Λ iff Λ is an instance of [Α/Β Β] (or an instance of [A/B/C C 5], etc.) The generalization about closeness to the head should apply only to elements that are "in" the same phrase: in [[Α/Β (A/B)\(A/B)] Β], Β is a complement in the phrase A, while (A/B)\(A/B) is an adjunct only in the phrase A/B.
62 6.
David. Dowty
To be sure, He always treated me can in fact be a grammatical string, but in the grammatical reading of it, treat does not have the same meaning as it does in treat me fairly, and the same is true for behave (badly). 7. Do not confuse this with theories in which all heads subcategorize for their adjuncts, as in some versions of HPSG. 8. Finnish, a non-Indo-European language, would at first seem devoid of this kind of connection, since it has three complete sets of four kinds of cases (which replace prepositions in that language): one set relating to enclosures (for "out of', "in", and "into"), a parallel set for surfaces ("off o f ' , "on", "onto"), a third set relating to proximate location ("at/near to", "away from", etc.) and a fourth, (morphologically distinct) set for abstract, non-locative transitions (non-locative Source, State, and Goal). However, it turns out that this last set is historically derived from an older set of case markers which signified locative transitions, before the first three sets came into the language. So even Finnish, through its history, reveals the same deeper connection between Local and corresponding Non-Local case marking as seen elsewhere. 9. Although this structure is actually perfectly adequate for both the semantics and syntax of complement reanalysis, there is no reason why it could not be further simplified, if desired, to replace the adjunct category VPWP with a simple, non-adjunct category like PP - i.e. speak2 would also belong to VP/PPfro] here, and to to ΡΡ[τοί/ΝΡ, with to translating as the identity function, insofar as its adjunct meaning is otiose. 10. This claim about expressive advantages of complements actually only follows if we make some further (but plausible) formal limitation on adjunct meanings beyond that which is implicit in the standard semantic interpretation of CG categories A\A. Logicians and linguistic semanticists have traditionally treated most adjectives and adverbs as one-place predicates (i.e., a Republican senator is simply anyone who is both a Republican and a senator), hence the compositional semantic rule for Adj-N plus N, or for VP-Adv plus V, must yield the intersection of two predicate denotations. To be sure, Montague (1974) chose the type vp\vp (etc.) precisely because it allowed for intensional as well as setintersective modifiers to be accommodated, and Parsons (1980) showed exactly how the semantics of the intersective cases would work out. Nevertheless, the category VPKVP itself inherently allows both kinds of meanings, so a further limitation needs to be imposed: specifically, all modifiers must initially have the semantics of intersective modifiers. See Dowty (1997) for details. On a related topic, Kasper (1997) shows that there is apparently a fundamental problem with employing modifiers only in categories A\A that arises in instances of recursive modification (e.g. an unbelievably expensive hotel)', if unsolvable, this problem could put CG's whole compositional approach to modifiers in question. Fortunately, Whitman (2001) has devised a fairly simple way to circumvent the problem within standard CG, which has parallels to the way Kasper proposes to treat it within HPSG. 11. Mike Calcagno (p.c.) has observed the paradigm below, which shows in more detail that this restriction cannot be an artifact of the particular choice of ad-
The dual analysis of adjuncts/complements
12.
13.
14.
15.
in Categorial Grammar
63
verb; rather treat (in this sense) requires an adverb complement on its right, and a pre-verbal adverb cannot satisfy this subcategorization requirement: (i) a. They treated him harshly. They treated him cruelly. b. They harshly treated him cruelly. They cruelly treated him harshly. c. *They harshly treated him. *They cruelly treated him. It should be noted that Dowty (1979) actually proposed two analyses of this adverb problem; in addition to the complement/adjunct analysis (pp. 260-264), another analysis was entertained (pp. 264-269) that attributes the ambiguity to the category of the adverb; subsequent examination showed that the complement/adjunct analysis is the more viable one. When examining all examples of adjuncts and complements to transitive verbs, it is important to keep in mind that I am assuming a WRAPPING analysis of direct objects (cf. Bach): thus what I call a complement (or adjunct) to a transitive verb will never appear immediately adjacent to the transitive verb, but rather after the direct object. Thus, the combination of persuade with its complement to leave form a DISCONTINUOUS CONSTITUENT in persuade Mary to leave. Barker (1998) argues that the "double genitive" is actually a partitive reading {a book of Mary's = "a book of Mary's books"); if so, this is not an extrinsic reading but nonetheless still not an intrinsic reading either, but my general point still holds that genitive complements to relational nouns are syntactically distinct from other post-nominal genitives. See also Partee and Borschev (1998, this volume). In their book On the Definition of Word, Di Sciullo and Williams (1987) introduce the term lísteme for linguistic units that are thought to be "listed individually" (as opposed to generated "on-line"): their listemes include all root morphemes, most derived words, certain syntactic phrases (idioms and, probably, collocations) and a few sentences. Although this term does seem to draw the same distinction I am making here, Di Sciullo and Williams go on to deny that their 'listemes' have any relevance to linguistics at all, much less do they even raise the possibility of dual analyses for any one form, morphological or syntactic. Hence, I will not adopt their term lísteme here.
References Anderson, John M. 1971 The Grammar of Case: Towards a Localistic Theory. (Cambridge Studies in Linguistics 4.) Cambridge: Cambridge University Press. Barker, Chris 1991 Possessive descriptions. Dissertation, University of California, Santa Cruz.
64
David Dowty 1995 1998
Possessive Descriptions. Stanford: CSLI Publications. Partitives, double genitives, and anti-uniqueness. Natural and Linguistic Theory 16: 679-717.
Language
Bresnan, Joan 1982 Polyadicity. In The Mental Representation of Grammatical Relations, 149-172. Cambridge, MA: MIT Press. Carpenter, Bob 1998 Type-Logical Semantics. Bradford Books. MIT Press. Chomsky, Noam 1970 Remarks on nominalization. In Readings in English Transformational Grammar, Roderick Jacobs and Peter Rosenbaum (eds.), 184-221. Waltham, MA: Ginn and Company. Clark, Eve V., and Katherine L. Carpenter 1989 The notion of source in language acquisition. Language 65(1): 1-30. Cresswell, M. J. 1985 Adverbs and Events. (Studies in Linguistics and Philosophy 28.) Dordrecht: Kluwer. Di Sciullo, Anna-Maria, and Edwin Williams 1987 On the Definition of Word. (Linguistic Inquiry Monographs 14.) Boston, Massachusetts: MIT Press. Downing, Pamela 1977 On the creation and use of English compound nouns. Language 53: 810-842. Dowty, David 1979 Word Meaning and Montague Grammar. (Studies in Linguistics and Philosophy 7.) Dordrecht: Reidel. 1982 Grammatical relations and Montague grammar. In The Nature of Syntactic Representation, Pauline Jacobson and Geoffrey Pullum (eds.), 79-130. Dordrecht: Reidel. 1989 On the semantic content of the notion "thematic role". In Properties, Types and Meanings. Vol. II. Gennaro Chierchia, Barbara Partee, and Ray Turner (eds.), 69-130. Dordrecht: Kluwer. 1993 Adjunct-to-argument reanalysis in a dynamic theory of grammar. Unpublished paper, Ohio State University. Various versions presented as talks at the University of Rochester, University of Illinois, inter alia. 1996 Non-constituent coordination, wrapping, and multimodal categorial grammar. In Structures and Norms in Science, M. L. Dalla Charia et al. (eds.), 347-368. (Proceedings of the 1995 International Congress of Logic, Methodology, and Philosophy of Science, Florence.) 1997 Adjunct-to-argument reanalysis in a dynamic theory of grammar: The problem of prepositions. Paper presented at the Blaubeuren Semantics Conference, University of Tübingen, March 2, 1997. 2001 Adjuncts and complements in Categorial Grammar. (In preparation.)
The dual analysis of adjuncts/complements in Categorial Grammar
65
Fabricius-Hansen, Cathrine 1983 Wieder ein wieder? Zur Semantik von wieder. In Meaning, Use and Interpretation of Language, C. Schwarze, R. Bäuerle, and A. von Stechow (eds.), 97-120. Berlin: Walter de Gruyter. Fillmore, Charles 1968 The case for case. In Universals in Linguistic Theory, Emmon Bach and Robert T. Harms (eds.), 1-90. New York: Holt, Rinehart and Winston. Gruber, Jeffrey S. 1965 Studies in lexical relations. Cambridge, MA: MIT dissertation. Published in revised form in Gruber 1976. Hankamer, Jorge 1977 Multiple analyses. In Mechanisms of Syntactic Change, Charles N. Li (ed.), 583-607. Austin: University of Texas Press. Harris, Alice, and Lyle Campbell 1995 Historical Syntax in Cross-linguistic Perspective. (Cambridge Studies in Linguistics 74.) Cambridge: Cambridge University Press. Jackendoff, Ray S. 1972 Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press. Karttunen, Lauri 1989 Radical lexicalism. In Alternative Conceptions of Phrase Structure, Mark Baltin and Anthony Kroch (eds.), 43-65. The University of Chicago Press. Kasper, Robert 1997 The semantics of recursive modification. Unpublished paper, Ohio State University, available from ftp://ling.ohio-state.edu/pub/kasper/jling-rev.dvi.ps. Keenan, Edward L. 1985 Passive in languages of the world. In Syntactic Typology and Linguistic Field Work, Timothy Shopen (ed.), Cambridge University Press. Kroch, Anthony S. 1989 Reflexes of grammar in patterns of language change. Language Variation and Change 1: 199-244. Ladusaw, William, and David Dowty 1988 Toward a nongrammatical account of thematic roles. In Thematic Relations. Vol. 21. Wendy Wilkins (ed.), 62-74. Academic Press. Lees, Robert B. 1960 The Grammar of English Nominalization. The Hague: Mouton. Levi, Judith N. 1975 The syntax and semantics of non-predicating adjectives in English. Dissertation, University of Chicago. Montague, Richard 1974 Formal Philosophy. Yale University Press.
66
David Dowty
Morrill, Glyn 1994 Type Logical Grammar. Kluwer. Parsons, Terence 1980 Modifiers and quantifiers in natural language. Canadian Journal of Philosophy Supplementary 6: 29-60. 1990 Events in the Semantics of English: A Study in Sub-Atomic Semantics. (Current Studies in Linguistics Series 21.) MIT Press. Partee, Barbara 1997 Uniformity vs. versatility: the genitive, a case study. In Handbook of Logic and Language, J. van Benthem and A. ter Meulen (eds.), 464470. Amsterdam: Elsevier. (Originally submitted in 1983.) Partee, Barbara, and Vladimir Borschev 1998 Integrating lexical and formal semantics: Genitives, relational nouns, and type-shifting. In Proceedings of the Second Tbilisi Symposium on Language, Robin Cooper and Thomas Gamkrelidz (eds.), 229-241. Tbilisi: Center on Language, Logic, Speech, Tbilisi State University. 2003 Genitives, relational nouns, and argument-modifier ambiguity. In this volume. Stechow, Arnim von 1996 The different readings of wieder 'again' - a structural account. Journal of Semantics 13: 87-138. Thomason, Richmond 1974 Some complement constructions in Montague grammar. In Proceedings of the Tenth Annual Meeting of the Chicago Linguistics Society, 712722. Chicago Linguistics Society. Venneman, Theo, and Ray Harlow 1977 Categorial grammar and consistent vx-serialization. Theoretical Linguistics 4: 227-254. Whitman, Philip Neal 2002 Category neutrality: A type-logical investigation. Ph.D. dissertation, Ohio State University. (http://www.ohiolink.edu/etd/view.cgi7osul023679306)
Genitives, relational nouns, and argument-modifier ambiguity1 Barbara H. Partee and Vladimir Borschev
Abstract The argument-modifier distinction is less clear in NPs than in VPs since nouns do not typically take arguments. The clearest cases of arguments in NPs are found in certain kinds of nominalizations which retain some "verbal" properties (Grimshaw 1990). The status of apparent arguments of non-deverbal relational nouns like sister is more controversial. Genitive constructions like John's teacher, team of John's offer a challenging testing ground for the argument-modifier distinction in NPs, both in English and cross-linguistically. In the analyses of Partee (1983/1997) and Barker (1995), the DP in a genitive phrase (i.e. John in John's) is always an argument of some relation, but the relation does not always come from the head noun. In those split approaches, some genitives are arguments and some are modifiers. By contrast, recent proposals by Jensen and Vikner and by Borschev and Partee analyze all genitives as arguments, a conclusion we no longer support. In this paper, we explore a range of possible approaches: argument-only, modifier-only, and split approaches, and we consider the kinds of semantic evidence that imply that different approaches are correct for different genitive or possessive constructions in different languages. For English, we argue that a split approach is correct and we offer some diagnostics for distinguishing arguments from modifiers.
1.
The argument-modifier distinction in NPs
The argument-modifier distinction is less clear in NPs than in VPs since nouns do not typically take arguments. The clearest cases of arguments in NPs are found in some nominalizations (Grimshaw 1990). Non-deverbal relational nouns like sister, mayor, enemy, picture, edge, height in some sense also seem to take arguments. C. L. Baker (1978) proposed a test using English one anaphora whereby one substitutes for N-bar, which obligatorily includes all of a noun's arguments. By that test, to Oslo in (la) is a modifier, while of Boston in (lb) is an argument. But neither this, nor any other known test, has seemed conclusive and the question of whether and in what sense "true nouns" take arguments remains controversial.
68 (1)
Barbara H. Partee and Vladimir Borschev a. The train to Oslo takes longer than the one to Stockholm. b. *The mayor of Boston has more power than the one of Baltimore.
Genitive2 constructions like those in (2a, b) offer an interesting test-bed for the argument-modifier distinction in NPs, in English and Russian and, indeed, cross-linguistically. (2)
a. Engl.: John's teacher, John's chair, friend of John's b. Russ.: Masin ucitel, Masin stul, drug Masi Masa-poss-M.sG. teacher, Masa-Poss-M.sG chair, friend Masa-GEN 'Masa's teacher', 'Masa's chair', 'Masa's friend'
Many, perhaps all, genitives seem to have some properties of arguments and some of modifiers, yet some seem more like arguments and some more like modifiers. Recent proposals by Jensen and Vikner (J994), Vikner and Jensen (2002), Partee and Borschev (1998), Borschev and Partee (1999a, b) analyze all genitives3 as arguments, a conclusion we are no longer sure of for English (see Partee and Borschev 2001). While we now doubt that such an analysis is correct for all kinds of genitives in all languages, we do believe that it is correct for some kinds of genitives in some languages. It is not easy to settle the question of whether there is a substantive difference between these two roles of genitives and it may well be the case that all or many genitives play both roles at once. In both English and Russian there are several constructions which may in some (possibly metaphorical) sense express possession; and in each language there seem to be several different kinds of meanings for constructions which may be considered genitive (genitive morphology in Russian, the morpheme '-s in English). The correlation between constructions and meanings is not transparent. Major questions about genitive constructions, then, are the following: Are all, some, or no genitives arguments of nouns, and if so, which ones (and how can we tell?), and of what kind, and at what level of analysis? Are some genitives able to get argumentai interpretations without actually being arguments in the structural sense of being syntactic complements of the noun and/or of having function-argument structure reflected typetheoretically? In this paper, we examine semantic aspects relating to the question of whether all genitives can and should be given a uniform approach, or whether we can find a satisfying way of accommodating a split approach, while remaining as neutral as possible throughout on the syntactic aspects of the question.
Genitives, relational nouns, and argument-modifier ambiguity
2.
69
Genitives and related constructions: The challenge
The terminology surrounding "possessives" and "genitives" is confusing, since the correspondences among morphological forms, syntactic positions, grammatical relations, and semantic interpretations are complex and subject to debate. Further, there is much variation cross-linguistically. For clarification, let us distinguish at least the following:4 (3)
a. Possessive pronouns: E. my, his\ R. moj 'my', ego 'his'; E. predicative forms mine, his; postnominal forms of mine, of his. b. English "Saxon genitives": John's·, the postnominal Saxon genitive of John's. c. English PP with of+ NPAcc· d. Russian postnominal genitive NP: Mendeleeva 'of Mendeleev', tigra 'of a/the tiger'. e. Russian prenominai possessive: Masin dom 'Masha's house'.
Some problems of the semantics of genitives affect all of the constructions listed in (3), while some problems require more fine-grained distinctions to be made. Very similar problems arise in corresponding constructions in many other languages, and related problems arise with the English verb have and its lexical and constructional counterparts in other languages (Bach 1967; Freeze 1992; Landman and Partee 1984; Szabolcsi 1994; Jensen and Vikner 1996; Partee 1999b). The present work concerns the possible need for a distinction between genitives as modifiers and genitives as arguments, and the role that predicate possessives may play in resolving that issue. We leave out of discussion the clear modifier genitives that occur in compounds like a boys' club, although Munn (1995) has shown that the line between those and other genitives is not as sharp as had been thought. Our starting point is the following data from Partee (1983/1997: 464): (4)
a. John's team b. a team of John's c. That team is John's.
(5)
a. John's brother b. a brother of John's c. (#) That brother is John's.
(6)
a. John's favorite movie b. a favorite movie of John's c. (#) That favorite movie is John's.
70
Barbara H. Partee and Vladimir Borschev
Informally, a unified interpretation of the genitive phrase "John 's", applicable to all cases in (4)-(6), is that the genitive phrase always expresses one argument of a relation, for which we will use the descriptive term "genitive relation", following Jensen and Vikner (1994). However, the relation can come from any of three sources: (i) the context, as in (4) ("plays for", "owns", "is a fan o f , etc.); this happens when the noun is a plain one-place predicate; (ii) an inherently relational noun like brother, (iii) an inherently relational adjective like favorite. Following Partee (1983/1997), we initially refer to case (i) as the "free R" reading, and to cases (ii) and (iii) as "inherent R" readings. (In later parts of the paper, the classification will be revised.) The puzzles include these: -
Can (and should) examples (4a) and (5a) be given a uniform approach ? If so, how? - Or does the genitive construction combine differently with plain and relational nouns ? And if so, are these differences predictable from some general principles? - Should the first case be split into two distinct cases, one being a default preference of the genitive construction itself for a genitive relation in the family of "owns", "possesses", "controls", possibly with a distinct syntactic source from the context-dependent "free R" readings? - Does the analysis of genitives require that phrasal as well as lexical categories be able to take complements? The examples in (6) show that argument genitives cannot always simply be analyzed as complements of a lexical noun, since it is the whole N-bar favorite movie that provides the relation of which John is an argument.5 The Russian genitive constructions exemplified in (7) present similar challenges, showing a similar range of genitive relations, with a similar range of relational and plain nouns, although there are also differences between English and Russian to account for. (7)
a. ljubitel'
kosek
Íover-NOM.SG Cat-GEN.PL
'lover of cats, cat-lover' b. rost celoveka height-NOM.sG man-GEN.SG
'height of the/a man' c. nozka stola leg-NOM.sG table-GEN.sG
'leg of the table, table leg'
Genitives,
d. krug
relational
nouns, and argument-modifier
ambiguity
71
syra
circle-NOM.sG cheese-GEN.SG
'circle (wheel) of cheese' e. stakan moloka glaSS-NOM.SG m i l k - G E N . S G
'glass of milk' f. portret Peti portrait-NOM.sG Petja-GEN
'picture of Petja' g. sled tigra track-NOM.SG tiger-GEN.sG
'track of the/a tiger' h. sobaka doceri dog-NOM.SG daughter-GEN.sG
'the daughter's dog' i. nebo Andreja Bolkonskogo sky-NOM.sG Andrej-GEN Bolkonsky-GEN
'Andrej Bolkonsky's sky' For Russian, the question of whether the examples in (7) all instantiate a single construction is also difficult and is not identical to the corresponding question for English since there is a many-to-many correspondence between the Russian and the English constructions. The uses of the Russian genitive NP cover uses analogous to the English Saxon genitive, to English of + NPAcc, and, in some cases, to English noun-noun compounds. English Saxon genitives may translate to Russian as genitives, as prenominai possessives, or as denominal adjectives. At a descriptive level, virtually all authors who have grappled with the semantics of genitive constructions are in agreement that in some cases the genitive NP seems like an argument and in other cases it seems like a modifier. The argument status of at least some genitives is clearest in the case of certain deverbal nouns, those called "Complex Event Nomináis" by Grimshaw (1990) and Schoorlemmer (1995), "Derived Nomináis" by Babby (1997), and "Process Nomináis" by Rappaport (1998). To clarify our relatively neutral, assumed syntax for the first of these constructions, and for Russian postnominal genitives, we give the syntactic structure in (8) below, a linearized form of the schematic phrase structure tree of Borschev and Partee (1999b): (8)
[N Ν N P g e n ], where Ν is a cover term for N° and non-maximal Nbar (= CN and CNP in Montague (1973)), and NP is a cover term for both N m a x and DP.
72
Barbara H. Partee and Vladimir Borschev
The semantic question is: Do the genitive constructions [N Ν NPGEN ] have a uniform compositional interpretation?
3.
Uniform approaches and split approaches
As we will illustrate in Section 4, given the possibilities that have been raised by work on type-shifting in the past decade or so, it seems that the semantics of any simple "NP's N" or "N NPGEN" construction could be given either an analysis in which the genitive NP is an argument or one in which it is a modifier. In this paper we are not trying to settle all the relevant arguments for even one such construction. Rather, we wish to explore the available alternatives from a semantic point of view. A full analysis of any genitive construction in any language requires greater syntactic specificity than we are providing here, as well as a theory of the interaction among lexical, structural, and contextual factors. Moreover, relevant evidence may be of many kinds, including binding and extraction facts, behavior in coordinate constructions, iterability, word order constraints, and quantificational properties. There are, by now, many proposals for many such constructions in different languages in the literature, in a variety of theoretical frameworks, and we will not enter into the sometimes crucial syntactic debates that are involved in some of the competing approaches. However, with little more than the minimal syntactic assumptions noted above, we can address some of the central issues of semantics and compositionality. To illustrate our concerns with a concrete example, let us discuss approaches to the semantics of the English genitive construction illustrated by the phrase book of John's.6 There are in principle three possibilities: a split approach and two kinds of uniform approach. (i) One possibility is to split the construction into two different genitive constructions, treating "inherent R" genitives (brother of John 's) as typeraised arguments and "free R" genitives (team of John's) as (intersective) modifiers (Partee 1983/1997; Barker 1995). This approach starts from the intuition that some genitives are arguments and some are modifiers, as will be illustrated in Section 4 below. If no uniform approach can be made to work (for a given genitive construction in a given language), a split approach may be necessary. One of our main points here will be, however, that raw intuitions of ambiguity or of argumenthood vs. modifierhood do not constitute real evidence.7 Most linguists would tend to prefer a uniform analysis if it can be made to work but, as Dowty (1997, this volume) argues, that is not an uncontroversial position. In the subsequent sections of this
Genitives, relational nouns, and argument-modifier
ambiguity
73
paper, we explore empirical arguments for and against the ambiguity of various genitive constructions. (ii) One possibility of a uniform approach is to assimilate all cases to the "inherent R" reading, treating all genitives as arguments, or as type-lifted arguments. This option was introduced by Jensen and Vikner (1994), and further explored in Partee and Borschev (1998), Borschev and Partee (1999a, b), and Vikner and Jensen (2002). We describe this approach in Section 4 below, and show some empirical advantages of this approach over a split approach. In Section 5, we review arguments from Partee and Borschev (2001) to the effect that, in spite of these attractions, this uniform approach is not correct for all genitive constructions in all languages, although it may well be correct for some. These conclusions open up interesting typological questions and invite the task of finding more kinds of evidence for true arguments of nouns. (iii) Another possibility of a uniform approach is to assimilate all cases to the "free R" reading. A variant of that option was proposed by Hellan (1980). Partee (1983/1997) argued against it on the basis of the contrast among the (c) examples in (4-6), but we return to it in Section 6. On this kind of analysis, all genitives are modifiers. Within approaches to modifier genitives, recent work by Kolliakou (1999) shows the need for a further distinction between genitives as predicates of type , i.e. as intersective modifiers, and genitives as non-intersective intensional modifiers of type « e , t > , >. As we discuss in Section 6, challenges to treating all genitives as modifiers include the obligatoriness or near-obligatoriness of a genitive complement with some relational nouns and the apparent systematicity of argument-inheritance with some kinds of de verbal nouns. For the treatment of genitives as intersective modifiers, another problem is the apparent impossibility of some genitives in predicate position, as illustrated by the contrasts in (4-6) above. New evidence from ellipsis ambiguity militating against a modifier-only approach is also introduced in Section 6. We conclude that we cannot support a modifier-only approach but we believe that more work on such a possibility would be worthwhile.
4.
Two theories of genitives
4.1. The early Partee split approach Partee (1983/1997) proposed two distinct genitive constructions with relational and non-relational nouns, the latter incorporating a free relation vari-
74
Barbara H. Partee and Vladimir Borschev
able R whose value must be supplied by context. On the other hand, (a modified version of) Jensen and Vikner (1994) offers a uniform interpretation of the genitive, with coerced type-shifting of the N-bar to a relational reading when necessary. The investigation of the differences between these two approaches, in part through an ongoing dialogue between Borschev and Partee and Jensen and Vikner over the past several years, has led us to an appreciation that the problem of the semantics of the genitive construction^) is a much richer domain of inquiry than we had originally imagined, and to convergence on some issues and new questions on others. A note concerning notation: in what follows we use CN for a ("plain") N-bar of type (one-place predicate, with only a "referential" θ-role (Williams 1981; the R role of Babby 1997), and TCN for a ("transitive" or "relational") N-bar of type like father, favorite movie. We sometimes use CNP and TCNP for phrasal constituents of those types. The analysis of Partee (1983/1997) posits a split in the construction, with the N-bar supplying the relation if it is relational, and with the construction supplying a "free relation variable" if the N-bar is not relational. We illustrate the postnominal genitive, as in (4b), (5b), (6b), which Partee (1983/1997) analyzed as more basic than the prenominai genitive, treating the prenominai genitive in (4a), (5a), (6a) as a composition of the postnominal genitive with an implicit definite determiner. Postnominal genitive (of John's) combines with CN or TCN to make a CN. When a genitive NP combines with a plain CN, type : the construction provides a "free R", a variable of type which we write as
(9)
of John's·. team of John's:
λΡλχ[Ρ(χ) & Ä,(John)(jc)] Xx[team(x) & /?,(John)(jt)]
When a genitive NP combines with a TCN, type , the TCN provides its "inherent R". (10) of John's·. AJ?[Xjt[fl(John)C*:)]] or equivalently, Àfl[fl(John)] teacher of John's·. Xjt[teac/ier(John)(;t)]] Compositionally, these are derived as follows. For the modifier genitive that combines with a plain CN, as in (9), the basic type of its genitive morpheme 's is , as in (11a). It combines first with the e-type possessor9 John to form the one-place predicate genitive John's as in (lib); this is the form that occurs predicatively in (4a). We assume that of is a semantically empty element inserted for syntactic reasons in postnominal genitives, and that the
Genitives, relational nouns, and argument-modifier
ambiguity
75
postnominal modifier genitive of John's in (11c) is derived from the predicative form by a simple type-shift, analogous to the way a predicative adjective red of type may be lifted to become an adnominal modifier red of type « e , t > , < e , t » (see Partee 1995). The modifier genitive then combines with a plain noun such as team as shown in (9) above. (11) a. 's: b. John's:
XyXx[Rj(y)(x)] (predicative) À^Àx[^,(y)(jc)](John) = AJC John) (x) ] (predicate) c. of John's: λΡλχ[Ρ(χ) & /J,(John)(x)] (postnominal modifier CN/CN)
The "inherent R", or argument genitive, is built from the homophonous genitive morpheme 's, shown in (12a) of type < e , « e , < e , t » , < e , t » > . It combines with the e-type possessor John to give a "detransitivizing modifier", a function from type to type , i.e. a type-lifted argument, as shown in (12b). When this argumentai genitive combines with a relational noun like teacher, the result is as shown in (10), making John the first argument of the noun. (We again assume that of is semantically empty and is purely syntactic in motivation.) (12) a. '5: b. of John's·.
λ>'λ/?[λτ[/?0')(χ)]] or equivalently XyXÄ[/?(y)] (argumentai) ÀyÀ/?[Àx[/?(y)(:t)]](John) = XR[hc[R(Joho)(x)]] or XR[R(John)]
4.2. The Jensen and Vikner uniform approach with coercion Jensen and Vikner (1994) propose that an analysis which incorporates coerced type-shifting in the sense of Partee (1987) should be able to do without two separate rules for the genitive. On their alternative analysis, which builds on the framework of Pustejovsky (1993, 1995), the genitive must always combine with a relational common noun (phrase), coercing a oneplace predicate noun to a two-place relational meaning ("team" to an appropriate sense of "team-of ')· Their analysis corresponds to the "inherent R" case of Partee (1983/1997) and, with a relational noun like teacher, the two analyses agree. The difference arises with a plain one-place CN like chair or team which, on their analysis, is coerced to a TCN interpretation. Jensen and Vikner follow Pustejovsky in appealing to the qualia structure of the lexical entry to guide the coercion, so that, for instance, the telic role of
76
Barbara H. Partee and Vladimir Borschev
chair ("chairs are to sit in") licenses the shift of CN chair to TCN chair illustrated below. (13)
CN chair. TCN chair:
λx[chair(x)] Xykx[chair(x) & sits-in(x)(y)]
Initially, we had some important disagreements with Jensen and Vikner concerning the degree to which lexical meaning drives coercion. In Vikner and Jensen (2002) and Partee and Borschev (1998), there is agreement that, on the most general version of their approach, the genitive construction should always demand a TCN to combine with, and if it finds instead a CN it will coerce it by whatever means are available and natural, sometimes lexical, sometimes pragmatic. (We make a less sharp distinction between lexically and contextually supplied shifted meanings than Jensen and Vikner do, because of the outlook on the integration of information from lexical and other sources described in Partee and Borschev (1998), Borschev and Partee (1998).) A "pragmatic" coercion is seen as shifting the noun to a relational reading that incorporates the free relation variable of Partee (1983/1997) into the shifted noun meaning.10 (14)
TCN
team:
Xykx[team(x) & Ri(x)(y)]
As in Partee's analysis, a felicitous use of an expression with a free variable requires that the context make salient a particular choice of value for the variable. Partee and Borschev (1998, 2000a) and Borschev and Partee (1999a) propose extensions to Jensen and Vikner's coercion approach to cover also the "contextual" cases. We also pointed to a need for more finegrained coercion principles to cover phenomena involving the relational adjective favorite and the difference in preferred relation in the interpretation of John's movie and John's favorite movie.11
4.3. Comparison of the two approaches One main difference between the two approaches concerns where a "free relation variable" is added in a case in which context is driving a pragmatically based coercion. Let us suppose that team of Mary's is such a case. (15) Jensen and Vikner (1994): of Mary's: (shifted) team: team of Mary's:
XR[hc[R( Mary)(x)]] Xy[Ajt[team(x) & /?,(y)(x)]] Xx[team(*) & i?,(Mary)(x)]
Genitives, relational nouns, and argument-modifier
ambiguity
77
(16) Partee (1983/1997): of Mary 's: λΡλχ[Ρ(χ) & Ä,(Mary)Ct)] (non-shifted) team: team team of Mary's: tar[team(x) & /?,(Mary)(x)] The final result is the same but for Jensen and Vikner the free relation variable comes in as part of the meaning of the shifted noun, while for Partee (1983/1997) it comes in as part of the meaning of the genitive construction itself. Does this difference in where the free relation variable is situated ever make a detectable difference? It does. Partee and Borschev (1998) give an empirical argument in favor of Jensen and Vikner's approach, based on an analysis of the example Mary's former mansion, suggested to us by Norvin Richards (p.c.). The argument rests on the four assumptions spelled out in (17). (17)
Assumptions: (i)
mansion is lexically a one-place noun.
(ii)
former is an endocentric modifier, lexically a CN/CN, shiftable to a TCN/TCN. former as CN/CN: former monastery, former dancer, former as TCN/TCN: former owner, former friend. (iii) The "free relation" variable in this case has as one of its most salient values something like "owns" or "lives in". (iv) Mary's former mansion has two readings: Reading A: "a former mansion (perhaps now just a ruin) that is (now) Mary's". Reading B: "something that was formerly Mary's mansion; it may still be a mansion, but it's no longer Mary's". On the Partee (1983/1997) account, there is no motivation for any typeshifting to occur and the free relation "owns" will be introduced with the genitive Mary's, after former has combined with mansion. This means that the free relation ("owns") in the interpretation of the genitive Mary's will never be under the scope oí former. As a result, Partee (1983/1997) can derive Reading A above, but not Reading B. The tree (18) shows the compositional structure of Mary's former mansion on the account of Partee (1983/1997).
78
Barbara H. Partee and Vladimir Borschev
NP
(18) NP's Mary's
CNP CN/CN
CN
former
mansion
free R introduced here Jensen and Vikner's account, with coercion of CN to TCN, does provide derivations for both readings, which Partee's account cannot. For Jensen and Vikner, Mary's coerces former mansion to a relational TCN. Given our assumptions, there are two ways that former mansion could shift to a TCN: (i) Initially leave mansion as a CN, treat former as CN/CN, combine them to form a CNP, as on the Partee account; then shift that CNP to a TCNP, bringing in the free variable at that stage to get the shifted meaning of former mansion, shown below in (19): (19)
'ky[λ*[former(mansion)(*)
& R¡(y)(x)]] [R¡: "is owned by"]
This corresponds to Reading A above, with the free R introduced at the point where the CNP shifts to become a TCNP. The compositional structure would be almost identical to that in tree (18), differing only in where the free R is introduced. (ii) Or shift mansion to a TCN, and former to a TCN/TCN, combine them to form a TCNP as shown below in (20): (20) Àj[Aj;[former(mansion-of)(x)(y)]], where mansion-of is short for Ày[ÂJc[mansion(x) & Ä,(y)(*)]] This corresponds to Reading Β above, with the compositional structure as in (21) below.
Genitives, relational nouns, and argument-modifier
(21)
ambiguity
79
NP NP's Mary's
TCNP TCN/TCN
TCN
CN/CN
CN
former
mansi free R introduced here
We assume that both of these ways of coercing the phrase former mansion are structurally available; different choices of lexical items or different contexts may favor one over the other, but since both are consistent with all the principles that we are aware of, the Jensen and Vikner approach successfully predicts the ambiguity and therefore has a clear empirical advantage over the Partee (1983/1997) approach.12
5.
Problems for the uniform "argument-only" approach
In spite of the theoretical appeal of the uniform approach and its ability to solve the problem of Mary's former mansion, we are still not convinced that it is correct for English. Interestingly, the arguments against a uniform analysis for English genitives do not apply to Russian genitives. Russian seems to show a clearer split between a genitive construction which does indeed seem to be uniformly argumentai13 and a prenominai possessive which is a modifier (but perhaps also ambiguous). One of our main worries, discussed in Partee and Borschev (2001), concerns predicate genitives and our earlier observation that predicate genitives seem to favor "free R" interpretations, together with the fact that predicate genitives are not in a structural argument position unless one posits an empty head noun accompanying them. As we examine predicate genitives and contrast them with the better candidates for argument genitives, it will emerge that the semantics of the clearest cases of predicate genitives seems to center on the notion of possession. Thus, the key distinction may not, after all, be "free R" vs. "inherent R" but, rather, a distinction between "possessive" modifiers and genitive arguments. We address this issue further in Sections 6 and 7.
80
Barbara H. Partee and Vladimir Borschev
5.1. Predicate genitives: A problem for the uniform approach? If some genitives can occur as basic predicates, that would suggest that when those same genitives occur inside the NP, they are basically modifiers, and not arguments, returning us to a version of the distinction posited in the earlier Partee (1983/1997) approach. If there are no genitives that demand a treatment as basic type predicates, that would be an argument in favor of treating all genitives occurring inside an NP within the uniform approach of Jensen and Vikner. If we find, on the contrary, that in some languages there are systematic differences in form and/or interpretation between certain genitives that occur only NP-internally and others that occur both predicatively and NPinternally, that would present a serious challenge to the uniform approach, at least for those languages. The issue is, however, empirically complex for at least two reasons: (i) there may be independent reasons (syntactic or morphological) why some kinds of genitives (e.g. Russian genitives) cannot occur as predicates; (ii) some predicate genitives may be elliptical full NPs; it is not always easy to tell. Much of what follows is concerned with this problem. In the following sections, we look at evidence concerning predicate genitives in English, Russian, German, and Polish. The evidence supports the idea of two semantically different kinds of genitives, with some forms, such as English Saxon genitives, used for both. One kind of genitives are argument genitives, which fit the Jensen and Vikner analysis. These occur in constructions with a relationally interpreted noun (or with an adjective like favorite plus a noun). Argument genitives do not occur in type so, when they occur alone, they are interpreted as elliptical NPs with a relational noun implicitly understood. The Russian genitive appears to be of this type and we consider the Jensen and Vikner analysis correct for the Russian genitive construction. The other kind of genitives are true predicative genitives, basically of type , interpreted approximately as in the corresponding analysis of Partee (1983/1997), but with the "free R" preferentially interpreted as some kind of "possession" or "control". To represent the way this distinction differs from the original distinction of Partee (1983/1997), we will stop referring to the "free R" and refer instead to Rposs• When this kind of genitive occurs inside an NP, it is a modifier rather than an argument. We believe
Genitives, relational nouns, and argument-modifier
ambiguity
81
that the Russian prenominal genitive forms discussed in Section 5.1.2. are of this type. Since the English Saxon genitives, as well as possessive pronouns in all four of the languages looked at here, have both uses, we conclude that the uniform approach cannot be correct for those constructions. But we are left with a puzzle concerning the large proportion of cases which could seemingly be analyzed either way: are they all "ambiguous"? We will return to this puzzle, which remains open, in Section 7.
5.1.1. Predicate genitives in English The nature of predicate genitives is less clear in English than in some other languages. It is difficult to be sure whether an apparent predicate genitive like John's in (4c), repeated below, is a simple one-place predicate with an Rposs or "possession" reading, or is an argument genitive occurring as part of an elliptical NP, i.e. with John's implicitly in construction with another occurrence of team. (4c)
That team is John's.
But note the following, where the judgments marked concern the possibility of construing the predicate genitive as involving a relation corresponding to the noun in the subject NP. (22) a. b. c. d. e. f.
WThatfather is John's. WThat favorite movie is John's. That teacher is John's. His [pointing] father is also John's. Dad's favorite movie is also mine. IThat father is John's father.
The good examples in (22), namely (22c, d, e), all have predicate genitives that may be interpreted as elliptical NPs:14 John's teacher, John's father, my favorite movie. The bad examples (22a, b, f) all have intrinsically relational head nouns (or common noun phrase in the case of (22b)) that have to be interpreted non-relationally in the subject but relationally in the predicate, assuming that (22a, b) have elliptical predicate genitives. The head noun in the subject in examples (22a, b, f) must shift to a non-relational reading in order to be compatible with the demonstrative determiner that}5 It may be that there is a restriction (perhaps a processing restriction) on shifting an expression away from its basic meaning and then back again.
82
Barbara H. Partee and Vladimir Borschev
(The "bad" sentences are presumably not ungrammatical, but are nearly impossible with respect to the intended readings "John's father", etc.) In the good examples (22d, e) we have the relational readings of the head noun (phrases) in both the subject and the (elliptical) predicate. Example (22f), with two overt occurrences of the relational noun, the first shifted and the second not, is less bad than (22a), and is probably only awkward; it would presumably be acceptable in the kind of context suggested by example (ii) in note 15. The relevant difference between the good (22c) and the bad (22a) may be that teacher, unlikt father, is lexically supplied with equally salient and closely related relational and non-relational readings, so that one would not have to suppress the relational reading by shifting in order to interpret teacher in the subject NP non-relationally. The data above, reinforced by the Dutch data mentioned in note 14, strongly suggest that predicate genitives may sometimes be elliptical (Determiner-only) NPs. And if all bare genitives in all languages could be interpreted as elliptical NPs, then predicate genitives would not pose a problem for the uniform "argument-only" approach; the difference between possessive or genitive forms that can and that cannot occur "bare" as predicates would simply reflect constraints on NP ellipsis. We believe, however, that not all predicate genitives are elliptical. We do not have conclusive arguments for English; there are several complicating factors, including problems in the analysis of copular sentences (Williams 1983; Partee 1987; Moro 1997; Heycock and Kroch 1998, 1999; Partee 1999a). So rather than try to support our intuitions about the English examples, we turn to some languages where we have found some syntactic and/or morphological distinctions that provide evidence for a distinction between modifier genitives and argument genitives.16
5.1.2. Russian prenominai possessives vs. genitives In Russian, possessive pronouns and the normally prenominai quasiadjectival possessive forms can occur in predicate position but genitive NPs cannot.17 This suggests that Russian genitive NPs are always arguments, and that the Jensen and Vikner uniform analysis with coercion of CNs to TCNs (extended to Russian in Borschev and Partee (1999a, b)) is correct for the Russian genitive construction. It also suggests that the Russian prenominai possessive forms, and possessive pronouns (see 5.1.4.), are at least sometimes modifiers.
Genitives, relational nouns, and argument-modifier
ambiguity
83
The Russian prenominal possessive construction studied by KoptjevskajaTamm and Smele ν (1994) and by Babyonyshev (1997) is illustrated in (23) and the genitive construction in (24). (23) a. Petin
stul
Petja-POSS-M.SG. chair-M.SG.
'Petja's chair' b. Mamin
portret
Mama-POSS-M.SG. portrait-M.SG.
'Mama's portrait' (24) a. stul
Peti
chair-M.SG. Petja-GEN.SG.
'Petja's chair' b. portret mamy portrait-M.SG. Mama-GEN.SG.
'Mama's portrait' In these examples, both constructions can be used in describing the same range of cases; the possible relations of Petja to the chair or of Mama to the portrait are as various as with the English prenominal genitive. But the meanings do not "feel" identical. In the possessive construction in (23), we would like to claim (as did Schoorlemmer (1995)) that the possessive Petin, mamin acts as a modifier of the head noun. We believe that the prototypical interpretation of the possessive modifier is indeed possession. To maintain such a claim, "possession" must be understood in a broad sense to apply to a diverse range of relations; see Heine (1997). Thus in example (23b), possession may be possession proper, "authorship", or the relation of "being portrayed". But the possibility of expanding the sense of "possession" is evidently not unlimited. Thus "murderer of Petja" can be expressed in Russian by (25a) but not by (25b). (25) a. ubijca
Peti
murderer-M.SG. Petja-GEN.SG.
'Petja's murderer' (murderer of Petja) b. Petin ubijca Petja-POSS-M.SG. murderer-M.SG.
#'Petja's murderer' [ok only as e.g. 'a murderer Petja has hired']
84
Barbara H. Partee and Vladimir Borschev
In the genitive construction in (24a), we analyze Peti as an argument of the relation which connects it to stul. In the given case, the most salient relation could alternatively be seen as some kind of possession as well; but possession proper is not the prototypical interpretation for the genitive construction. The range of possible relations expressed with a genitive is extremely broad (cf. Knorina 1985, 1988, 1990, 1996; Borschev and Knorina 1990; Partee and Borschev 1998; Borschev and Partee 1999a, b). While this set of data is not completely conclusive, it supports the hypothesis that the Russian genitive construction is correctly analyzed as uniformly argumentai, i.e. that Jensen and Vikner's approach to English genitives is correct for Russian genitives. Further, we believe that the Russian prenominai "adjectival" possessives are basically modifiers, with the "free" Rposs as the core of their meanings (see the analysis in (33) below). However, the high overlap in possible interpretation of the two constructions, as illustrated in (23) and (24), is a puzzle.
5.1.3. German possessive pronouns Tony Kroch (p.c.) suggested looking for languages that would give evidence from agreement behavior as to whether predicate genitives are more like simple (adjectival) predicates or more like full NPs. Sten Vikner (p.c) observed that German is a language that gives some evidence: Predicate adjectives in German do not agree with subjects, but predicate possessives do, suggesting that predicate possessives are indeed more like elliptical NPs than like simple predicates.18 (26) Diese
Bücher
These-N.PL books-N.PL
(27) Diese
Bücher
These-N.PL books-N.PL
sind alt/ *alte. are old/ *old-PL
sind meine/ *lmeìn. are mine-PL/ *mine
This would suggest that the uniform approach may be correct for German, if all apparent predicate possessives give morphological evidence of being elliptical NPs. But it was further observed by Hans Kamp (p.c.) and others that actually the non-agreeing form can sometimes be used. It is used only in "standard" German, not in colloquial German, and it has an "archaic" flavor. Most interestingly, it seems that there are semantic differences between the agreeing and the non-agreeing predicate possessive.
Genitives, relational nouns, and argument-modifier
(28) a. Diese
Bücher
sind meine.
These-N.PL books-N.PL
ambiguity
85
(can be any relation)19
are mine-PL
b. Diese Bücher sind mein, These-N.PL books-N.PL are mine
(archaic, "possession" only) (no agreement)
Further examples are given in (29) and (30). A newly naturalized citizen might say (29a), but (29b) suggests a conqueror is speaking. Any relation is possible in (30a), with the most likely possibility being the parent-child, but (30b) suggests a custody fight, i.e. a dispute about who is to be in "possession" of the children. (29) a. Das
Land
ist (jetzt) meins.
The-N.SG land-N.SG is ( n o w ) mine-N.SG
b. Das
Land
The-N.SG land-N.SG
ist jetzt mein. is n o w m i n e
(30) a. Die Kinder sind meine. The children are mine-PL b. Die Kinder sind mein. The children are mine.
(no agreement)
(no agreement)
In (28b), (29b) and (30b), the form which shows absence of agreement, in the way a predicate adjective would, is limited in its interpretation to "possession". In other words, the form in which the possessive pronoun appears to be a simple predicate of type is interpreted in terms of a relation that appears to be associated with the genitive construction itself rather than with the semantics of any governing noun. In contrast, the forms which appear to be elliptical NPs have a range of interpretations including possession but also including relations typical of argument genitives, where the relevant relation is determined principally by the noun to which the genitive supplies an argument. Typical choices for the genitive relation for the argument genitive interpretations in (28a), (29a) and (30a) are authorship, citizenship, and the parent-child relation, respectively. Of course, "possession" itself can have metaphorical extensions, so the "possession" cases do not always have to be about ownership in a literal sense. But these distinctions nevertheless provide important evidence for the idea of two distinct genitives.
86
Barbara H. Partee and Vladimir Borschev
5.1.4. Russian and Polish possessive pronouns In Russian, in the past tense, predicate nomináis may be in the Instrumental case, particularly when indicating temporary relations. Babby (1973), Siegel (1976) and others have used case and other agreement phenomena to argue that some predicative adjectives are elliptical NPs and others are simple APs. The following data may provide a basis for distinguishing between predicate possessive pronouns that are, and those that are not, elliptical NPs. (31) a. Èta
strana
byla
kogda-to moej.
That-F.NOM.SG C0untry-F.N0M.SG was-F.SG o n c e
my-F.iNSTR.SG
'That country was once mine.' ("possession" or citizenship) b. Èta strana byla kogda-to moej stranoj. That-F.NOM.SG country-F.NOM.SG was-F.SG once my-F.iNSTR.SG Country-F.INSTR.SG
'That country was once my country.' ("possession" or citizenship) (32) a. Èta
strana
byla
kogda-to moja.
That-F.NOM.SG C0untry-F.N0M.SG was-F.sG o n c e
my-F.NOM.SG
'That country was once mine.' ("possession" only) b. *Éta strana byla kogda-to moja
strana.
That-F.NOM.SG country-F.NOM.SG was-F.sG once my-F.NOM.SG country-F.NOM.SG
'That country was once my country.' A full predicate nominal is impossible in the nominative in the context of (32b), and in the same context, a nominative possessive pronoun can be interpreted only as a possessive, not as an argument genitive (even with a seemingly "free" relation). Thus the possessive in (32a) cannot reasonably be analyzed as an elliptical NP but must be a simple predicate. Further, it is this occurrence of the predicate possessive that unambiguously denotes "possession". These data are similar to the German data, supporting the idea that there is a "possessive" predicate of type instantiated at least by some possessive pronouns in German and Russian and possibly also by some predicative "NP's" forms in English, distinct from other cases of predicate possessives which are elliptical full NPs and in which the possessive may be an argument of an implicit relational noun. Wayles Browne (p.c.) suggested that we should extend our data to include Polish because, in Polish, NP - BE - NP requires Instrumental on the predicate NP, whereas in Russian the predicate NP may or may not be Instrumental. Further, in Polish, NP - BE - Adj requires Nominative on the Adjective, whereas in Russian the predicate AP may be (i) short-form Ad-
Genitives, relational nouns, and argument-modifier
ambiguity
87
jective, (ii) long-form Nominative Adjective, or (iii) long-form Instrumental Adjective. The corresponding Polish data are as follows.20 (33) a. Ten
kraj
byl
kiedys moim. once my-M.INSTR.sG 'That country was once mine.' ("possession" or citizenship)
That-M.NOM.SG country-M.NOM.sG was-M.sG
b. Ten
kraj
byl
kiedys once
That-M.NOM.SG country-M.NOM.sG was-M.sG
moim
krajem.
my-M.INSTR.SG country-M.INSTR.SG
'That country was once mine.' ("possession" or citizenship; citizenship preferred) (34) a. Ten
kraj
byl
That-M.NOM.SG country-M.NOM.sG was-M.sG
kiedys mój. once
my-M.NOM.sG
'That country was once mine.' ("possession" only) b. *Ten
kraj
byl
That-M.NOM.SG country-M.NOM.sG was-M.sG
mój
kiedys once
kraj.
my-M.NOM.SG c o u n t r y - M . S G
'That country was once my country.' (ungrammatical) c. Ten
kraj
to byl
That-M.NOM.SG country-M.NOM.sG PRT was-M.sG
mój
kiedys once
kraj.
my-M.NOM.SG c o u n t r y - M . S G
'That country was once my country.' ("possession" or citizenship) The Polish data confirm the hypothesis that when a predicate possessive pronoun allows an argument reading, it is the remnant of an elliptical NP, and when it does not, it is not. The "possession" reading, which seems to be emerging as the clearest case of a non-argumental (or modifier) reading, can occur either in a remnant of an NP or as a bare predicate. This reinforces the idea that a genitive inside an NP can be either an argument or a modifier. However, a genitive which is an predicate in a predicational construction cannot be an argument, presumably because it is not in construction with a head of which it could be the argument.
88
Barbara H. Partee and Vladimir Borschev
5.1.5. Conclusions about predicate genitives We now believe that some predicate genitives really are plain predicates, and that those have just a possession/control reading, which we take to be the semantics of the genitive, as shown in (35) below. (Having reached this conclusion, we now prefer to refer to this type as "predicate possessives".) Other predicate genitives may be elliptical NPs whose interpretation may have the full range of possibilities that would be displayed by a full NP with a prenominal genitive occurring in such a position. (Note that a full NP may itself have meanings of types e, , or , depending on both its internal makeup and the position in which it occurs, so the study of the full range of meanings of bare genitives as elliptical NPs requires further research.) (35)
[John YIpred:
X*r[ÄPoss(John)(x)]
type:
This conclusion supports the proposal that, in the case of argument genitives, the genitive relation comes principally from the relational noun, whereas in the case of the modifier genitive, whose prototypical interpretation is possession, the genitive relation comes from the genitive construction itself. The cases analyzed as "free R" in Partee (1983/1997) therefore should be split into two kinds. One kind should be assimilated to the Rposs of the "possessive" genitive, and the other treated as in Vikner and Jensen (2002) and Borschev and Partee (1999a, b), as incorporated into a coerced relational reading of the head noun.
5.2. Other problems for the uniform "argument-only" approach A second and related argument concerns acquisition. Children may acquire some kinds of genitives before they show clear mastery of relational nouns. Mine! is one of the early expressions small children learn. At this stage, it seems to mean "control" or possession, compatible with an reading, although we do not know how one could completely rule out the possibility that it is elliptical for something like My (mine) blanket! We believe that this usage pre-dates any evidence of children's understanding of relational nouns like daddy, brother as relational. We are not sure whether genitive NPs like Bobby's occur at this early stage. M. Tomasello (p.c.) suggests that it is only personal pronouns that are seen in early predicative uses. If it is indeed the case that children acquire "possessive" genitives before they acquire relational nouns with relational type , then the uni-
Genitives,
relational nouns, and argument-modifier
ambiguity
89
form argument approach would have to posit later reanalysis, while a split approach would say that that the earlier form persists and that the argument genitive is added later. We assume that accretion is easier than reanalysis, so that would be an argument in support of the split approach. Another problem for any uniform approach, either modifier-only or argument-only, comes from the complex patterns of constraints on multiple genitives found with many genitive constructions in various languages. While the data are complex and often controversial, at least some of the data suggest that the number of argument genitives that can occur with a given noun is rarely more than one (except in the case of deverbal nouns, which we are neglecting in this paper), and that when two or more genitives are able to occur with a noun, at least one of them must be a "possessive". This would be easiest to explain if the possessive is a modifier rather than another argument. The typical pattern of constraints suggests that a noun can have at most one genitive argument21 (although Babby (1997) and a few others have argued for two genitive argument positions in the Russian noun phrase). One would expect that a noun can have any number of modifiers but, if genitive modifiers are all of the same kind, "possessive", then a restriction to just one genitive modifier would be similar to the blocking of multiple adverbials sharing the same semantic function on a single verb.
6.
A possible uniform "modifier-only" approach
In this section we describe a possible uniform "modifier-only" approach to the English genitive and other genitives which appear to have both modifier and argument uses. Such an approach could, in principle, preserve the insights of Jensen and Vikner's uniform "inherent-R" approach and might also help to provide a semantic perspective on the notion of "quasiargumental modifier" that has been proposed by Grimshaw and others. This approach is similar in some important respects to that of Hellan (1980) and it appears to be subject to some of the same potential problems. Ultimately, we argue against it as a uniform treatment of English genitives, although we believe that some of the ideas sketched here have application to at least some parts of English and other languages and therefore deserve further research. In Section 6.1., we show how such an approach might work, briefly and with some gaps. In Section 6.2., we discuss compositionality issues from the perspective of this kind of approach. In Section 6.3., we argue that even within a "modifier-only" approach, we will need to distinguish the simply predicative "possessives" from other "relational" genitives. At this point, it
90
Barbara H. Partee and Vladimir Borschev
begins to look as though even if a uniform modifier approach can, in principle, be made to work, it may not be the right answer for languages like English. Indeed, in Section 6.4., we present a new argument we have found, based on the distribution of readings in certain elliptical constructions, in favor of maintaining a distinction between argument genitives and modifier genitives, at least for English. 6.1. Steps toward a uniform modifier analysis Suppose we would like team of Mary's, teacher of Mary's, brother of Mary's, height ofMary('s), sky of Mary's all to look like instances of intersective modification by an predicate.22 Then we might represent them as in (36). However, more must then be said about how the formulas in (36) are to be interpreted. (36)
a. b. c. d. e.
Xx[team (x) and R G EN(Mary)(x)] Àx[teacheri (χ) and R G EN(Mary)(x)] Xx[brotherj (x) and R G E N(Mary)(x)] Xx[heighti (x) and R G E N(Mary)(x)] Xx[sky (x) and R G E N(Mary)(x)]
Formula (36a), for instance, can be read informally as the property that something has if it is a team and it is Mary's, i.e. if it is a team and it stands in the relation R G e n with Mary. The meaning of ( o f ) Mary's as a basic intersective modifier of type that is used in these formulas is as shown in (37). (37)
Xx[RGEN(Mary)(x)]
We also need axioms to tell us what sorts of relations can be "genitive rela23
tions". We sidestep this important issue here and simply make the assumption (38) that teacher2 can be an appropriate value for R G e n · (38) teacher2 can be
RGEn-
Then we have to answer several questions. One concerns the interpretation of the one-place predicates in the representations above; another is the nature of R G e n (is it a variable or a constant?) and its place in the grammar. A third is the question of compositionality: how are such meanings derived from the meanings of the parts? Let us try to approach answers to these questions in several steps.
Genitives, relational nouns, and argument-modifier
ambiguity
91
Step 1: Let us focus on the sortal part of the meaning of a relational noun. We can exploit the fact that every noun has a basic sortal part in its meaning. We can even define it, at least in some cases, as the projection onto the xE-axis of the "whole" meaning of the noun, where the XE-argument is the "external" argument, the "referential" argument. Note that this can be done whether or not the noun can ever be used as a plain sortal noun (as teacher, nose, portrait easily can be, and brother, favorite movie, edge normally cannot be), since even those for which an internal argument is obligatory still have this sortal part of their meaning. For "plain" (sortal) nouns, the sortal part of the meaning is the whole meaning. We will refer to this definable kind of sortal meaning as the first projection of the relation denoted by the relational noun: (39)
SoTibrother = λχ . 3y (brother2 (jy)(x)) = first projection of brother2
Two important parameters of semantic differences 24 among relational nouns are the following: (a) whether the noun has a "normal" independent use as a plain sortal noun (of course in strong enough context, any noun can have a one-place use) and (b) if so, whether the sortal (one-place) variant of the noun has a meaning which amounts to more than just the first projection of the relational meaning (as teacher, lawyer does and brother does not). Earlier examples suggested that if the meaning of a relational noun's one-place variant was nothing more than the first projection of its relational meaning, then that noun would not normally be usable as an independent one-place predicate. However, further examples make it clear that even mere first projections can be used independently if that property has cultural importance. In our society, being a mother or a parent is important, being a brother or an uncle normally is not. It is not only for nouns like teacher that sentences like (40a) are good; (40b, c) are also fine, but (40d, e) are not.25 (40)
a. b. c. d. e.
Many teachers voted for John. Many mothers voted for John. Many parents voted for John. # Many brothers voted for John. # Many uncles voted for John.
We suggest that the one-place predicates in (36) are related to the basic noun meanings in one of three ways: (i) The one-place predicate may itself be the basic noun meaning, as in (41a). (ii) It may be an independently established one-place alternate of a two-place relational noun, as in (41b).
92
Barbara H. Partee and Vladimir Borschev
(iii) Otherwise (not counting the influence of strong contexts) by default it will be the first projection of the relational noun, as in (41c). (41) a. Àx[team (χ)], Àxfsky (χ)]: the meanings of the plain CNs team, sky b. Xx[teacheri (x)] : generic agentive noun, "one who teaches" c. Àxfbrotheri (χ)], Xx[heighti (χ)] : first projections of the TCNs brother, height2 The one-place predicate teacheri in its most basic use does not seem to be elliptical (as one-place friendi usually seems to be) and is not simply the first projection of the TCN teacher 2 , but rather the name of a profession, much like surgeon, actor. Step 2: We should compositionally derive the sortal part of the meaning of a phrasal NP (CNP). In simple cases, it will just be the sortal part of the meaning of its head noun, but more work is needed to identify the principles which specify the effects of non-subsective adjectives and of adjectives like favorite. Modifiers may also further specify sortal information by way of their selectional restrictions and/or their content. As a first approximation, but not an adequate general account, it is probably reasonable to assume (42). (42) Sorte»»» = λχ . 3y (CNP'(y)(x)) = first projection of CNP' Step 3: In order to unify the combination of a genitive phrase with CN and TCN, we need to assume a natural kind of "polymorphism", something we need for all sorts of noun-modifiers and verb-modifiers. We want to be able to say that adnominal (of) Mary's can take any kind of a CNP as argument, whether one-place or two-place or in principle η-place. The essence of the analysis will then be as in (43^44) (using Ν as a cover variable for any lexical or phrasal CN(P) or TCN(P)): (43) The genitive modifier (of) Mary's takes any N-type argument, keeps the sortal part of the Ν meaning and adds a free R G E N for the relation. (44) of Mary's: λΝ : Ν is a noun-meaning . λχ. [(Sort(/V))(X)
&
RGEN
(Mary)(;t)]
To further generalize this polymorphic operator to the 0-place case, we can follow the strategy of Montague (1970) and treat predicates as though they
Genitives, relational nouns, and argument-modifier
ambiguity
93
are modifiers of an empty noun entity. Since entity denotes a predicate true of everything in the domain, the predicative meaning given in (45) is reducible to that given in (46). This is one normal way for adjectives not originally of intersective type to shift to intersective modifiers. (45) λ*, [entity'(χ) & RGEN (Mary) (x) ] (46) λ*.
RGEN
(Mary) (x)
For a plain CN(P), the sortal part of the meaning is simply the meaning; for a TCN(P), it is the sortal "part" of the meaning as discussed above. Step 4: In the fourth step, in which we identify RGEN, we are influenced by Optimality Theory and by the work of Dölling (1992, 1997), Bierwisch (1989), and Hobbs et al. (1993). What we need are principles that say that if the noun already had a relational part of its meaning, then that should normally be used, and the more obligatorily relational the noun is, the more strongly that inherent relation is preferred. There should be such a principle in some very general terms, something about "using all the meaning" or at least using all the relevant parts of the meaning. There are also principles like those proposed by Frosch (1999) about RGEN being salient, being shared information, having suitable uniqueness properties. Further, there are principles relating to the content of the genitive relation, explored by Jensen and Vikner (1994, 1996), Vikner and Jensen (2002), and Partee and Borschev (2000a) - RGEN likes to be agentive, it likes to be part-whole, it does not like to be telic26 in the sense of Pustejovsky (1995). Rakhilina (2001) argues that in Russian, a genitive relation should be a relatively "stable" relation, not an ephemeral one, and should not be the kind of relation normally expressed with dative or instrumental case (insofar as those can be semantically described.)
6.2.
Compositionality issues
6.2.1. The basic cases How do we put together the meaning of brother of Mary's and team of Mary's on this view? We are moving towards a view that blends unification with ordinary function-argument application: the genitive modifier acts as a polymorphic function that applies "alike" to both one-place CNs and twoplace TCNs, yielding a one-place CNP which preserves the sortal part of the
94
Barbara H. Partee and Vladimir
Borschev
meaning of the noun and intersects it with the "genitive" predicate meaning shown in (46). If the meanings of brother and of Mary's are as in (47a, b), functionargument application would give (47c). From there, (i) an axiom analogous to that in (38) would tell us that an available value for the variable RGEN is brother2, and (ii) there should be a general principle to the effect that if the sortal part of brother is not a salient property on its own, any value for RGEN other than brother will yield an anomalous (or at least very hard to interpret) reading. With the help of such principles we would arrive at (47d). (47)
a. brother: b. of Mary's:
XyXx[brother2 (y)(x)] λΝ : Ν isa. noun-meaning . λχ. [(Sort(A0)(x) & RGEN (Mary)(x)] c. brother of Mary's·, λχ. [(Sort(brother2))(x) & RGEN (Mary)(x)] d. = (by principles above) λχ. [brotheri(x) & brother 2 (Mary)(x)]
Analogously, if the meaning of team is as in (48a), function-argument application will give (48c). Since team is already a one-place predicate, its sortal part is simply team. And for the variable RGEN in this case, the context, including information in the lexical meaning of team, should provide a value, e.g. "plays for", "coaches", etc. (48)
a. team: b. of Mary's:
Xx[team(x)] λΝ : Ν is a noun-meaning . λχ. [(Sort(/V))(x) & RGEN (Mary)(x)] c. team of Mary's: λχ. [(Sort(team))(x) & RGEN (Mary)(x)] d. = (by principles above) λχ. [team(x) & RCEN(Mary)(x)]
Note the contrast between the English of+ NPAcc construction {portrait of John), which is strictly argumentai, and the postnominal genitive {portrait of John 's), which allows any relational reading except that expressed by portrait of John. This contrast shows that there are evidently some "Blocking" principles: the reason that portrait of John's cannot usually mean what portrait of John must mean is presumably the very existence of portrait of John with its more specific meaning. There is no inherent prohibition of such a meaning, or even a dispreference for it, but it is blocked by the existence of the more specific alternative. One argument for this approach is that there is no such effect in prenominai position {John's portrait), where there is no alternative expression. Such blocking principles need to be explored further as a potentially important part of the explanation of the typological differences across lan-
Genitives, relational nouns, and argument-modifier ambiguity guages in the range of relations expressed by genitive constructions in "competition" with other constructions in the same language. For example, English genitives are not used for some of the relations expressed by genitives in Russian, apparently because of competition from the Noun-Noun compound construction in English (see the glosses of the examples in (7)).
6.2.2. Mary's former mansion revisited Since much was made of the example Mary's former mansion in our earlier arguments in favor of Jensen and Vikner's uniform "argument-only" approach, we should examine how such an example might be handled, if it can be, on an alternative "modifier-only" approach. As a first observation, we note that former affects the asserted part of the meaning of its modifiee and not the presupposed parts, a distinction we have not been explicitly representing. A former bachelor is normally interpreted as someone who is still adult, male, human but no longer unmarried.27 Suppose also that only lexical nouns can shift from CN to TCN (a welcome assumption, but one that was violated in the earlier account in Section 4). Then we could not assume that former and CN mansion could combine and then type-shift, as we did in (19) in Section 4.3. If former combines with CN mansion, and if we continue to assume that the most salient value of R G E N in this case is "possess", then we straightforwardly get Reading A: "a former mansion that is now Mary's". In order to get Reading B:, "something that was formerly Mary's mansion", mansion would have to shift to a TCN or relational reading before it combines with former, but without Jensen and Vikner's uniform "argumentonly" approach, we cannot appeal to coercion to account for such a shift. Leaving open the question of what independent motivation, if any, can be found for such a shift, let us assume for the sake of pursuing this approach that such motivation can be found. Then relational mansion may be represented as follows: (49) XyXxfmansioni(x) and RcEN(y)(x)] R G E N is "possess"
where the most salient value of
Applying former to this TCN mansion could then in principle target either part, depending on what was presupposed and what was focussed in the given context. Structurally, former could always apply just to the noun, ending up with either "formerly a mansion" (yielding a second derivation
95
96
Barbara H. Partee and Vladimir Borschev
for "reading A") or "formerly owned by y", the reading shown in (50) below. (50)
former mansion [assuming R G E N is the focussed part of XyXx[mansioni(x) and PAST(possessed-by(y)(x))]
(49)]
:
However, there is still a problem in how the possessive Mary's combines With former mansion on this approach. Suppose we try to follow the model shown in (47), the derivation of brother of Mary's, With former mansion in place of brother. (51) a. former mansion: XyXx[mansioni(x) and PAST(possessed-by(y)(x))] b. of Mary's·.
λΝ : Ν is a noun-meaning . λχ. [(Sort(/V))(x) & R G E N (Mary)(x)]
c. former mansion of Mary's: Ajc.[(Sort(51a))(jc) &
RGEN
(Mary)(x)]
What remains to be worked out for this approach is the further interpretation of (51c). It should be straightforward to derive mansioni as the sort in (51c); the greater challenge is to identify principles according to which the choice of R G E N in (51c) would be PAST(possessed-by(y)(x)). That is, to maintain the modifier approach, it is not enough to show that we can derive a reasonable relation-modifying interpretation for former in former mansion, but that when of Mary's combines with the result, the choice for R G E N should be the unusual relation "formerly owned". This contrasts with the straightforward function-argument combination of TCNP former mansion with Mary's as a genitive argument as shown in tree (21) in Section 4.3. When the genitive can be interpreted as an argument, it is straightforward to get it under the scope of former; whereas if it is to be uniformly interpreted as an intersective modifier, it is not straightforward.28 At this point, while we have done our best to imagine how a modifieronly approach might work, we are not sanguine about the prospect of finding good independently motivated solutions to all of the problems we have noted along the way. In the next section we nevertheless proceed to the issue of predicate genitives from the perspective of this approach. But in Section 6.4. we present a new argument in favor of going back to a split approach (at least for English).
Genitives, relational nouns, and argument-modifier
ambiguity
97
6.3. Predicate genitives again Since the uniform meaning proposed above amounts to a type-raised "predicate-conjunction" meaning, it should be based on a simple predicative meaning (type ) as shown in (46), repeated below as (52); this is equivalent to the meaning derived by the strategy of Montague (1970) given in (45). (52)
MARY'spRED : λχ.
RGEN
(Mary) (x)
But at this point we should probably bear in mind the "Janus-faced" nature of the genitives that we noted in Section 5: for "pure" non-elliptical predicate genitives, it may not be right to call this a "genitive" relation at all; this is where the distinction between "genitive" and "possessive" may become important. (53) MARY'spRED : λ*. RPOss (Mary) (.χ) It is, in our minds, a question for further research how to argue for a distinction between two classes of potentially "free" relations; we suspect that the distinction will be one of prototypical preferences (cf. Dowty 1989) rather than an absolute one. Possibly, Rposs should just be thought of as one of the most salient relations (or family of relations) accessible when there is no salient sortal information in the construction: not only in the predicative case, but in cases like anything of mine, all this stuff of John's, where the head noun has minimal lexical content. To say all these things, we need R G E N as a notion; the grammar (and Universal Grammar) has to be able to talk about it, and it has to be able to describe constraints and preferences. So it is not just the bare logical notion of a two-place relation; it is a two-place relation "template" that is part of the interpretation of a particular construction. The approach described here, while not fully worked out and still facing substantial problems, can be seen to differ in certain crucial ways from Partee (1983/1997) as well as from Jensen and Vikner (1994) and Vikner and Jensen (2002) (and from Borschev and Partee (1999a, b) insofar as we have been following Jensen and Vikner). We summarize the differences below:
98
Barbara H. Partee and Vladimir Borschev
(54) a. Partee (1983/1997): Two distinct constructions (i) with inherent-R nouns, Mary's is ÀRÀx[R(Mary)(x)] (a lifted argument). (ii) with sortal nouns Mary's is λΡλχ [P(x) & RcEN(MaryXx)] (a predicate lifted to become an intersective modifier). b. Jensen and Vikner (1994): All as lifted arguments, forcing plain nouns and NPs to shift to relational meanings. The genitive "wants" a relational TCN(P) to combine with, "wants" to give it an argument. c. This proposal: Related in part to Hellan (1980): Assimilate all to free-R case, by (a) splitting relational nouns into a "sortal part" plus a relation, (b) making Mary's a polymorphic function, and (c) having principles which help make sure that the "inherent R" of an inherently relational noun cannot easily be ignored. On the current proposal, all genitives could be viewed as modifiers. There are remaining conceptual problems, particularly for the "inherent R" case. The goal would be to have enough general principles at work that one could simply say RGEN(Mary)(x) and have all the rest follow. Even if the conceptual problems can be solved, any uniform approach will have to wrestle with the problem noted earlier of the limitations on the occurrence of more than one genitive with a single noun, limitations which may be better described in terms of co-occurrences of distinct genitive structures.
6.4. Another argument in favor of ambiguity We continue to wrestle with the issue of whether a uniform analysis is really correct for English genitives. After developing some steps towards a Hellanstyle unification and becoming increasingly skeptical of such an approach, we have come up with a new argument in favor of keeping possessives and argument genitives distinct. The strategy behind the search for new evidence is as follows: we consider contexts that favor an elliptical NP analysis of bare genitives and compare the behavior of bare genitives in such contexts with their behavior in predicate position where we may or may not have an elliptical NP. If the behavior is systematically different, that could provide possible evidence that not all predicate bare genitives in English are elliptical NPs. The particular evidence we present here in fact provides evidence that the genitive relation is located differently in the two cases: as part of the meaning of
Genitives,
relational
nouns, and argument-modifier
ambiguity
99
possessive Mary's in the case of an possessive (this is the kind of "free R" that we call RPOSS), and as part of the noun (possibly after coercion) in the case of argument genitives (this is our "RGEN", whether free ("pragmatic", "contextual") or inherent). First we consider a sentence in which the NPs are in argument positions (not predicate position), so that we know that a bare genitive is a remnant of NP ellipsis. (55) Sanderson's portraits are mostly better than his wife's. Here, the genitive relation can be any of owner, artist, subject (we limit our attention to those three possibilities, ignoring further possible contextual relations), but it must be same relation in Sanderson's portraits and his wife's [portraits]. That would follow if the RGEN relation is packed into the meaning of the noun portraits (in at least all cases except the possessive Rposs)29 and there is a deleted identical noun (whether or not it is a syntactic deletion; identity of semantic content is required in any case). That would NOT follow so clearly if RGEN were always part of the meaning of the genitive, although it does not directly argue against that because there could be "parallel structure" effects. Now we consider the hypothesis that when a bare genitive occurs in predicate position, it may or may not be a remnant of NP ellipsis (in English). Our new evidence for this hypothesis is based on examples like (56) below. (56) If Kandinsky's portraits had all been Gabriele Miinter's, then I suppose they would all be in Munich now?0'31 We believe that the predicate genitive in (56) has exactly the following possible readings:32 (a) Independently of how we interpret Kandinsky's portraits, Miinter's can express possession (ownership). (b) If Kandinsky's is interpreted as one of the inherent relations (artist, subject), then Miinter's can also express that same relation, but not a different inherent relation·, i.e., if we interpret Kandinsky's portraits as portraits by Kandinsky, then Miinter's can be interpreted as portraits by Munter but NOT as portraits ofMiinter. And conversely. These judgements can be accounted for on the following assumptions:
100
Barbara H. Partee and Vladimir Borschev
(i)
The predicative Gabriele Münter's can be either a simple predicate or an elliptical NP. (ii) A simple predicate possessive expresses "possession" (we assume that "agent", and "the one portrayed" are inherent relations, not cases of "possession"). (iii) A prenominai genitive can express either possession or any inherent relation. (iv) Inherent relations reside in the noun, either lexically or via coercion. So the choice in (i) leads to two possibilities: possession, or "same as the relation expressed in the antecedent genitive construction". Note that the ambiguity of (56) helps to show that "parallel structure" alone does not force identity of interpretation of the genitive relation, which in turn gives greater significance to the non-ambiguity of (55). This new evidence leads us to the conclusion that even if an argumentonly or a modifier-only approach is in principle possible, the facts of English nevertheless force us to accept a split approach.
7.
Speculative hypotheses and remaining puzzles
7.1. Two competing prototypes? It has often been pointed out that an argument genitive is most like a direct object, an "internal argument", most intrinsic to relational nouns. A "possessor" genitive, on the other hand, is most subject-like, agent-like, less like an internal argument, more independent; perhaps with more work it can be shown to follow that it is hence more easily a predicate. In cases where we can distinguish all three possibilities of possessor, subject, and object (e.g. John's portrait), the possessor seems to be even more external than the subject, as evidenced by well-known hierarchies of interpretive possibilities of e.g. Russian Mamin portret Ivana 'Mama's portrait of Ivan('s)', where Mama must be higher than Ivan in the hierarchy Possessor > Agent > Theme. We started from the idea that genitives with relational nouns are basic, and have been trying to figure out what adjustments take place when a genitive is used with a plain sortal noun. Heine (1997) starts from the other end, so to speak, with have sentences as primary concern and predicate genitives as secondary, and adnominal genitives as a tertiary interest. Inherent relations have a subordinate place in
Genitives, relational nouns, and argument-modifier ambiguity
101
the discussion; various notions of control and "possession" are at the forefront. This makes us see genitives as Janus-faced. From our perspective, the deverbal nouns are in a sense archetypal relational nouns, with genitives most clearly arguments: John's arrival, the city's destruction. From Heine's perspective, the use of a have-like construction or of a genitive construction with deverbal nouns is more like the grammaticization of a metaphorical extension of possession and inalienables like Mary's hand are closer to the core. Perhaps the child's early That's mine! is even more core-like. For genitive constructions which include the kind of possessive predicative readings discussed in Section 5, it seems clear that they are not to be treated as uniformly arguments. We have tried in the first parts of Section 6 to develop a version of the proposal of Hellan (1980) which could preserve many of the properties of Jensen and Vikner's uniform argument approach within a uniform modifier approach. But we argued that in the end it is preferable for genitive constructions like those in English to go back to a split approach, acknowledging that genitives may arise from either of two different prototypes, though with a wide overlap in the result.
7.2. Hypotheses and puzzles We summarize below some of our specific hypotheses about particular genitive and genitive-like constructions in English and Russian. 1.
The English of + N P A C C construction (portrait of John) is strictly argumentai.
2. The English Saxon genitive (John's) can be used as a predicate, type . 3. The English of+ NP's construction {portrait of John's) is ambiguously argumental/non-argumental. 4. The English prenominai NP'Í neutralizes the distinction between postnominal of + N P A C C and of + N P ' S . It can also be either argumentai or non-argumental. 5. The Russian genitive (Masi), always postnominal, is always an argument. It can never be used as a predicate (caveats). (But it can be used with plain nouns to express all kinds of relations including possession, as predicted by Jensen and Vikner's coercion analysis.)
102 6.
Barbara H. Partee and Vladimir Borschev The Russian prenominai possessive {Masin, -a) can be used as a predicate, it has certain limitations on its use as an argument, and it is either sometimes or never structurally an argument, although it can certainly fill argument roles.
The puzzle that emerges is that there seem to be argumentai genitive constructions and modifier possessive constructions that have a very great overlap in what they can express. If this is correct, it means that we cannot use "intuitions" of argumenthood as a good guide to whether something is "really" an argument at a given level of structure. As Dowty (1997, this volume) has argued, the distinction between modifiers and arguments need not be inherently sharp. Fleshing out more specific proposals about the relevant structures is necessarily a theory-dependent matter and we do not intend to undertake it without the collaboration of syntacticians. There are many different proposals in the literature for different argument and nonargument positions/sources for genitives and possessives in English, Russian, and other languages. The bottom line seems to be that type-shifting and lexical meaning shifts make many compositional routes available to very similar net outcomes. The line between arguments and modifiers is not intrinsically sharp in terms of what is being expressed, and can only be investigated in theorydependent ways. It is hard to find sharp differences between a theory in which the genitive construction itself contributes a "possessive" relation and a theory in which the genitive construction causes the head Ν or N-bar to shift to a relational interpretation possibly involving a "possessive" relation as one of its "preferred" relations. At this point we believe that both kinds of analyses have their place; the arguments we have found for a split approach for English and for the languages discussed in Section 5 are arguments that indeed genitives are sometimes modifiers and sometimes arguments. Genitives are evidently a domain of great semantic flexibility, where we have to find detailed language-particular evidence to try to sort out how lexical semantics, compositional semantics, and type-shifting possibilities interact in each particular construction.
Notes 1.
The authors wish to thank many colleagues for suggestions and discussion, especially Carl Vikner, Per Anker Jensen, Elena Paducheva, Ekaterina Rakhilina, and two anonymous referees. In addition to the conference in Oslo for which this paper was prepared, parts of this material were presented by one or
Genitives, relational nouns, and argument-modifier ambiguity
2.
3. 4. 5.
6.
7. 8.
103
both authors in graduate courses in Leipzig, Potsdam, Kolding, Moscow, Prague, and Sao Paulo, in a reading group at UMass Amherst, and in lectures in Berlin, Munich, Kleinwalsertal, Austria, at ESCOL 1999, in Bloomington, Swarthmore, Tel Aviv, Stanford, and at Sinn und Bedeutung 2000 in Amsterdam. We are grateful to members of those audiences for useful suggestions. The first parts of this paper overlap substantially with the first parts of Partee and Borschev (2001), and the whole paper is a revised version of Partee and Borschev (2000b). This material is based upon work supported by the National Science Foundation under Grant No. BCS-9905748. As noted in the next section, there is no perfect term to cover the whole range of "genitive" and "possessive" constructions. We use "genitive" as our neutral cover term, reserving "possessive" for notional possessives. The first two examples in (2b) are not morphological genitives; see (3d-e) in the next section. For Russian, this applied to the postnominal genitives illustrated below in (3d) but not to the prenominai possessives of (3e). We use English and Russian for illustrative purposes, abbreviated below as E andR. We are grateful to Marcel den Dikken (p.c.) for suggesting that one should explore a possible approach on which the genitive in (6) is a complement of the lexical adjective favorite, so that genitives, when complements, would always be complements of some lexical item. That could certainly be made to work semantically, as long as the adjective favorite is always a function applying to the noun's meaning. As den Dikken notes, "it does complicate the syntax at first blush"; we suspect that a fuller investigation might best be carried out in connection with a study of the interaction of genitives with superlative and superlative-like constructions as in John's best picture, John's first picture. There is already a problem in using this construction for illustration, since a number of authors, including Barker (1995), have argued that the English postposed genitive is a reduced partitive, book of John's books, and that there is therefore no simple construction of the form [N Ν NPGEN ] IN English. The reason we are not using the construction John's book for our basic case is that the prenominai genitive in English seems to combine the "basic" genitive with an implicit definite article. We are assuming here that the postposed genitive is a basic construction in English (see also Lyons 1986), but the general points we make would also hold for the prenominai genitive "minus the meaning of the definite article". Thanks to Michael Brody (p.c.) for noting that one should of course explore the "underlying position" of the prenominal genitive, which may move into a determiner position from somewhere else, at least in theories with syntactic movement. This point is made more systematically for a wide range of constructions in Dowty (this volume). As with the use of free variables like x¡ to represent pronouns used without linguistic antecedent, we assume as a felicity condition on the use of free R¡
104
9.
10.
11. 12.
13.
14.
15.
Barbara H. Partee and Vladimir Borschev that the context should make it sufficiently clear to the hearer what particular relation the speaker has in mind. We do not discuss quantifier possessors in this paper. As a first approximation, we would follow the analysis of Bach and Partee (1980) which gives quantifier possessors like every student's widest scope within the noun phrase but does not allow a quantifier possessor to have scope independently of its noun phrase. See also Vikner and Jensen (2002). But see Storto (2000) for observations about the distribution of contextually supplied relations that challenge some of the assumptions made in our work. We do not address those challenges here. Vikner and Jensen (2002) address the issue of favorite in a manner consistent with the points raised here. An alternative analysis of the ambiguity, based on different assumptions which we do not share, has since been offered by Larson and Cho (1999). As noted by Marcel Den Dikken (p.c.), one non-standard assumption we are making is that phrasal categories (like TCNP) can take arguments; this is a standard assumption in Categorial Grammar (see e.g. Bach 1980) but not in most other frameworks. Our claim that the Russian genitive uniformly has argument status has been challenged by some colleagues; we acknowledge the existence of some problematic data but need to study it further before trying to respond. We thank Ash Asudeh (p.c.) for example (22c), and Ekaterina Rakhilina and Elena Paducheva for examples (22d, e). We are also grateful to Per Anker Jensen for similar examples, and to all of them for helpful discussion of the possible differences between the good and bad examples. We thank M. den Dikken for pointing out that in Dutch, the predicate possessive in example (22c) is even more clearly an elliptical NP than in English and that Dutch, furthermore, is a language which clearly distinguishes elliptical from non-elliptical predicate possessives. In the Dutch rendition of (22c), the d-word die, signalling the presence of nominal structure, is obligatory, as shown in (i). (i) Die decent is *(die) van Jan. That teacher is *(that) of Jan 'That teacher is Jan's.' By contrast, in (ii) both options are possible. (ii) Die auto is (die) van Jan. That car is (that) of Jan 'That car is Jan's.' An anonymous referee suggests that (22a, b, f) are bad simply because father and favorite movie cannot be used non-relationally, as would be required for occurrence in the subject position of these sentences. We only partly agree. We believe that they can be used non-relationally, with a corresponding meaningshift, as in (i-iii) below but that, once they have been shifted to a nonrelational meaning, they cannot support ellipsis with their original relational reading, which is what would be required for (22a, b, f) to be good.
Genitives, relational nouns, and argument-modifier ambiguity
105
(i) Some fathers are stricter than others. (ii) That father over there in the playground isn 7 having much fun. (iii) Very few favorite movies come out of Hollywood anymore. (= "very few of anyone's favorites") In a context in which (ii) above would be used, some speakers may be able to accept (22a), but we believe that even in such a context, John's in (22a) could not be understood as John's father. Apparent intuitions to the contrary may involve a genitive relation corresponding to something like "assigned to", "associated with", "paired with", which might in the given context be extensionally equivalent to the "father o f ' relation. This issue, which often arises in discussions of the data, deserves further investigation. 16. The material in this section of the paper is drawn in large part from Partee and Borschev (2001). 17. Caveats must be put on the statement that genitive NPs cannot occur in predicate position in Russian; but the conditions under which they can occur are relatively special. 18. Further evidence that these predicate possessives are elliptical NPs was provided by Sigrid Beck and Irene Heim (p.c.): the possessive pronoun in (27) can be followed by adjectives (i.e. there can be ellipsis of just the head noun), while the adjective in (26) and the adjective-like possessive pronoun in (28b) cannot be. Thanks to Claudia Maienbora for correcting the mistakes in our earlier rendition of these examples. (i) Diese Bücher sind meine alten. These-N.PL. book-N.PL a r e
(ii)
my-N.PL. old-N.PL.
'These books are my old ones.' * Diese Bücher sind teuer
neu(en).
These-N.PL. book-N.PL. a r e e x p e n s i v e n e w
'These books are expensive new ones.' 19. One anonymous referee considers "any relation" too strong a statement. We have encountered considerable speaker variation on these examples, but the intuitions we report seem to be in the majority. 20. Thanks to Ania Lubowicz and Anita Nowak for judgments. For (33a), Anita reports no preference for one reading or the other, while for (33b) she reports a preference for the "citizenship" reading. Both rejected (34b) as ungrammatical; Ania suggested that it should be corrected to (34c), which she finds possibly ambiguous. Both agreed that (34a) is unambiguously "possession" only, whereas (33a) allows either reading. The basic judgments given above in the text for (33a, b) and (34a, b) were further confirmed by Janusz Bieñ, Bozena Cetnarowska (and by a substantial majority of a group of 12 students of hers), Bozena Rozwadowska, Piotr Bariski, and Joanna Blaszczak, to all of whom we are grateful. 21. We are grateful to an anonymous referee for reminding us that this statement is not fully general and that it makes a difference how many different kinds of genitives a language has. As the referee notes, in German it is fine for a dever-
106
22.
23.
24. 25.
26.
27. 28.
29.
30.
Barbara H. Partee and Vladimir Borschev bal noun to have both a prenominai (Saxonian) possessive/"subjective" and a postnominal "objective" genitive as well. We use subscripts 1 and 2 to represent the one-place predicate and two-place relation versions of nouns. Thus, teacheri is of type , while teacher2 is of type . We discuss the meanings of one-place versions of normally two-place nouns below. We are grateful to Ekaterina Rakhilina (p.c. and lecture in Apresjan's seminar in 2001) for pointing out the need to uncover the principles that regulate the choice of e.g. genitive vs. dative for "argumentai modifiers" in Russian. Principles are also needed concerning the choice of e.g. genitive modifier vs. noun-noun compound in English. Thanks to Ash Asudeh (p.c.), who first brought the importance of these issues to our attention with examples using the noun teacher. We mark the "bad" examples here with the symbol "#", indicating that they are normally anomalous, but not ungrammatical. As usual, a sufficiently strong context can render them fully felicitous. This provides an additional argument for treating the "kind-modifying" possessives studied by Munn (1995) and Strauss (ms. 2002) as a distinct construction, since "telic" readings are the first choice for such possessives. Compare the two readings of children's poems·, if we mean poems of some particular children, the agentive reading is most salient, but as a kind of poems, a "for" reading is most salient. The analogous observation in the case of sentential negation is a standard test for presupposition vs. assertion. There is, however, a possibility that former could apply to mansion of Mary's and that this order of semantic combination might even be possible in the case of the prenominai genitive in Mary's former mansion if the visible surface structure is not isomorphic with the semantic structure. More work needs to be done on this issue. We assume that an possessive can act as an ordinary intersective modifier in an NP; in that case the Ν may remain one-place and not shift to a "transitive" reading. So, if Sanderson's is an possessive, deletion of an identical one-place noun portraits will mean that his wife's can also only be an possessive. In all the other cases, the meaning of some particular relation will be packed into the noun portraits. Kandinsky and Münter were both artists, both did some portraits of each other and featured in portraits by other people as well, and he left a lot of paintings with her when he left Munich, and she eventually gave those paintings to the city of Munich. So we hope that all sorts of counterfactuals involving who painted what and of whom, and who was in possession of whose paintings, might all be pragmatically reasonable. (Some of the paintings were sufficiently abstract that one could also imagine there being uncertainty as to who was the one portrayed, if anyone. We do not think there is actually any confusion about that, but it is not unimaginable.)
Genitives, relational nouns, and argument-modifier ambiguity
107
31. It is not easy to find plausible examples in which we can have potential ambiguity among possession and some inherent-R reading while having one of the possessives in a predicate position. We have purposely put them in the antecedent of a counterfactual to try to help make plausible a reading where we can imagine portraits having a different artist or a different subject in addition to the possibility of different owner. The latter is normally by far the easiest to imagine being different, which can obscure judgments about other possibilities. 32. Thanks to the English-speaking participants in our March 2002 Mathesius Institute lectures for judgments.
References Babby, Leonard H. 1973 The deep structure of adjectives and participles in Russian. Language 49: 349-360. 1997 Nominalization in Russian. In Formal Approaches to Slavic Linguistics, The Cornell Meeting 1995, W. Browne, E. Dornisch, N. Kondrashova, and D. Zee (eds.), 54-83. Ann Arbor: Michigan Slavic Publications. Babyonyshev, Maria 1997 The possessive construction in Russian: A crosslinguistic perspective. Journal of Slavic Linguistes 5(2): 193-230. Bach, Emmon 1967 Have and be in English syntax. Language 43(2): 462—485. 1980 In defense of passive. Linguistics and Philosophy 3: 297-341. Bach, Emmon, and Barbara H. Partee 1980 Anaphora and semantic structure. In Papers from the Parasession on Pronouns and Anaphora, Jody Kreiman and Almerindo E. Ojeda (eds.), 1-28. Chicago: Chicago Linguistic Society. Baker, Carl L. 1978 Introduction to Generative Transformational Syntax. Englewood Cliffs, NJ: Prentice-Hall. Barker, Chris 1995 Possessive Descriptions. Stanford: CSLI Publications. Bierwisch, Manfred 1989 Event nominalizations: Proposals and problems. In Wortstruktur und Satzstruktur, W. Mötsch (ed.), 1-73. Berlin: ZISW Borschev, Vladimir B., and L. V. Knorina 1990 Tipy realij i ix jazykovoe vosprijatie [Types of entities and their perception in language]. In Language of Logic and Logic of Language, 106134. Moscow.
108
Barbara H. Partee and Vladimir Borschev
Borschev, Vladimir B., and Barbara H. Partee 1998 Formal and lexical semantics and the genitive in negated existential sentences in Russian. In Formal Approaches to Slavic Linguistics 6: The Connecticut Meeting 1997, Zeljko BoSkovic, Steven Franks, and William Snyder (eds.), 75-96. Ann Arbor: Michigan Slavic Publications. 1999a Semantika genitivnoj konstrukcii: raznye podxody k formalizacii [Semantics of the genitive construction: different approaches to formalization]. In Typology and Linguistic Theory: From Description to Explanation. For the 60th birthday ofAleksandr E. Kibrik, Ekaterina V. Rakhilina, and Yakov G. Testelets (eds.), 159-172. Moscow: Jazyki Russkoj Kul'tury. 1999b Semantic types and the Russian genitive modifier construction. In Formal Approaches to Slavic Linguistics: The Seattle Meeting 1998, K. Dziwirek et al. (eds.), Ann Arbor: Michigan Slavic Publications. Dolling, Johannes 1992 Flexible Interpretation durch Sortenverschiebung. In Fügungspotenzen, I. Zimmermann and A. Strigin (eds.), 23-62. Berlin: Akademie Verlag. 1997 Semantic form and abductive fixation of parameters. In From Underspecification to Interpretation. Working Papers of the Institute for Logic and Linguistics, R. van der Sandt, R. Blutner, and M. Bierwisch (eds.), 113-138. Heidelberg: IBM Deutschland. Dowty, David 1989 On the semantic content of the notion "thematic role". In Properties, Types and Meanings. Vol. II. G. Chierchia, B. Partee, and R. Turner (eds.), 69-130. Dordrecht: Kluwer. 1997 Adjunct-to-argument reanalysis in a dynamic theory of grammar: The problem of prepositions. Paper presented at the Blaubeuren Semantics CCG/adj-arg Conference, University of Tübingen, March 1997. 2003 The dual analysis of adjuncts/complements in Categorial Grammar. In this volume. Freeze, Ray 1992 Existentials and other locatives. Language 68: 553-595. Frosch, Helmut 1999 German "attributive" genitives as adjuncts. Paper presented at the Conference "Approaching the Grammar of Adjuncts", 22-25 Sep 1999, Oslo. Grimshaw, Jane 1990 Argument Structure. Cambridge, MA: MIT Press. Heine, Bernd 1997 Possession: Cognitive Sources, Forces, and Grammaticization. Cambridge: Cambridge University Press.
Genitives, relational nouns, and argument-modifier ambiguity
109
Hellan, Lars 1980 Toward an integrated theory of noun phrases. Dissertation, Trondheim University. Heycock, Caroline, and Anthony Kroch 1998 Inversion and equation in copular sentences. In ZAS Papers in Linguistics 10, A. Alexiadou, N. Fuhrhop, U. Kleinhenz, and P. Law (eds.), 71-87. Berlin: Zentrum für Allgemeine Sprachwissenschaft. 1999 Pseudocleft connectivity: Implications for the LF interface level. Linguistic Inquiry 30: 327-364. Hobbs, Jerry J., Mark Stickel, Douglas Appelt, and Paul Martin 1993 Interpretation as abduction. Artificial Intelligence 63: 69-142. Jensen, Per Anker, and Carl Vikner 1994 Lexical knowledge and the semantic analysis of Danish genitive constructions. In Topics in Knowledge-based NLP Systems, S. L. Hansen and H. Wegener (eds.), 37-55. Copenhagen: Samfundslitteratur. 1996 The double nature of the verb have. In LAMBDA 21, OMNIS Workshop 23-24 Nov. 1995, 25-37. Handelsh0jskolen i K0benhavn: Institut for Datalingvistik. Knorina, L. V. 1985 Ob interpretacii genitivnyx konstrukcij [On the interpretation of genitive constructions]. Theses of the workshop "Semiotic foundations of intellectual activity". Moscow: VINITI. 1988. Klassifikacija leksiki i slovarnye definicii [Lexical classification and dictionary definitions]. In Nacional'naja specifika jazyka i eë otrazenie ν normativnom slovare, ed. Ju.N. Karaulov, 60-63. Moscow: Nauka. 1990 Narusenija soéetajemosti i raznovidnosti tropov ν genitivnoj konstrukcii [Violations of cooccurence and varieties of tropes in the genitive construction]. In Contradictions and Anomalies of Text, 115-124. Moscow. 1996 The range of biblical metaphors in smikhut. Moscow Linguistic Journal 3: 80-94. Kolliakou, Dimitra 1999 De-Phrase extractability and Individual/Property denotation. Natural Language and Linguistic Theory 17(4): 713-781. Koptjevskaja-Tamm, Maria, and Aleksej ámelev 1994 Ale§ina s Masej stat'ja (o nekotoryx svojstvax russkix "pritjaZatel'nyx prilagatel'nyx") [Alesina and Masa's article (on some properties of Russian "possessive adjectives")]. Scando-Slavica 40: 209-228. Landman, Fred, and Barbara H. Partee 1984 Weak NPs in HAVE sentences. Draft abstract. University of Massachusetts, Amherst.
110
Barbara H. Partee and Vladimir Borschev
Larson, Richard, and Sungeon Cho 1999 Temporal adjectives and the structure of possessive DPs. In Proceedings of WCCFL 18, S. Bird, A. Carnie, J. D. Haugen, and P. Norquest (eds.), 299-311. Cambridge: Cascadilla Press. Lyons, Christopher 1986 The syntax of English genitive constructions. Journal of Linguistics 22: 123-143. Montague, Richard 1970 English as a formal language. In Linguaggi nella Società e nella Tecnica, B. Visentini et al. (eds.), 189-224. Milan: Edizioni di Comunità. Reprinted in Montague (1974), 188-221. 1973 The proper treatment of quantification in Ordinary English. In Approaches to Natural Language, K. J. J. Hintikka, J. M. E. Moravcsik, and P. Suppes (eds.), 221-242. Dordrecht: Reidel. Reprinted in Montague (1974), 247-270. 1974 Formal Philosophy: Selected Papers of Richard Montague, edited by Richmond Thomason. New Haven: Yale University Press. Moro, Andrea 1997 The Raising of Predicates. New York: Cambridge University Press. Munn, Alan 1995 The possessor that stayed close to home. In Proceedings of the Western Conference on Linguistics (WECOL 24), V. Samiian and J. Schaeffer (eds.), 181-195. Partee, Barbara H. 1987 Noun phrase interpretation and type-shifting principles. In Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers (GRASS 8), J. Groenendijk, D. de Jongh, and M. Stokhof (eds.), 115-143. Dordrecht: Foris. 1983/1997 Uniformity vs. versatility: The genitive, a case study. Appendix to Theo Janssen (1997), Compositionality, In The Handbook of Logic and Language, Johan van Benthem and Alice ter Meulen (eds.), 464—470. New York: Elsevier. 1995 Lexical semantics and compositionality. In An Invitation to Cognitive Science, 2d ed. Vol. 1 : Language, Lila Gleitman and Mark Liberman, (eds.), 311-360. Cambridge, MA: MIT Press. 1999a Copula inversion puzzles in English and Russian. In Formal Approaches to Slavic Linguistics: The Seattle Meeting 1998, K. Dziwirek et al. (eds.), Ann Arbor: Michigan Slavic Publications. Also in UMOP 23: Issues in Semantics and its Interface, Kiyomi Kusumoto and Elisabeth Villalta (eds.), 183-208. Amherst: GLSA Publications 2000. 1999b Weak NP's in HAVE sentences. In JFAK: A Liber Amicorum for Johan van Benthem on the occasion of his 50th Birthday, CD-Rom, Jelle Gerbrandy, Maarten Marx, Maarten de Rijke, and Yde Venema (eds.), Amsterdam: University of Amsterdam. Accessible at: http://www.illc. uva.nl/j50/.
Genitives, relational nouns, and argument-modifier ambiguity
111
Partee, Barbara H., and Vladimir Borschev 1998 Integrating lexical and formal semantics: Genitives, relational nouns, and type-shifting. In Proceedings of the Second Tbilisi Symposium on Language, Logic, and Computation, R. Cooper and Th. Gamkrelidze (eds.), 229-241. Tbilisi: Center on Language, Logic, Speech, Tbilisi State University. 2000a Possessives, favorite, and coercion. In Proceedings of ESCOL 99, Anastasia Riehl and Rebecca Daly (eds.), 173-190. Ithaca, NY: Cornell University Linguistics Students' Association. 2000b Genitives, relational nouns, and the argument-modifier distinction. In ZAS Papers in Linguistics 17, Cathrine Fabricius-Hansen, Ewald Lang, and Claudia Maienborn (eds.), 177-201. Berlin: Zentrum für Allgemeine Sprachwissenschaft, Sprachtypologie und Universalienforschung. 2001 Some puzzles of predicate possessives. In Perspectives on Semantics, Pragmatics and Discourse: A Festschrift for Ferenc Kiefer, István Kenesei and R. M. Harnish (eds.), 91-117. Amsterdam: John Benjamins Publishing Company. Pustejovsky, James 1993 Type coercion and lexical selection. In Semantics and the Lexicon, J. Pustejovsky (ed.), 73-94. Dordrecht: Kluwer. 1995 The Generative Lexicon. Cambridge, MA: MIT Press. Rakhilina, Ekaterina 2001 Pokazateli posessivnosti i ix funkcii ν russkom jazyke [Indicators of possessivity and their function in the Russian language]. In Issledovanija po jazykoznaniju, S. Subik (ed.), 197-207. St. Petersburg: St. Petersburg University Press. Rappaport, Gilbert 1998 The Slavic noun phrase. Position paper for Comparative Slavic Morphosyntax. Available at: http://www.indiana.edu/~slavconi7 linguistics/download.html [1999, Jan. 4]. Schoorlemmer, Maaike 1995 Participial passive and aspect in Russian. Ph.D. dissertation, Utrecht: OTS (Utrecht Institute of Linguistics). Siegel, Muffy 1976 Capturing the Russian adjective. In Montague Grammar, B. Partee (ed.), 293-309. New York: Academic Press. Stockwell, Robert P., Paul Schachter, and Barbara H. Partee 1973 The Major Syntactic Structures of English. New York: Holt, Rinehart and Winston. Storto, Gianluca 2000 On the structure of indefinite possessives. In SALT X: Proceedings of the Tenth Conference on Semantics and Linguistic Theory 2000, Brendan Jackson and Tanya Matthews (eds.), Ithaca, N.Y.: CLC Publications, Department of Linguistics, Cornell University.
112
Barbara H. Partee and Vladimir Borschev
Strauss, Uri 2002
Individual-denoting and property-denoting possessives. Paper presented at the Workshop on the Syntax and Semantics of Possessives, University of Massachusetts, May 2002. Szabolcsi, Anna 1994 The Noun Phrase. In The Syntactic Structure of Hungarian, F. Kiefer and K. É. Kiss (eds.), 179-275. (Syntax and Semantics 27.) New York: Academic Press. Vikner, Carl, and Per Anker Jensen 2002 A semantic analysis of the English genitive. Interaction of lexical and formal semantics. Studia Linguistica 56:191-226. Williams, Edwin 1981 Argument structure and morphology. The Linguistic Review 1: 81-114. 1983 Semantic vs. syntactic categories. Linguistics and Philosophy 6: 423424.
Heads, complements, adjuncts: Projection and saturation Manfred Bierwisch
Abstract Given the generally accepted distinction between Head, Complement, and Adjunct as a relational syntactic classification, my claim is that there is a corresponding distinction between Argument and Modifier that draws on the semantic function of Complement and Adjunct, respectively. The main proposal is to account for the Head-Complement/Adjunct distinction in terms of two basic operations that are applied to two expressions X and Y: (i) projection of category features and (ii) discharging of thematic roles/argument positions. Syntactically, the Head X projects its categorizing features, thereby specifying the syntactic properties of [ X Y ]. Semantically, either the Head X discharges a thematic role to the Complement Y, or the Adjunct Y discharges a thematic role to the Head X. Discharging an argument position of a Head to a Complement is technically realized by functional application. Discharging an argument position of an Adjunct to its Head must provide for different types of modification. Two basic possibilities are to be considered. In so-called Extensional Modification as in green table, the Semantic Form (SF) of the complement is added to that of the Head. Technically, it is realized as an absorption of the Adjunct's argument position by one of the Head's argument positions, inducing appropriate variable substitution in the Adjunct's SF. So-called Intensional Modification as in former president restricts the validity of the Head's predication by means of the Adjunct's SF. Formally, it comes down to functional composition of the Adjunct taking the Head as argument. It follows from my analysis that Adjuncts are syntactically optional for principled reasons. As arguments surface either obligatorily or optionally (subject to partially systematic, partially idiosyncratic, conditions) there is some indeterminacy in cases, where e.g. locatives can be construed as either (optional) Complements or as Adjuncts. Finally, particles like again, even (usually classified as Adjuncts) are shown to be a particular type of Intensional Modification with characteristic properties concerning both feature projection and argument saturation.
1.
Introduction
The problem addressed in this paper is the distinction between Complementation and Adjunction. Directly connected to this issue is the relation between the notions Adjunct and Modifier and perhaps also Adverbial. The
114
Manfred Bierwisch
same issue arises wrt. the relation between their syntactic counterparts Complement and Argument (including hybrid notions like ArgumentAdjunct). These distinctions, in turn, have consequences for their common opposite, viz Head. The problem is not just a matter of terminology, but concerns non-trivial factual clarification and hence needs a certain amount of theoretical background. The common property of Complements and Adjuncts seems to be fairly clear: Both must be attached to a head on which they depend in some way. What deserves clarification, though, is the nature of this attachment: whether and how it is determined by the Head; how, in particular, it differs for Complements and Adjuncts; and how it relates to the notions Argument and Modifier. As a first approximation, one might suggest that Argument and Modifier indicate the semantic aspect of constituents that are syntactically classed as Complement and Adjunct, respectively. The term Head, on the other hand, seems to define their opposite primarily wrt. the syntactic aspect. This leads to the following provisional schema, where A indicates the status of the nucleus as opposed to its companion B, which is either selected or free, distinguished as Β1 and B2, respectively: (1)
A
Head
B1
B2
Complement
Adjunct
Argument
Modifier
Under this classification, Adverbial must presumably be considered as a hybrid notion referring to Adjuncts with additional conditions concerning the category of the Head as well as that of the Adjunct. Similar provisos apply to the term Attribute (in the sense relevant in this context). The classification in terms of nucleus, companion, selection, or the distinction of the upper vs. the lower row of the B-elements by means of syntax and semantics are highly provisional and in need of clarification. This clarification is the aim of the paper.
Heads, complements,
2.
adjuncts: Projection
and saturation
115
Heads, complements, adjuncts
It is well-known that the notions Head, Complement, and Modifier are not defined in terms of syntactic categories like noun, verb, adjective, determiner, etc. (or the features from which these categories are made up) nor by semantic types like Predicate, Argument etc. As a matter of fact, verbs, nouns, adjectives, or prepositions can all be Heads of the same companion, e.g. a constituent headed by a preposition, as illustrated in (2): (2)
a. b. c. d.
He [ [ worked ] v [ in the garage ]PP ]yp The [ [ car ]N [ in the garage ]PP W He is [ [familiar ]A [ with the plan ]PP ]AI> He is [ [ up ]P [ in the mountains ]PP ]PP
One and the same element can be Head, Complement, or Adjunct, as for instance the adjective in (3) or the NP in (4): (3)
a. [ [ tall ]A [ for a teenager ]PP ]AI> b. [ [ become ] v [ tall ]A ]VP c. one of the [ [ tall ]A [ teenagers ]N ]NP
(Head) (Complement) (Adjunct)
(4)
a. [ [ two hours ]NI> [ before the departure ]PP ]PP b. [ [ FOR ]p [ two hours ]NP ] pp c. he [slept ] v [ two hours ]MP ]vp
(Head) (Complement) (Adjunct1)
Hence the notions Head, Complement, and Adjunct must be defined by functional or relational criteria.2 This observation does not exclude certain elements or even categories from being restricted to one particular function. Thus particles like again, almost, also can show up only as Adjuncts3, while most functional categories can only serve as Heads. The criterion usually assumed to necessarily characterize the Head X of a construction φ is that X determines the properties of φ. Thus the properties of a directional PP like into the woods are clearly determined by the preposition into, but not by its Complement the woods. The properties of the Head, however, can obviously not be identical with those of φ in general. Thus into is "incomplete" in a way in which into the woods is not. The addition of the complement the woods furthermore adds specific semantic properties which the Head by itself does not provide. Hence exactly which properties of a construction φ can and must be determined by its Head X cannot be specified in terms of simple observational criteria and will be a matter of concern as we proceed. However, even without simple and obvious criteria,
116
Manfred Bierwisch
the distinction between Head and companion seems to be fairly clear in principle.4 This does not hold for the distinction between Complement and Adjunct. Two criteria are usually considered as relevant for this distinction: First, Adjuncts are optional, while Complements usually are not, and second, Complements are selected by the Head, Adjuncts are free. It is not quite clear, however, whether these are independent criteria, since being selected is sometimes considered as tantamount to being obligatory. The problem is illustrated by the following minimal pairs: (5)
a. Everybody signed the petition. b. He remained inside the circle. c. Eve is younger than her sister.
a.' Everybody signed. b.' He remained inside. c.' Eve is younger.
According to standard opinions, the object the petition is selected by the verb sign in (5a), but it is optional if (5a') is to be considered as complete and grammatical. In this case, Complements are selected, but can still be optional. Alternatively, one might stipulate two verbs sign, one that selects an object, while the other is just an intransitive verb (with something like a semantically incorporated object). Similar considerations apply to the preposition inside in (5b) and the comparative younger in (5c). Hence either one recognizes optional Complements, accepting the consequence that optionality does not discriminate between Complements and Adjuncts any more, or one tries to rescue the criterion, but then one has to stipulate separate lexical items each time apparent optional Complements are to be accommodated, which comes close to making the criterion circular, however. Further difficulties arise if one compares cases like those in (6), where the same constituent hinter dem Schloß is - according to received opinion selected in (6a) through (6c), and optional in (6c) through (6e), hence both optional and selected in (6c). (6)
a. Er ist hinter dem Schloß gewesen. 'He was behind the castle.' b. Er hat hinter dem Schloß gewohnt. 'He lived behind the castle.' c. Er hat hinter dem Schloß gestanden. 'He stood behind the castle.' d Er hat hinter dem Schloß geschlafen. 'He slept behind the castle.' e. Er hat hinter dem Schloß geschrieben. 'He wrote behind the castle.'
Heads, complements, adjuncts: Projection and saturation
117
To put it differently, it is by no means obvious in which sense stehen 'stand' like wohnen 'live', 'reside' selects a locative Complement, while schlafen 'sleep' does not. Additional differences might be observed in the way in which the PP of copulative constructions like (6a) combines with its Head, or adverbials like those in (6e) and (6d) modify their Heads. Problems of different sorts can be easily be added. Compare, for instance, cases like (7a) and (7b), where the PP mit unseren Freunden 'with our friends' is an optional Adjunct in (7a), but perhaps an optional Complement in (7b), while the directional PP an/über die Grenze is selected in (7a) - similar to the locative PP in (6c) - while streiten in (7b) selects the PP as a so-called Prepositional Object with rather different properties, suppressing not only the locative semantics of Ρ but, furthermore, restricting the choice of the PP's Head to über. (7)
a. Wir fuhren (mit unseren Freunden) an/über die Grenze. 'We drove (with our friends) to/across the border.' b. Wir stritten (mit unseren Freunden) *an/über die Grenze. 'We argued (with our friends) *to/ about the border.'
As the examples in (8) show, these conditions are preserved, moreover, under nominalization in spite of the fact that Complements of nouns usually are optional and are hence often classified by the hybrid notion "Complement-Adjunct": (8)
a. die Fahrt (mit unseren Freunden) an/über die Grenze 'the ride (with our friends) to/across the border.' b. der Streit (mit unseren Freunden) *an/über die Grenze 'the argument (with our friends) *to/about the border.'
Examples like (2) to (8), to which further intricacies could easily be added, should suffice to show that intuitive notions like optionality and selection cannot be used to distinguish Heads, Complements, and Adjuncts without a more systematic, theoretical foundation. In what follows, I will discuss a proposal to that effect relying as far as possible on familiar assumptions, in terms of which, however, a range of relevant facts can be accounted for provided these assumptions are made explicit. Suppose first, with respect to Heads, that [X] is a set of morpho-syntactic features representing the categorization of a given constituent X. Then according to familiar notions, X projects its categorization [X] to φ, if X is the Head of φ, such that [X] becomes the syntactic characterization of φ. In
118
Manfred Bierwisch
other words, φ consists of X and its companion Y, with XY (or YX)5 as a whole being categorized as [X].6 Second, according to equally familiar assumptions with respect to Complements, the Head X selects Complement Y by means of conditions that Y must meet in order to form a complex expression XY. The precise nature of the conditions, however, is a matter of dispute and will be taken up below. As a first step in that direction, we might assume that the Complement Y must be compatible with, and therefore saturates, an argument position of the Head X. In this sense, then, 'X selects Y' is equivalent to 'X discharges an argument position to Y \ Third, deviating from the standard view on Adjuncts, according to which an Adjunct Y is free wrt. its Head X in the sense that no selection takes place, I will assume that the characteristic property of an Adjunct Y is to discharge an argument position to its Head X, not determining the morphosyntactic properties of the combination XY, though. This assumption has important consequences to be explored below. In short, I will assume that selection (construed as discharging an argument position) and projection of morpho-syntactic features are independent, though interacting, conditions. We thus get the two possibilities schematically summarized in (9), where X » Y indicates discharging of an argument position of X to Y: (9)
a. Complementation
b. Adjunction
M [Χ] I χ
[Y]
[X]
I
I > γ
χ
2hours]
Here, Udo is the Agent and a certain novel is the Theme of a process which lasts at least two hours and which forms a constituent part of a potential reading event, the Agent and the Theme of which are likewise Udo and the novel, respectively.8 For a sentence like (3) in which, again, an accomplishment comes in combination with a durative adverbial, a process-related interpretation is also possible. (3)
Anna öffnete das Fenster fünf Minuten lang. 'Anna opened the window for five minutes.'
While a continuous reading of the VP das Fenster öffnen ('open the window') seems to be adequate only under particular contextual conditions, an iterative interpretation is quite feasible. If, however, such an understanding is not explicitly suggested by the context, sentence (3) will have a clear preference for a third kind of interpretation on which the adverbial specifies the duration of the state brought about by the event of opening the window. In this case, (3) conveys that Anna opened the window and the resulting state of its being open lasted at least five minutes. This reading is represented in (3a), where s is used as a variable for states, RES and HD as predicates for the relations 'RESulting state of and 'the HolDer o f , respectively.9
Flexibility in adverbal modification (3)
515
a. 3e [AG(anna, e) & OPEN(e) & TH(window, e) & 3s [RES(s, e) & + OPEN(s) & HD(window, s) & T(s) > 5min]]
Unlike the cases considered so far, (3) in this understanding requires the adverbal modifier to be shifted in its meaning in order to meet the conditions of the verbal expression. The use of durative adverbials such as fünf Minuten lang ('for five minutes'- a literal equivalent is missing in English, see below) has hardly been mentioned in the literature, which mostly deals with English.10 A first proposal for explaining such resulting state-related reinterpretations was made in Dolling (1998), which served as a starting point for Piñón (1999) who, however, argues against the necessity of a meaning transfer. Instead, Piñón assumes that the argument structure of a resultative verb like öffnen ('open') contains an additional state variable with which the durative adverbial can immediately link up in modification.11 As a consequence, it seems that the adverbials under review here might be treated analogously to adverbials like für fünf Minuten or their English equivalents for five minutes but have to be restricted to specifying the duration of resulting states. For several reasons, I consider such an approach unacceptable. First, this approach does not take into consideration the fact that the behaviour of a resultative verb may vary when modified by different resultoriented modifiers. For example, the verbs zerbrechen ('break'), essen ('eat') und zerstören ('destroy') are compatible with wieder ('again') in its restitutive reading, although the resulting state induced by these verbs cannot be temporally restricted by other durative modifiers. Since, according to Piñón's proposal, such verbs thus cannot have an additional state variable, it remains unclear how the modifier wieder could be linked up with such a variable.12 Second, Piñón's proposal should also account for cases in which the putative state variable in the argument structure is not required for adverbal modification. In fact, it includes the assumption that whenever the variable remains unused in this sense, then a special semantic operator takes over the linking. However, since this is evidently the standard case, the proposal requires an additional step in the compositional derivation of these verbs that can scarcely be motivated. Third, Piñón's approach reveals the need for a more general procedure that enables us to capture another type of occurrence of modifiers for which no verbal linking site exists. As we shall see, modifying expressions such as unauffällig ('unobtrusively'), elegant ('elegantly') or korrekt ('correctly') can be used in such a way as not to characterize the resulting state associated with the verb. Rather, these adjuncts relate to objects which emerge as
516
Johannes Dölling
a result of the respective event. Thus, it appears to be mistaken to assume that such resultative verbs have a further argument position for resulting objects. It will be shown later that the solution based on Dölling (1998) not only avoids the problems mentioned above but that it also has the advantage of being an instantiation of a more general approach. In sentence (4), the adverbial drei Wochen lang ('for three weeks') certainly neither implies that Jutta arrived three weeks late nor does it specify the duration of a single arrival of Jutta. (4)
Jutta kam drei Wochen lang zu spät an. 'Jutta arrived (too) late for three weeks.'
However, contrary to the cases of reinterpretation adduced above, in (4) it is highly improbable that the modifier is used to characterize a process of arriving late on any occasion within three weeks. Rather, in view of our standard experience, in the given use the achievement zu spät ankommen ('arrive late') should be understood in the habitual reading.13 Thus, (4) refers to a habitual state of Jutta which lasted at least three weeks and which was realized by repeated, but not immediately successive, situations of arriving late. (4)
a. 3s [HD(jutta, s) & Vb [REAL(b, s) - > ARRIVE_TOO_LATE(b)] & T(s) > 3weeks]
Here, b is a variable for borderline situations, or more simply, borders14, as characterized, for example, by the verb ankommen 'arrive' while REAL stands for the relation 'REALization o f . Let me now turn to the analysis of cases where time-span adverbials occur as modifiers of achievements, states or activities. Since, for example, den Gipfel erreichen ('reach the summit'), as well as ankommen, denotes a property of borders, in a sentence like (5), the adverbial in zwei Tagen ('in two days') cannot serve to modify the VP in its original meaning. (5)
Ede erreichte den Gipfel in zwei Tagen. 'Ede reached the summit in two days.'
However, sentence (5) can be understood such that Ede was the Agent of an event which ended within two days by Ede's reaching the summit and thus culminating in it. Using CULM as a predicate for the relation 'the CULMination o f , the content of (5) can be represented as in (5a).
Flexibility in adverbal modification (5)
517
a. 3e [AG(ede, e) & 3b [CULM(b, e) & REACH(b) & TH(summit, b)] & T(e) < 2days]
The core of this event-related, reading is that the VP den Gipfel erreichen changes from a predicate of borders into a predicate of events that terminate in such borderline situations.15 A sentence like (6) can be treated in a similar way. (6)
Sarah war in fünf Minuten wach. 'Sarah was awake in five minutes.'
(6)
a. 3e [TH(sarah, e) & 3s [RES(s, e) & + AWAKE(s) & HD(sarah, s)] & X(e) < 5min]
As follows from (6a), Sarah is characterized as the Theme of an event which results in her being awake within five minutes. This interpretation of (6) requires the expression wach sein ('be awake'), which originally denotes a property of states, to be changed into a predicate of events that may have the pertinent resulting state.16 It is somewhat more complicated to assign an event-related interpretation to a sentence like (7). (7)
Peter rannte in fünfundvierzig Sekunden. 'Peter ran in forty-five seconds.'
Here, it would be necessary to construe the process predicate rennen ('run') as a predicate that can describe an event, the developmental phase of which is formed by a quantum of the running process. Then, the content of (7) can be identified with (7a) where the predicate SUBST denotes the relation 'SUBSTratum of between processes and events. (7)
a. 3e [AG(peter, e) & 3p [SUBST(p, e) & RUN(p) & AG(peter, p)] & T(e) < 45 sec]
Obviously, such an understanding is justified only in contexts from which a suitable culmination can be drawn - in (7) by way of identifying a certain running distance.
518
Johannes
Dölling
3. Reinterpretation by sort coercion? Meaning transfers that occur in connection with modification by durative or time-span adverbials have already been documented more or less extensively in the literature, and various proposals for their explanation have been advanced. Basic deliberations can be found in Moens and Steedman (1988), where a first systematic, albeit informal, analysis of reinterpretations in temporal modification is presented. Moens and Steedman analyze temporal adverbials (as well as aspectual auxiliaries) as functions which, under particular conditions, induce changes in the meaning of the verbal expressions they modify. The change involves that the verb's reference to situations of one sort gets transformed into a reference to situations of another sort. Such meaning adaptations based on a correspondingly differentiated network of ontological relationships is referred to as type coercion.17 How the relevant shifts are to be accomplished in detail, however, still requires explication. It can be assumed that such adverbials trigger semantic operations by means of which the verbal expressions are directly reinterpreted in a suitable way thereby creating the prerequisites for suitable modifications. So, if a conflict arises between the sortal selection restrictions of an adverbal modifier and the semantic sort of its argument, a specific operator applies to the verbal predicate to achieve sort matching. For example, the reinterpretation that takes place in (2) can be explained simply thus: utilizing a special coercion operator which meets the requirements of the adverbial, the meaning of the VP den Roman lesen gets transferred from a predicate of events to a predicate of processes. However, this mechanism of direct semantic adaptation leaves a number of questions unsettled. As discussed with respect to (2) above, the occurrence of a sortal conflict between temporal adverbial and verbal expression does not at all predetermine the form of its solution by the underlying conceptual ontology. A first problem is how, out of the set of conceptually possible operators, and in both systematic and economical a way, we can pick out exactly those operators that will each time provide the adequate reinterpretations. It is obvious that this choice cannot be made without resorting to world and discourse knowledge and without allowing for specific pragmatic restrictions. A second, and more serious, problem follows from the fact that by inserting such adaptation operators, contextually determined parts of meaning are introduced into an otherwise compositional derivation. Obviously, under this condition, the general validity of the principle of semantic compositionality cannot be upheld any longer.18 In view of the fact that we so far lack any convincing alternative to this principle, renouncing a strictly
Flexibility in adverbal modification
519
regulated calculation method of context-independent meaning of expressions is not acceptable. As a possible way out, some authors have offered a two-step approach according to which necessary reinterpretations are realized as follows: In a first step, a semantic representation of a given utterance is constructed in terms of compositionality. If, in the derivation of the context-independent meaning, a sortal conflict arises, it is resolved by inserting an underspecified coercion operator. In a second step, attempts are made to justify this hypothetic sortal adaptation by exploiting world and discourse knowledge in order to contextually specify the semantic representation. Thus, it is only in this step that a proper meaning transfer, if it is possible, gets realized.19 Taking up this idea for the reinterpretations discussed above, it seems that only two underspecified adaptation operators are required: one for constructions involving durative adverbials and one for those involving timespan adverbials. Given the sortal selection requirements of adverbials, the first operator should allow transferrai of predicates of events, borders or moments to predicates of processes or states; the second operator transfers predicates of borders, processes or states to predicates of events. These conditions are largely met by the operators proposed in (8) and (9) where e/b/m, p/s and b/p/s are provisional variables for situations of the respective supersorts, β varies over the quantifiers V and 3, C varies over the connectors & and —* and R is a parameter for relations between situations of various sorts. (8)
λΡλρ/s. ße/b/m [«(e/b/m, p/s) C P(e/b/m)]
(9)
λΡλβ. ßb/p/s [7?(b/p/s, e) C P(b/p/s)]
Now, if, for example, (8) is used in the compositional construction of the semantic representation of (10), the structure given in (10') can be assumed to be the result of this derivation (further provisionals included). (10) Ilse spielte die Sonate einen Tag lang. 'Ilse played the sonata for one day.' (10') Bp/s [AG/HD(ilse, p/s) & ße [Ä(e, p/s) C PLAY(e) & TH(sonata, e)] & T(p/s) > lday] Then, conceptually possible specifications of (10') will result in (10a) to (10c).
520
Johannes Delling
(10) a. 3p [AG(ilse, p) & Ve [CONST(e, p) —> PLAY(e) & TH(sonata, e)] & τ(ρ) > lday] b. 3p [AG(ilse, p) & 3e [COMPL(e, p) & PLAY(e) & TH(sonata, e)] & τ(ρ) > lday] c. 3s [HD(ilse, p) & Ve [REAL(e, s) PLAY(e) & TH(sonata, e)] & T(s) > lday] Which of the alternatives eventually provides the specified content of an utterance of (10), i.e. whether it refers to a process of successively repeated playing the sonata concerned (iterative reading), to a process being only part of an individual playing event (continuous reading) or to a state realized by repeated but not uninterrupted playing the sonata (habitual reading) has to be decided depending on world knowledge and other contextual informa.· 20 tion. However, an approach like this, in which semantic sort adaptation and context-related reinterpretation are separated, will also lead to difficulties. First, it has to meet the condition that meaning transfers can proceed only in one direction.21 The inadequacy of this condition emerges from sentences like (3) where in addition to the reinterpretation of the verbal expression, reinterpretation of the modifying expression is also possible. Therefore, the starting point of a required meaning transfer is not at all clearly determined a priori. Thus, it has to be decided to which of the involved expressions an adaptation operator is to be applied. However, decisions of this kind are not compatible with a strictly compositional semantic derivation because they require extra-linguistic knowledge to be taken into account. Second, under this approach, the fact that not every meaning transfer in adverbal modification has to result from a direct conflict of sorts is left out of consideration.22 For example, in (11), joggen ('jog') fulfills the sortal selection restriction of durative adverbials insofar as this verb represents a predicate of a process. (11)
Renate joggte zehn Jahre lang. 'Renate jogged for ten years.'
Accordingly, (11) can mean that Renate's activity of incessant jogging lasted at least ten years. (11) a. 3p [AG(renate, p) & JOG(p) & τ(ρ) > lOyears]
Flexibility in adverbal modification
521
Unless the person in question has extraordinary abilities, our standard experience of jogging will make us sceptical about the justification of this process reading. Therefore, it has to be concluded that (11) refers to Renate's state as realized by suitable activities of jogging and asserting that this state lasted ten years. Such a habitual interpretation is represented in (lib). (11) b. Bs [HD(renate, s) & Vp [REAL(p, s) & T(s) > lOyears]
JOG(p) & AG(renate, p)]
Sentence (11) can be construed in this way only if the verb is subject to a shift from a predicate of processes to a predicate of (habitual) states.
4.
Reinterpretation as enrichment of the Inflected Semantic Form
I will now develop an approach that, unlike previous attempts, might be considered adequate from the perspective of both content and methodology. In particular, the analysis to be proposed has to meet the following, partly interrelated, requirements: First, in keeping with the present state of research, any treatment of adverbal modifications should strictly obey the principle of semantic compositionality. Second, reinterpretations in modifying adjuncts should not be reduced to just those which display conflicts between the semantic sorts of the expressions involved. Third, a mechanism as general as possible should be provided by means of which meaning transfers of both modified expressions and modifiers can be performed. Investigating various kinds of systematic meaning variation, I have developed an approach that meets these requirements, see Dölling (1997) and subsequent works. The basic idea of the multi-level model is that the conceptual information an utterance may convey has to be spelled out over several levels of meaning representation.23 My assumption is that the starting point of understanding an utterance is formed by the level of its Semantic Form (SF) representing the contextindependent meaning of the utterance.24 There are two crucial properties of this kind of conceptual structure: First, SF representations are built up strictly compositionally, i.e. they are calculated exclusively in accordance with the syntactic structure of the expressions involved. Thus, any interference with the autonomously organized semantic structure by making reference to elements of extra-linguistic knowledge is excluded. Second, SF representations are radically underspecified insofar as they contain a lot of parameters, whose contextually determined fixing allows for considerable
522
Johannes Dölling
variation in the meaning of utterances. It is crucial to my approach that such SF parameters occur not only as elements of semantic entries of lexical units. Rather, in semantic composition, this primary variation potential of meaning is systematically extended by adding further SF parameters under precisely defined conditions. Given this, two subtypes of SF can be distinguished. (12)
a. The Basic Semantic Form SFB of an expression is that SF which is associated with a lexical item or with a syntactically complex expression as a direct result of the meaning combination of its parts. b. The Inflected Semantic Form SF¡ of an expression results from its SF b by introducing additional parameters by means of operations - so-called SF inflections - that are obligatorily applied on the given SFB.
As will be shown, it is the extended variation potential given by SF[ that allows for the type of meaning transfers considered here. 25 The SF of an utterance provides the basis to which various kinds of interpretative operations may apply. By means of the latter, the SF gets step by step contextually specified by exploiting world and discourse knowledge and by utilizing pragmatic principles. What emerges at the end of this process is what I call the Prepositional Content (PC) of the utterance. In calculating this level of meaning representation, the procedure of abductive interpretation plays a major role. According to this procedure, the information conveyed by the utterance is reconstructed by 'explaining' its SF via deduction from a suitable conceptual knowledge base. 26 An intermediate result of this derivation is what might be called a Parameter-fixed Structure (PFS) of this utterance. It is conceived as a level of meaning representation which immediately succeeds the level of SF but differs from it in that the parameters of SF are now replaced by actually instantiable conceptual units. Thus, PFS results from the first step in the course of specifying the contextindependent meaning of the utterance concerned. The conceptual structure provided by PFS represents the very level on which systematic meaning variations take place. Let me now characterize the operators to be used in SF inflection in more detail. In the papers mentioned above, I have advanced several proposals in search of schemata that, on the one hand, are sufficiently specific to furnish the necessary salient points for the PFS desired and that, on the other, are general enough to cover all cases of systematic meaning transfer observed so far. The inflection operator met proposed in Dölling (2000)
Flexibility in adverbal modification
523
seems to be a suitable candidate for reinterpreting any expression of the type by fixing the parameters that have been introduced by SF inflections.27 In particular, the operator met enables us to treat the reinterpretations in adverbal modification as cases of contextual enrichments which apply to inflected semantic forms SFj. For expository reasons, I will use the operator met in a slightly simplified version. In (13) I present the inflection operator met', where χ and y are individual variables and Qn, Cn and Rn are parameters for the quantifiers 3 and V, for the connectors & and —» and for relations between elements of ontological sorts, respectively.28 (13) met'.
λΡλχ. Qny [Rn(y, x) C„ P(y)]
According to condition (14), met' is to be applied to every one-place predicate that occurs as SFB of an expression a. (14) SFB(OC) of type has to be transferred to SF¡(A) such that holds: SFI(A) = MEI'(SF B (A)).
The fixing conditions of SF! of a given in (15) determine in which way special parameters are substituted for the SF parameters introduced by means of met': (15) SFi(a) can be specified to PFS(a) as follows: (i)
Qn and C„ in SFj(a) are fixed by Ξ and & or by V and —», respectively;
(ii) Rn in SF[(a) is fixed by = or by some other general relation holding between elements of two ontologically distinct sorts; (iii) in the case of default, Qn, C„ and Rn are fixed by 3, & and =, respectively. Condition (iii) warrants that whenever there is no reason for a meaning transfer of a, the contribution of met' to the interpretation of α in PFS amounts to zero.
524 5.
Johannes Dölling Demonstration of a reinterpretation
I will now illustrate the application of the inflection operator met' and the possibilities of fixing its parameters by sentence (3), repeated as (16) here. (16) Anna öffnete das Fenster fünf Minuten lang. 'Anna opened the window for five minutes.' In (17a), the relevant parts of the SF derivation for the VP fünf Minuten lang das Fenster öffnen ('open the window for five minutes') are shown. (17) a. das Fenster öffnen-, SFB: λχ. OPEN(x) & TH(window, χ)
I I met'. λΡλχ. Qiy [Rj(y, x) C, P(y)] 1/ das Fenster öffnen- SF^ λχ. Q¡y [R¡(y, χ) C¡ OPEN(y) & TH(window, y)] fünf Minuten lang·, SFB: λχ. τ(χ) > 5min
I I met'·. λΡλχ. Q2y [/?2(y, x) C2 P(y)] 1/ fünf Minuten lang-, SFi: λχ. Q2 y [^(y, x) C2 X(y) > 5min]
I I MOD. λζ)λΡλχ. P(x) & Q(x) 1/ fünf Minuten lang·, SF: λΡλχ. P(x) & Q2 y [Ä2(y, x) C2 T(y) > 5min] / fünf Minuten lang das Fenster öffnen·, SFb: λχ. Qiy [Rj(y, X) C, OPEN(y) & TH(window, y)] & Ö2y [fi2(y, χ ) C2 T(y) > 5min] met'. λΡλχ. Q3y [R¿y, χ) C3 P(y)] / fünf Minuten lang das Fenster öffnen; SFj: λχ. Q3y [R3(y, x) C3 Q¡ζ [Rj(z, y) Cj OPEN(z) & TH(window, ζ)] & Q2ζ [R2(z, y) C2 τ(ζ) > 5min]]
Flexibility in adverbal modification
525
(17a) invites some comments. First, the derivation shows that for SF, a representation format in which no sorted individual variables and, hence, no variables for situation sorts occur, is to be preferred.29 The differentiation of ontological sorts is accounted for in terms of axioms for the constants involved. Second, the three occurrences of met' indicate that in the SF derivation exactly as many predicates appear in the role of a SFB and therefore, in agreement with (14), require an according number of operator applications. The last application of met' takes place for the sake of completeness only. SF parameters that are introduced by met' are relevant to the possible reinterpretations of the results of modification but not to those of their components. Third, a special operator MOD for type coercion is used. By means of MOD, expressions of the type predicate can be transferred to predicates of the type modifier. In this sense, the application of the MOD operator forms a crucial condition for realizing adverbal modifications.30 Starting from the result of (17a), we may assume a SF for sentence (16) as represented in (16a), which is simplified in several respects. (16) a. SF: 3x [ θ (anna, x) & Q3y [J?5(y, x) Cj Q,ζ [R,(z, y) C¡ OPEN(z) & TH(window, z)] & g 2 z [R2(z, y) C2 T(z) > 5min]]] θ is an additional SF parameter which has to be fixed by participation relations like AG, HD or TH. The parameter θ is part of the particular coercion operator SUBJ as given in (18). (18) SUBJ:
λΡλγλχ.
θ (y, χ) & Ρ(χ)
My assumption is that SUBJ serves to extend the SF of the given verbal expression by one argument place for the SF of grammatical subjects.31 The compositionally calculated SF of (16), that is (16a), has now to be interpreted against the background of contextual knowledge (in the broadest sense); the first step in doing this consists in fixing the parameters that occur in (16a). Obviously, the knowledge required to do this is obtained from diverse sources. First of all, it includes axioms like (19a), (19b) and (20), which provide the conditions of use for more specific conceptual units and configurations. (19) a. D V x [OPEN(x)
3yAG(y,x) & ΞζΤΗ(ζ,χ)]
b. D V x [OPEN(x)
3y [RES(y, x) & + OPEN(y)]
(20)
D V x [3y [τ(χ) > y] —» EVENT(x) ν STATE(x)]
526
Johannes
Dölling
Thus, (19a) characterizes every opening as an eventuality that involves an Agent and a Theme as participants; (19b) lays down that every opening implies a resulting state of being open. The axiom in (20) may be considered a condition which restricts the use of durative adverbials. Moreover, certain axioms of conceptual ontology are required as well. They serve to characterize the basic properties and relations of various sorts of situations. Such general fixings may have the form of, for example, (21a)-(21d) or (22a) and (22b). (21) a. b. c. d.
DVxVy [RES(x, y) - » STATE(x) & CHANGE(y)] D V x [CHANGE(x) 3y [STATE(y) & RES(y, x)]] D V x [CHANGE(x) EVENT(x)] DVxVyVz [RES(x, y) & (TH(z, y) ν AG(z, y) & -3z TH(z, y)) HD(z, x)]
(22) a. DVxVy [CONST(x, y) PROCESS(x) & EVENT(y)] b. D V x [EVENT(x) 3y [PROCESS(y) & CONST(y, x)]] Axioms (21a)-(21c) define a change as an event such that there is a certain state resulting from it.32 From (19a) and (21b), it follows that OPEN and + OPEN are predicates of changes and of states, respectively. Axiom (21d) in connection with (21a) - ensures that the Theme or - if there is no Theme - the Agent of a change is also the Holder of its resulting state. Finally, (22a) and (22b) give a first characterization of the relation of constitution which links processes with events. Needless to say, if we were to describe the interpretation at issue in more detail, further axioms would have to be added as part of the conceptual knowledge base.33 Given a sufficient number of fixings of the kind outlined above, we obtain the following conceptually possible specifications of SFj of fünf Minuten lang das Fenster öffnen·. (17) b. PFS i : = c. PFS2: =
λχ. 3y [=(y, x) & 3z [=(z, y) & OPEN(z) & TH(window, ζ)] & 3z [RES(z, y) & τ(ζ) > 5min]] λχ. OPEN(x) & TH(window, χ) & 3y [RES(y, x) & T(y) > 5min] λχ. 3y [=(y, x) & Vz [CONST(z, y) - > OPEN(z) & TH(window, ζ)] & 3z [=(z, y) & τ(ζ) > 5min]] λχ. Vy [CONST(y, x) OPEN(y) & TH(window, y)] & τ(χ) > 5min
Flexibility in adverbal modification d. P F S 3 :
= e. P F S 4 :
=
527
λ χ . 3 y [=(y, Χ) & ΞΖ [ C O M P L ( Z , y) & O P E N ( Z )
& TH(window, y)] & Ξζ [=(z, y) & τ(ζ) > 5min]] λχ. 3y [COMPL(y, x) & OPEN(y) & TH(window, y)] & τ(χ) > 5min λ χ . 3 y [=(y, X) & V Z [ R E A L ( z , y ) - > O P E N ( Z )
& TH(window, y)] & 3z [=(z, y) & τ(ζ) > 5min]] λχ. Vy [REAL(y, x) ->· OPEN(y) & TH(window, y)] & X(x) > 5min
Each of these PFSs involves a meaning transfer in one of the two components of the verb-adverbial construction: in PFSj, the adjunct fünf Minuten lang is reinterpreted as a predicate of resulting states, in PFS2 to PFS4, the VP das Fenster öffnen is reinterpreted as a predicate of processes - either in terms of iterations (17c) or of developmental phases of events (17d) or as a predicate of (habitual) states (17e). Of course, given our everyday knowledge of typical events like opening a window and of the situations connected with them, the four possible specifications differ in likelihood. So, an interpretation of (16) as yielded by (17e) can be ruled out under normal conditions. Interpretations that draw on (17c) or (17d) seem to be more probable but still marginal. As mentioned above, under normal conditions we will prefer the interpretation of the VP fünf Minuten lang das Fenster öffnen that is represented in (17b). Thus, the PFS to be assumed for (16) as the most likely one is the one given in (16b). (16) b. PFS: 3x [AG(anna, x) & OPEN(x) & TH(window, x) & 3y [RES(y, x) & T(y) > 5min]] Beside the parameters introduced into the PFS of (16) by fünf Minuten lang das Fenster öffnen, θ is fixed as Agent due to (19a). After further steps of enrichment which, among others, include backtracking to axioms like (19b) and (2 Id), the process of interpretation ends in yielding the propositional content PC of (16). In a somewhat simplified form, this PC can be identified with the structure in (16c). (16) c. PC:
3x [AG(anna, x) & OPEN(x) & TH(window, x) & 3y [RES(y, x) & +OPEN(y) & HD(window, y) & T(y) > 5 min]]
528
Johannes
Dölling
Compared with (16b), the structure (16c), which represents the fullspecified meaning of the utterance (16), is determined more exactly in that (i) the resulting state is shown to be that of being open; and (ii) the holder of this state is shown to be identical with the Theme of the change (cf. also (3a)).
6.
Further adverbal modifications with reinterpretation
Contrary to what is generally assumed in the literature, reinterpretations of the kind under consideration are not confined to modification by temporal adverbials. First, it has to be stated that the use of manner adverbials may also involve a transfer of meaning of the verbal expression modified. For example, in analogy to the interpretation of (5), sentence (23) has to be understood as characterizing a change, the Agent of which was Claudia and the culmination of which is Claudia's finding the flat. (23)
Claudia fand die Wohnung schnell. 'Claudia found the flat quickly.'
Thus, for simplicity's sake again using sorted variables as a means of representation, the structure in (23a) can be assumed to be the PFS of (23).34 (23) a. PFS: 3c [AG(claudia, c) & 3b [CULM(b, c) & FIND(b) & TH(flat, b)] & QUICK(c)] It is part of the interpretation of (23) that - as a result of enriching its SFj the VP die Wohnung finden ('find the flat') denotes, deviating from its original meaning, not a property of borders but a property of changes, as noted in (24b). (24) a. SFi: b. PFS:
λχ. Qky [Rk(y, x) Ck FIND(y) & TH(flat, y)] λβ. 3b [CULM(b, c) & FIND(b) & TH(flat, b)]
It is only under such a precondition that schnell ('quickly') in (24) can reasonably be used as a manner adverbial. A meaning transfer of the modified expression can also be observed in sentences in which an instrumental PP occurs as adverbal modifier, as in (25).
Flexibility in adverbal modification 529 (25) Stefan war mit dem Auto in der Stadt. 'Stefan went to town by car.' (lit. Stefan was with the car in the town) In parallel with the interpretation of (6), (25) asserts a change that results in a state of being in town. 5 More specifically, (25a) and (25b) can be considered the PFS and the PC of (25), respectively, where INSTR denotes the relation 'instrument o f . (25) a. PFS: 3c [AG (stefan, c) & 3s [RES(s, c) &+IN_THE_CITY(s)] & INSTR(car, c)] b. PC: 3c [AG (stefan, c) & 3s [RES(s, c) & +IN_THE_CITY(s) & HD(stefan, s)] & INSTR(car, c)] In order to specify Stefan's state indirectly by determining the instrument used for its coming about, the copula-predicative construction in der Stadt sein ('be in town') has to be shifted from a predicate of a state to one of a change. In addition, in (25b), the fact that Stefan is the holder of the state is inferred on the basis of axiom (2Id). Unlike (25), sentence (26) below is an example in which an originally change-related or process-related PP, in this case the manner adverbial mit Begeisterung ('with enthusiasm'), is reinterpreted in such a way that it is compatible with an expression that denotes a property of states. (26) Peter war mit Begeisterung Angler. 'Peter was an angler with enthusiasm.' Accordingly, Peter was in a habitual state of being an angler such that he performed the events or processes that realize this state with enthusiasm. Using e/p as provisional variable for events and processes, (26) then has the following PC: (26) a. PC:
3s [HD(peter, s) &+ANGLER(s) & Ve/p [REAL(e/p, s) - > WITH_ENTHUSIASM(e/p) & AG(peter, e/p)]]
Based on a suitable fixing of the SF parameters that occur in (27a), the PP mit Begeisterung ('with enthusiasm') contributes the PFS given in (27b).
530
Johannes
(27)
a. SFÍ
Dölling
λχ. Qky [Rk(y, χ) Q WITH_ENTHUSIASM(y)]
b. PFS: Às. Ve/p [REAL(e/p, s)
WITH_ENTHUSIASM(e/p)]
In what follows, it will become clear that reinterpretations of adverbal modifiers are not at all exceptional. Most of the examples below are cases in which the meaning of the modifying adjunct is subject to various kinds of transfer. In Eckardt (1998) it is argued that sentences like (28) and (29) do not just permit an interpretation according to which the manner adverbial specifies the described event as unobtrusive and elegant, respectively. (28) Anna frisierte Max unauffällig. 'Anna dressed Max's hair unobtrusively.' (29) Maria kleidete Hans elegant. 'Maria clothed Hans elegantly.' The adverbials at issue can also specify a result that is achieved by the action concerned. It seems to be obvious that they are interpreted, in analogy to the temporal adverbial in (3), as making a predication about states.36 If so, the second interpretation of (28) would imply that Anna dressed Max's hair and that Max's state resulting from this was unobtrusive. However, this assumption cannot be upheld as the analysis of sentences (30) and (31) shows.37 (30) Der Student übersetzte den Brief korrekt. 'The student translated the letter correctly.' (31)
Die Bibliothekarin stapelte die Bücher ordentlich. 'The librarian piled up the books neatly.'
Clearly, (30) does not mean that the letter was in a correct state as a result of its translation by the student concerned. Rather, (30) conveys that the translation of the letter that resulted from this event, i.e. the object produced in this way, was correct.38 Supposing that RES_OBJ stands for the relation 'the resulting object of and that o is a variable for objects, such a resulting object-related interpretation of the adverbial korrekt ('correctly') will yield the PFS given in (30a). (30) a. PFS: 3c [AG(student, c) & TH(letter, c) & TRANSLATE(c) & 3o [RES_OBJ(o, c) & CORRECT(o)]]
Flexibility in adverbal modification
531
In a similar way, the adverbials unauffällig ('unobtrusively'), elegant ('elegantly') and ordentlich ('neatly') may assert a property of Max's hair-style (28), of Hans's clothing (29) or of the pile of books (31).39 But to be applicable as modifiers, these object predicates have to become predicates of changes at the level of PFS in order to be provided with a suitable site of application in the meaning structure of the sentences in question. Suppose that, in (32a), the SFi of unauffällig indicates that the predicate UNOBTRUSIVE is left unspecified wrt. its applicability to objects or to situations. (32) a. SF¡: λχ. Qky [Rk(y, χ) Ck UNOBTRUSIVE(y)] Then, via specification, at least two PFSs can be obtained for this expression, on which the two possible interpretations of (28) can be based. (32) b. PFS,:Àc. 3 c ' [ = ( c ' , c ) & UNOBTRUSIVE(c)] =
Àc. U N O B T R U S R V E ( c )
c. P F S 2 : AC. 3O [RES_OBJ(O, C) &
UNOBTRUSIVE(o)]
Since in (32b) the contribution of met' to the interpretation is reduced to zero, it is only the change described by (28) which is left to be characterized by the adverbial unauffällig. In contrast, (32c) involves a transfer of unauffällig from a predicate of an object to a predicate of a change, which enables the adverbial to specify the object resulting from this change. To sum up, in terms of RES_OBJ all sentences in (28) to (31) make reference to an ontological relation by means of which object-related predicates that surface as adverbal modifiers can be interpreted properly. We will see below that quite a number of similar cases of reinterpretation in adverbal modification have to be taken into account. Findings presented in Maienborn (2001, this volume) prove that not all adverbial occurrences of locative PPs may be interpreted as localizing the situation to which the respective sentence immediately refers. Sentence (33), for example, can be understood in two ways. (33) Die Bankräuber flüchteten auf Fahrrädern. 'The bank robbers escaped on bicycles.' (33) can be understood as a description of a bizarre scenario in which the bank robbers escaped not by cycling but while being situated on (perhaps oversized) bikes - the bikes as such not being relevant to the escaping. In
532
Johannes Dölling
addition to this situation-related interpretation, there is another reading, which in view of our everyday knowledge obviously has to be preferred, namely the object-related interpretation on which the locative modifier auf Fahrrädern ('on bicycles') specifies the bank robbers' location in a way that is relevant to their act of escaping. The two interpretations of (33) can be represented by the following PFSs: (33) a. PFSj: 3p [AG(robbers, p) & ESC(p) & 3p'[=(p',p) & LOCQN(P'. bicycles)]]
= 3p [AG(robbers, p) & ESC(p) & LOC ON (p, bicycles)] b. PFS2: 3p [AG(robbers, p) & ESC(p) & 3o [AG(o, p) & LOCON(°> bicycles)]]
As we may rule out the case that two different objects play the role of the same participant, the identity of the localized Agents can be directly inferred from PFS2. Moreover, as demonstrated by Maienborn, the objectrelated interpretation of the modifier permits us to infer, due to additional axioms, that the bikes served as instruments of escape. Thus, the structure in (33c) can be assumed to render the PC 2 of (33). (33) c. PC2: 3p [AG(robbers, p) & ESC(p) & LOCON(robbers, bicycles) & INSTR(bicycles, p)] The second interpretation of sentence (33), however, is possible only by transferring auf Fahrrädern from an object predicate to a predicate of processes. In analogy to the sentences considered before, such a reinterpretation supplies the precondition for combining the object-localizing occurrence of the adverbial PP with the verb flüchten ('escape'). 40 (34) exemplifies a sentence that allows at least three different readings of the locative PP used as a modifier.41 (34) Der Koch bereitete das Hähnchen in einer Marihuana-Tunke zu. 'The cook prepared the chicken in a marijuana sauce.' Again, the PP in einer Marihuana-Tunke ('in a marijuana sauce') can be seen as an adverbial of localizing the event to which (34) refers. In this case, the PFS concerned permits, depending on the conceptual knowledge involved, alternative inferences as to whether only the chicken or - under quite bizarre circumstances - also the cook is localized at the given place as objects participating in the event. Second, the modifying expression can be
Flexibility in adverbal modification
533
considered to relate exclusively to the chicken. Thus, the Theme, but not the entire situation, is localized in a marijuana sauce. Additionally, in analogy to the second interpretation of (33), the local content of the PP is overlaid with an instrumental reading that specifies the medium of preparation. Third, there is also the possibility to construe the PP as specifying the place the cook took while preparing the food.42 Note that the latter objectlocalizing interpretation of the modifier belongs to what is usually subsumed under the term secondary predication. Before turning my attention to this issue, I will briefly discuss a case in which a directional PP occurs as adverbal modifier. (35) Fred tanzte in den Saal. 'Fred danced into the hall.' Sentence (35) refers to a change that is performed by Fred and that results in Fred's being in the hall. Here, the PP in den Saal ('into the hall') accomplishes the task of providing the process of dancing with a result state and thus of transferring it to a change.43 The PFS given in (36) can be assumed to be the meaning representation of this adverbial. While the second conjunct represents its locative part, i.e. 'being in the hall', the first one stands for its resulting part.44 (36) PFS: Àc. 3s [RES(s, c) & Vo[HD(o, s)
L O C ^ o , hall)]]
As can be seen from (36), the modifying combination of the directional PP with tanzen ('dance') requires the verb - in parallel with, for example, rennen in sentence (7) - to become a change predicate in the course of specification of its SFj. The PFS arising therefrom can be identified with the structure given in (37). (37) PFS: Xc. 3p [SUBST(p, c) & DANCE(p)] If, in addition, the parameter θ gets fixed by AG, we obtain the following PFS for (35): (35) a. PFS: 3c [AG(fred, c) & 3p [SUBST(p, c) & DANCE(p)] & 3s [RES(s, c) & Vo[HD(o, s) - > L O C ^ o , hall)]]]
534
Johannes Dölling
The remaining parts of the PC of (35), including the assertion that Fred is both Agent of the dancing process and Holder of the state of being in the hall, can be inferred from axiom (21d) in conjunction with additional axioms for DANCE and SUBST.
7. Secondary predications as adverbal modifications Following current views, the semantic difference between a secondary predicate and an adverbial is based on the condition that the former, in contrast to the latter, does not relate directly to a verbal expression but to a DP in the sentence.45 Two subtypes of secondary predicates are distinguished: Depictive predicates refer to an additional property which pertains to one of the participants during the situation denoted by the verb; resultative predicates, however, refer to a state which results from the event covered by the verb. Examples of sentences containing secondary predications are (38) to (40). (38) Der Koch bereitete das Hähnchen roh zu. 'The cook prepared the chicken raw.' (39) Der Koch bereitete das Hähnchen betrunken zu. 'The cook prepared the chicken drunk.' (40) Der Koch bereitete das Hähnchen knusprig zu. 'The cook prepared the chicken crisp.' Under standard conditions, roh ('raw') in (38) is used as a depictive predicate that relates to the grammatical object, betrunken ('drunk') in (39) as one that relates to the grammatical subject, whereas knusprig ('crisp') in (40) is used as a resultative predicate that relates to the grammatical object. The remaining part of the paper will outline how secondary predications can be treated within the model of multi-level meaning representation. Starting with an analysis of depictives, let me first consider sentence (38) that, as an approximation, can be paraphrased by (38'). (38') Während der Koch das Hähnchen zubereitete, war es roh. 'While the cook was preparing the chicken, it was raw.' It is crucial for the understanding of (38) that the characterization of the chicken as being raw does not just refer to a state that covers the temporal
Flexibility in adverbal modification
535
interval required for preparing the chicken but, more precisely, to a state that is to be considered a concomitant circumstance (of secondary import) to this process. Using CIRC as a shorthand predicate denoting this particular relation, the structure given in (38a) represents the PC of (38). (38) a. PC:
3c [AG(cook, c) & PREPARE(c) & TH(chicken, c) & 5s [CIRC(s, c) & HD(chicken, s) & Vo [HD(o, s) RAW(o)]]]
The axioms (41) and (42) hold, among others, for CIRC, whereby ν is a variable ranging over situations. (41) (42)
DVsVv [CIRC(s, v) x(s) D τ(ν)] ClVsVvVo [CIRC(s, v) & (AG(o, ν) ν TH(o, ν) ν HD(o, v)) HD(o, s)]
Now, how can (38a) be derived? In what follows, I assume that depictive predications can be considered adverbal modifications, in which reinterpretations of the expression used as a modifier occur as the general rule and, hence, quite regularly.46 Related to (38), this implies that roh ('raw') is combined with the verb zubereiten ('prepare') in a modifying way and, for this reason, is shifted in the process of parameter fixing from an object predicate to a change predicate. Supposing (43a) as PFS of the adjective in its primary meaning, (43b) shows the PFS of roh resulting from this meaning transfer.47 (43) a. PFS: λο. RAW(o) b. PFS: Xc. 3s [CIRC(s, c) & Vo [HD(o, s)
RAW(o)]]
It is evident that a SFi representation obtained by means of the meVoperator used so far will not be sufficient as the basis for a more complex transfer. Therefore, a revision of (13) is inevitable. In approximation to the more general operator of SF inflection developed in Dölling (2000a), the complex character of which is accounted for by the occurrence of chains of métonymie interpretation, the operator met" given in (44) shall therefore be used below. (44) met".
λΡλχ. 3y [R2n(y, x) & Qnζ [R1n(z, y) Cn P(z)]]
536
Johannes Dölling
While the application condition of the inflection operator agrees with that assumed in (14), the conditions of parameter fixing for met" in (15) have to be modified in such a way that in transition to PFS, two parameters R'N and R2n can now be suitably fixed by = or by another general relation holding between elements of two ontological sorts. Such a use of met" does not lead to any problems in the cases considered earlier since the contribution of the newly introduced components will prove empty at the PFS level there. As can be seen from (43c), the SFj of roh, which is derived compositionally with met", contains all parameters required for the interpretation. (43)
c. SFj:
λχ. 3y [tf2*(y, x) & Qkz [R\(z, y) CK RAW(z)]]
In analogy, this holds for the SF of the entire sentence (38) that - again highly simplified - can be given with (38b). (38)
b. SF:
3x [θ (cook, x) & PREPARE(x) & TH(chicken, x) & 3y [R\(y, x) & Qkζ [R'k(z, y) Ck RAW(z)]]]
After fixing all remaining SF parameters, the following PFS results for (38): (38) c. PFS: 3c [AG(cook, c) & PREPARE(c) & TH(chicken, c) & 3s [CIRC(s, c) & Vo [HD(o, s) RAW(o)]]] Finally, the PC given by (38a) is obtained by axiom (42), which makes the chicken - as the Theme of preparing - into the holder of the state that is concomitant with the preparation. The type of depictive exemplified by (39) differs from the one considered above only in the fact that now the object denoted by the grammatical subject, but not by the grammatical object, is the holder of the state in question. Thus, (39a) can be assumed to be the PC of (39). (39)
a. PC:
3c [AG(cook, c) & PREPARE(c) & TH(chicken, c) & 3s [CIRC(s, c) & HD(cook, s) & Vo [HD(o, s) - > DRUNK(o)]]]
A consequence of this difference, which is confined to PC, is that a sentence like (45) has only one PFS, although it permits two interpretations of the depictive predicate traurig ('sad') - one interpretation related to the subject DP and one to the object DP.
Flexibility in adverbal modification 537 (45) Peter verließ Maria traurig. 'Peter left Mary sad.' (45) a. PFS: 3b [AG(peter, b) & LEAVE(b) & TH(maria, b) & 3s [CERC(s, b) & Vo [HD(o, s) - > SAD(o)]]] Accordingly, the PC given in (45b) as well as that given in (45c) is deductively derivable from (45a) by means of axiom (42). (45) b. PCi: 3b [AG(peter, b) & LEAVE(b) & TH(maria, b) & 3s [CIRC(s, b) & HD(hans, s) & Vo [HD(o, s) SAD(o)]]] c. PC2: 3b [AG(peter, b) & LEAVE(b) & TH(maria, b) & 3s [CIRC(s, b) & HD(maria, s) & Vo [HD(o, s) SAD(o)]]] Which of the two possible PCs of (45) is to be viewed as adequate for the respective utterance is decided on the basis of additional knowledge of the situation referred to. An example for the use of a PP as a depictive predicate is given by the third interpretation of sentence (34) discussed in Section 6 and repeated here. (34) Der Koch bereitete das Hähnchen in einer Marihuana-Tunke zu. Suppose that the locative PP in einer Marihuana-Tunke in its objectlocalizing meaning has the following PFS: (46)
a. PFS: λ ο . LOCHM(O, marijuana_sauce)
Then, by fixing parameters occurring in the SFi, its depictive use is represented by (46b). (46) b. PFS: Xc. 3s [CIRC(s, c) & Vo [HD(o, s) LOCIN(O, marijuana_sauce)]]
As can be seen from (46b), it denotes a property of changes which have a concomitant state such that its holder is localized in a marijuana sauce.48 Finally, the PC representing the reading concerned indicates that in parallel with (39), the Agent of the change is also the holder of the respective state.
538
Johannes Dölling
(46) c. PC:
Xc. 3s [CIRC(s, c) & HD(cook, s) & Vo [HD(o, s) —> LOCnv(o, marijuana_sauce)]]
Like (38), sentence (47) involves a depictive predicate related to its grammatical object. (47) Maria aß das Brot in Scheiben. 'Maria ate the bread in slices.' Characterizing a state of the object as playing the role of the Theme of Maria's eating, the PP, however, includes an occurrence of in which has lost its locative meaning. The fact that in Scheiben ('in slices') is replaceable with the clear-cut manner adverbial scheibenweise ('slice by slice') does not just confirm the metaphoric use of the preposition. It also delivers an additional argument in support of my assumption that secondary predications can be viewed as a special kind of adverbal modification.49 Turning now to resultati ve predicates, I do not see any reason to treat this type of secondary predication in a principally different way. Such cases are evidently also instances of adverbal modification which, like depictives, always involve reinterpretations of the secondary predicate. For example, sentence (40) differs from (38) and (39) only insofar as knusprig ('crisp') does not specify a state that accompanies, but one that results from, the preparation of the chicken.50 Once more, the SF¡ of the adjective derived with met" contains all parameters the fixing of which supplies the resultative predicate. (48) a. SF¡:
λχ. 3y [tf2*(y, x) & Qkζ [r\(z,
y) Ck CRISP(z)]]
As can be seen from (48b), its PFS differs from that of a depictive predicate, above all, in involving RES in the place of CIRC. (48) b. PFS: Xc. 3s [RES(s, c) & Vo [HD(o, s) - > CRISP(o)]] The three levels of the meaning representation of (40) relevant to our purposes are given in (40a)-(40c). (40) a. SF:
3x [θ (cook, χ) & PREPARE(x) & TH(chicken, χ) & 3y [R2k(y, x) & Qkζ [r\(z,
y) Ck CRISP(z)]]]
Flexibility in adverbal modification 539 b. PFS: 3c [AG(cook, c) & & 3s [RES(s, c) c. PC: 3c [AG(cook, c) & & 3s [RES(s, c) & Vo [HD(o, s)
PREPARE(c) & TH(chicken, c) & Vo [HD(o, s) CRISP(o)]]] PREPARE(c) & TH(chicken, c) & HD(chicken, s) - > CRISP(o)]]]
The statement contained in PC, on the basis of which the Theme of the change is determined to also be the holder of its resulting state, follows again from axiom (21d). A case of a resultative predicate in which not only the modifying AP but also the modified verb is reinterpreted can be found in (49). (49) Gerda wischte den Tisch sauber. 'Gerda wiped the table clean.' Here, sauber ('clean') — in analogy to knusprig in (40) - is transferred from a predicate of states into a predicate of changes by fixing the parameters occurring in its SF r However, since wischen ('wipe') is one of those process verbs which, when connected with a quantized object DP, do not necessarily result in an accomplishment,51 the verb also has to be transferred to a predicate of change - in analogy to tanzen in (35). More specifically, by fixing the parameters in the SF¡ of wischen in (50a), we get the PFS given in (50b). (50) a. SFi: λχ. 3y [R2k(y, x) & Qkζ [r\(Z, y) Ck WIPE(z)]] b. PFS: Àc.3c'[=(c',c) & 3p [SUBST(p, c') & WIPE(p)]] = Xc. 3p [SUBST(p, c) & WIPE(p)] Finally, the structure in (48a) obtains as the PC of (48). (48) a. PC:
3c [AG(gerda, c) & TH(table, c) & 3p [SUBST(p, c) & WIPE(p) & AG(gerda, p) & TH(table, p)] & 3s [RES(s, c) & HD(table, s) & Vo [HD(o, s) CLEAN(o)]]]
Sentence (51) exemplifies the kind of resultative secondary predication using a locative PP.
540
Johannes
Dölling
(51) Peter versteckte das Geld in der Kiste. 'Peter hid the money in the box.' Suppose that in parallel with (46a), the structure in (52a) is the PFS of the PP in der Kiste ('in the box') in the object-localizing reading. (52) a. PFS: λο. LOC^o, box) Then, its resultative interpretation which emerges from meaning transfer in secondary predication is represented as follows: (52) b. PFS: Xc. 3s [RES(s, c) & Vo [HD(o, s)
L O C ^ o , box)]]
By recourse to axiom (21d), the money is identifiable as the holder of the resulting state. Thus, the structure given in (51a) can be assumed as the PC of (51). (51) a. PC:
3c [AG(peter, c) & HIDE(c) & TH(money, c) & 3s [RES(s, c) & HD(money, s) & V o [ H D ( o , s)
L O Q N ( O , b o x ) (o)]]]
Crucially, resultative predications by locative PPs have to be clearly distinguished from directional PPs used as arguments or as modifiers of verbal expressions. To take an example closely related to (51), consider sentence (53). M
(53)
Peter legte das Geld in die Kiste. 'Peter put the money into the box.'
The conceptual information conveyed by (53) is to be represented as follows: (53) a. PC:
3c [AG(peter, c) & PUT(c) & TH(money, c) & 3s [RES(s, c) & HD(money, s) & V o [ H D ( o , s)
L O Q N ( O , b o x ) (o)]]]
As can be seen from a comparison with (51a), the PCs of sentences (51) and (53) turn out to be structurally identical. But their derivations are distinct in one important respect. Since legen ('put') is an explicitly directional motion verb, it requires an argument like in die Kiste ('into the box') to account for
Flexibility in adverbal modification 541 the direction of the transport. Evidently, the PFS to be assigned to the PP is the same as the one given in (52b). However, whereas the locative PP in der Kiste does not have this meaning structure prior to acquiring a resultative reading by specifying its SFi, the resultative reading of the PP in die Kiste is inherited from a corresponding interpretation of the preposition in. Finally, I briefly discuss cases which may appear to be problematic for the approach proposed here. Unlike the ('weak') resultati ves analyzed so far, so-called 'strong' resultatives give the impression of resisting an analysis as adverbal modifications.53 This is based on the fact that, in such cases, the resultative predicates - as exemplified in (55) - do not relate to a DP subcategorized by the verb. (55) Der Gast trank das Glas leer. 'The guest drank the glass empty.' Intuitively, the sentence implies that the guest concerned drank something, which was the content of the glass in question and that, as a result, this glass was empty. Therefore, the structure given in (55a) can be assumed to be the PC of (55), where CONT stands for the relation 'content o f . (55) a. PC:
3c [AG(guest, c) & 3o [CONT(o, glass) & TH(o, c)] & DRINK(c) & 3s [RES(s, c) & HD(glass, s) & Vo[HD(o, s) EMPTY(o)]]]
Obviously, the interpretation of (55) involves a métonymie interpretation of the DP das Glas ('the glass').54 Thus, although we have to resort to more complex interconnections, I suppose that resultatives of this kind can also be explained in the framework proposed above. This, however, has to be left to future work.
8.
Conclusions
The subject of my discussion was several forms of reinterpretation as observed in various occurrences of adverbal modification. Essentially, I did not just consider meaning shifts that come with temporal and non-temporal adverbials modifying verbal expressions. Rather, it was demonstrated that by allowing reinterpretation, so-called secondary predications can also be understood as a special kind of adverbal modification. As a suitable framework for analysis, I presented a multi-level model of meaning representation in which reinterpretations proved to be the result of contextual enrichments
542
Johannes Dölling
of an underspecified, yet strictly compositionally structured, Semantic Form (SF). A crucial ingredient of this approach is the use of obligatory inflection operations that systematically extend the lexically given potential of meaning variation by introducing additional parameters. The paper concentrates particularly on the formal possibilities offered by the level of SF for realizing the pertinent meaning transfers. In contrast, syntactic preconditions and some 'spell-out' steps of interpretation in deriving the Propositional Content (PC) of utterances are only touched on briefly. Moreover, psycholinguistic aspects of the topic concerned were neglected. 55 It remains the task of further investigations to formulate sufficiently precise grammatical, pragmatic, as well as conceptual, conditions of systematic meaning variation and to analyse their realization in processing. Although, admittedly, the approach proposed has partly programmatic features, its fertility as a general device for explaining meaning flexibility in adverbal modification should be clear.
Notes *
1.
2.
3.
4. 5.
I wish to thank especially Ewald Lang and an anonymous referee for productive comments on an earlier draft of the paper. Thanks for helpful discussion are also due to Manfred Bierwisch, Markus Egg, Stefan Engelberg, Wilhelm Geuder, Manfred Krifka, Claudia Maienborn, Barbara Partee, Chris Piñón, Anita Steube, and Ilse Zimmermann. Traditionally, compatibility with temporal adverbials is considered a crucial criterion for classifying verbal expressions into states, activities, accomplishments and achievements (Vendler 1967; cf. also Dowty 1979). According to this, durative adverbials may modify only states or activities but not accomplishments or achievements. In contrast, time-span adverbials permit only a modification of accomplishments. Not least because of 'exceptions' resembling those to be discussed here, the justification of this classification has often been questioned (see e.g. Smith 1991; Klein 1994). Following the basic conception advanced in Mourelatos (1978) and Bach (1986), I take situations or, according to Bach's terminology eventualities, to comprise at least processes, events and states as forming pairwise distinct ontological subsorts that are systematically related to each other. Later on, I will distinguish between some further sorts of eventualities. Sentences emerging from (1) and (2) by deletion of the adverbials lend themselves also to a corresponding reinterpretation. As we will see, such facts support the approach proposed here. However, in the cases considered, the readings of the sentences are coerced by the respective modifiers. Cf. Smith (1991). Accordingly, the general ontological assumption to be made here is that a process can be composed of a number of temporally connected moments of the same kind. Like most other conditions on the conceptual ontology that under-
Flexibility in adverbal modification
6.
7.
8.
9.
10. 11. 12.
13. 14.
15.
543
lies our speaking of eventualities, this assumption is in need of further elaboration which, however, cannot be done here. In the following, I will represent the meaning structure of verbal expressions in a neo-Davidsonian format as used in Parsons (1990) or Krifka (1992). For the treatment of measure phrases, cf. Krifka (1992, 1998) and Kamp and Reyle (1993), except for some simplifications I will make for expository reasons. My provisional assumption with respect to the ontological status of events draws on ideas that emerged in several versions, e.g. Bach (1986), Moens and Steedman (1988), Parsons (1990), Kamp and Reyle (1993) and Piñón (1996). Without going into the details, I prefer the suggestion by Piñón (1996) that events are constituted by spatio-temporally superposed processes. Note that I use the Ξ-quantifier without existential commitments and, thus, only in the sense of 'for a certain ... it is true that'. As a consequence, using 3 does not presuppose the entity at issue to exist in the actual world. For the understanding of states and their holders see Parsons (1990, 2000), Kratzer (1996) and Dölling (1998, 1999). In terms of + OPEN, a 'blocking' manner of representation is used for the complex state predicate proper. For comments see the running text below. One of the first semantic analyses of this use in German is given in Worm (1995). For a similar analysis of resultative verbs cf. Kratzer (2000). In Dölling (1998) an explanation for modification with restitutive wieder 'again' is proposed which corresponds to that for modification with durative adverbials. Such an interpretation has been suggested in Smith (1991) and de Swart (1998). For an analysis of achievements as predicates of such temporally atomic situations that form the beginning and the end of states, processes and events thus limiting them, see Piñón (1997). This is questioned by an anonymous reviewer who points out that there is a difference between the PP nach etlichen Strapazen (literally 'after several strains') in (i) and (ii): (i) Ede hat den Gipfel in zwei Tagen nach etlichen Strapazen erreicht. 'Ede reached the summit in two days after considerable physical strain.' (ii) Ede hat den Gipfel in zwei Tagen nach etlichen Strapazen bestiegen. 'Ede scaled the summit in two days after considerable physical strain.' While in (ii), the 'physical strain' must have already arisen before Ede's scaling of the summit, the PP in (i) can refer to the time of the scaling itself. I am unable to follow the reviewer's argumentation that this shows the reinterpretation postulated in (5) to be untenable. Rather, it must be assumed that in (i) the extended VP den Gipfel nach etlichen Strapazen erreichen is shifted to an event-predicate. The conditions on application of the time-frame adverbial are thereby met and, futhermore, the PP nach etlichen Strapazen characterizes activities/states which precede the actual reaching of the summit.
544
Johannes Dölling
16. Arguments for analyzing copula-predicative constructions like wach sein ('to be awake') as predicates of states are provided in Dölling (1999) (cf. also Parsons 1990). As I maintain the arguments presented there, let me add only some comments. In its basic meaning, an adjective like wach is represented as an object predicate Xo.AWAKE(o), where o is a variable for objects. To combine with the copula, the adjective has to be reinterpreted by means of a procedure assumed for any predicatives as state predicate Às.Vo[HD(o, s) —> AWAKE(o)]. The latter structure can be abbreviated Às.+AWAKE(s), which in turn is used in (6a) in the 'blocking' representation used for wach sein; cf. also the predicate + OPEN which occurs in (3a). 17. The concept of 'type coercion' of an argument by its functor is also dealt with, from a more general view point, in Pustejovsky (1991, 1995, 1998). In these works, however, reinterpretations in adverbal modifications play only a minor role. Following the tradition of Logical Semantics, I prefer to use the term sort coercion rather than type coercion. In my opinion, it is obvious that the phenomena under consideration are related not to the problem of separating expressions into semantic types but to the problem of additionally separating them into semantic sorts. For the use of operators of type coercion in the strict sense, see e.g. Partee (1987, 1995), Dölling (1995, 1997), and the main text. 18. Indeed, Jackendoff (1997) - cf. also Jackendoff (2002) - sees in the required enrichment in reinterpretations an important argument against the standard hypothesis of "syntactically transparent semantic composition" (p. 48). Referring to deliberations as can be found in Pustejovsky (1991, 1995), Jackendoff pleads instead for treating the meaning of a complex expression as a function of the meanings of its parts and the way they are combined syntactically only as a default in a wider range of options. 19. Generally, such a concept is advocated e.g. in Dölling (1992) and in Hobbes et al. (1993). In the field of modification by temporal adverbials, this course is first followed in Worm (1995). The approach is presented in Pulman (1997) and de Swart (1998) in a more detailed way. 20. With respect to the alternatives considered, an anonymous referee remarks that the most probable reading is not given here and is perhaps not possible in this approach, namely, the reading that Use was playing various parts of the sonata the whole day through, without necessarily playing the whole sonata at least once without intermediate repetitions or repeatedly as a whole - but just in the way musical students practise. This, the reviewer claims, would require two steps of reinterpretation (an iteration [1] of different parts [2] of the sonata). Although I wonder why it should be the most probable interpretation, I agree that this interpretation of (10) exists. At the moment, I am not able to make a proposal for representing it. 21. In particular, this assumption is made in Moens and Steedman (1988), Bierwisch (1989), Pustejovsky (1991, 1995, 1998), Jackendoff (1991, 1997, 2002), Worm (1995), Pulman (1997) and de Swart (1998). For the general possibility of different starting points and, hence, directions in reinterpretations see Nunberg (1995) and Dölling (2000a).
Flexibility in adverbal modification
545
22. This erroneous assumption is shared by almost all authors concerned with the phenomenon discussed here. 23. For similar considerations see Carston (1999). Cf. also the general assumptions made in Löbner (2002). 24. Cf. e.g. Bierwisch (1988, 1989), Bierwisch and Lang (1989), Lang (1994), and Maienborn (1996, 2001, this volume). 25. In a similar way, Maienborn (2001, this volume) assumes that, under certain conditions, new SF parameters may be introduced in a compositional derivation independent of whether or not there is a semantic incompatibility. The possibilities of meaning transfer thus given are, however, only partial in nature insofar as a systematic extension of the interpretation potential is limited to socalled internal adverbal modifiers, cf. the respective notes in Section 6. The approach I am advocating also has some similarity to the concept of reinterpretation proposed by Egg (2000). On this approach, an underspecified semantic description formalism marks specific sites in the meaning structure of expressions, in which, by means of certain operators, material can be inserted to mediate between semantically conflicting constituents. It is an advantage of this procedure that it permits an integrative treatment of very different kinds of meaning flexibility, among them also scopai ambiguities; for the latter, see Pinkal (1996). However, there are also weaknesses in Egg's approach. First, the principles that govern the systematic marking of the relevant insertion sites remain obscure; second, the mere marking of insertion sites does not sufficiently constrain the admissible structure of the material to be inserted. 26. This mechanism was proposed by Hobbes et al. (1993). As it still has to be elaborated in the future, I will not deal with it here. An overview of how it works within the multi-level model of meaning representation is given in Dolling (1997), for further demonstrations see Dolling (1998) and Maienborn (2001, this volume). 27. The term met is to indicate that the operator at issue provides the prerequisites to explain, within a uniform formal framework, métonymie and metaphoric interpretations as basic types of meaning transfer. 28. Cf. also Dölling (1998, 1999). As will be shown, this hypothetically assumed operator has to be modified in order to also cover other cases of reinterpretation in adverbal modification. 29. The reason for this is twofold. On the one hand, the network of ontological sorts is much too differentiated to allow for an assessment of the number of variables to be admitted in SF. On the other hand, the possible presence in SF of sorted variables would impair the use of general operators like met'. 30. Cf. e.g. Partee (1987, 1995), Zimmermann (1992), Wunderlich (1997), Dölling (1998) and Maienborn (2001, this volume). It is conceivable that in calculating modifications, instead of using the Boolean conjunction '&', the use of a noncommutative restriction operation will turn out to be more appropriate. For suggestions in that direction, see Bierwisch (1989) or Zimmermann (1992).
546
Johannes Dölling
31. In this respect, I follow an idea of Kratzer (1996) where the category VOICE is used to induce subject argument places in corresponding structures. See also Dölling (1999). 32. Axioms (21a)-(21c) allow to account for the fact that, in contrast to a widespread view, not all events are changes of states. Egg (1995) proposes to separate events of the latter sort from intergressives as denoted by predicates like ein Lied singen ('sing a song') or einhundert Meter schwimmen ('swim a hundred metres'). Piñón (1999) argues for explicitly characterizing expressions of change by including a component of resulting state in their semantic representation and, accordingly, for supplementing their argument structure by a state variable. In Section 2,1 expressed some doubts about this proposal. 33. It will be a crucial task for future conceptual analysis to investigate the various fields of knowledge and their interaction in greater detail. 34. In order to be more precise, in what follows, c, c' etc. are used as variables for changes. 35. Once more, the analysis of the copula sentences (25) and (26) is based on Dölling (1999). For an alternative approach, see Maienborn (2002). 36. Cf. Dölling (1998). This idea was suggested much earlier in Parsons (1990). 37. This observation has been made in Dölling (2000b). For a more detailed discussion of the topic, see Geuder (2000). Cf. also Eckardt (this volume). 38. Cf. Bierwisch (1988) for the proposal to assume for a nominalization like Übersetzung ('translation'), in addition to the basic meaning as an event predicate, a derived meaning as a predicate for objects that are brought about by the event at issue. 39. As adverbs do not have any special morphological marking in German, the question may arise whether it is really the adverbial use of adjectives that we face here. Relying on parallel English sentences, this question can be answered in the affirmative. Parsons (1990), however, assesses the use of the ending -ly in these constructions, e.g. she dressed elegantly, as "a mere case of compensating hypercorrectness" and, therefore, as unjustified in the strict sense. 40. Distinct from my approach, Maienborn (2001, this volume) assumes a special mechanism for deriving the object-related reading of locative PPs. The starting point of her deliberations is the observation that such an interpretation is permitted only if the respective expression is in a position adjacent to the verb. This, in turn, is explained by the fact that different operations are used depending on whether the locative adjunct applies to a VP or to a V constituent. While cases of VP adjunction follow the 'usual' pattern, cases of Vmodification require a special semantic operation that paves the way for suitable contextual specifications. It is no doubt an asset of Maienborn's proposal that, in this way, syntactic as well as prosodie restrictions on reinterpreting adverbal modifiers are accounted for. But this proposal has not only the drawback that it can hardly be extended to occurrences where the meaning of the modified expression is transferred. The proposal also seems to be problematic insofar as other possibilities of meaning transfer in VP-modifiers are ruled out.
Flexibility in adverbal modification
41. 42. 43.
44.
547
I think that a more general approach to systematic meaning variation in adverbal modification should be preferred. This example is also drawn from Maienborn (2001). By taking into account this reading, I deviate from the analysis proposed for the sentence in Maienborn (2001). Basically, similar considerations can be found in Pustejovsky (1991) where, however, motion verbs like to drive or to walk which are implicitly directional are treated in this manner. Crucially, in contrast to to dance, directional PPs do not play the role of adjuncts to but of complements of such verbs. My assumption is that the directional interpretation of the preposition in ('into') can be represented by the PFS XoXc. 3s [RES(s, c) & Vo'[HD(o\ s) —» LOCjn(o\ o)]]. Applying this to the object-referring DP der Saal ('the hall') will immediately produce the structure in (36).
45. See, among others, the proposals in Koch and Rosengren (1995), Maienborn (1996), Wunderlich (1997) and Kaufmann and Wunderlich (1998). That adjectives that function as heads of secondary predicates are not used as adverbs can be directly seen in the English examples by the absence of -ly. 46. Here, I follow the basic understanding of depictives as stated in Zimmermann (1992) and Steube (1994). For the use of past-participle constructions as depictive predicates, not allowed for here, see Zimmermann (this volume). 47. It should be recalled that the second conjunct is to be understood as a representation of that part of meaning which can be abbreviated, in a simplifying way, also with ^RAWfs). 48. By way of contrast, the PFS being derivable for in einer Marihuana-Tunke in the second interpretation of (34) is represented by Ac. 3c' [= (c\ c) & 3o [TH(o, c') & LOCdm(o, marijuana_sauce)]] or, after logically simplifying, by Xc. 3o [TH(o, c) & LOCin(o, marijuana_sauce)]. I assume that it is only possible to understand locative PPs in the sense of such a direct object localization if, similar to auf Fahrrädern in (33c), by inferring an instrument - in the broad sense - an additional participant of the situation can be identified. 49. I am indebted to Ewald Lang for this observation. 50. At this point, the difference to modifications by manner adverbials, as discussed wrt. the sentences (28)-(31), should be recalled. There, properties of objects resulting from the events but not properties of resulting states are specified. 51. The characteristics of such verbs are explicated in, among others, Engelberg (2000). For the concept of quantized nominal predicates, cf. Krifka (1992, 1998). 52. The need to realize this distinction has been pointed out to me by Ewald Lang. 53. For the distinction of these two kinds of resultative predication see Kaufmann and Wunderlich (1998). 54. See Dolling (1995, 1997, 2000a). 55. Cf. Piñango, Zurif, and Jackendoff (1999) and Dölling (to appear).
548
Johannes Bölling
References Bach, Emmon 1986 The algebra of events. Linguistics and Philosophy 9: 5-16. Bierwisch, Manfred 1988 On the grammar of local prepositions. In Syntax, Semantik und Lexikon, Manfred Bierwisch, Wolfgang Mötsch, and Ilse Zimmermann (eds.), 1-65. Berlin: Akademie Verlag. 1989 Event nominalization: Proposals and problems. In Wortstruktur und Satzstruktur, Wolfgang Mötsch (ed.), 1-73. (Linguistische Studien 194.) Zentralinstitut für Sprachwissenschaft, Berlin. Bierwisch, Manfred, and Ewald Lang (eds.) 1989 Dimensional Adjectives: Grammatical Structure and Conceptual Interpretation. Berlin: Springer. Carston, Robyn 1999 The semantics/pragmatics distinction: A view from Relevance Theory. In The Semantics/Pragmatics Interface from Different Points of View, Ken Turner (ed.), 85-125. Oxford: Elsevier. Dölling, Johannes 1992 Polysemy and sort coercion in semantic representations. In Discourse and Lexical Semantics, Peter Bosch and Peter Gerstl (eds.). (Arbeitspapiere des SFB 340 "Sprachtheoretische Grundlagen für die Computerlinguistik", Nr. 30.) Stuttgart. 1995 Ontological domains, semantic sorts and systematic ambiguity. International Journal of Human Computer Studies 43: 785—807. 1997 Semantic form and abductive fixation of parameters. In From Underspecification to Interpretation, Rob van der Sandt, Reinhard Blutner, and Manfred Bierwisch (eds.), 113-139. Working Papers of the Institute for Logic and Linguistics, IBM Deutschland, Heidelberg. 1998 Modifikation von Resultatszuständen und lexikalisch-semantische Repräsentationen. In Lexikalische Semantik aus kognitiver Sicht - Perspektiven im Spannungsfeld linguistischer und psychologischer Modellierungen, Petra Ludewig and Bart Geurts (eds.), 173-206. Tübingen: Narr. 1999 Kopulasätze als Zustandsbeschreibungen. In Kopula-PrädikativKonstruktionen als Syntax/Semantik-Schnittstelle, Ewald Lang and Ljudmila Geist (eds.), 95-122. (ZAS Papers in Linguistics 14.) Zentrum für Allgemeine Sprachwissenschaft, Berlin. 2000a Formale Analyse von Metonymie und Metapher. In Meaning Change Meaning Variation, Regine Eckardt and Klaus von Heusinger (eds.), 31-54. (Arbeitspapiere des Fachbereichs Sprachwissenschaft 106.) Universität Konstanz.
Flexibility in adverbal modification 2000b
549
Uminterpretationen bei adverbaler Modifikation: Ein generelles Herangehen. In Prosodie - Struktur - Interpretation, Johannes Dölling and Thomas Pechmann (eds.), 271-302. (Linguistische Arbeitsberichte 74.) Universität Leipzig. to appear Aspectual (re-)interpretation: Structural representation and processing. In Mediating between Concepts and Language-processing Structures, Holden Härtl and Heike Tappe (eds.). Berlin/New York: Mouton de Gruyter. Dowty, David 1979 Word Meaning and Montague Grammar. Dordrecht: Kluwer. Eckardt, Regine 1998 Events, Adverbs, and Other Things. Issues in the Semantics of Manner Adverbs. Tübingen: Niemeyer. 2003 Manner Adverbs and information structure: Evidence from the adverbial modification of verbs of creation. In this volume. Egg, Markus 1995 The Intergressive as a new category of verbal Aktionsart. Journal of Semantics 12: 311-356. 2000 Flexible semantic construction: The case of reinterpretation. Habilitation Thesis, Universität des Saarlandes, Saarbrücken. 2001 Beginning novels and finishing hamburgers - remarks on the semantics of to begin. In Ereignisstrukturen, Johannes Dölling and Tatjana Zybatow (eds.), 295-319. (Linguistische Arbeitsberichte 76.) Universität Leipzig. Engelberg, Stefan 2000 Verben, Ereignisse und das Lexikon. Tübingen: Niemeyer. Geuder, Wilhelm 2000 Oriented adverbs. Issues in the lexical semantics of event adverbs. Doctoral dissertation, Universität Tübingen. Hobbs, Jerry, Mark Stickel, Douglas Appelt, and Paul Martin 1993 Interpretation as abduction. Artificial Intelligence 63: 69-142. Jackendoff, Ray 1991 Parts and Boundaries. Cognition 41: 9-45. 1997 The Architecture of the Language Faculty. Cambridge: MIT Press. 2002 Foundations of Language. Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press. Kamp, Hans, and Uwe Reyle 1993 From Discourse to Logic. Introduction to Modeltheoretic Semantics of Natural Language, Formal Semantics and Discourse Representation Theory. Dordrecht: Kluwer. Katz, Graham 2000 Anti neo-Davidsonianism: Against a Davidsoninan semantics for state sentences. In Events as Grammatical Objects, Carol Tenny and James Pustejovsky (eds.), 393^416. Stanford: CSLI.
550
Johannes Dölling
Kaufmann, Ingrid, and Dieter Wunderlich 1998 Cross-linguistic Patterns of Resultatives. (Arbeiten des SFB 282 "Theorie des Lexikons", Nr. 109.) Universität Düsseldorf. Klein, Wolfgang 1994 Time in Language. London: Routledge. Koch, Wolfgang, and Inger Rosengren 1995 Secondary predicates: Their grammatical and conceptual structure. Sprache und Pragmatik 35: 1-100. Lund. Kratzer, Angelika 1996 Severing the external argument from its verb. In Phrase Structure and the Lexicon, J. Rooryck and L. Zaring (eds.), 109-137. Dordrecht: Kluwer. 2000 Building Statives. Ms., University of Massachusetts at Amherst. Krifka, Manfred 1992 Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, Ivan Sag and Anna Szabolcsi (eds.), 29-53. Stanford: CSLI. 1998 The origins of telicity. In Events and Grammar, Susan Rothstein (ed.), 197-235. Dordrecht: Kluwer Academic Press. Lang, Ewald 1994 Semantische vs. konzeptuelle Struktur: Unterscheidung und Überschneidung. In Kognitive Semantik / Cognitive Semantics. Ergebnisse, Probleme, Perspektiven, Monika Schwarz (ed.), 25-40. Tübingen: Narr. Löbner, Sebastian 2002 Understanding Semantics. London: Edward Arnold. Maienborn, Claudia 1996 Situation und Lokation. Die Bedeutung lokaler Adjunkte von Verbalprojektionen. Tübingen: Stauffenburg. 2001 On the position and interpretation of locative modifiers. Natural Language Semantics 9: 191-240. 2002 Die logische Form von Kopula-Sätzen, (studia grammatica 56.) Berlin: Akademie Verlag, (in press) 2003 Event-internal modifiers: Semantic underspecification and conceptual interpretation. In this volume. Mourelatos, Alexander 1978 Events, processes, and states. Linguistics and Philosophy 2: 415-434. Moens, Mark, and Mark Steedman 1988 Temporal ontology and temporal reference. Computational Linguistics 14: 19-28. Nunberg, Geoffrey 1995 Transfers of meaning. Journal of Semantics 12: 109-132.
Flexibility in adverbal modification
551
Parsons, Terence 1990 Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge: MIT Press. 2000 Underlying states and time travel. In Speaking of Events, James Higginbotham, Fabio Pianesi, and Achille C. Varzi (eds.), 81-94. New York/Oxford: Oxford University Press. Partee, Barbara 1987 Noun phrase interpretation and type-shifting principles. In Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers, Jereon Groenendijk, Dick de Jongh, and Martin Stokhof (eds.), 115-143. Dordrecht: Foris. 1995 Lexical semantics and compositionality. In Invitation to Cognitive Science, Part I: Language, L. Gleitman and M. Liberman (eds.), 311-360. Cambridge: MIT Press. Pinkal, Manfred 1996 Radical underspecification. In Proceedings of the 10th Amsterdam Colloquium, Paul Dekker and Martin Stokhof (eds.), 587-606. Amsterdam: ILLC. Piñango, Maria, Edgar Zurif, and Ray Jackendoff 1999 Real-time processing implications of enriched composition at the syntax-semantics interface. Journal of Psycholinguistic Research 28: 395414. Piñón, Chris 1996 An Ontology for Event Semantics. UMI, Ann Arbor. 1997 Achievements in an event semantics. In Proceedings of SALT VII, A. Lawson (ed.), 276-293. Cornell University, Ithaca. 1999 Durative adverbials for result states. In Proceedings of the 18th West Coast Conference on Formal Linguistics, Sonya Bird, Andrew Carnie, Jason D. Haugen, and Peter Norquest (eds.), 420-433. Somerville, MA: Cascadilla Press. Pulman, Stephen 1997 Aspectual shift as type coercion. Transactions of the Philological Society 95(2): 279-317. Pustejovsky, James 1991 The syntax of event structure. Cognition 41: 47-81. 1995 The Generative Lexicon. Cambridge: MIT Press. 1998 The semantics of lexical underspecification. Folia Linguistica 32: 323347. Smith, Cariota 1991 The Parameter of Aspect. Dordrecht: Kluwer. Steube, Anita 1994 Syntaktische und semantische Eigenschaften sekundärer Prädikationen. In Zur Satzwertigkeit von Infinitiven und Small Clauses, Anita Steube and Gerhild Zybatow (eds.), 221-241. Tübingen: Niemeyer.
552
Johannes Dölling
Swart, Henriette de 1998 Aspect shift and coercion. Natural Language and Linguistic Theory 16: 347-385. Vendler, Zeno 1967 Linguistics in Philosophy. Ithaca: Cornell University Press. Wunderlich, Dieter 1997 Argument extension by lexical adjunction. Journal of Semantics 14: 95-142. Worm, Carsten Lorenz 1995 Interpretation und Uminterpretation natürlichsprachlicher Beschreibungen von Zeitobjekten: Repräsentation und Inferenz• Working Papers of the Institute for Logic and Linguistics, No. 10, IBM Deutschland, Heidelberg. Zimmermann, Ilse 1992 Der Skopus von Modifikatoren. In Fügungspotenzen, Ilse Zimmermann and Anatoli Strigin (eds.), 251-279. Berlin: Akademie Verlag. 2003 German participle II constructions as adjuncts. In this volume.
Secondary predication and aspectual structure* Susan Rothstein
Abstract This paper presents an analysis of secondary predicates as aspectual modifiers and secondary predication as a summing operation which sums the denotation of the matrix verb and the secondary predicate. I argue that, as opposed to the summing operation involved in simple conjunction, secondary predication is subject to a semantic constraint involving the TPCONNECT relation, where ei is TPCONNECTed to e 2 with respect to an argument y if e! and £2 have the same running time and share a grammatical argument y. In depictive predication, the constraint is simply that the event introduced by the matrix verb and the event introduced by the secondary predicate must be TPCONNECTed. Resultative predication differs from depictive predication in that the TPCONNECT constraint holds between the event which is the culmination of the event introduced by the matrix predicate ei and the event introduced by the resultative e2. Formally, while depictive predication introduces the statement TPCONNECT(ei,e 2 ,y), resultative predication introduces the statement TPCONNECT(cul(e!),e2,y). I show that this is all that is necessary to explain the well-known properties of resultative predication.
1.
Introduction
This paper presents a compositional analysis of the semantics of secondary predication. I argue that secondary predicates, both depictives and resultatives, as illustrated in (1) and (2) respectively, are aspectual modifiers in the sense that they introduce a new event and define a relation between it and the event introduced by the main predicate. (1)
a. John, drove the car drunk¡. b. Mary drank the coffee¡ hot¡.
(2)
Mary painted the house¡ redi.
I begin by distinguishing explicitly between secondary predicates and nominal modifiers, on the one hand, and between secondary predicates and adverbials on the other, and I present some of the properties of secondary predication which any account has to explain; these include the much-
554
Susan Rothstein
discussed differences between depictive and resultative predication, in particular, the so-called 'direct object restriction' on resultative predication. I argue that secondary predicates are related to the matrix eventuality via an operation of event summing, based on the summing operation which Lasersohn (1992) argues is the core of the conjunction relation. This operation sums the events introduced by the verb and the secondary predicate under the condition that they are connected via a relation which I call 'TPCONNECT', which holds if the two events coincide temporally, and share a participant. I present a compositional interpretation of secondary predication structures which uses this operation. I go on to show that the differences between depictive and resultative predication derive from a minimal difference between the summing operation used in each case; specifically, in depictive predication the TPCONNECT relation holds between the matrix event and the event introduced by the secondary predicate, while in resultative predication this relation holds between the culmination of the matrix event and the event introduced by the secondary predicate. This paper is primarily concerned with the structural aspects of secondary predication, that is to say how to interpret the syntactic structures compositionally and how to account for the restrictions on argument sharing between the matrix predicate and the secondary predicate. Because of space limitations, I shall not address a number of important related issues, in particular the question of where the 'result' meaning comes from in resultative constructions. I believe that this is essentially a pragmatic issue because of examples like (3), where there is no causative relation between the event denoted by the matrix verb and the 'result-state' or 'end-state' given by the secondary predicate: (3)
a. The crowd cheered the gates open. b. Every night the neighbour's dog barks me asleep. c. On May 5 1945, the people of Amsterdam danced the Canadians to Dam Square. d. Mary drank John under the table//herself sick/dizzy.
I discuss where the result/causative meaning comes from in detail in Rothstein (2001, to appear), where I argue that it derives from the semantics of accomplishments, and where I situate the analysis presented in this paper in the context of a much wider research project.
Secondary predication and aspectual structure 2.
555
The syntax of secondary predication
2.1. The data The basic data that we have to deal with were given in (1) and (2) above, and I repeat them here: (1)
a. Johtii drove the car drunk¡. b. Mary drank the coffee¡ hot¡.
(2)
John painted the house¡ redi.
The example in (la) is an instance of subject-oriented depictive predication, and means roughly "John drove the car when John was drunk", (lb) and (2) are object-oriented predicates, (lb) is a depictive, and means roughly "Mary drank the coffee when the coffee was hot", and (2) is a resultative, meaning more or less, "John painted the house and at the end of the painting, the house was red". These examples have been discussed in the general literature at least since Halliday (1967), and in the generative literature since Dowty (1979) and Simpson (1983). Simpson pointed out that resultatives can be directly predicated only of direct objects, so that (4a/b) are ungrammatical with the purported readings "John painted the house and as a result he was red" and "John laughed and as a result he was sick"; however, with intransitives like 'laugh', a so-called 'fake reflexive' may be used, so that (4b) has the grammatical counterpart in (4c). Furthermore, these 'nonthematic' objects, as I shall call them, need not be reflexives, as (4d) shows: (4)
a. b. c. d.
*John¡ painted the house red¡. *John¡ laughed sick¡. John laughed himself¡ sick¡. John sang the babyi asleep ¡.
There has been much discussion of the properties of resultatives and of the direct object restriction in the literature, including Dowty (1979), Tenny (1987, 1994), Levin and Rapoport (1988), and others: some more recent papers, including Wechsler (1997), Levin and Rappaport Hovav (1999), Rappaport Hovav and Levin (1999) have also questioned the validity of the direct object restriction, and suggested that there are counter-examples which indicate that the generalisation is incorrect. As far as I know, there has been no explicit discussion of the impossibility of analogous nonthematic objects occurring with depictives. (5) cannot have the reading
556
Susan Rothstein
"John ran while Mary was drunk", and I will show that it is instructive to consider why. (5)
*John ran Mary drunk.
2.2. Some syntactic structures First, let us look at syntactic structures. The structures for the two types of secondary predicates are given in (6). I assume that object-oriented secondary predicates (both resultatives and depictives) are generated under V' while subject-oriented depictives are generated under VP (following arguments for VP structure in Andrews (1982)). (6)
a. John¡ [[drove the car]v· drunki]vp b. Mary [[drank the coffee¡ hot¡]v]vp c. John [[painted the house¡ red¡]y]vp
This can be shown by using standard tests of fronting, and pseudo-clefting: subject-oriented secondary predicates may be stranded by these tests, but need not be, showing that they are daughters of VP, while object-oriented secondary predicates may never be stranded, showing that they are daughters of V : (7)
a. b. c. d. e. f.
What Mary did was paint the house drunk. What Mary did drunk was paint the house. What Mary did was drink the coffee hot. *What Mary did hot was drink the coffee. What Mary did was paint the house red. *What Mary did red was paint the house.
Second, secondary predicates may stack, as shown in (8), with the proviso that the resultative must be lower than any depictives, that object-oriented predicates are lower than any subject-oriented predicates (broken in (8c/d) is to be interpreted as a depictive), and that these depictives may be predicated of direct objects, but not goals, or prepositional objects (Williams 1980).
Secondary predication and aspectual structure
(8)
a. b. c. d. e. f.
557
Billj [[drove the car¡ broken¡]v drunkj]vp Janej [[painted the car¡ red¡]v drunkjJvp ?Janej [[painted the car¡ red¡ broken¡] ν · drunk¡]vp *Janej [[painted the car¡ broken¡ redi]v drunkj]vp John visited Mary¡ drunk¡. *John spoke to Mary¡ drunk¡.
That secondary predicates are not predicated of objects of prepositions means that they cannot be predicated of arguments of either nominal or adjectival heads. Third, secondary predicates do not form a constituent with their subject. This is obvious for subject-oriented depletives, as the stranding facts in (7) show. It also holds for object-oriented predicates, and this can be shown via contrasts with small clause predicates. If an object-oriented predicate and its subject formed a constituent, then that constituent would be the direct object of the matrix verb, and this is exactly what happens with small clause predicates such as those in (9) (see Rothstein (2000a) for a detailed discussion). (9)
a. Mary considers [John intelligent]Sc b. Mary made [it seem that John was on time]Sc
Yet in these constructions, the entailments are very different from those in secondary predicate constructions, as the following data show. (10a/lla) do not entail (10b/llb), while (12a/13a/14a) do entail the (b) examples, and the contrast between the examples in (15) demonstrates the same point. (Re. (lib): see Higginbotham (1983) and others for arguments that seeing an event does not entail seeing its participants). (10)
a. Mary believes/considers b. Mary believes/considers
John foolish. John.
(11)
a. Mary saw the president leave. b. Mary saw the president.
(12)
a. Mary drank her coffee hot. b. Mary drank her coffee.
(13)
a. Mary painted the house red. b. Mary painted the house.
(14)
a. Mary drove the car drunk. b. Mary drove the car.
558
Susan
Rothstein
(15) a. #Mary drank her coffee hot though she never drank her coffee. b. John believes Bill a liar, and he doesn 't believe Bill. Fourth, secondary predicates are optional (and again the contrast is with small clause predicates): (16) a. * I thought/believed that problem. b. Mary drank her coffee/drove the car/painted the house. Fifth, secondary predicates assign a thematic role to their arguments (subjects). There is no morphological difference between secondary predicates and small clause predicates and they are subject to the same structural condition on predication (see Rothstein 2001), and we assume that this indicates that in both constructions they have the same thematic properties. I assume also, following Higginbotham (1983), Parsons (1990), Kratzer (1995), Greenberg (1998) and Rothstein (1999, 2000a) that adjectival predicates introduce some kind of eventuality argument into the representation. For simplicity, I will assume that this is an e variable, and not introduce the distinction between mass-eventualities denoted by adjectives and count eventualities denoted by verbs which I argue for in Rothstein (1999). Assuming, then, a neo-Davidsonian framework in which verbs and adjectives denote sets of events, and thematic roles introduce functions from events to participants (Parsons 1990; Landman 2000), the AP predicate drunk, as it occurs in both (17a) and (17b) will translate as an expression like (18): (17) a. I consider Mary drunk. b. I met Mary drunk. (18) drunkAp: XxXe.DRUNK(e) Λ Arg(e)=x
2.3.
What secondary predicates are not
2.3.1.
Secondary predicates are not nominal modifiers
That secondary predicates are not nominal modifiers is shown through pronominalisation tests and through testing entailments. First, I will consider entailments. When an AP is used as a secondary predicate then the property that it expresses must hold of the denotation of its subject for the whole time that the matrix event is going on (for depictives) or for the whole time that
Secondary predication and aspectual structure
559
the culmination of the matrix event is going on (for resultatives). With nominal modifiers this is not so. (19) a. / met the drunk man again, but this time he was sober. b. #/ met the man drunk again, but this time he was sober. (20) a. The drunk man drove the car home, after he had sobered up. b. #The man drove the car drunk, after he had sobered up. (21) a. They paint the red house once every year. Last year they painted it white and this year they painted it green. b. #They paint the house only once a year, and they always paint it red. Last year they painted it white and this year they painted it green. Nominal modifiers are part of the NP, combining with Ν to form a Common Noun expression, and they are not temporally related to the matrix verb at all. The fact that they are syntactically part of the nominal argument expression, while secondary predicates are not, is shown by the fact that pronominalisation replaces the expression containing the nominal modifier, while it does not affect the secondary predicate at all. (22) a. I met the drunk man today. ENTAILS: I met him today DOES NOT ENTAIL: I met him today and he was drunk when I met him. b. I met the man drunk today. ENTAILS: I met him today and he was drunk when I met him.
2.3.2.
Secondary predicates are not adverbs
We can show that secondary predicates are to be distinguished from adverbs again via comparing entailments. (23a) entails that John was drunk, (23b) is compatible with no-one being drunk. And as a correlate, as (24) shows, the secondary predicate, but not the adverb, needs a lexically expressed subject. (23) a. a.' b. b.'
John drove the car drunk. #John drove the car drunk, although he was sober. John drove the car drunkenly. John drove the car drunkenly, although he was sober.
560
Susan Rothstein
(24) a. The car went (drunkenly) round the corner (drunkenly). b. #The car went round the corner drunk. I conclude that secondary predicates must be predicated of a subject, and that they assign a thematic role to that subject, whereas adverbs do not do so. If we make this the litmus test for distinguishing between adverbs and secondary predicates, then an obvious question is what about subjectoriented adverbs, such as enthusiastically or reluctantly, as illustrated in (25), which appear also to assign some sort of thematic role to the subject: (25) John greeted Mary
enthusiastically/reluctantly.
It seems to me that, although these adverbs are subject-oriented (or more properly, agent-oriented), and must introduce a relation between the denotation of the subject and the event, this orientation is not equivalent to predication. The function of these adverbs is to add information about how the Agent of the matrix verb performed the relevant action, e.g. in an enthusiastic or reluctant way, but they do not entail that this Agent had the property of being himself enthusiastic or reluctant. Thus (26a) entails that John was reluctant about something, but not that he was enthusiastic about anything, and the converse is true of (26b). Similarly, (26c) is not a contradiction, and neither is (26d), where the AP is used as a secondary predicate: (26) a. John greeted Mary enthusiastically, although he was secretly very reluctant to meet her. b. John greeted Mary reluctantly because he was so shy, although he was secretly very enthusiastic about meeting her. c. John welcomed Mary enthusiastically, although he was not enthusiastic about welcoming her. d. John greeted Mary drunkenly, although he did not, in fact, greet her drunk. I assume that the grammatical relation of α assigning a theta-role to an argument β, or standing in a thematic relation to β, is interpreted as the denotation of β being assigned a particular participatory role in the event introduced by a. Where a is adjectival, the role assigned to the external argument (or 'Argj', as I shall call it) is that of bearer of the property expressed by a. Under this assumption, the data in (26) are sufficient basis for asserting that adjectives, or rather APs, do assign a thematic role to their external arguments, but that subject- or agent-oriented adverbs do not assign thematic roles.
Secondary predication and aspectual structure
561
Following Davidson (1967), Parsons (1990), Landman (2000), we treat adverbials as modifiers of events. (23b), "John drove the car drunkenly", is represented as (27), with the adverb drunkenly translating as the expression in (28), where VP is the denotation of VP. (27)
3e [DRIVE(e) Λ Ag(e)=JOHN Λ Th(e)=THE CAR Λ DRUNKENLY(e)]
(28)
drunkenlyAov: XVPXe.VP(e) Λ DRUNKENLY(e)
This contrasts with the translation for drunk given above in (18), and repeated here, where the adjective introduces its own event argument and dictates what participants there must be in that event: (18)
drunkAp: Xîde.DRUNK(e) Λ Argi(e)
The contrast between the denotations of adverbs and adjectives correctly predicts that adverbs, unlike adjectives, can never be predicated of syntactic arguments in predication structures, as in (29): (29)
a. */ consider the driving drunkenly. b. I consider the driver drunk.
3. Secondary predication as a summing operation 3.1. Evidence for event summing On the assumption that (18) represents a typical translation of an adjectival predicate, whether used as a primary (i.e. clausal) or secondary predicate, what kind of operation must secondary predication be? Lasersohn (1992) argues on the basis of conjoined predicates such as hot and cold alternately that conjunction of predicates must involve a summing operation rather than event intersection. I shall argue that secondary predication also involves a summing operation of the same kind. One possibility for interpreting secondary predicates is to assume that an operation of intersection is involved. On this reading, (30) will have the interpretation in (31): (30) John drove the car drunk.
562
Susan Rothstein
(31 ) 3e[DR0VE(e) Λ Ag(e)=JOHN Λ Th(e)= THE CAR Λ DRUNK(e) Λ Argi(e)= JOHN] However, this entails that the event of John driving the car and the event of John being drunk were the same event, and this is not a desirable conclusion. Lasersohn (1992), in his discussion of event conjunction, shows that expressions like hot and cold alternately entail that the instances of being hot and the instances of being cold are temporally distinguishable, and are thus distinguishable events. Since secondary predicates are crucially not temporally independent of the main verb, this kind of evidence is not available for our structures. However, we can still argue that the events introduced by the matrix verb and the secondary predicate must be distinguished using the arguments from finegrainedness presented in Parsons (1990), from which it follows that the representation in (31) cannot be correct. Parsons argues that different event predicates which hold of an argument at the same run time can be modified by contradictory modifiers. So suppose with one stroke of the broom I sweep away both a pile of dirt and an earring, then it can be true that I intentionally swept away the pile of dirt and accidentally swept away an earring. But since an event cannot be both intentional and accidental at the same time, Parsons argues, the two expressions swept away the pile of dirt and swept away an earring must be descriptions of different sweeping events, distinguished by the fact that they have different participants, which hold at the same time. A similar argument involves identification of events via their participants. We have just shown in the previous section that AP predicates, unlike adverbs, introduce thematic roles; this means that they denote entities which have participants, which means that they denote events which can be identified via their participants. So, while an event of driving is an event which must have two participants, an Agent and a Theme, an event of being drunk must have one participant, which we have called for convenience the 'bearer' of the property, or possibly the 'experiencer', but which crucially is not an Agent. A being drunk event can thus be distinguished from a driving event both in terms of how many participants it has, and what their relationship to the eventuality is. We can make this argument more clearly by looking at examples like those in (32). (32) a. John drove the car drunk from the cognac. b. John drove the car scared out of his wits. In (32a) where the adjectival head of AP introduces two thematic roles, we can clearly distinguish the event introduced by drove, which has John as the Agent and the car as the Theme, and thus denotes an event with John and
Secondary predication and aspectual structure
563
the car as participants, and the event introduced by drunk, which has John as its external argument and the cognac as the internal argument, and thus denotes an eventuality with John and the cognac as participants. At this point we can see that the two events belong to two different aspectual classes: JOHN DRIVE THE CAR is an activity while JOHN DRUNK FROM THE COGNAC is a state. But if these are the appropriate distinctions to make, then the reading in (31), which asserts that there was one event of which both these predicates can be predicated, will just be false. Similarly, in (32b) John may be driving the car of his own free will, but may be sacred out of his wits despite himself. If we grant that relative to one event he cannot be both a willing Agent and an unwilling experiencer, we must assume a distinction between the event of driving the car and the event of being scared out of his wits, and ascribe him the willing Agent role as participant of the first, and the unwilling experiencer role as participant in the second. I assume, following Lasersohn (1992), Krifka (1992, 1998), Landman (2000), that the domain of events has a part structure: i.e. it forms a Boolean semilattice, with the sum operation, U, and the part-of relation, E, defined in the usual y, such that χ Ç y iff χ U y = y. Secondary predication will involve a summing operation, which sums the denotation of two event expressions a(ei) and ß(e2); unlike the summing operation in Lasersohn and Landman, I assume that the operation yields a singular event rather than a plurality. I will assume further the theory of predication developed in Rothstein (1999, 2000a), in which VPs and APs denote sets of events (i.e. expressions of type ), and (usually) contain a distinguished free variable χ of type d, as the value of the external thematic role, (where e is the type of events and d is the type of individuals). Predicate formation operates on the denotation of the maximal projections, i.e. VPs and APs, by lambda abstracting over the distinguished χ variable, and raising the value of the XP from to . For details see Rothstein (2000a, Chapter 6). Secondary predicate formation, then, involves an operation of the form in (33), where the superscript lS ' on the sum operation indicates that the output of the operation is singular. (33) SUM[a( ei ), ß(e2)] = X e . B e ^ f e ^ e . U e i ) A a(e0 Λ ß(e2)] This gives (34a) as the translation of the VP drive the car drunk, and (34b) as the translation of (30): (34) a. Xe.ae^ezte^eiUez ) Λ DRIVE(ei) A Ag(e0=x A Th(e0= THE CAR A DRUNK(e2) A Arg(e2)=JOHN] b. 3e3e,3e 2 [e= s (eiUe 2 ) A DROVE(ej) A Ag(e,)=JOHN A Th(d)= THE CAR A DRUNK(e2) A Arg(e2)= JOHN]
564
Susan Rothstein
(34b) is true if there is an event which has both an event of John driving the car as part and an event of John being drunk as a part. It is the VP in (34a) which, in the framework of Rothstein (2000a), is the input to further operations such as predicate formation. This also predicts that this VP should be the input to adverbial modification, and indeed this is the case. (35) a. John drove the car drunk twice. b. John drove the car drunk for an hour. c. John drove the car drunk soberly. The adverbials in bold in (35) all modify the VP drive the car drunk with the meaning in (34a) and, as we would predict, have the meanings in (35') respectively. (Thanks to Manfred Krifka for suggesting that I make this explicit.) (35') a. There were two distinct events that are sums of a driving event and a being drunk event. b. There was an event which was the sum of a driving event and a being drunk event which lasted an hour. c. There was an event which is the sum of a driving event and a being drunk event which took place in a sober manner. As follows from the discussion of (23)-(28) above, (35c) entails that there was a drunk participant in the event, in this case John, and that the event took place in a sober manner, but not that there was a sober participant. So (35c) is not contradictory.
3.2. Semantic constraints on the secondary predication operation I assume then that the summing operation is the basic mechanism involved in interpreting secondary predicates. However, as it stands in (34), it is not enough. There are constraints on the summing relation, in addition to the fact that the output is a singular event, which distinguish summing via secondary predication from simple predicate conjunction with and. Furthermore, there are a series of questions about the nature of secondary predication, and we would like the answers to fall out from the properties of the operation. The following issues arise:
Secondary predication and aspectual structure
565
1. What are the constraints on the summing relation which distinguish secondary predication from event conjunction? 2. Why are there no intransitive depictives? i.e. why does I sang the baby asleep not have the reading "I sang while the baby was asleep"? 3. Why are the two kinds of secondary predicates depictive and resultati ve (e.g. why are there no 'inceptives')? 4. Why are resultatives not predicated of subjects? 5. What are the effects on aspectual class of adding a secondary predicate? 6. How can we account for the restricted set of examples discussed in Wechsler (1997) and Rappaport Hovav and Levin (1999) which are purported to be subject-oriented resultatives? I shall argue that there are two constraints on the relation between the summed events. The first is a constraint about temporal dependency, and the second about shared participants. We look first at the temporal dependency constraints.
3.2.1.
Temporal dependency
The summing operation as it stands is not enough. If we look at a simple example such as (30), it does not merely assert that an event occurred which had a driving part and a being drunk part, but more strongly, that these events were going on at the same time. Simple conjunction does not require this. Yesterday I wrote a letter and read a book does not imply that these events were going on simultaneously. This becomes even clearer if we look at (35b) which requires that the events of driving and being drunk which are summed must be related in such a way that every part of the hour-long event must have a driving part and a drunk part. Contrast (35b) with (36): (36) I played with the child and read her stories for an hour. This requires that the complex event which had a playing part and a reading her stories part went on for an hour, but it does not specify what the temporal relation between the two subevents is. It can be true if we play Happy Families for ten minutes and then read for 50 minutes, or any other combination. (35b) does not allow this. The contrast between (35b) and (36) shows that the temporal dependency constraints on the interpretation of (35b) follow from the secondary predication relation and not from any homogeneity requirements of durative
566
Susan Rothstein
adverbials. The same contrast between (37a) on the one hand and (37b/c) on the other shows that the temporal dependency constraint in secondary predication does not follow from any independent constraint such that conjoined matrix sentential predicates must each be independently marked for tense, or that non-verbal predicates may not be. The complement in (37a) is a secondary predicate construction involving temporal dependency between the driving and the being drunk event. (37b/c) are true if Mary made there be a sum of events which had an event of John driving and an event of John being drunk as a part, but there is no indication of any temporal relation between these events; the first can precede the second or vice versa, or the first can be contained in the second or vice versa, or one can overlap the other. (37) a. Mary made John drive the car to Tel Aviv drunk. b. Mary made John drive to Tel Aviv and be drunk. c. Mary made John drive to Tel Aviv and John be drunk. We assume a temporal trace function 'τ' as defined in Krifka (1998), which maps an event e onto its running time such that T(eiUe2) = τ(βι)υτ(β 2 ) (the run time of the sum of ei and e 2 is the sum of the run time of ei and the run time of β2). Thus if ei is 'John sing' and e 2 is 'John dance', the running time of the plural event 'John run U John dance' is the time it took for both these events to go on, i.e. x(John run) U x(John dance). Secondary predication imposes an additional constraint on the event summed, namely that their run times must coincide. (38)
SUM[a( ei ), ß(e2)] = Xe.3 ei 3e 2 [e= s (eiUe 2 ) Λ a(eO Λ ß(e2) α τ(βι)=τ(β2)
(30) will have the denotation in (39): (30) John drove the car drunk. (39) 3e3ei3e 2 [e= s (eiUe 2 ) Λ DRIVER) A Th(e0= the car A Ag(eO=John A DRUNK(e2) A Arg(e2)=John Α τ(βι)=τ(β2)] (39) asserts that there occurred an event which was the sum of a 'John driving the car' event and a 'John drunk' event, where both events were going on at the same time. It need not be the biggest event of its kind and, crucially, it can be part of another event of John driving the car, which is not an event of his being drunk, or it can be part of another event of his being drunk which is bigger than the event of his driving the car.
Secondary predication and aspectual structure
567
The formalisation in (39) indicates that the temporal dependency between ei and e 2 is symmetrical. Examples like (40) indicate that perhaps it should be asymmetrical, since clearly it is the event introduced by the verb, namely the event of driving from Jerusalem to Tel Aviv, which determines the run time of the summed event. (40) John drove the car from Jerusalem to Tel Aviv drunk. However, this asymmetry follows not from the semantics of the secondary predication operation, but from the (independent) distinction between quantized and non-quantized verbal predicates in the sense of Krifka (1992). In (30) both drove the car and drunk are homogenous predicates, the first an activity and the second a state. This means that for any event e¡ in the denotation of drove the car, any part of that event will also be an event of driving the car and, similarly, for any event e2 in the denotation of drunk, any part of e2 will also be a drunk event. So for any two events of John driving the car and John being drunk which have overlapping running times, any part of that overlap can contain an event which we describe as John drove the car drunk. However, in (40), the matrix event is an accomplishment, drove the car from Jerusalem to Tel Aviv, which is quantized, which means that for any event in the denotation of this verbal expression, no part of it can be in the denotation of the same verbal expression (in the normal case; I will not discuss exceptions here). This means that the run time of the event introduced by the verbal predicate e¡ will determine the size of the summed event s(eiUe2). The fact that the secondary predicate is constrained for independent reasons (in English) to be an AP (or, rarely, a DP) but is never verbal means that the secondary predicate is always homogenous and thus never dictates the size of the summed event.
3.2.2.
The shared participant constraint
In addition to the constraint of temporal dependency, there is a well-known constraint that the secondary predicate and the matrix verb must share a thematic argument; c.f. Williams (1980), Rothstein (1983). It is this constraint which rules out intransitive depictives such as (41), with the reading "John drove while Mary was drunk". (41)
John drove Maryi drunk¡.
568
Susan Rothstein
Under this reading, 'John' is the single argument of drove, while 'Mary' is the single argument of drunk, and the two predicates drunk and drove do not share an argument. I suggest that the two constraints, the constraint of temporal dependency and the constraint that e¡ and e2 share an argument, combine to form the content of a relation which we can call 'Time-Participant Connected', or TPCONNECT for short, which holds between two events and an individual such that TPCONNECT(ei,e2,y) holds in the following circumstances: (42) TPCONNECT(ei,e2,y) iff: (i) τ(βι) = τ(β2) (i.e. the run time of ei is the same as the run time of e2); (ii) ei and e 2 share a participant y This relation is reflexive and symmetric, but of course not transitive, since if e¡ and e2 share a run time I¡ and a participant x, and e2 and e3 share the same run time /, but a participant y, then e¡ and e3 will share a run time, but not a participant. TPCONNECT thus does not define a partial order/equivalence class. The TPCONNECT relation is a symmetric version of the temporally asymmetric PART-OF condition on secondary predication which I formulated in Rothstein (2000b, 2001). The idea was then, and still is, to capture a relation between eventualities which reflects some intrinsic connection between events more strongly than the relation between elements in an equivalence class. The analogy from the domain of individuals is the way in which the set of cells making up John's body is related to John, although both are singularities with respect to the coordination relation. It is clear that while the set of cells making up John's body is part of John in a very fundamental way, the relation between these two elements is not an equivalence relation since it is obviously non-transitive; if the set of cells making up John's body is part of John and John is part of the debating team, it does not mean that the set of cells making up John's body is part of the debating team. The set of cells making up John's body is part of John in the sense that they both share 'stuff but, despite this inherent relation between them, John and the set of cells making up his body remain independent and the grammar treats them as such; for example they can be conjoined in the appropriate circumstances. Imagine that John is visiting a holistic doctor who says (43) to him: (43) I can't just treat the set of cells making up your body. I have to treat the set of cells making up your body and you.
Secondary predication and aspectual structure
569
It is this kind of non-transitive relation, reflecting an inherent connection between two entities, that I claim holds between the eventualities involved in secondary predication. When we assert that "John drove the car drunk", we assert that there is a sum of two events, the driving the car event and the being drunk event, which do not just occur at the same time but which are inextricably attached to each other since they share a participant which is involved in both these events at the same time. Suppose that John is driving the car and at the same time he is missing Mary who has gone to a conference in another town. Then the event e¡ of John driving the car is inherently connected to the event e2 of John missing Mary, and the event of John missing Mary is inherently connected to the event e¡ of Mary being away at a conference, but the event of John driving the car is not inherently connected to Mary being away at the conference. As a correlate, the grammaticality judgements are as in (44): (44) a. John drove (the car) missing Mary. b. John missed Mary, away at a conference. c. *John drove the car Mary away at a conference. Notice that (ii) of the TPCONNECT relation is formulated in terms of a thematic participant y. This is because (45) is ungrammatical on the intransitive reading of drive, with the purported reading "John drove when he was asleep". (45)
*John drove himself asleep.
This shows that the shared participant condition is actually a grammatical condition on sharing an argument. In (45), the two events mentioned, John driving and John being asleep, do share a 'real-world' participant, namely John, but the sentence is still ungrammatical.
4.
The semantic interpretation of depictive predication
4.1. Object-oriented depictive predication Based on the above, I assume that the secondary predication operation is an operation which sums two predicates of the same type with the constraint that the two stand in the TPCONNECT relation with respect to a thematic participant. (46) gives the operation for object-oriented depictive secondary
570
Susan Rothstein
predication which, in the framework of Rothstein (2000a), sums predicates at type . (46)
Summing operation for object-oriented depictive secondary predication: OSUM[a, Β] = ÀyXe.3ei3e2[e=s(eiUe2 ) Λ a(ei,y) Λ ß(e2,y) Λ TPCONNECT(ei,e 2 ,y)]
This is the operation which is used in the interpretation of object-oriented depictive predication such as the examples in (47): (47)
a. Mary drank her coffee¡ hot¡. b. The police arrested John¡ drunks
In interpreting (47b), we use the operation in (46) to sum the predicates in (48a) to give the VP in (48b). The derivation is given below: (48)
a. λyλe.ARREST(e) Λ Ag(e)=x A Th(e)=y Xx.Xe.DRUNK(e) A Arg(e)=x b. ÀyXe3ei3e 2 [e= s (eiUe 2 ) A ARREST(ei) A Ag(e0=x A Th(e0=y A DRUNK(e 2 ) A Arg(e2)=y A TPCONNECT(e 1 ,e 2 ,y)]
The police arrested John i drunk¡ [arrest]Ν
-> λyλe. ARREST(e) A Ag(e)=x A Th(e)=y
[drunk]A
-> λ ε ^ υ Ν Κ ( ε ) a Arg(e)=x
[drunk]AP —> XxXe.DRUNK(e) a Arg(e)=x (by predicate formation) [arrest John! drunk]]y - » OSUM([arrest] v , [drunk]Ap)(JOHN) = OSUM[XyXe.ARREST(e) A Ag(e)=x A Th(e)=y, Xxle.DRUNK(e) A Arg(e)=x] = XyXe.3ei3e2[e=s(eiUe2 ) A ARREST(ei) A Ag(e0=x A Th(e0=y A DRUNK(e 2 ) A Arg(e2)=y A TPCONNECT(ei,e 2 ,y)] (JOHN) = λβ.Ξβι36 2 [ε= 5 (βιυε 2 ) A ARREST(ei) A Ag(e0=x A Th(e0=JOHN A DRUNK(e 2 ) A Arg(e 2 )=JOHN A TPCONNECT(e!,e 2 ,JOHN)]
Secondary predication and aspectual structure
571
[arrest Johni drunkjvp—» XxXe.aeiae2[e=s(eiUe2) Λ ARREST(ei) A Ag(eO=x A T h f o ^ J O H N A DRUNK(e2) A Arg(e2)=JOHN A TPCONNECT(ei,e2,JOHN)] (by predicate formation) [arrested Johni drunker XxXe.3e!3e2[e=s(eiUe2) A ARREST(e,) A Ag(eO=x A Th(ei)=JOHN A DRUNK(e2) A Arg(e2)=JOHN A TPCONNECT(e, ,e2,JOHN) A PAST(e)] [The police arrested Johni d r u n k s AxXe.3ei3e2[e=s(eiUe2) A ARREST(ei) A Ag(eO=x A ThCe^JOHN A DRUNK(e2) A Arg(e2)=JOHN A TPCONNECT(ei,e2,JOHN) A PAST(e)] (THE POLICE) = λβ.3β13β2[β=8(β1υβ2 ) A ARREST(ei) A Ag(e,)=THE POLICE A Th(ei)=JOHN A DRUNK(e2) A Arg(e2)=JOHN A TPCONNECT(ei,e2,JOHN) A PAST(e)] Existential quantification leads to: 3e3e,3e 2 [e= s (eiUe 2 ) A ARREST(e,) A Ag(ei)=THE POLICE A Th(e,)=JOHN A DRUNK(e2) A Arg(e2)=JOHN A TPCONNECT(e Ι ,e2,JOHN) A PAST(e)] "There was a singular event which was the sum of an event of the police arresting John and John being drunk where the arresting was timeparticipant-connected to the drunk event." With regard to the shared participant condition, note that the formulation of the operation in (46) in fact guarantees that there will be a shared participant, since it applies both predicates simultaneously to the same argument. So instead of the time-participant connect relation expressed in (46), we could have an explicit condition concerning identity of run times and rely on the functional application operation to guarantee the shared participant. However, I am not going to take this step because I want it to be explicit that it is not accidental that (46) is formulated the way it is, and that the shared time-participant condition is a crucial element in licensing the relation. The shared time-participant condition does not follow from the formulation of the operation; instead the operation is formulated the way it is because of the shared time-participant condition.
572
Susan Rothstein
4.2. Subject-oriented secondary predication Subject-oriented depictive secondary predication involves conjunction at the VP level and, assuming the predication theory in Rothstein (2000a), this means conjoining at the level before predication formation has taken place. The depictive predicate is already a predicate and thus is at type ; before it can be conjoined with the verbal predicate it must first be applied to a distinguished variable χ to bring it to the right type (see Rothstein (2000a) for details of this operation, which is I call there 'predicate absorption'): (49)
Summing operation for subject-oriented depictive secondary predication: SSUM[a, β λyλe.DRIVE(e) A Ag(e)=x A Th(e)=y (THE CAR) = Xe.DRIVE(e) A Ag(e)=x A Th(e)=THE CAR [drunk]A ^ [drunk]Ap
Xe.DRUNK(e) A Arg(e)=x —> λχ Xe.DRUNK(e) A Arg(e)=x (by predicate formation)
[drive the car drunk] Y —> SSUM([drive the car]v·, [drunk]AP(X)D = SSUM([Xe.DRTVE(e) A Ag(e)=x A Th(e)=THE CAR] V , ^xle.DRUNK(e) A Arg(e)=x]Ap(x)]) = Xe.3ei3e 2 [e= s (eiUe2 ) A DRIVE(ei) A Ag(e0=x A Th(ej)=THE CAR A DRUNK(e 2 ) A Arg(e 2 )=x A TPCONNECT(ei, e 2 , x)] [drive the car drunk]vp—> λ χ λ β . Ξ ε ^ ε ^ ^ υ β ; , ) A D R I V E R ) A Ag(e0=x A Th(e0=THE CAR A DRUNK(e 2 ) A Arg(e 2 )=x A TPCONNECT(ei, e 2 , x)] (by predicate formation) [drove the car drunk] XxleaeiBezte^ieiUez) A DRIVE(ei) A Ag(ej)=x A Th(e0=THE CAR A DRUNK(e 2 ) A Arg(e 2 )=x A TPCONNECT(ei, e 2 , x) A PAST(e)]
Secondary predication and aspectual structure
573
[John drove the car drunk] n>—» kAe.3ei3e2[e=s(eiUe2) a DRIVE(ei) a Ag(eO=x λ Th(e,)=THE CAR a DRUNK(e2) a Arg(e2)=x a TPCONNECT(ei, e2, x) a PAST(e)] (JOHN) = X e . a e ^ e ^ e i U e a ) λ DRIVE(ei) a Ag(e,)= JOHN a Th(ei)= THE CAR a DRUNK(e2) a Arg(e2)=JOHN a TPCONNECT(ei, e2, JOHN) a PAST(e)] Existential quantification leads to: Be3ei3e2[e=s(eiUe2 ) a DRTVEfo) λ Ag(e,)= JOHN λ Th(ei)=THE CAR a DRUNK(e2) λ Arg(e2)=JOHN a TPCONNECT(e,, e2, JOHN) a PAST(e)] "There was a singular event which was the sum of an event of John driving the car and John being drunk where the driving event was time-participant connected to the drunk event".
5.
The semantics of resultatives
5.1. The interpretation of simple resultatives The account given above is sufficient to get us the essential semantics of depictive predicates. The next stage is to extend the account to explain how resultatives work. The goal is to formulate an account which is minimally different from the account of depictive predication, thus maintaining the idea that secondary predication is a unitary phenomenon, while explaining the crucial differences between depictive and resultative predication. As background, I assume Dowty's (1979) analysis of aspectual classes, which I reformulate in an event-style framework. The basic structures of the aspectual classes are as in (50). Here I treat accomplishments as culminating activities, which is compatible with Dowty's analysis, although (50b) is not the decomposition that he gives. I discuss the question of what precisely accomplishments are in Rothstein (2001, to appear). (50) a. b. c. d.
States: Activities: Achievements: Accomplishments:
Xe.P(e) Xe.(DO(P))(e) λe.(BECOME(P))(e) Ie.3f 1 Bf 2 [e= s (f 1 Uf 2 ) a (DO(P))(fi) a cul(e!)=f2]
574
Susan Rothstein
Deriving simple resultatives such as (2), repeated here as (51), is straightforward. (51) Mary painted the house¡ redi • We assume that the basic summing operation is the same that is used in depictive predication. Resultative predication differs from depictive predication because in resultative predication, the TPCONNECT relation holds between the culmination of the event introduced by the matrix verb and the event of the adjectival predicate. The difference is summed up in (52): (52) depictives: resultatives:
... ATPCONNECT(ei,e2) ... A TPCONNECT(cul(e1),e2)
Resultative conjunction is object-oriented, and thus the process conjoins expressions at type : (53) Summing operation for resultative secondary predication: RSUM[a, β] = λyλe.ΞelΞe2[e=s(elUe2) A a(ei,y) A ß(e2,y) A TPCONNECT(cul(ei),e2,y)] Resultative summing in the interpretation of (51) sums the verbal predicate in (54a) and the adjectival predicate in (54b): (54) a. [paintlv -> ÀyÀe.PAINT(e) A Ag(e)=x A Th(e)=y b. [red]Ap—> AxXe.RED(e) A Arg(e)=x Derivation: Mary painted the house¡ red¡ [paintlv -> ÀyÀe.PAINT(e) A Ag(e)=x A Th(e)=y [red]A
-> Àe.RED(e) A Arg(e)=x
[redjAP —> ÀxAe.RED(e) a Arg(e)=x (by predicate formation) [paint the house red]v -> RSUM( [paint]v , [redW) (THE HOUSE) = RSUM[ÀyÀe.PAINT(e) A Ag(e)=x A Th(e)=y, ÀxÀe.RED(e) A Arg(e)=x] (THE HOUSE) = ÀyXe.3ei3e2[e=s(eiU e2) A PAINTieO A Ag(eO=x A Th(eO=y A RED(e2) A Arg(e2)=y A TPCONNECT(cul(e,),e2,y)] (THE HOUSE)
Secondary predication and aspectual structure
575
= Xe.3ei3e2[e=s(eiUe2) A PAINT(ei) A Ag(ei)=x A Th(ei)=THE HOUSE Λ RED(e 2 ) A Arg(e 2 )=THE HOUSE A TPCONNECT(cul(ei),e2, THE HOUSE)] [paint the house red]yp—> XxÀe.3e13e2[e=s(eiUe2) A PAINT(ei) A Ag(e t )=x A Th(e,)=THE HOUSE A RED(e 2 ) A Arg(e 2 )=THE HOUSE A TPCONNECT(cul(e, ),e2, THE HOUSE)] [painted the house red]r—» XxÀe.3ei3e2[e=s(eiU e 2 ) A PAINT(ei) A Ag(ei)=x A Th(eO=THE HOUSE A RED(e 2 ) A Arg(e 2 )=THE HOUSE A TPCONNECT(cul(e1),e2, THE HOUSE) A PAST(e)] [Mary painted the house red]n>—» Àe.3ei3e 2 [e= s (eiU e 2 ) A PAINT(ei) A Ag(ei)=MARY A Th(eO=THE HOUSE A RED(e 2 ) A Arg(e 2 )=THE HOUSE A TPCONNECT(cul(ei),e2,THE HOUSE) A PAST(e)] Existential quantification leads to: 3e3ei3e 2 [e= s (eiU e 2 ) Α ΡΑΙΝΤ(βΟ A Ag(ei)=MARY A Th(e,)=THE HOUSE A RED(e 2 ) A Arg(e 2 )=THE HOUSE A TPCONNECT(cul(ei),e2,THE HOUSE) A PAST(e)] 'There was a singular event which was the sum of an event of Mary painting the house and the house being red where the culmination of the painting event was time-participant connected to the red event." We can collapse OSUM and RSUM (= depictive and resultative predication respectively at the type) in the following way: (55)
SUM[a,ß] ^yXe.3ei3e 2 [e= s (eiUe 2 ) A a(ei,y) A ß(e2,y) A TPCONNECT(A(ei),e2,y)] if Δ = id (i.e. the identity function on events) then the interpretation is depictive if Δ = cul then the interpretation is resultative.
576
Susan Rothstein
5.2. Some answers to some questions We are now in a position to answer some questions. We have seen what depictive and resultative predication have in common, namely the SUM operation in (55). I suggest that these are the only two forms of secondary predication available in English because the verb class distinctions make available only the 'highest' e, and two kinds of subevents in (50d). If the sum operation relates the highest e and the secondary predicate, then we have depictive predication. If the sum operation relates cul(e) in (50d) and the secondary predicate, then we have resultative predication. In (50d) the event variable introduced by (DO(P)) is also, in principle, available for secondary predication, but since cul(e) is a near-instantaneous event part of e, the activity part of e is equal in time to the whole accomplishment. Thus a TPCONNECT relation between the activity subevent in (50d) and the predicate is equivalent to a TPCONNECT relation between the whole accomplishment and the predicate, and the effect is depictive predication.
5.3. The direct object restriction A central fact about the resultative construction is that there is a direct object restriction; we have no reading for Mary painted the house red in which red is predicated of the subject, as in (56), although an assertion that Mary became red as a result of painting the house is perfectly plausible. (56) Mary i painted the house red¡. The reason for this is as follows. TPCONNECT(cul(ei),e2,y) requires cul(ei) and e 2 to share a thematic argument y. The culmination of an accomplishment is determined by what happens to its Theme: Mary built a house culminates at the point at which a house 'comes into existence' or 'becomes built'. So, the argument of the culmination event, y, is the incremental Theme of the matrix verb. Since TPCONNECT requires the secondary predicate to share the argument of the culmination event, the secondary predicate must have as its argument the incremental Theme of the matrix verb. So the direct object restriction reduces to the restriction that resultatives must be predicated of the Theme of the matrix verb, and this itself derives from the interpretation rule. This predicts that subject-oriented resultatives can occur when subjects are incremental Themes, since the subject is then the argument of the culmination event. For this reason, we find
Secondary predication and aspectual structure
577
subject-oriented resultatives with unaccusative and passive verbs, but not with unergatives. (57) a. b. c. d.
The house¡ was painted red¡. The riveri froze solid¡. Maryi grew up smart,·. *John¡ ran tired¡.
We can tell that the subjects in (57a-c) are incremental Themes because of the lack of entailments between the progressive versions of (57a-c) and the simple pasts (i.e. the test of the imperfective paradox) while in (57d) the entailment goes through. The house was being painted does not entail the house was painted; the river was freezing does not entail the river froze', and Mary was growing up does not entail Mary grew up. In contrast, John was running does entail John ran. Note that according to the rule for resultative predication, the resultative does not give the culminating event, which is defined in terms of the incremental Theme, but gives a property of the culminating event. In (51a) and (57a), cul(e) is defined by when the house gets painted, and the secondary predicate gives a property of that event: it is time-participant connected to the event of the house being red.
6.
Non-accomplishment resultatives
6.1. Type shifting in non-accomplishment resultatives The rule we have so far allows us to add a resultative predicate to a verb whose meaning includes a culmination event in its denotation, namely an accomplishment verb. However, as is well-known, resultatives can be added with great freedom to VPs headed by activity verbs (see the examples in Levin and Rappaport Hovav (1995)). Some examples of this are given in (58): (58) a. b. c. d.
Mary hammered the metal flat. John sang the baby asleep. John sang himself asleep. *John sang asleep.
In (58a), the resultative predicate has been added to a VP headed by a transitive activity verb, and the predicate is predicated of the verb's Theme ar-
578
Susan Rothstein
gument. In (58b), the matrix V is an intransitive activity, and the resultative is predicated of an argument which has been added as direct object, although it is apparently not thematically related to the verb. The contrast between (58c) and (58d) shows that, while a resultative cannot be predicated of the subject of an unergative, a 'fake reflexive' (to use the term from Simpson (1983)) can be added in direct object position, and the result is grammatical. So we need to answer the following questions: (i) Given the rule in (53), how can we add a resultative to a nonaccomplishment verb which does not have a culmination subevent without violating the TPCONNECT condition. (ii) Why does (58b) not doubly violate the TPCONNECT condition, since the matrix verb is not thematically related to the subject of the resultative and so does not share an argument with it? (iii) If we can add resultatives to non-accomplishment verbs, then why only to activities? (iv) Why is the fake reflexive necessary in (58c)? Beginning with (i), we can make the issue more specific. The interpretation of a resultative requires the matrix verb to have a culmination, as expressed in the rule in (53), yet the examples in (58) indicate that the resultative is added to activities which do not already have culminations. We do not want to analyse the resultative rule applying to (58) as adding a culmination. This is because we are assuming that there is a single resultative rule which applies in all resultative constructions. Since an event can, by definition, have only one culmination point, and since resultatives occur with lexical accomplishments, for which the culmination is lexically specified, resultatives cannot in general introduce culmination points. In (56), the accomplishment VP paint the house defines when the culmination of the painting event occurs, namely when the house is or becomes painted, and the resultative adds a property of the culmination, namely that it is part of the event of the house being red. On the assumption that there is only one resultative rule, then even in (58) the resultative will only be able to give a property of the culmination and not add the culmination itself. So the grammar must allow us to 'add' a culmination to the activity verbs in (58), thus making the resultative possible. I hypothesise that the resultative predication operation triggers a typeshifting operation which adds a culmination to an activity. I assume, following Krifka (1992, 1998) and others, that processes are not inherently telic or atelic, but telic or atelic under a particular description, and that we can mod-
Secondary predication and aspectual structure
579
ify our descriptions of them. Thus the same event can be described as a running (atelic) or a running to the store (telic). Let us assume a culmination modifier of the form in (59): (59)
λΕλβ.Ε(β) A3e'[cul(e)=e' A Arg(e')=Th(e)]
It has the following properties: -
It is of type « e , t > , < e , t » .
-
Like a thematic role/function, it assigns an event a location, manner or time. Its domain is restricted to the set of activity eventualities, and it denotes a function from sets of activities onto their culminations. It specifies that the argument of cul(e) is the incremental Theme of e.
-
Resultative predication forces the cul(e) modifier to be added if the matrix event lacks a culmination. Thus in (58a), resultative predication forces a shift from the activity verb meaning of hammer to an accomplishment reading: (60)
SHDT(XyXe.HAMMER(e) Λ Ag(e)=x A Th(e)=y) = XyXe.HAMMER(e) A Ag(e)=x A Th(e)=y A 3e'[cul(e)=e' A Arg(e')=Th(e)]
The interpretation of (58a) is then as in (61): (58a) Mary hammered the metal flat. (61)
Ξ β Ξ β , Ξ β ζ ^ ί β ι ϋ β ΐ ) A HAMMER(ei) A Ag(d)=MARY A Th(ei)=THE METAL A 3e 1 '[cul(ei)=ei' A Arg(eO=Th(ei) A FLAT(e 2 ) A Arg(e 2 ) = THE METAL A TPCONNECT(cul(ei),e2, THE METAL)]]
"There was an event which was the sum of a hammering event and an event of the metal being flat and the culmination of the hammering event was time-participant connected to the event of the metal being flat." Since TPCONNECT(cul(ei),e 2 ,y) forces cul(ei) and e 2 to share an argument, the culmination of the hammering event must have THE METAL as its argument. When the verb is intransitive, the culmination modifier introduced by the shifting process effectively adds an argument:
580
Susan Rothstein
(62) SHIFT^e.SING(e) a Ag(e)=x) = X)ß,e.SING(e) λ Ag(e)=x λ 3e'[cul(e)= e' λ Arg(e')=Th(e)] (58b) John sang the baby asleep. (63) Be3e13e2[e=s(eiUe2) a SINGLO a A g f o ^ J O H N a 3ei'[cul(ei)=ei' a A r g í e i ^ h í e j ) a ASLEEP(e2) a Arg(e2) = THE BABY a TPCONNECT(cul(ei),e2, THE BABY)]] (63) says: -
-
-
The singing event (eO was assigned a telic point (ei') and the telic point of the singing event was time-participant connected to the event (e2) of the baby being asleep. The semantics of TPCONNECT tells us that if cul(e,) is TPCONNECTed to e2, then they share a thematic argument. The argument of cul(ei) must be the argument of e2, the baby. The resultative tells us that at cul(ei) the baby is asleep.
The argument the baby is the incremental Theme of the verb in the sense that if the telic point is determined by what happens to the baby, then in some sense the event of singing is 'measured' by what happens to the baby. I elaborate on this point and on the details of the SHIFT operation in Rothstein (2001, to appear). However, there is some straightforward linguistic evidence that the added argument in (58b) is indeed the Theme of the verb. It is well-known (Verkuyl 1972; Tenny 1987, 1994; Krifka 1992, 1998 and others) that the quantized or non-quantized status of the Theme determines the telicity of the VP when the verb is an accomplishment. In (58b) the telicity of the VP is determined by the quantized or non-quantized status of the added Theme, the baby: (63) a. John sang babies asleep for hours/*in an hour last night. b. John sang 3 babies asleep *for hours/in an hour last night. But note that this Theme has a different 'thematic status' from other arguments which is plausible, since it is introduced via the shifting process and not lexically. We can see this since it does not license theta-government, hence contrasts such as (64), noted originally in Rothstein (1992):
Secondary predication and aspectual structure
581
(64) a. Which table,· did you ask whether John wiped t¡ clean? b. HWhich babyi did you ask whether John sang ti asleep? To sum up then, resultative predication triggers a type-shifting operation on activities, in which activities shift to an accomplishment reading through the imposition of a culmination modifier. I give the general operations in (65): (65) a. Shifted transitive activity ( —> ): ÀyÀe.ACnvrrY(e) Λ Ag(e)=x Λ Th(e)=y -> ÀyÀe.ACTrVHY(e) A Ag(e)=x Λ Th(e)=y Λ 3e'[cul(e)=e' A Arg(e')=Th(e)] b. Shifted intransitive activity ( —» ): Xe. ACTrvrrY(e) A Ag(e)=x -> ÀyÀe.ACTrvrrY(e) A Ag(e)=x A Th(e)=y A 3e'[cul(e)=e' A Arg(e')=Th(e)]
6.2.
Remaining questions
6.2.1. De-thematicised resultatives If the resultative predicate is predicated of the Theme of the matrix verb, then how can we derive sentences like (66), where the matrix verb drink has a Theme, the liquid consumed, yet John is the subject of the secondary predicate? (66) Mary drank John under the table. I assume that here two processes have taken place. First, the transitive verb drink is intransitivised and the internal argument is existentially quantified over. At that point, it is no longer available as the incremental argument, and we will call it 'Patient' to distinguish it from the incremental Theme. We thus have Xe.3z[DRINK(e) A Ag(e)=x A Patient(e)=z] as its meaning. Resultative predication forces a shift which adds a new incremental argument, as in (67a), giving (67b) as the meaning of (66): (67) a. XyXe.3z[DRINK(e) A Ag(e)=x A Patient(e)=z] A 3e'[cul(e)=e· A Arg(e')=Th(e)] b. 3e3ei3e23z [e=s(eiUe2) A DRANK(ei) A Agent(e,)=MARY A Patient(eO=z A UNDER-THE-TABLE(e2) A Arg(e2)=JOHN A TPCONNECT(cul(ei),e2.JOHN)]
582 6.2.2.
Susan Rothstein Why do resultatives not occur with achievements or states?
The type-shifting operations in (65) are designated to apply to activities, which means that resultative predicates can occur only with accomplishments, which naturally have culminations, or with activities, which are shifted into this form via (65). And indeed we want this restriction, since the examples in (68), where the matrix predicates are stative and achievement predicates respectively, do not have grammatical resultative readings, but only depictive readings: (68) a. John¡ was happy drunk¡. b. John, arrived late¡. Does the restriction of (65) to activities have to be stipulated or does it follow from something else? I assume that it follows from the properties of the aspectual classes themselves. For an event to have a culmination requires that it be non-static, since culmination is inherently connected to the notion of development. Adding a culmination to an activity, as we do via (65), involves imposing a developmental structure on the activity. Since states are static, no such developmental structure can be imposed on them. An achievement, on the other hand, consists only of a culmination. Predicating a property of the culmination of an achievement is equivalent to predicating it of the achievement itself, hence the depictive reading of (68b).
6.2.3.
Fake reflexives
The example in (58c), repeated here as (69a) with some analogous cases, involves a so-called 'fake reflexive'. Intuitively, the fake reflexives allow a resultative to be indirectly predicated of a non-theme subject, via a direct object argument. We can show now exactly how this works and why the examples in (70) are ungrammatical. (69) a. John sang himself asleep. b. Bill laughed himself sick. c. Mary wrote herself into a corner. (70) a. *John laughed sick. b. *John sang asleep. c. *Mary wrote into a corner.
Secondary predication and aspectual structure
583
(70) is ruled out by the theta-criterion. I am being unspecific about which version, not because I want to be imprecise but because a central generalisation that all formulations of the theta-criterion want to capture, though they express it in different ways, is that a verb has as many syntactic arguments as it has thematic arguments or, in other words, that two thematic roles cannot be assigned to the same argument by the same lexical head. The resultative rule requires the secondary predicate to share the Theme argument of the main predicate. In (69) the matrix verbs do not have a Theme argument, but only an Agent. Agents are crucially non-incremental, as we saw in the discussion of (57) above, and so the Agents in (69) cannot be the arguments of cul(ei), the culmination event of the matrix verb. The examples in (70) are ungrammatical in the same way that John saw is ungrammatical on the meaning "John saw himself'. However, in the same way that we use a reflexive in John saw himself to allow the same real-world entity to be both Agent and Theme participant in the event through the mediation of two grammatical arguments, we use a reflexive to allow the same entity to be both Agent and incremental argument of an intransitive verb in an intransitive resultative construction. Note also that the 'uniqueness' part of the theta-criterion, the claim that there can be only one incremental Theme per verb, rules out pseudothematic resultatives with unaccusatives in (71). Since the verb already assigns a Theme role to its subject argument, as we saw in (57b, c), repeated here, there is no available Theme role to be assigned to the reflexive. (71) * The river¡ froze itself solid. (57b) The river¡froze solid¡. (57c) Mary ι grew up smarti.
7.
Subject-oriented resultatives
There have been a number of recent works which have argued that the direct object restriction, whether it is phrased in terms of direct objects or incremental Themes, is not correct, and that there exists a class of subjectoriented resultatives which are not predicated of incremental Themes. Most prominent among these works are Wechsler (1997), who offers (72a-c) as evidence, Verspoor (1997) who offers examples (72d/e) and Rappaport Hovav and Levin (1999).
584
Susan Rothstein
(72) a. b. c. d. e. f.
The wise men followed the star out of Bethlehem. The sailors caught a breeze and rode it clear of the rocks. He followed Lassie free of his captors. The children played leapfrog across the park. John walked the dog to the store. John danced mazurkas across the room.
Hoekstra (1988) and Levin and Rappaport Hovav (1995) argue also that verbs of manner of motion and verbs of sound emission occur both in intransitive, object-oriented resultatives and as apparent subject-oriented intransitives. The examples in (73) are taken from Rappaport Hovav and Levin (1999): (73) a. b. c. d.
Dan ran/hopped/jogged/danced to the station. She started to run the hangover out of her system. The elevator creaked to the ground floor. The alarm clock buzzed them awake.
They point out that sometimes minimal pairs are possible, as in (74), although not always, as (75) shows: (74) a. b. c. d.
One of the race cars wiggled loose inside the transporter. The snake wiggled itself loose. She danced across the room. She danced herself across the room.
(75) a. b. c. d.
She wiggled herself comfortable in the chair. *She wiggled comfortable in the chair. She danced herself dizzy. *She danced dizzy.
Rappaport Hovav and Levin (1999) and Wechsler (1997) give competing analyses of how to explain when and why subject-oriented resultatives are possible, and I discuss their accounts in detail in Rothstein (to appear). There is no space to go into the details of the discussion here, but since it is an obvious and crucial prediction of my analysis that subject-oriented resultatives are not possible except where the subject is an incremental Theme, I do want to say something about how the apparently subject-oriented examples above should be analysed. The most pertinent observation concerning the apparently subjectoriented resultatives is that the XPs which are supposed to be result predi-
Secondary predication and aspectual structure
585
cates are restricted to expressions of location and, more precisely, of direction. Rappaport Hovav and Levin point this out, commenting that subjectoriented resultatives are restricted to expressions which denote 'result locations' rather than states. The PPs which occur include across the room, out of Bethlehem, and to the store, and the APs too are expressions which can express a direction with respect to a fixed point such as clear of the rocks, free of his captors and loose. Crucially, a non-directional expression such as comfortable cannot be a subject-oriented resultative, and we have minimal pairs with (74c) and (75d), where dance allows a subject-oriented directional expression across the room, but disallows the non-directional dizzy in the same position. I suggest that apparently subject-oriented result predicates are not resultative predicates at all, but are internal path arguments of the verb, in the sense of Krifka (1998). A path argument can be, and usually is, the incremental Theme, and Krifka shows that what defines path arguments is precisely that as the matrix event grows temporally, the portion of the path which is the argument of the event grows too. Thus in an example like John danced across the room, the verb dance is supplied with an incremental path argument across the room. The effect is analogous to a resultative predicate because the event denoted by dance across the room reaches its telic point when the path is 'used up' and that of course will be when John is across the room. This is of course the same situation as the one that occurs at the telic point of John danced himself across the room, which asserts that there is an event of dancing whose culmination point is coincidental with the event of John being across the room. There are various questions that are answered by this account of the examples in (72/75) which makes the account convincing. First, we explain Rappaport Hovav and Levin's observation that (so-called) subject-oriented resultatives denote result locations and not other kinds of states. Since they are in fact path arguments, the telic point of the event will be when the subject is at the location designated by the end of the path - and this will be a 'result location'. Second, we explain why subject-oriented resultatives are temporally dependent, to use Rappaport Hovav and Levin's words, by which they mean that the result event "unfolds" at the same rate as the matrix event. If the locational expression denotes a path which is the incremental argument of the verb, and which is 'used up' gradually as the event unfolds, then of course progress along the path will be temporally dependent on the progress of the matrix event. Third, although there are minimal pairs such as (74c/d), we see that when the PP is directional but non-telic, in other words when it determines a non-bounded path, the object-oriented version is not as good, as in (76a/b):
586
Susan Rothstein
(76) a. John danced about the house. b. IJohn danced himself about the house. This is because the object-oriented versions are true resultatives, and nontelic directional phrases do not easily denote result states. Fourth, Wechsler (1997) observes that PPs can have a 'metaphorical' reading when they are object-oriented, as in (77a), but not when they are subject-oriented, as in (77b). (77) a. She danced herself over the edge. b. She danced over the edge. This is because in the subject-oriented constructions, the PP denotes a spatial path and must thus be interpreted spatially, whereas in the objectoriented constructions, the resultative is not constrained to have a spatial interpretation. Clearly, there is a lot more to say about this topic and the formal details of the analysis of so-called subject-oriented resultatives still have to be worked out. There is no space for this here, but I hope I have shown the direction in which I think an analysis of these putative counterexamples should go.
8.
Summary and the next set of questions
I have argued that secondary predication involves a summing operation which sums eventualities which are closely related to each other temporally and which share a participant. Depictive readings occur when the secondary predicate eventuality is cotemporal with the event introduced by the matrix predicate, while resultative predication occurs when the culmination of the matrix predicate is cotemporal with the eventuality introduced by the secondary predicate. The paper has focussed on structural issues: what are the formal syntactic structures which are interpreted, what are the rules which interpret them, and how does this explain a number of facts about secondary predication, especially the central fact that a resultative interpretation is possible only when the predicate is predicated of an incremental Theme, while there is no such restriction on depictive predication. I have not discussed additional issues concerning depictive predication, in particular what is the basis of the intuition that depictive predicates are very preferably noninherent and transitory. I assume that this question can be discussed within the framework presented here.
Secondary predication and aspectual structure
587
The analysis presented here raises a number of larger questions, which space constraints do not allow me to discuss in this paper, but which I take up in Rothstein (2001) and at length in Rothstein (to appear). I want to mention briefly here what I take these issues to be. There are two central issues, which are closely related. The first is what exactly does it mean for a nominal such as the baby to be the incremental Theme of John sang the baby asleep. A related question to do with 'taking away' rather than 'adding' an incremental Theme is raised by the analysis I have given of Mary drank John under the table·, this analysis implies, and in fact requires, that when the Theme argument of drink is existentially quantified over, it loses the property of being the incremental argument. I think this is correct, since Mary drank, which has the interpretation in (78a), is atelic, as (78b) shows: (78) a. 3eHz[DRINK(e) Λ AG(e)=Mary Λ Th/Patient(e)=z] b. Mary drank for hours/*in an hour. However, this means that the incremental argument is not necessarily the argument/participant directly affected by the activity determined by the verb. The question of what exactly the incremental argument is is tied up with an account of the semantics of accomplishments, and this is a question which I begin to explore in the works cited above. The second obvious question which is unanswered so far is where the 'result' meaning of resultatives comes from. The analysis I have given of, for example, John sang the baby asleep requires that the baby be asleep at the culmination of John's singing, but does not require John's singing to cause the baby to be asleep. But informants often feel strongly that the causative element is part of the meaning of the sentence and, further, it has always been assumed that the result/cause element in the meaning is what is responsible for making examples like (79) unacceptable on the resultative reading. (79) #John painted the house dirty. But, as I pointed out with respect to the examples in (3), repeated here, there is no necessary causal element in the meaning of resultatives: (3)
a. The crowd cheered the gates open. b. Every night the neighbour's dog barks me asleep. c. On May 5 1945, the people of Amsterdam danced the Canadians to Dam Square. d. Mary drank John under the table//herself sick/dizzy.
588
Susan Rothstein
I argue at length in Rothstein (2001, to appear) that the result meaning in most resultatives is a fall-out from the meaning of the accomplishment. The structure of an accomplishment involves measuring the progress of an activity in terms of an extended change-of-state event, which has a culmination at which the change-of-state is reached. Constraints on how the activity and change-of-state events are matched make it frequently the case that the change-of-state is caused by the activity, but this is not necessarily the case. The result aspect of the resultative derives from this causal relation and, when it is absent, the result meaning in resultatives is absent too.
Notes *
The material in this paper has been presented at several conferences and colloquia since it was first presented at the Oslo Conference on Adjuncts in September 1999. These include the 16th Annual meeting of the Israel Association for Theoretical Linguistics, The Linguistics Department colloquium at Tel Aviv University, the Workshop on Predication held at ZAS in September 2000, and the Seminar on Predication held at Trondheim University in October 2001. I would like to thank participants at all these events for their questions and comments, and especially Manfred Krifka. As usual, I would like to thank Fred Landman for much helpful discussion along the way.
References Andrews, Avery 1982 A note on the constituent structure of adverbials and auxiliaries. Linguistic Inquiry 13: 313-317. Carrier, Jill, and Janet Randall 1992 The argument structure and syntactic structure of resultatives. Linguistic Inquiry 23: 173-234. Davidson, Donald 1967 The logical form of action sentences. In The Logic of Decision and Action, N. Rescher (ed.), 81-95. Pittsburgh/Pa.: University of Pittsburgh Press. Dowty, David 1979 Word Meaning and Montague Grammar. Dordrecht: Kluwer. 1982 Grammatical relations and Montague Grammar. In The Nature of Syntactic Representation, Pauline Jacobson and Geoffrey K. Pullum (eds.), 79-130. Dordrecht: Kluwer. 1991 Thematic proto-roles and argument selection. Language 67: 546-619.
Secondary predication and aspectual structure
589
Greenberg, Yael 1998 An overt syntactic marker for genericity in Hebrew. In Events and Grammar, Susan Rothstein (ed.), 125-144. Dordrecht: Kluwer. Halliday, Michael 1967 Notes on transitivity and theme in English, Part I. Journal of Linguistics 3: 37-81. Higginbotham, James 1983 The logic of perceptual reports: An extensional alternative to situation semantics. Journal of Philosophy 80: 100-127. 1985 On semantics. Linguistic Inquiry 16: 547-593. Hoekstra, Teun 1988 Small clause results. Lingua 74: 101-139. Kratzer, Angelika 1995 Stage and individual level predicates. In The Generic Book, G. Carlson and F. Pelletier (eds.), 125-175. Chicago: Chicago University Press. Krifka, Manfred 1992 Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, Ivan Sag and Anna Szabolsci (eds.), 29-53. Stanford: CSLI. 1998 The origins of telicity. In Events and Grammar, Susan Rothstein (ed.), 197-235. Dordrecht: Kluwer. Landman, Fred 2000 Events and Plurality: The Jerusalem Lectures. Dordrecht: Kluwer. Lasersohn, Peter 1992 Generalized conjunction and temporal modification. Linguistics and Philosophy 15: 381—410. Levin, Beth, and T. R. Rapoport 1998 Lexical subordination. Proceedings of the 24th Annual Meeting of the Chicago Linguistic Society, 275-289. Levin, Beth, and Malka Rappaport Hovav 1995 Unaccusativity: At the Syntax-Lexical Semantics Interface. Cambridge, Mass: ΜΓΓ Press. 1999 Two structures for compositionally derived events. Proceedings of SALT 9: 199-223. Cornell University, Ithaca, NY: Cornell Linguistics Circle Publications. Parsons, Terry 1990 Events in the Semantics of English. Cambridge, Mass.: MIT Press. Rappaport Hovav, Malka, and Beth Levin 1999 A re-evaluation of the direct object restriction on English resultatives. Ms., Bar-Ilan University and Stanford. Rothstein, Susan 1983 The Syntactic forms of predication. Ph.D. dissertation, MIT, Cambridge, Mass.
590
Susan Rothstein
1992
Case and NP licensing. Natural Language and Linguistic Theory 10: 119-139. 1999 Fine-grained structure in the eventuality domain: The semantics of predicate adjective phrases and be. Natural Language Semantics 7: 37420. 2000a Predicates and their Subjects. Dordrecht: Kluwer. 2000b Secondary predication and aspectual structure. In Approaching the Grammar of Adjuncts. Papers from the Oslo Conference, Sept 22-25, 1999, C. Fabricius-Hansen, E. Lang, and C. Maienborn (eds.), 241264. (ZAS Papers in Linguistics 17.) Berlin: ZAS. 2001 What are incremental themes? In Papers on Predicative Constructions, G. Jaeger, A. Strigin, C. Wilder, and N. Zhang (eds.), 139-157. (ZAS Papers in Linguistics 22.) Berlin: ZAS. to appear Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Simpson, Jane 1983 Resultatives. In Papers in Lexical-Functional Grammar, L. Levin, M. Rappaport, and A. Zaenen (eds.), 143-157. Bloomington, IN: Indiana University Linguistics Club. Tenny, Carol 1987 Grammaticalizing aspect and affectedness. Ph.D. dissertation, MIT, Cambridge, Mass. 1994 Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. Verkuyl, Henk 1972 On the Compositional Nature of the Aspects. Dordrecht: Kluwer (Reidei). Verspoor, C. M. 1997 Contextually dependent lexical semantics. Ph.D. dissertation, University of Edinburgh, Edinburgh, UK. Wechsler, Stephen 1997 Resultative predicates and control. Texas Linguistics Forum 38: 307321. Williams, Edwin 1980 Predication. Linguistic Inquiry 11: 203-238.
Real adjuncts in the Instrumental in Russian* Assinja Demjjanow and Anatoli
Strigin
Abstract An adjunct-DP in the Free Instrumental case in Russian may occur in a number of surface positions where it (i) is syntactically optional, (ii) does not depend on any element in the sentence, and (iii) may have a number of different interpretations. We exploit Bailyn's proposal of a syntactic environment for the Instrumental case to cover some adjunct uses of instr and propose a uniform semantics of these DPs to accompany it. The proposed semantics can accommodate the different interpretations as it allows contextual expansion. Starting with the hypothesis of Roman Jakobson about context-dependence of the semantics of the Instrumental case, we formulate a semantic interpretation theory based on abduction. Abduction serves as a means to implement context dependence via inference. The sources for the inference are assumed to be given in the situation description provided by the sentence and by general world knowledge. Finally, we discuss some questions and objections concerning the treatment proposed in this paper.
1.
Explaining the problem: How are free DPs in the Instrumental interpreted in Russian?
Russian has six morphologically distinguishable cases.1 One of them is used to mark instruments, among other things, and is hence called Instrumental. The internal and the external verb arguments are usually marked by the accusative and by the nominative, respectively, and are interpreted basically as (generalized) quantifiers which bind the occurrence of the corresponding argument variable of the verb. The Instrumental case occurs in a number of surface positions where a DP marked by it is regularly syntactically optional. This use may be termed free (DPs in the) Instrumental case. The syntactic role of a free DP in the Instrumental is often claimed to be that of an adjunct. The semantics of such free DP adjuncts in the Instrumental presents a problem, since there seem to be a number of other relations associated with this case in addition to the generalized quantifier interpretation. Consider some such occurrences of the Instrumental illustrating the difficulty. We shall abbreviate the name of the morphological case-form to instr and the corresponding bunch of morpho-syntactic features as INSTR. The examples in (1) are taken from Jakobson (1936), reprinted in Jakobson
592
Assinja Demjjanow
and Anatoli
Strigin
(1990: 378). The translations of the DPs in the Instrumental are in italics. The brackets introduce the terminology. (1)
a. On el rebënkominstr ikru. He ate child-mjfr· caviar 'He ate caviar as a child. ' b. On el pudamiinstr ikru. He ate pud-instr caviar 'He ate caviar by the pood.' (= 36 lbs) c. On el lozkojinsir ikru. He ate spoon-iníír caviar 'He ate caviar with a spoon. ' d. On el dorogojins,r ikru. He ate road-instr caviar 'He ate caviar on the way. ' e. On el utrominstr ikru. He ate morning-iníír caviar 'He ate caviar in the morning.'' f. On el gresnyminstr delominstrr ikru. He ate sinful-insír matter-msfr caviar 'He ate caviar I am sorry to say.'
(temporal 1)
(manner)
(instrument)
(path)
(temporal 2)
(idiomatic)2
The italicized prepositions clearly show that different relations between the DP in the Instrumental and the rest of the sentence are involved. All these uses seem to be syntactic adjuncts.3 The problem of interpretation posed by these free DPs may be summed up by the question: Where do all these different semantic relations come from? The number of different possible meanings of such free DP adjuncts in the Instrumental is great and we do not want to ascribe every such meaning to each DP in instr, thereby creating an unmotivated extreme polysemy. The move to assign some kind of meaning to the Instrumental case, similar to the meaning of a preposition, is slightly better, but it simply shifts the problem of extreme polysemy to this case meaning; Wierzbicka (1980), for instance, argues that there are seventeen very general meanings of the Instrumental case. We see this problem of interpretation as a part of the larger problem of how to construct meanings in context. Consequently, we will propose a treatment which constructs the meanings of the free adjuncts in the Instrumental from different ingredients in the context of their interpretation. We will use an abductive theory of interpretation to characterize this construction.
Real adjuncts in the Instrumental in Russian
593
Jakobson proposed an interesting hypothesis to the effect that, in Russian, the meaning of the Instrumental arises partly from its opposition to other case forms, and partly from the interaction with the context: The Instrumental itself denotes nothing more than peripheral status; it occupies the same position among the peripheral cases that the Nominative does among the full cases: that of the unmarked category. ... Everything other than peripheral status is given in individual uses of the I by the actual meaning of its referent and by the context, but not by the case form. (Jakobson 1936/1990: 356)
We do not seek to explain all the uses of the Instrumental by this hypothesis in this paper. Instead, we will investigate what one implementation of this hypothesis amounts to in cases which can be semantically treated as intersective modifiers. We will define the term in a moment. Other uses are treated in Demjjanow and Strigin (2000) and Demjjanow and Strigin (2001). Following Jakobson, we assume that all grammatical cases of Russian are sorted into two groups, the central and the peripheral cases. We wish to avoid formally reconstructing Jakobson's ideas on this issue here, but see Demjjanow and Strigin (2000) for a partial attempt. The essence of Jakobson's ideas is this: ... what is specific to the peripheral cases is not that they indicate the presence of two points in the utterance, but only that they render one peripheral with respect to the other. ... [the peripheral point] could be omitted without impairment to the central one, as is the effect of the peripheral cases. (Jakobson 1990: 352)
The uses we consider here are Instrumental of Transport (not listed in (1)), Instrumental of Path, and the two temporal uses in (1). Instrumental of Transport (2)
On exal poezdom. He drove train-msir 'He was going by train.'
Instrumental of Path (a spatial predicate) (3)
On sei dorogoj. He went To&d-instr 'He was going on the road.'
594
Assinja Demjjanow and Anatoli Strigin
Temporal adverbial (1 and 2) (4)
Rebënkom on bolel. Child-instr he UI.PAST 'He was ill as a child.'
(5)
Letom on bolel. Summer-insír he ILL.PAST 'He was ill in summer.'
It is sometimes objected that examples like (3) are hardly acceptable. The impression vanishes if a suitable context is provided. In particular, a contrastive context which implies alternatives renders the example fully acceptable. (6)
Do reki on sei dorogoj. Tam ona koncilas '. 'To the river, he walked on the road.' 'There it ended.'
There is also a difference between our example with a verb of motion (3) and Jakobson's example with a verb unrelated to motion (Id). We will return to this effect in Section 4.2. There we shall also discuss the restrictions on DPinstr in these constructions. A theory which ascribes case meanings to the Instrumental must postulate at least three different meanings associated with INSTR to account for these examples and must also provide some kind of meaning shift function which maps people onto times when they were children to account for the first temporal use. We will argue that an abductive theory of interpretation allows us to treat all these uses uniformly as instances of predication on different discourse referents figuring as participants in the situation within a given context. So the context-independent meaning of INSTR merely indicates predication. The specific meaning of INSTR is determined by the respective context (including systematic world knowledge about types of situations). The context provides additional semantic relations which make up the diversity of the meanings of the Free Instrumental. We also choose a syntactic proposal which postulates a uniform environment for all these uses of instr to go with our semantics. We first discuss this proposal and formulate our technical semantic interpretation apparatus which is based on abduction. Then we formalize some selected uses of DPs in the Instrumental and explain those properties of them which seem to be amenable to the treatment proposed. In the concluding part of the paper, we discuss some possible alternatives and ramifica-
Real adjuncts in the Instrumental
in Russian
595
tions as well as questions and objections raised with respect to the treatment proposed.
1.1. Where to assign the Free Instrumental case? A syntactic unification Bowers (1993) argued that some interesting syntactic consequences follow if we adopt a special functional projection which is responsible for the predication in the sentence. He calls this projection Pr for Predicative element and assumes (7) as the structure for simple clauses of English. It is immaterial here whether or not I(nfl) is split into a number of separate phrases. (7)
IP
Here, SpecPr is the subject of the clause and SpecV is the direct object of the clause. SpecPr is the external argument of the verb, SpecV is its internal argument. Bailyn (1995) has applied this hypothesis to Russian to account for all non-idiosyncratic uses of the Instrumental by claiming that INSTR is the case assigned by Pr°, i.e. it is also the case of secondary predication. An example of an object-oriented depictive then has the structure given in (8), with PrP being a V'-adjunct small clause (the second V' is omitted).
596
Assinja Demjjanow and Anatoli Strigin
(8)
I.NOM
found
him.ACC
pjanym drunk.lNSTR
We shall adopt this hypothesis and extend it to cover the syntactic structures of our examples. This may be controversial, but it should be noted that this is the only proposal known to us which provides a uniform syntactic environment for all the uses of the Instrumental, which is an interesting hypothesis in its own right. Since PrP is the uniform structure of predication, we have to provide a uniform semantics for it.4 We will explore the straightforward semantic proposition that all of the uses of the Instrumental which interest us are basically predicative on some discourse referent in the situation described by the sentence via its identification with PRO. In which case, we have to stipulate that Russian has a kind of semantically defined control of PRO. 5 Ordinarily, PRO is considered to be controlled either by the subject or by the object or to have arbitrary interpretation with a kind of generic reading. We have to say how this rather semantic notion of control which we need works in Russian. We intend to analyze these three uses of instr as adjunct small clauses PrP. For the moment, we may also accept the assumption that these small clauses are adjuncts to IP in (8).
Real adjuncts in the Instrumental in Russian
597
1.2. Interpretation of the Free instr We are interested in the semantic and in some systematic pragmatic aspects of the meaning of a sentence S containing a DP instr . Since our conception of interpretation emphasizes the similarities between pragmatics and semantics, paying attention to both aspects is quite natural. An interpretation of a sentence S with a free DP instr therefore includes at least -
the determination of the relation between the interpretation of DP instr and the interpretation of the rest of the sentence; the determination of the information status of this relation and of the ||DP i n s t r ||.
We understand the hypothesis put forward by Jakobson as stating that the interpretation of DP instr is determined by inference from the context. Hence, we (i) consider interpretation to be an instance of inferential activity, and (ii) assume that the interpretation of DPjnstr proceeds from some contextindependent semantic contribution of DPinstr. The kind of inferential activity we mean is hypothetic inference, often termed abductive inference or abduction (Peirce 1992; Hobbs et al. 1993). The context of inference includes the representational description of the situation that the sentence characterizes. Our general position can be summed up as in (INT) below. Note that it is no longer the interpretation of INSTR itself which is inferred, but rather the interpretation of its environment. (INT)
The hypothesis of interpretation: In all cases under investigation we have an abductive interpretation of the relation between the predication relation introduced by [prp PRO [pr·, Pr° DPinstr]] which embeds a DPjnstr and the rest of the sentence.
More explicitly, we assume that DP ins t r has a context-independent semantics and it is the interpretation of [prp PRO [jy, Pr° DP ins , r ]] that supplies a contextually inferred relation. Thus, there is no meaning of INSTR independent of the structure in which it is assigned. Now, what is this inferred relation? It comprises two aspects. First, we propose that the uses of DP iaslr in (2)—(5) can be basically described in terms of intersective modification on situations. It seems to be the default interpretation of modification adjuncts in an event-based semantics anyway. This move provides a simple interpretation of the free clausal adjunct with DPinstr· Second, we propose that the other aspect of this relation is inferred
598
Assinja Demjjanow and Anatoli Strigin
by choosing some appropriate relation in the situation that the adjunct is modifying. The implementation of this idea uses the choice of a discourse referent in the situation rather than the choice of a particular relation. This discourse referent is to be the implicit controller of PRO, and it is in some relation to the eventuality of the situation. Therefore, it is only indirectly the choice of this relation itself. Situations are theoretical entities. For the moment we may think of them as states of the world being described by sentences and consider the terms containing them to refer to - perhaps partial - models of our semantic representations. But as they must also provide the sources for the inference, we shall have more to say about them in Section 4.4. Intersective modification on situations is defined in (9). (9)
A DP in instr in a sentence S is an intersective modifier on situations, if it is interpreted by (||[m. PRO [jy, Pr° DP instr ]] || & 11S ΊI ) ( s ) where S' is S without UP PRO [ft. Pr° DP instr ] ].
An intersective modifier on situations is then simply a predicate on situations. We thus assume that the default interpretation of the clausal adjunction is a conjunction of two predicates on situations. This will be made more precise below. We shall often ignore the complexities of the syntax, and speak of DP¡nstr (or ||DP¡nstr||) as a modifier, although it only provides the individual predicate (or a quantifier) of the clause. Thus, the modifier by train is a predicate on situations of transport including those in which someone is going by train. The matrix sentence Peter is going is a predicate on situations in each of which Peter gets somewhere by some means. So the interpretation of the sentence with the intersective modifier is made up of the conjunction of the two components shown in (9). If Peter is going by train, then any situation making the sentence true would contain Peter who is going somewhere and is doing so by train. Obviously, this definition will make sense only if some kind of temporal indexing is involved in the notion of situation. We assume such an indexing implicitly for the moment. The definition provides some semantic properties we should look for in the sentences in question. Thus, we should have (HDPinstr II & ||S'||)(s) |= HS'll(s) since from a conjunction either conjunct follows. That the modifier itself does not constitute a complete sentence is not detrimental. We may use some paraphrase, e.g. it was by train, as in Peter was going to London. It
Real adjuncts in the Instrumental in Russian
599
was by train. Note that we must consider '&' to be extensional wrt. the situation in order to get (llDPmstr II & llS'HXs) |= I I D P ^
||(s) & ||S'||(s)
as an intermediate step. So HDPinsü-IKs) is actually a clausal type, which is exactly what we want it to be in our full-scale syntax. On this understanding of situations as models, their explicit mentioning is probably not essential, because explicit statements about models are made in the meta-language. The situation argument (s) is indeed usually dropped for this reason. But, as we shall see, we will need slightly different properties of situations, therefore we shall continue to add this argument in our representations. How good a rendering of our uses of the Instrumental does the intersective modification give? To begin with, (2) and (3) seem to satisfy our expectations about entailments. The example with the train (2) has already been discussed above. Example (3) also fits in: if someone is walking on a road, he is walking and he is on the road. In both cases, we must use the temporal index of the situation to relativize the assertions to it. Consider (4) now. If someone was ill as a child, s/he was ill at some time in the past and s/he was a child at this time. The latter entailment is somewhat tautological for people, but we may substitute direktor ('director') for rebënok ('child'), and thus obtain (10). (10) Direktorom Director-ms/r
onbolel. he ill.PAST
'He was ill when he was the director.' With this sentence, the entailment seems to be more readily obtained: he was a director at some time or other. Intuitively, both entailments seem to be obtainable. But now let us add a quantifying adverb, e.g. (11). (11)
Rebënkom on casto bolel. Child-instr he often ÜI.PAST 'As a child, he was often ill.'
Here, we have a difficulty with the expected entailments: the sentence (12), which is (11) without the DPinstr modifier, does not follow from (11).
600
Assinja Demjjanow and Anatoli Strigin
(12)
On casto bolel. He often ÌII.PAST 'He was often ill.'
Clearly, if someone was ill as a child, s/he was ill at some time under some circumstances. But if someone was often ill as a child under the circumstances, s/he need not have been often ill, in general. Childhood is only one part of a life, and there might be plenty of time for a person to recover physically prior to the reference time of the situation. The problem seems to be this: a quantificational adverb needs a restrictor. We seem to implicitly change the restrictor of the quantificational adverb casto ('often') in passing from the example sentence to its entailments. The only observable change between (11) and (12) is the absence of the modifier. Therefore, we must conclude that the Temporal Instrumental constrains the restrictor clause in the quantificational structure of the adverb. Our entailment test is not applicable to this case, hence it is actually vacuously satisfied. But the problem of the place of ¿nsír-modifiers in the quantificational structure of the proposition remains, and we shall have to consider it. This will lead us to determining the pragmatic status of the interpretation of intersective modifier D P ^ . Now we come to the second, constructional aspect of the meaning of the INSTR.
2.
The theory of inferential interpretation
2.1. Abductive inference For simplicity we will ignore here the contexts of interpretation which go beyond simple clauses. The formalization draws on Poole (1989) and (1988). Let Γ be a set of sentences which we consider to be true in the context, let Π be a set of formulas and Ρ a subset of ground instances of these.6 We consider Π to be a set of hypothesis rules and hypotheses. We consider Ρ to be an explanation of φ according to (13). (13) Γ u Ρ explains φ if the following holds: (i) Ρ υ Γ | = φ (ii) Ρ υ Γ is consistent This is a very weak notion of explanation. We should exclude trivial explanations, e.g. φ as an explanation for φ itself, and probably provide other
Real adjuncts in the Instrumental in Russian
601
limiting conditions which would help us to select one explanation from a number of alternatives. We might look for basic explanations, minimal explanations, most specific explanations or combinations thereof. But we will not discuss the choice criteria here. The formal basis can be found in the cited works by Poole and in Hobbs et al. (1993). Consider Γ in (14). This is a set of background facts. (14) rained - last - night => grass - is - wet,
Γ = sprinkler - was -on=$
grass - is - wet,
grass - is - wet => shoes - are - wet
If we observe φ = shoes-are-wet and want to explain it in this technical sense, we could have two explanations. The two sets of hypotheses (the explanations) of shoes-are-wet
are Pi = {rained-last-night}
and P2 =
{sprinkler-was-on}. We can choose one of them. The hypotheses may include rules, i.e. implications. If we agree to use rules as hypotheses, whenever consistent, though subject to competition and choice, we can model the concept of a default rule. Hypotheses in general are used when there is evidence for them, i.e. some observation which requires an explanation. Defaults are simply hypotheses which are used whenever possible. Consider (15). The hypothesis can be treated as a default, and if we inquire what can be said about the flying abilities of something called bob, we have to use the default. (15) Abirds = (Γ, Π) Π = {bird(X)
=
=>
flies(X)},
(VXXemuiX)
=>
(\/xXemu(X)
=»
(bird(X)\ -flies(X)\
emu(polly), bird(tweety)
Since we know nothing else about bob, we only get a conditional assertion. We know that tweety is a bird, so we can explain its flying. We cannot explain the flying of polly, since this would contradict the facts. This simple sketch of the use of abduction suffices for the moment to draw a sketch of interpretation by abduction.
602
Assinja Demjjanow and Anatoli Strigin
2.2. Interpretation by abduction In general, what is interpreted by abduction is an underspecified semantic representation which is constructed with the help of the syntactic structure. The contribution of grammar to semantics is expressed at the representational level of Logical Form. The term "the logical form of the sentence" thus denotes the representation of the sentence at this level. Logical Form specifies the propositionally relevant aspects of syntactic structure. Logical forms are further converted to representations in propositional format, i.e. in some language with rules of inference defined for it. These representations are abstract semantic values of logical forms which we will call semantic forms of sentences. The corresponding representational module may be called Semantic Form (SF), by analogy. We will not specify the conversion algorithm LF - SF in any detail, but we will use the Discourse Representation Theory ( DRT) of Kamp and Reyle to write semantic values, implicitly assuming that it is possible to use discourse representations as an input to inference (see Kamp and Reyle 1993; Kamp and Roßdeutscher 1994). That the syntactic representations which serve as an input for constructing discourse representation structures are logical forms was proposed in Szabolcsi (1997). We consider DRSs to be predicates over situations and we treat a conjunction of two predicates of the same situation as a merger of the two corresponding DRSs. To be able to use syntactic information during inferential interpretation, we will assume two things. First, the fact that some syntactic relation holds at Logical Form between two entities of this level is coded in the semantic form of the sentence. Thus we may have a relation like subject in the semantic form of a sentence which has the syntactic subject-of-the sentence relation at LF, syntactically coded perhaps in structural terms. Second, a grammatical relation of a logical form (i.e. subject-of-the sentence) may be further associated with its own semantic form or with a special inference pattern. Thus the contribution of the syntax is split into two parts. To be more precise, if the structurally defined syntactic relation subjectof-the-sentence holds between two nodes of a logical form, the relation gfsubject(i, j) is introduced to state this fact. The terms i and j are the discourse referents associated with the relevant nodes. This fact is added to the set Γ used in the interpretation. The associated semantic value of the relation subject-of-the-sentence will be specified as argsubject(i, j). This relation can be described as "i is that argument of j which stands in the subject-of-the-sentence relation to it in the syntax". In interpreting syntactic structure, such semantic values of relevant syntactic relations are part of the observational input to the abductive component of inference which fur-
Real adjuncts in the Instrumental in Russian
603
ther hypothetically specializes them using the resources of the context they are in, e.g. the semantics of the verb which provides the referent j. Thus, the grammatical context enters the interpretation as facts of the kind gfsubject{i, j). Its semantic forms, e.g. argsubject(i, j), if made available by the lexical entry of the verb, enter the context as observations to be explained abductively. Our defaults which are involved in the interpretation should obviously be indexed by contexts, since two different verbs usually provide different explanations for their subjects. We will use two simplifications throughout this paper. We consider any sentence to introduce two inferentially relevant referents, each of which may define a context: a situation referent (introduced by the I node) and an eventuality referent introduced by the verb. In cases where there is no need to distinguish the two, we simply assume that the verb itself introduces the situation argument. Under these assumptions, (17) is approximately the logical form of (16), and (18) is its semantic representation in the DRT format. The syntactic index i makes up the DR u, and index k is rendered by the DR s. Semantic forms will mention only DRs, not the corresponding syntactic indices. (16)
Jabolel. I
ill.PAST
Ί was ill.' (17)
IP
bolelk
Vo
tk I.NOM
ill.PAST
604
Assinja Demjjanow and Anatoli Strigin
(18)
u s, ts now before(ts, now) myselfiu) [argsubject(u, s)] ill(s) gfsubject(u, s) This semantic representation records that the situation is the one of being ill, ill(s), but does not assume that what it means to be ill has to be more specifically inferred. The condition before(ts, now), where ts introduced by INFL, is the time of the situation. The condition myselfiu) is introduced by DPnom· The DR is the subject argument. This is expressed by argsubject(u, s). Note that we show the observation status of this expression by putting it into square brackets. It is the evidence which has to be explained abductively. This evidence is there because the logical form of the sentence asserts that the subject relation holds between the two referents indicated, gfsubject(u, s), and the verb specifies that the subject is a semantic argument. The only underspecified interpretation to be explained is argsubject(u, s). The interpretation is done by abductive inference. We may specialize argsubject{u, s) in the context of the situation s, characterized as ill(s), by hypothesizing that this relation can be explained by the statements that I myself am the theme of the situation of being ill, i.e. the person who is ill. This is done by using first-order theories like (19) and (20). We assume that predicates like theme(E, X) refer to eventualities which figure in the situation rather that the situation itself. (19)
(20)
Afberœ= (U) Π = \theme(E, X ) = > argsubject(S,
X )}
Am = (Γ) _ |i7/(s)=> is - ill(X,E),
1
[is - ill(x, E) => theme(E, X )J The theory in (19) contains a default rule to the effect that themes are subject arguments. Being a default, this rule can be overridden. The implication characterizing the active voice in English that whenever an agent is present in a situation, it will be realized as subject argument, hence as subject syntactically, may be added as a fact. Then our realization of theme is a default and is rejected in the presence of an agent in the situation provided that the sentence is in the active voice.7 The theory in (20) says that a situation s,
Real adjuncts in the Instrumental in Russian
605
labelled i//(s), is built up around an eventuality (i.e. process or state) in which someone is ill. We represent the situation by a theory which is labelled ill(s) and which specifies what the ingredients of the situation are. Among them is is-ill(X, E), introducing the Patient-Theme X. Moreover, it must be further specified which abstract thematic role this individual is to play. Thematic roles characterize predicates from the point of view of thenpossible realization in the syntax. The Patient-Theme individual X of ill(s) plays a role known as Theme, theme(E, X). The latter condition may also be rendered by theme(E) = X, if the relation is conceived of as a functional one. The explanation of argsubject(u, s) is then given by theme(E, u)=> argsubjectiu, s) on the assumption that ill(s) holds, since (21) holds. Our semantic form (17) states that i//(s) indeed holds. (21)
{is—ill(u, E), theme(E, u) => argsubject(u, s)} |= argsubject(u, s)
Since we are indeed in the situation ill(s), is-ill(u, E) helps to explain argsubject(u, s). That is to say, the sentence is interpreted as I am ill. The theory of interpretation sketched here is presented in more detail in Strigin (2000). We are actually more interested in the consequences of ill(s) which make our explanations applicable, than in the explanations themselves, i. e. we are more interested in is-ill(u, E) than in the corresponding default itself.
3.
Interpreting instr abductively
The syntactic structure of the free ¿/uir-adjunct defines the relation of predication which assigns INSTR to the nominal predicate. We postulate that it is basically this syntactic relation of predication assigning INSTR which allows us to identify the subject of predication. The identification consists in hypothetically finding the antecedent of PRO. This hypothetical inference is therefore associated with INSTR. Let HDP^H be the predicative interpretation of DPinsu·. Predicating DP^tr of PRO gives X = χ & HDP^H (x), where χ is simply the contribution of PRO, and X = χ is a hypothesis scheme which states that X is the antecedent referent of PRO. Actually, not any X is to be considered, but only the referents (i) which belong to the situation and (ii) which are characterized as belonging to the group which is usually marked by peripheral cases. We shall mark this situation context by prefixing s: to the hypotheses, i.e. the argument of the situation that the adjunct modifies.
606
Assinja Demjjanow
and Anatoli
Strigin
The expression ||DPinstr II stands for the interpretation of D P ^ , capital X may be identified with any DR in the domain of the representation of situation s. (22) Ainstr — (Γιπ^Γ,Πιηί/Γ) rinstr={s:\p^\\(x)} n i n s t r = {s : peripheral(X)
=> X =
This claim depends on the characterization of what is peripheral(w), since it provides an interpretation schema for all referents u G U of the discourse structure with the universe U interpreted in the domain Doms, of situation s. We shall proceed on the assumption that such a characterization is possible, we will adopt this course of investigation and apply it to the three uses of instr.
4.
The three cases of instr
4.1. Instr of transport This is probably the easiest case. Conjoining a situation predicate (i.e. a DRS) with a DRS amounts to simply importing its predicates into the latter DRS. If a situation characterized by the DRS contains the referent for the means of transport, this referent can be hypothetically taken to be the value of X. Consider (22) and (23). The interpretation is given within the situation description movebysomething(s). The verb exal (here: 'went/was going') is partly rendered by the predicate move(e, u, Is, lg). This states that event e is an event of movement from source location to goal location lg, with u being the bearer of the event, the moving object.8 The verb requires an argument which is the moving object, say movingobject(e, u). This argument is classified as Theme, so we should have something like movingobject(e, u) => theme(e) = u, and theme(e) = u => argsubject{u, s). We therefore would have something like movebysomething
)
move(e, u, Is, lg ),
move(e,u,Is,lg)=>
movingobject(e,u),
Fexat, = move(e,u,ls,lg)=>
msoftransport(e,z),
msoftransport(e, z) => train{w)
peripheral{z),
Real adjuncts in the Instrumental in Russian movingobject(e,u)=$
theme(e) = u,
Η exal ' = theme{e) = u => argsubjiu, peripheral(z)=>
607
s),
ζ=w
We may simplify again, and write the intermediate result of the abductive inference explaining argsubject(u,s), theme{e)=u in order not to be too detailed. The interpretation we normally get is movingobject(e, u), of course. The situation may have a means of transport, hence msofiransport(e, z). The inferential interpretation of PRO consists in adopting the hypothesis ζ = w, since w is its context-free interpretation, and train(w) is provided by the Free Instrumental DP: (23)
On exal poezdom. He drove train-insfr 'He was going by train.'
(24) s, ts, now before(ts, now)
u, e, z, Is, lg, w he(u) move(e, u, Is, lg) theme(e) = u msoftransportée, ζ) Ζ=w train{w)
The proposal predicts that the reading is only possible with situations which already have the appropriate referent. We can check this prediction in (25). (25)
*On spai poezdom. He slept train-instr 'He slept while being transported by train.'
The sentence is unacceptable. It is, of course, quite possible to characterize the situation with the help of a locative PP. (26)
On spai ν poezde. He slept in train 'He slept on the train.'
The reason for the difference under our theory is the difference in the interpreting relations: locatives relate events within the situation to a location,
608
Assinja Demjjanow and Anatoli Strigin
whereas the Instrumental of Transport merely identifies two referents, one of which is a means of transport. There is another curious fact which can probably be explained under the predication interpretation of instr. If a quantified distributing DP is put into predicational instr, the result is unacceptable, cf. (27). If we manage to indicate that there is a need for the wide scope of kazdym poezdom, as in (28), the sentence becomes acceptable. (27)
*On exal kazdym poezdom. He went every tiam-instr 'He drove on every train.'
(28)
On exal kazdym poezdom dva casa. He went every train-instr for two hours 'He drove two hours on every train.'
Similar effects are known for copula structures in English where quantifying-in renders some sentences acceptable.9
4.2. Instr of Path The treatment of the Instrumental of Path is essentially the same as for the Instrumental of Transport. Some new points of interest arise, however. We have (30) as a partial representation of (29). (29)
On sèi dorogoj. He went road-instr 'He was going on the road.'
(30) s, ts, now before{ts, now)
u, e, z, Is, lg, w he(u) move(e, u, Is, lg) theme(e) = u path(e, ζ) ζ = w road(w)
The availability of the Path discourse referent in the representation of the situation is a necessary prerequisite, as the pair (31) and (32) shows.
Real adjuncts in the Instrumental in Russian (31)
Onsël asfaltovoj dorogoj. He went asphalt road-instr 'He was going on the asphalt road.'
(32)
*On spai asfaltovoj dorogoj. He slept asphalt road-instr 'He was sleeping on the asphalt road.'
609
The example (33) and the example in (Id) seem to contradict this generalization. (33)
On spai dorogoj. He slept road-instr 'He was sleeping on the (road)way.'
It can be argued, however, that dorogoj (wayjnstr) is an adverb. The semantics of this adverb is a generalization of the part of any situation of movement which contains the referent for the Path. 10 The accommodation of such an adverb in the case of (33) rests on extending the representation of any situation which allows some participant to move simultaneously with the main eventuality of the situation. The extension affects that part of the movement situation which is associated with the adverb. There are some restrictions on what can be a path in this use of Instrumental, but they are difficult to state. Paths in Instrumental should be more or less natural. Thus, if the movement is within a city, the city provides a natural path. If, as in case of perfective verbs, we are interested in the state at the end of the path, a city is no longer a good path, whereas a road still is, cf. (34), (35) and (36). (34)
On sël Parizem. He went Paris-instr 'He was going/walking through Paris.'
(35)
*On prisel Parizem. He arrived on foot Paris-insír 'He arrived through Paris.'
(36)
On prisël dorogoj. He arrived on foot road-instr 'He arrived via a road.'
610
Assinja Demjjanow and Anatoli Strigin
A similar case can be observed with voda ('water') in (37). There is nothing wrong, in principle, with water being the surface on which the transportation takes place, as (38) shows. (37)
*Oni dobralis ' do goroda vodoj. They reached to town water-instr 'They reached the town by water.'
(38)
Oni dobralis' do gorodapo vode. They reached to town on water-prep 'They reached the town by/through water.'
The restrictions on possible Paths become explainable, if considerations of conceptual characterizations are involved in deciding whether to choose the referent as a good hypothesis. Thus, Paris probably ceases to be a good hypothesis in the context of a telic verb, because it cannot be portioned into pieces with a declared end. The same applies in the case of water, but not of roads. Quite in parallel to the use of the Instrumental to mark means of transportation, distributive quantification with narrow scope is bad with the Instrumental of Path, but not in general for paths, as (39) and (40) show. (39)
*On proexal kazdym He went through every 'He went through every town.'
gorodom. town-instr
(40)
On proexal po kazdomu He went through upon every 'He went through every town.'
gorodu. town-dat
4.3. Temporal instr 1 The temporal use of the Instrumental presents more difficulties. We assumed at the start that the default mode of combination of a PrP with the matrix sentence is that of simple conjunction. The temporal use is difficult, as (41) shows, for, although the predication is of the subject, the sentence does not assert merely the simple conjunction of the matrix sentence and the predicate expressed by the DP instr . Thus, (41) does not simply mean that at some time in the past he was a child and was ill.
Real adjuncts in the Instrumental in Russian (41)
611
Rebënkom on bolel. Chi\d-instr h e ill.PAST 'He was ill as a child.'
The correct interpretation seems to be derived by constructing a temporal characteristics for any model which is relevant to the evaluation of the sentence on the basis of the direct predication. To construct the temporal characteristics, we restrict our attention to the time at which the ||DPLnstrj| is true, i.e. we restrict the situation (the model) to that time, and then assert the matrix sentence relative to this restricted situation. This assertion relative to a time cannot be adequately rendered by the simple truth-functional conjunction. This point requires some elaboration. First, consider another possible way to represent the temporal reading of the Instrumental. We may postulate a regular lexical process that forms temporal predicates from temporally dependent nouns. It is immaterial here that we resort to lexical processes, since we could express them via inference in context. The derived predicates should then be used like temporal adverbs, e.g. yesterday or on May 21st. The interpretation would also be similar e.g. the time which is May 21st vs. the time when χ was a child. The conjunction of the matrix sentence and the adverb would yield an interpretation like there is a time (i) which satisfies the temporal predicate and (ii) at which the matrix sentence is true. The point is that this treatment would not be adequate. First, we noted earlier that (11) violates our expectations about the entailments, and suggested that DP¡ nstr provided a condition for the restrictor of the adverb of quantification casto ('often'). Second, even if one of the conjuncts gets a different status (e.g. We make '&' a dynamic conjunction, under which one of the conjuncts is tested first, and the other is tested only if the first is satisfied), we suggest that restrictor relativization is not always a conjunct formally speaking. We consider the proposal of Hajicová et al. (1998) that some propositions have quantificational structure to be applicable in a broad variety of cases. Indeed, we assume that there is a certain quantificational structure of the proposition even if there is no adverb of quantification. The structure of a proposition is then a restrictor and a nuclear scope, i.e. a predication. Now take some different temporally dependent noun which does not imply that there is a single homogeneous time period associated with every member of the extension of the noun11, e.g. (10), repeated here as (42). (42) Direktorom on bolel. Director-míír he ill.PAST 'He was ill as a director/whenever he was a director.'
612
Assinja Demjjanow and Anatoli Strigin
The time period of being a child associated with a person is homogenous. Not so the period of being a director. There may be several periods when the he of (42) was a director which are separated by times when he was not. Now, what (42) may mean is that at least some times when he was a director he was ill, but it may also mean that each time he was a director he was ill. The second reading is no longer a conjunction, but rather a conditional. We seem to relativize the assertion that he was ill to either some or to all relevant periods. The accompanying change seems to be from the conjunction to the conditional. This change is easily explainable if we note that we have a distribution of j|DPinSU.||. In other words, if the restriction of the situation can distribute, we get a conditional, if not, a conjunction. Such effects are well-known with definite plural DPs. The interesting thing is that we do not have a plural here. Second, there is a certain pragmatic implicature in the case of (41). This sentence is perfectly acceptable only when the person referred to by he is not a child at the time of utterance! Otherwise it is infelicitous. This implicature is unexpected if we are dealing with a simple conjunction. However, it can be explained by pragmatic factors if we assume that the temporal interval provided by the predicate in the Instrumental should play a role different from the one played by the time of the utterance or the time of the situation which sets the index of the model, and should restrict the situation. To render these intuitions about restrictions formal, we need the possibility of referring to separate temporal stages of the same individual plus the reference time of the situation, ts, and not simply the time of the situation. The difference is this: whereas we took the time of situation to be simply the temporal index of a model, the reference time of a situation is a restriction on this index for the purpose of confining some part of the predication in the situation to the restricted index. We will retain the notation ts for the reference time, and will not explicitly specify the time of situation any longer, since reference times of situations seem to be sufficient. We shall comment on our use of the term "situation" in a moment, and suggest a first approximation of the semantic representation of (41). We must restrict the situation in which there exists a person referred to by he to the time specified by the DPinsU. rebënok ('child'), and then evaluate the rest of the sentence with respect to the restricted temporal interval. To be able to accommodate the implicature that the person is no longer a child, we separate the universe of the discourse structure into two universes. One is the general universe of the situation, the other is the universe of the restrictor. This gives us a quantifier-like representation, in which the overall situation of the utterance is not represented by a discourse referent, whereas
Real adjuncts in the Instrumental in Russian
613
the restricted situation is represented. The restrictor is the left sub-box, the predication is the right sub-box. (43) u now s ts before(ts, ts = t childt(x)
now)
e heiu) be-ill (e, u) theme(e) = u χ = u
Here t in the abbreviation childt(x) denotes the time when child(x) is true. Evidently, the additional predication ts=t is a new hypothesis. The interpretation of the whole structure is as follows: the sentence is true in a model if the DR is embeddable. It is embeddable, if (i) (ii)
the restrictor is embeddable, and the embedding of the restrictor may be extended to that of the predication.
What happens is the following: we hypothesize that the subject is the referent of the predication associated with the Instrumental. Presumably because there is already one predication structure for the subject (built on the main verb), and the addition of a new parallel predication structure is achieved via a conjunction, a different hypothesis is put forward. This hypothesis is to take temporally dependent DPs as restrictors of the situation. It yields a quasi-quantificational structure of the situation. Note that this cannot be simply a conjunction, because this is a different hypothesis. In fact, as we shall discuss at the end of the next section, the restrictor is a kind of presupposed information. The relation between the reference time of the situation and the time of the event associated with the verb is indirect. On ('he') in (41) need not be ill all through the time he was a child. Therefore, we need an additional event referent. The identification of the two discourse referents for the individuals is possible given some additional assumptions, e.g. that a child grows up to become an adult, but remains the same person. Now the use of the term "situation" deviates from that in the standard DRT apparatus, and we will try to clarify our use of the term.
614
Assinja Demjjanow and Anatoli Strigin
4.4. Situations There are different traditions of the use of situations in semantics. One approach is to conceive of them as total models, relativize all the pertinent semantic relations to a model, if necessary, and not refer to situations explicitly from the language which uses the model, e.g. Chierchia and McConnell-Ginet (1990). This is the classical Fregean approach. Situation theory, with its starting point in Barwise and Perry (1983), is a radical departure from the bulk of assumptions of Fregean-type natural language semantics. Situations are conceived of here as information objects. Yet another way of using situations is to take them to be a kind of object in itself, a kind of individual in the sorted domain of different kinds of individuals. This use is found in Kratzer (1989) and Berman (1987). If semantic interpretations of situations are used at all, the latter use is the closest to our demands. However, we need inferential aspects of situations. Situations should specify the information available in them, though we would not like to identify them with that information. We therefore divorce situations from information and make them entities in the domain, in keeping with Kratzer or Berman. But, on the other hand, we will simultaneously treat them as contexts which constrain the inference by providing a limited amount of information for that purpose. Situations provide all kinds of informational anchors, either explicitly or by inference, but they also limit this inference. Consider (44) below. Here, the date provides a temporal anchor for the event of Peter's sleeping. (44) On March 21st, 1990, Peter slept. If such anchors are not available explicitly, they are obtained by inference. If the information which can be inferred is not sufficient, the sentence sounds strange or infelicitous. Thus (45) uttered out of the blue, could be baffling if the temporal anchors are not provided by the situation and we cannot resolve the temporal ambiguity. (45) Peter slept. Note that the date in (44) is not necessarily predicated of the time of sleep. The exact nature of the relation between the temporal anchor and the time of sleep is not sufficiently specified. The sleep need not have taken place during the whole time characterized as March, 21, 1990, and the relation may be more like that of inclusion. We may resort to situation and assume this,
Real adjuncts in the Instrumental in Russian
615
hence fixing the context of our assumption, and thus say that the time of this situation was on March 21st, 1990, and that Peter slept at some time during the situation. We thus temporally specify a context of inferential interpretation. It might seem that this move is representationally superfluous, but actually this indexing by a context has interesting consequences. Further discourse may shift the context by extending the situation with material from common ground or in other ways which involve temporal indexing. Therefore, inferences about temporal indexes could become dependent on the situation. Let (46) be the continuation of the story of Peter in (45). (46)
This was a good thing to do because March 21st, 1990, was a bleak and cold day.
We could now perhaps infer that Peter also slept some time during the day which was part of March 21st, 1990, and not only in the night. We cannot consider the inference about Peter's sleeping during the day as generally given whenever we know that he is asleep. Our inference depends on the assumption that the causative relation should connect relevant statements, i.e. that bleak days are good for sleeping only if the sleeping takes part on them, and also depends on the particular context obtaining in (46). Our knowledge is bound to the situation. So we treat the situation as a kind of context and limit our inference that Peter slept during the day to this context. Anchors thus fix the context, sometimes interacting with other information available there. We will not speculate here on precisely what information from the sentence can serve as an anchor beyond the temporal and the existential. In any case, we would like to be able to collect all relevant anchors from a sentence separately. The rest of the representation of the sentence is then taken relative to these anchors, hence relative to the situation. To give an example, (47) is the first attempt to partition the representation for (44) into the anchors and the assertion part. Here, s is the discourse referent (DR for short) of situation, now is the time of now, ρ is the DR for Peter. (47) now ρ s ts
e
before(ts, now) peterip) sleep(e,p) Marchix (ts)
616
Assinja Demjjanow and Anatoli Strigin
We have two kinds of anchors here: the temporal information and the condition introduced by the proper noun. We also implicitly index the predicate sleep(e, ρ) to the time of situation ts, i.e. we put it into a context. Putting things into contexts is what situations are for. The notion of situation is thus dependent on the notion of context, which is just as vague and in need of precision. The theory of context we would like to have should be modelled after McCarthy (1993) and McCarthy and Buvac (1997). 12 For theoretical reasons, these two papers make a distinction between contexts and situations, but make it possible to assign a context to any situation. We see no need to follow them in this respect. Collecting some information about a situation as an anchor gives this information a kind of presuppositional status: whether we assert something about the situation anchored in a particular way or deny some particular aspect of it, it should remain the same situation due to the anchoring. We may now identify the anchors with the restrictor on situations which we needed earlier to account for the temporal 1 use of the Instrumental. This move has some explanatory power, because in the case of the temporal use of instr, there is evidence to support the presuppositional status of this information. The temporal restriction by a DPmstr resembles a presupposition. A denial of the assertion still refers to the period when the person was a child as can be seen in (48). (48)
On rebënkom ne bolel. He child-insir not ÌII.PAST 'He was not ill as a child.'
This fact is reminiscent of Frege's argument about existential presuppositions of proper names. Frege argued that if the names did not presuppose their bearers, but rather asserted their existence, the denial of (49)
Kepler discovered Neptune
would have been equivalent to (50)
Kepler did not discover Neptune, or there was no Kepler
which is usually not the intended meaning. The same effect can be observed with (48) or (51). (51)
Ondirektorom ne bolel. He director-insir not ÌII.PAST 'He was not ill as a director.'
Real adjuncts in the Instrumental in Russian
617
The normal interpretation is the one which denies that he was ill when he was a director, and not the disjunction of the negations.
4.5. Temporal instr 2 There are some interesting problems with the temporal use of the Instrumental case. The most interesting one is that of temporal nouns in the Instrumental. We call these nouns distributive temporal predicates, for reasons which will immediately become obvious. A noun like le to ('summer') is predicated of a temporal discourse referent, and not of the subject. We consider this referent to be the reference time of the situation, i.e. a temporal anchor of the situation. The representation (53) is straightforward. (52) Letom on bolel. Summer-msfr he ill.PAST 'He was ill this summer/in summer.' (53) now u e ζ s ts before{ts, now) ts = t summer(t)
he{u) be-ill (e, u) th(e) = u
One curious thing about this use is that some otherwise similar temporal uses of nouns denoting temporal measure units are impossible. (54) *Casom ort citai. Hour-insfr he read 'He was reading for an hour/this hour.' If the unit is used in the Accusative, the sentence is acceptable with the durative reading of the DPacc· (55) Cas on citai. Hour-acc he read 'He was reading for an hour.'
618
Assinja Demjjanow
and Anatoli
Strigin
Distributive temporal predicates, e.g. den' ('day'), can be used in both ways, i.e. in the Instrumental or in the Accusative. However, in contrast to the Accusative use (57), in the use requiring the Instrumental, the temporal predicate cannot be modified by celyj ('whole'). (56) On citai (*celym) dnëm. He read (whole-msir) day-instr 'He was reading the whole day.' (57) On citai (celyj) den'. He read (whole-acc) day-acc 'He was reading (the whole)/for a day.' We proposed that DP instr in such uses are situation restrictors, i.e. anchors. The Accusative is then the case which is reserved for duratives. If this is the case, there must be some features based on which good temporal anchors are distinguished from duratives. Indeed, there is a substantial difference between the two kinds of temporal predicates. The ones we call "distributive" predicates are genuinely distributive. Any part of summer is summer. Units are not distributive: no part of a week is a week. The modifier celyj ('whole') disallows distribution. Distribution correlates with quantificational structures. So, if we assume that the anchoring function of temporal predicates (i.e. their functioning as restrictors) requires the preservation of the potential for distributivity, temporal units (e.g. a week) are excluded. However, one may expect that if temporal units are pluralized, they should acquire the ability to distribute if the theory of plural in Krifka (1989) is assumed. This seems to be borne out at first because temporal unit nouns in plural can be used in the Free Instrumental. (58) Casami on citai. Hour-msír he read 'He was reading for hours on end.' However, there is still a difference with distributive predicates. A plural temporal unit is probably best regarded as predicated of the event or state of the situation, and not of its reference time. This can be easily shown. Perfectivizing the verb immediately blocks the interpretation with the plural unit, but not with distributive temporal predicates in singular.
Real adjuncts
in the Instrumental
(59)
* Cas ami on procital knigu. HouT-instr he read.PERF the book 'He read the book in hours.'
(60)
Vecerom on procital knigu. Evening-msir he read.PERF the book 'He read the book (to the end) in the evening.'
in Russian
619
This complex, event : state after it, which is so characteristic of Russian perfectivization, is not distributive. It should be, however, to satisfy the homomorphism requirement associated with the distributive nature of the temporal referent of casami (hours-msir). This homomorphism is the cornerstone of the theory proposed in Krifka (1992). No homomorphism is required from temporal noun anchors, which are singular and distribute on demand, rather than maximally. Thus, (61) is acceptable, (62) is ungrammatical, but if we let dvazdy ('twice') have scope over nedel'ami (weeksinstr), the sentence is rendered felicitous with the durative reading for nedel'ami, cf. (63). (61)
Letom on dvazdy bolel. Summer-ms/R he twice ÌII.PAST 'He was twice ill this summer/in summer.'
(62)
*Nedel 'ami on dvazdy bolel. Weeks-instr he twice ÌII.PAST 'For weeks he was twice ill.'
(63)
On dvazdy bolel nedel'ami. He twice ÌII.PAST weeks-instr 'He was twice ill for weeks.'
The same operation can be performed on (61). The interpretation is that he was twice ill in summer (different summers), but not that he was ill twice during the whole summer (different or same). We may cautiously conclude: the additional hypothesis which interprets the DPjnstr which is temporally dependent on a distributive temporal predicate characterizes the reference time of the situation, and this time is not identical with the time of the event of the situation. What we still have not explained is the strange requirement on such D P ^ of distributivity on demand, which seems to be associated with this hypothesis. We do not have a good explanation at present.
620
Assinja Demjjanow and Anatoli Strigin
5. Discussion and conclusions We assumed the proposal of Bailyn that the three adjunct uses of D P ^ introduced at the beginning should be treated as having a uniform structure. This is the structure of secondary predication within a small clause. The choice is conducive to our attempt to provide a context-dependent interpretation for DPjnsfr. This move was intended to solve the problem of accounting for the polysemy in the three free DPu^-adjuncts. We proposed an abductive theory of interpretation which can handle this problem without assigning three different meanings to each D P ^ . We showed how to infer the referent within the description of the situation which can be the subject of predication associated with the Free Instrumental, possibly making additional temporal hypotheses. The theory seems to give a plausible picture, but there are some questions to be answered yet. Is there a better alternative using another syntactic structure? Sentences like (64) show that some syntactic constraints are operative,13 so the structure is important. In (64), there is no reading on which the advice was given when Peter was a child. Under the syntax of PrP, the Dative object Petru is the complement to Vo, and not the specifier of PrP, and thus cannot control PRO. (64)
*Onuze rebenkom¡ sovetoval Petru¡ begat'. He already child-insfr advised Peter to run/jog 'He already as a child advised Peter to jog.'
While the theory of Bailyn explains this, we are not aware of other comparable syntactic solutions which would explain this restriction whilst still treating the assignment of INSTR uniformly. If Bailyn's theory is adopted, however, we see no possibility of a lexical treatment of Instrumental adjuncts in the way Wunderlich (1997) proposes for secondary predication. Wunderlich's proposal is suggestive here because the Instrumental case can be the case of secondary predication too. Note that since the PrP clause responsible for the cases under consideration is attached to the IP node, we can speculate on the role of syntactic scope. When we spoke about semantic control in Russian, we actually meant the possibilities of the identification of the discourse referent introduced by PRO with some discourse referent in the domain of the DRS. We might define an accessibility order on the universe of the DRS which depends on syntactic scope, i.e. on the c-command relation. We can postulate that the identification of a DR u with the DR of PRO (i.e. semantic control) can take place only if the PRO-node discourse referent can access u in the c-
Real adjuncts in the Instrumental in Russian
621
command ordering. If this convention is adopted, then the attachment site at IP would only allow the identification of those discourse referents which have no syntactic realization. Case 1 of the temporal use, when the subject provides the DR to control PRO, would then have the same structure as (8). Indeed, this position might be a good alternative proposal for all kinds of PrP-adjuncts. We have not yet clarified the relative merits of the two proposals. Another problem we are still left with is that of the scope of the treatment. We proposed that any discourse referent in the domain of the situation can serve as the subject of predication. Some discourse referents in the domain of the discourse representation of the situation seem never to be able to be the subjects of insir-predication. This can be formally reflected, but is conceptually unsatisfactory without an explanation. We used discourse referents for the source and the goal of movement in movement situations, for instance, following Kamp and Roßdeutscher (1994). But these referents do not seem to enter the predication relation. However, it can probably be argued that Is and lg are not legal semantic referents for either the full or the peripheral grammatical cases since they are narrowly connected to PPadjuncts. This line of thought requires a more elaborate picture of situation types and their discourse referents (Strigin (2000) discusses some similar cases). If this proves to be viable, it could also be a contribution to a theory of semantic PRO-control in Russian. The three uses of the Free Instrumental are not the only ones, as (1) showed. We intend to extend the theory to all non-idiosyncratic uses of the Instrumental case, including the construction of secondary predication in Russian, but this is future work. Another major problem is the referential status of DP instr . We assumed that there are no restrictions on its logical type. Consequently, it can provide a predicate which has been converted from a generalized quantifier (Partee 1987). Such predicates tend to be unacceptable in primary predication sentences, as noted, as well as in some uses of DPj nstr , whereas they are perfectly acceptable in the instrument-denoting Instrumentals. We have no particularly interesting explanation of this fact to offer at present. What we hope to achieve as a side effect of providing an interpretation of sentences with a DPinstr is -
to contribute to the constructional conception of meaning (meaning in context); to contribute to the study of the interpretation of adjuncts; to further develop the inferential approach to semantic interpretation.
622
Assinja Demjjanow and Anatoli Strigin
Notes * 1. 2. 3.
4.
5. 6. 7. 8. 9.
10.
11.
12. 13.
We would like to thank Ilse Zimmermann for her comments on the draft. Opinions differ. Some, like Jakobson (1936/1990), see eight cases. The literal translation of the idiom would be 'it is a sinful matter that...'. We do not insist that they are simple DP-adjuncts, however. In fact, we will consider them to have a more complex syntactic adjunct structure of which the DP is only a part. Bowers (1993) provides a predication semantics for his proposal for English. The predication relation is defined in Property Theory. This semantics is clearly insufficient for our purposes, since the predication Bowers is concerned with covers only cases of secondary predication with the subject or with the direct object as a controller. We shall need more. Nichols (1982) has already proposed that there is a control relation at work in secondary predication in Russian. A ground instance of a formula is its substitution instance which contains no variables. Strigin (2000) presents a more detailed treatment of such a thematic role theory in an abductive framework. See Kamp and Roßdeutscher (1994) for the motivation of this treatment. Partee (1987) proposed a number of type-shifting operations to account for the semantic NP-type ambiguities. None of them would allow a distributive generalized quantifier like every to be a predicate. Examples like (i) seem to contradict this proposal. (i) This house has been every colour. Such examples motivated Partee to propose that nouns like colour are predicates of those properties which are among the entities of the domain of type e (individuals). Constructions like (i) are cases of quantifying-in into contexts forming predicates out of properties. Traditional Russian grammar often describes e.g. temporal uses of nouns in the Instrumental case, as in (4), as adverbs and speaks of 'adverbial derivation'. However, Isaôenko (1962) noted that this kind of derivation does not really allow the formation of new adverbs. He proposed characterizing the process of forming occasional adverbs as Entstehung ('coming into being, emergence') rather than derivation. Some uses of D P ^ gradually become adverbialized. Such development is a separate topic of investigation, however. A set S with the join operation ® is homogenous, iff for any two objects (i) if e¡ e S, e2 e S, then et Θ e2 e S (cumulativity), (ii) if e e S, and e = el Θ e2, then el e S, e2 e S ( d i s t r i b u t i v i t y ) . As usual, join is a binary commutative and associative operation. See Strigin (2000) for an attempt to integrate their theory into a linguistic description. We are working on the problem of how to handle these syntactic constraints in inference.
Real adjuncts in the Instrumental in Russian
623
References Bailyn, John F. 1995 Configurational case assignment in Russian syntax. The Linguistic Review 12: 315-360. Barwise, Jon, and John Perry 1983 Situations and Attitudes. Bradford Books, MIT Press. Berman, Steven R. 1987 Situation-based semantics for adverbs of quantification. In Issues in Semantics, J. Blevins and A. Vainikka (eds.). (UMOP 12.) University of Massachusetts, Amherst: GLS A. Bowers, John 1993 The syntax of predication. Linguistic Inquiry 24: 591-656. Chierchia, Gennaro, and Sally McConnell-Ginet 1990 Meaning and Grammar. An Introduction to Semantics. Cambridge, MA: The MIT Press. Demjjanow, Assinja, and Anatoli Strigin 2000 Case assignment to conceptual structures: The Russian Instrumental. In Papers on the Interpretation of Case, Marcus Kracht and Anatoli Strigin (eds.), 75-107. (Linguistics in Potsdam 10.) University of Potsdam. 2001 Measure Instrumental in Russian. In Papers on Predicative Constructions, Gerhard Jäger, Anatoli Strigin, Chris Wilder, and Niina Zhang (eds.), 69-97. (ZAS Papers in Linguistics 22.) Zentrum für Allgemeine Sprachwissenschaft, Sprachtypologie and Universalienforschung, Berlin. Hajióová, Eva, Barbara H. Partee, and Peter Sgall (eds.) 1998 Topic-Focus Articulation, Tripartite Structures, and Semantic Content. Dordrecht: Kluwer. Hobbs, Jerry R., Mark E. Stickel, Douglas E. Appelt, and Paul Martin 1993 Interpretation as abduction. Artificial Intelligence 63: 69-142. Isaöenko, Alexander V. 1962 Die russische Sprache der Gegenwart. Formenlehre. Halle (Saale): Max Niemeyer Verlag. Jakobson, Roman 1936 Beitrag zur allgemeinen Kasuslehre. Gesamtbedeutung der russischen Kasus. In TCLP VI, 240-288. 1990 Contribution to the general theory of case: General meanings of the Russian cases. In Roman Jakobson On Language, Linda R. Waugh and Monique Monville-Burston (eds.), 332-385. Harvard University Press. Kamp, Hans, and Uwe Reyle 1993 From Discourse to Logic. Dordrecht: Kluwer.
624
Assinja Demjjanow and Anatoli Strigin
Kamp, Hans, and Antje Roßdeutscher 1994 Remarks on lexical structure and DRS construction. Theoretical Linguistics 20: 97-164. Kratzer, Angelika 1989 An investigation of the lumps of thought. Linguistics and Philosophy 12: 607-653. Krifka, Manfred 1989 Nominalreferenz und Zeitkonstitution. München: Wilhelm Fink Verlag. 1992 Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, Ivan A. Sag and Anna Szabolcsi (eds.), 29-53. Stanford: CSLI. McCarthy, John 1993 Notes on formalizing context. In Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence. McCarthy, John, and Saga Buvaö 1997 Formalizing context (expanded notes). Ms., Stanford. Nichols, Johanna 1982 Prominence, cohesion, and control: Object-controlled predicate nomináis in Russian. In Studies in Transitivity, Paul J. Hopper and Sandra A. Thomson (eds.), 319-350. (Syntax and Semantics 15.) San Diego: Academic Press. Partee, Barbara Hall 1987 Noun phrase interpretation and type-shifting principles. In Studies in Discourse Representation Theory and The Theory of Generalized Quantifiers, J. Groenendijk, D. de Jongh, and M. Stokhof (eds.), 115143. (GRASS 8.) Dordrecht: Foris. Peirce, Charles Sanders 1992 Reasoning and The Logic of Things. The Cambridge Conferences Lectures of 1898. Cambridge, Mass.: Harvard University Press. Edited by Kenneth Laine Ketner. Poole, David 1988 A logical framework for default reasoning. Artificial Intelligence 36: 27-47. 1989 Explanation and prediction: An architecture for default and abductive reasoning. Computational Intelligence 5: 97-110. Strigin, Anatoli 2000 Constructing lexical meanings by hypothetical inference in context. Habilitationsschrift, Humboldt University, Berlin. Szabolcsi, Anna 1997 Strategies for scope taking. In Ways of Scope Taking, Anna Szabolcsi (ed.), 109-154. Dordrecht: Kluwer. Wierzbicka, Anna 1980 The Case for Surface Case. Ann Arbor: Karoma Publishers.
Real adjuncts in the Instrumental in Russian
625
Wunderlich, Dieter 1997 Argument extension by lexical adjunction. Journal of Semantics 14: 95-142.
German participle II constructions as adjuncts Ilse
Zimmermann
Abstract The present investigation is concerned with German participles II (past participles) as lexical heads of adjuncts. Within a minimalist framework of sound-meaning correlation, the analysis presupposes a lexicalist conception of morphology and the differentiation of Semantic Form and Conceptual Structure. It is argued that participles II have the same argument structure as the underlying verbs and can undergo passivization, perfectivization and conversion to adjectives. As for the potential of participles to function as modifiers, it is shown that attributive and adverbial participle constructions involve further operations of accommodation. Participle constructions are considered as reduced sentences. They do not have a syntactic position for the subject, for an operator (comparable to the relative pronoun in relative clauses), or for an adverbial relator (as in adverbial clauses). The pertinent components are present only in the semantic structure. Two templates are designed to account for the composition of modifiers - including participle constructions - with their modificanda. It is necessary to differentiate between modification which unifies two predicates relating to participants or to situations and frame setting modification where the modifier is given the status of a propositional operator. The proposed analysis shows that the high degree of semantic underspecification and interpretative flexibility of German participle II constructions resides (i) in their indeterminacy wrt. to voice and perfect, (ii) in the absence of certain constituents in the syntactic structure, and (iii) in the presence of corresponding parameters in the Semantic Form of the participle phrases.
1.
Introduction
This article refers to earlier research on the syntax and semantics of constructions with an adjective or a participle as lexical head and on modification (Zimmermann 1985, 1987, 1988a, 1988b, 1992). N o w I will propose certain refinements, which partly result f r o m the comparison of my analysis with the treatment of participle phrases by Fanselow (1986), Wunderlich (1987, 1997a), Bierwisch (1990), Kratzer (1994a, 1994b, 1998), and von Stechow (1999a, 1999b). More detailed versions of this reconsideration are published in Zimmermann (1999, 2000).* I shall concern myself with German participles Π (past participles) as lexical heads of attributive and adverbial phrases, as in ( l ) - ( 6 ) .
628
Ilse Zimmermann
(1)
die in meiner Heimat gleich nach Ostern geschorenen Schafe the in my home country right after Easter shorn sheep 'the sheep that are/were shorn in my home country right after Easter'
(2)
die trotz der Kälte schon geöffneten Apfelblüten the despite the cold already opened apple blossoms 'the apple blossoms that (have) already opened despite the cold'
(3)
der seit zwei Wochen verreiste Nachbar the since two weeks away neighbour 'the neighbour who has been away for two weeks'
(4)
Irene kann sich, endlich von ihrer Angst befreit, wieder besser konzentrieren. Irene is able, finally freed of her fear, again to concentrate better 'Finally freed of her fear, I. is once more able to concentrate better.'
(5)
Das Fleisch bleibt, im Römertopf gegart, schön saftig. the meat stays, in the chicken brick roasted, nice and juicy 'Roasted in the chicken brick, the meat stays nice and juicy.'
(6)
Mit ein paar Blumen geschmückt, sieht das Zimmer viel freundlicher aus. with a few flowers decorated, looks the room much more friendly 'Decorated with a few flowers, the room looks much more friendly.'
In the examples (l)-(3), we are dealing with modifiers used attributively which agree with the nominal head of the modificandum in gender, number and case. In (4)-(6), there is no morphologically indicated relation between the modifier and the modificandum. I regard these participle constructions as adverbial modifiers, which can be paraphrased as adverbial sentences. In many languages, there are special morphemes marking the adverbial form of the verb, the so-called adverbial participles (Haspelmath 1995; König 1995; Hengeveld 1998; V. P. Nedjalkov 1995; I.V. Nedjalkov 1995, 1998; Rüzicka 1978, 1982; Kortmann 1995). I will leave aside the characterization of participle constructions as secondary predicates. Further investigation must clarify whether there is a sharp delimitation between adverbial participle constructions and depictive secondary predicates.
German participle II constructions as adjuncts
629
The main concern of this paper will be the division of labour between morphology, syntax and semantics. The particular questions to be raised are the following: -
2.
Which morphosyntactic features characterize German participles Π? Which configurations and operations are involved in capturing the morphosyntactic and semantic polyfunctionality of participles Π? How do participle Π constructions get their status as attributive and adverbial modifiers?
The framework
Within a minimalist framework of sound-meaning correlation the analysis follows a lexicalist conception of morphology (Wunderlich 1997c) and the differentiation of Semantic Form (SF) and Conceptual Structure (Bierwisch 1987, 1997; Lang 1987, 1990, 1994; Dölling 1997, this volume; Maienborn 1997, this volume). A strict distinction is made between morphological marking and semantic interpretation of morphological forms. There are syntactic configurations which serve to check morphosyntactic features and to provide their semantic interpretation. This means that the relation between morphology and semantics is mediated by syntax in many cases. The semantic characterization of constituents can be underspecified. It is assumed that the Semantic Form of linguistic expressions involves parameters which are specified at Conceptual Structure. I will show explicitly in which respects participle Π constructions are semantically underdetermined. Any analysis of participles Π must take a stance on the nature of Tense, Aspect and Aktionsarten. Aktionsarten are semantic characteristics of verb phrases and depend on the semantics of the verb and of the modifiers and argument realizations. As regards Aspect, it is evident that German does not express Aspect morphologically. There is no systematic differentiation between Perfective and Imperfective Aspects. I assume that in German there are neither morphosyntactic features of Aspect nor an Aspect Phrase. At the level of Semantic Form, the aspectual relation as part of the verb meaning (see below) remains unspecified. As regards Perfect, I take it as a special time interval and will discuss whether it is necessary to assume a Perfect Phrase as von Stechow (1999a, 1999b) does.
630
Ilse Zimmermann
The syntactic structure of participial modifiers is sentence-like. Only the domains of the extended projection of verbs - ForceP, MoodP and TenseP are absent. The problem of whether there is a special XP on top of the participle construction will be discussed below. Participle constructions in the function of attributive or adverbial modifiers are - like all modifiers - syntactic adjuncts. They can be embedded into the matrix construction at those places where they are given the right interpretation according to their nature and with respect to scope relations (Grundzüge 1981; Maienborn 1996, 1997, this volume; Frey/Pittner 1998; Haider/Rosengren 1998). In contrast to Bailyn (1994), I do not assume a special ModP in syntax in order to integrate modifiers.
3.
The analysis
3.1. Lexical representation of participles Π as verb forms in the third status Words as syntactic atoms are fully inflected items. They enter syntactic representations with all affixes of word formation and inflection. The German participle Π as a non-finite verb form is composed of the verb stem, of the participle Π affix, and - for pre-nominal adjuncts - of adjectival inflection. It differs from the verb stem in the Phonetic Form (PF) and in the Morphosyntactic Characterization (MSC). Its Semantic Form (SF) is basically the same as the SF of the verb stem. (7)
a. /.../ b. +V-N, α sein, +infïn, +3 S, β part, γ A-Fl, δ pass, ε perf, ζ max >β = +) ( β = _ ^ γ = _ & ζ = + 5 δ = _ ^ ε = +, ζ = e. λχ η ... λχι λί Xs [[ Ts R a s p t ] & [ s INST [... x t ... x n ... ]]] ( T e ( α , i ) , a e { e, i }, RaSp e ( i, < i,t )), INST e < t, < e,t )))
(7a) represents the PF of the affixation process of participle Π formation, for instance, operiert 'operated', gelesen 'read' (without adjectival inflection) or operierte, gelesenem (with adjectival inflection). (7b) accounts for the morphosyntactic polyfunctionality of participles Π. It categorizes participles Π as a non-finite verb form ([+infin]), as third status ([+3S]) and as [-part] for the supinum or as [+part] for the participle (in the sense of Bech 1955/1957). [± A-Fl] is a morphological feature shared by adjectives, participles, determiners and certain numerals which can take adjectival inflection. [± max] serves to characterize the word structure level. [+sein] and [-sein] are selectional features of verbs forming the perfect with the auxil-
German participle II constructions as adjuncts
631
iary verb sein or haben, respectively. Furthermore, I assume that the participle Π is characterized by the morphosyntactic features [+pass] and/or [+perf], which are the basis for selection by auxiliary verbs and for semantic interpretation of participle constructions.1 The table in (8) shows the possible combinations of the features [±part], [±pass], and [±perfl. In the examples, '(-)' indicates that [+part] participles can have adjectival inflection or appear uninflected (in Bech's terms, with zero inflection). 3S + + +
+ + +
part +
pass
perf + +
-
+ + + +
+
-
+
-
-
+
-
+
-
vom Chefarzt operiert (-) vom Chefarzt operiert worden sein gern gelesen (-) gern gelesen werden gestern verreist (-) gestern verreist sein, gearbeitet haben
The semantic impact of these feature combinations will be accounted for by special rules of semantic interpretation. The participle affix alone does not alter the meaning of the verb. The SF of participles Π was given in (7c). I assume that the SF of verbs and of their participles is an xn+2-ary predicate with λχ„... λχι as argument positions for participants and λί as the argument position for time characterizations and Xs as the referential argument position. I shall leave open whether it is necessary to have verb semantics associated with possible worlds (i.e. to have one further position for possible worlds). 'INST' in (7c) reads as 'instantiates' and introduces the situation argument s for all lexical verbs (Bierwisch 1987). R¿,sp is a parametric relation between the time interval of the situation and a time interval 1.1 can be specified by perfect, tense and modifiers. In German, the aspectual relation Rasp remains unspecified on the grammatically determined level of SF.
3.2. Passivization and perfectivization In the following, we must decide how to capture the semantics of passivization and of perfectivization. In principle, there are two possibilities. We could simply formulate semantic interpretation rules for the constituents bearing the features [+pass] and/or [+perf] and indicate on which level of morphological or syntactic projection the corresponding semantics comes into play. I will call this method "affixless interpretation". The second pos-
632
Ilse Zimmermann
sibility is connected with the idea of feature checking in a certain syntactic configuration with a phonetically empty functional head which brings in the pertinent semantics. I call this method "affixal interpretation". It is evident that with the second solution, the syntactic structure is less economic. Therefore, I tend to prefer the first method of semantic interpretation, which, moreover, is flexible in allowing the pertinent semantic operation to work at different verb projections. In the following representations, I will enclose the functional PF and MSC information in parentheses, thereby indicating the omission of the zero head and of its projection. Passivization and perfectivization do not change the lexical category of the input. The two rules are mutually ordered. Like the auxiliaries in the verbal complex (for instance, gelesen worden sein 'have been read'), passivization - following the mirror principle - comes first.
3.2.1.
Passivization
As examples like (1) and (6) illustrate, there are attributive and adverbial participle Π constructions with passive voice semantics. I assume that constituents with participles e.g. gelesen 'read', or with the supinum gelesen in complex verb forms, e.g. gelesen wird 'is read', gelesen worden ist 'has been read', as lexical heads undergo the following rule of interpretation: (9)
Passive voice interpretation (PASS) (a. / 0 / ) (b. +pass) c. λΡ λί Às 3x [ Ρ χ t s ] +pass
The only condition for the rule to apply is that the morphosyntactic feature [+pass] in the MSC of the constituent be given its passive voice semantics. Passive voice semantically consists in existential binding of the highest argument for participants. (For selectional restrictions see Rapp 1997.) The rule is not limited to word structure. It can be freely applied at the level of phrase structure. 2 The same is true of perfectivization and of conversion to adjectives. (10) illustrates the operation of PASS for the participle Π geschorenen 'shorn' in (1) with the indicated morphosyntactic features. (10)
a. Input + V - N -sein +infin +3S +part +A-F1 +pass +/-perf +max (plus agreement features of the adjectival flexive) λχ 2 λχ' λί Xs [[ TS Rasp t ] & [ s INST [ Xi SHEAR' x 2 ]]]
German participle II constructions as adjuncts
633
b. Output +V-N -sein +infin +3S +part +A-F1 +pass +/-perf +max (plus agreement features of the adjectival flexive) λχ 2 λί Xs 3xi [[ Ts Rasp t ] & [ s INST [ xi SHEAR' x2 ]]]
3.2.2.
Perfectivization
Again, the rule of perfect interpretation applies to a constituent marked by a characteristic feature, in this case by [+perf]. Constituents with participles like verreist 'gone away [on a trip]', gelesen 'read', geschorenen 'shorn', or with the supinum in complex verb forms like verreist sein 'be away', gelesen haben 'have read', geschoren worden sein 'have been shorn' as lexical heads undergo rule (11). (11) Perfect interpretation rule (PERF) (a. / 0 / ) (b. +perf) c. λΡ Xt Xs 3t' [[ t'